REMOTE SENSING OF THE CANADIAN ARCTIC:
MODELLING BIOPHYSICAL VARIABLES
by
Nanfeng Liu
A thesis submitted to the Department of Geography and Planning
In conformity with the requirements for
the degree of Doctor of Philosophy
Queen’s University
Kingston, Ontario, Canada
(June, 2017)
Copyright © Nanfeng Liu, 2017
Abstract
It is anticipated that Arctic vegetation will respond in a variety of ways to altered temperature and precipitation patterns expected with climate change, including changes in phenology, productivity, biomass, cover and net ecosystem exchange. Remote sensing provides data and data processing methodologies for monitoring and assessing Arctic vegetation over large areas. The goal of this research was to explore the potential of hyperspectral and high spatial resolution multispectral remote sensing data for modelling two important Arctic biophysical variables: Percent Vegetation Cover (PVC) and the fraction of Absorbed
Photosynthetically Active Radiation (fAPAR). A series of field experiments were conducted to collect PVC and fAPAR at three Canadian Arctic sites: (1) Sabine Peninsula, Melville Island, NU; (2) Cape Bounty
Arctic Watershed Observatory (CBAWO), Melville Island, NU; and (3) Apex River Watershed (ARW),
Baffin Island, NU. Linear relationships between biophysical variables and Vegetation Indices (VIs) were examined at different spatial scales using field spectra (for the Sabine Peninsula site) and high spatial resolution satellite data (for the CBAWO and ARW sites). At the Sabine Peninsula site, hyperspectral VIs exhibited a better performance for modelling PVC than multispectral VIs due to their capacity for sampling fine spectral features. The optimal hyperspectral bands were located at important spectral features observed in Arctic vegetation spectra, including leaf pigment absorption in the red wavelengths and at the red-edge, leaf water absorption in the near infrared, and leaf cellulose and lignin absorption in the shortwave infrared.
At the CBAWO and ARW sites, field PVC and fAPAR exhibited strong correlations (R2 > 0.70) with the
NDVI (Normalized Difference Vegetation Index) derived from high-resolution WorldView-2 data.
Similarly, high spatial resolution satellite-derived fAPAR was correlated to MODIS fAPAR (R2 = 0.68), with a systematic overestimation of 0.08, which was attributed to PAR absorption by soil that could not be excluded from the fAPAR calculation. This research clearly demonstrates that high spectral and spatial resolution remote sensing VIs can be used to successfully model Arctic biophysical variables. The methods and results presented in this research provided a guide for future studies aiming to model other Arctic biophysical variables through remote sensing data.
Co-Authorship
This dissertation is based on the following three manuscripts:
Chapter 2: Liu, N., Budkewitsch, P., and Treitz, P. 2017. Examining spectral reflectance features related to Arctic percent vegetation cover: Implications for hyperspectral remote sensing of Arctic tundra. Remote
Sensing of Environment 192, 58-72. doi: 10.1016/j.rse.2017.02.002
Chapter 3: Liu, N. and Treitz, P. 2016. Modelling high arctic percent vegetation cover using field digital images and high resolution satellite data. International Journal of Applied Earth Observation and
Geoinformation 52, 445-456. doi: 10.1016/j.jag.2016.06.023
Chapter 4: Liu, N. and Treitz, P. 2017. Multi-scale remote sensing of Arctic percent vegetation cover and fAPAR. Submitted to International Journal of Applied Earth Observation and Geoinformation (in review).
For manuscript 1, Sarah Allux, Paul Treitz and Paul Budkewitsch developed the field sampling design and
Sarah Allux and Paul Budkewitsch collected the field data. I developed the research goals and objectives, and developed and implemented the analysis design. For manuscripts 2 and 3, I was responsible for the design and implementation of the study. I was also responsible for the writing of all manuscripts. My supervisor, Dr. Paul Treitz provided helpful guidance on the study and was consulted on the sampling designs and analysis methods. He also reviewed and edited the manuscripts.
iii
Acknowledgements
First, I would like to thank my supervisor Dr. Paul Treitz for your excellent supervision. Thank you for the invaluable patience, advice and guidance on my PhD study and research. Thank you for sharing your experiences of Arctic field work with me. Thank you for your detailed and critical comments which highly improved my manuscripts. Most importantly, your continuous encouragement has always boosted my confidence when I got stuck. I am very thankful for the opportunity to work with you.
I would also like to thank the committee members of my qualifying examination: Dr. Ryan Danby, Dr. Paul
Martin, Dr. Dongmei Chen and Dr. Neal Scott. Your insightful comments and suggestions helped narrow down my research questions and shape this final thesis. Special thanks to Dr. Dongmei Chen and Dr. Neal
Scott for their field sampling design advice. I would also like to thank Dr. Greg Henry for being my external reviewer.
Many thanks to my colleagues in the Laboratory for Remote Sensing of Earth and Environmental Systems
(LaRSEES). Especially, Amy Blaser and Rebecca Edwards: thank you for all your help during my Arctic field work; Karin van Ewijk: thank you for sharing the remote sensing lecture materials with me, which helped my teaching a lot. Many thanks to the staff of the Geography and Planning Department. John Bond, thank you for your IT support and help. Sheila-Rae MacDonald, Joan Knox, Sharon Mohammed and Kathy
Hoover, thank you for your patience and help. Joan, wish you get well soon. I would also like to thank my
Chinese friends at Queen’s University. Chen Shang, Pengpeng Ni, Tikang Li, Mengqi Yang and Yao Feng, thank you for your help and encouragement when I needed it. I enjoyed the great time spent with you at
Queens.
Financial support for this work was provided by NCE ArcticNet, the Norther Science Training Program
(NSTP), Polar Knowledge Canada, Polar Continental Shelf Project (PCSP), the Natural Sciences and
Engineering Research Council (NSERC), and Queen’s University. I would like to thank Dr. Scott
iv
Lamoureux and Dr. Melissa Lafreniere for their support of this research. I would also like to thank Sarah
Allux and Dr. Paul Budkewitsch for collecting the field data for Chapter 2.
Finally, I would like to thank my family – my parents and girlfriend, Zhihui Wang. I am most grateful to my parent’s support and unconditional love. Zhihui, I am sincerely grateful to you for patience, tolerance and most importantly your love. I love you all so much.
v
Statement of Originality
I hereby certify that all the work described within this thesis is the original work of the author. Any published (or unpublished) ideas and/or techniques from the work of others are fully acknowledged in accordance with the standard referencing practices.
(Nanfeng Liu)
(June, 2017)
vi
Table of Contents
REMOTE SENSING OF THE CANADIAN ARCTIC: MODELLING BIOPHYSICAL VARIABLES .... i Abstract ...... ii Co-Authorship...... iii Acknowledgements ...... iv Statement of Originality ...... vi Table of Contents ...... vii List of Figures ...... x List of Tables ...... xiii List of Abbreviations ...... xv Chapter 1 Introduction ...... 1 1.1. Arctic Vegetation and Environmental Change ...... 1 1.2. Remote Sensing of Arctic Vegetation ...... 2 1.2.1. Arctic Vegetation Classification ...... 2 1.2.2. Spatial and Temporal Dynamics of Arctic Vegetation ...... 4 1.2.3. Arctic Biophysical Variable Estimation ...... 4 1.3. Research Issues ...... 5 1.3.1. Research Objective 1 ...... 6 1.3.2. Research Objective 2 ...... 8 1.3.3. Research Objective 3 ...... 9 1.4. References ...... 10 Chapter 2 Examining spectral reflectance related to Arctic percent vegetation cover: Implications for hyperspectral remote sensing of Arctic tundra...... 16 2.1. Abstract ...... 16 2.2. Introduction ...... 17 2.3. Study Area ...... 20 2.4. Methods ...... 21 2.4.1. Field Sampling Design ...... 21 2.4.2. Percent Vegetation Cover ...... 23 2.4.3. Field Spectra ...... 24 2.4.4. Regression ...... 32 2.5. Results and Discussion...... 33 2.5.1. Percent Vegetation Cover ...... 33 2.5.2. Spectral Characteristics of Arctic Vegetation ...... 37
vii
2.5.3. Broadband VIs ...... 42 2.5.4. Narrowband VIs ...... 43 2.5.5. Conclusion ...... 48 2.6. Acknowledgement ...... 49 2.7. References ...... 49 Chapter 3 Modelling High Arctic Percent Vegetation Cover Using Field Digital Images and High Resolution Satellite Data ...... 57 3.1. Abstract ...... 57 3.2. Introduction ...... 58 3.3. Study Area ...... 61 3.4. Methods ...... 63 3.4.1. Field Sampling Design ...... 63 3.4.2. Percent Vegetation Cover ...... 66 3.4.3. Vegetation Indices...... 70 3.4.4. PVC-VI Regression Models ...... 70 3.5. Results and Discussion...... 71 3.5.1. Percent Vegetation Cover Estimation – A Comparison of Methods ...... 71 3.5.2. Temporal Patterns of Percent Cover ...... 73 3.5.3. PVC-NDSI relationships ...... 74 3.6. Conclusion ...... 77 3.7. Acknowledgement ...... 79 3.8. References ...... 79 Chapter 4 Remote Sensing of Percent Vegetation Cover and fAPAR on Baffin Island, Nunavut, Canada 85 4.1. Abstract ...... 85 4.2. Introduction ...... 86 4.3. Study Area ...... 89 4.4. Methods ...... 90 4.4.1. Field Sampling Design ...... 90 4.4.2. Percent Vegetation Cover ...... 92 4.4.3. The Fraction of Absorbed Photosynthetically Active Radiation ...... 94 4.4.4. Normalized Difference Vegetation Index ...... 95 4.4.5. Satellite Image Processing ...... 95 4.5. Results and Discussion ...... 96 4.5.1. Temporal Patterns of PVC and fAPAR ...... 96 4.5.2. Comparison of PVC between the point-frame and image classification methods ...... 98
viii
4.5.3. PVC/fAPAR-VI relationships ...... 100 4.5.4. Comparison with MODIS LAI/fAPAR product ...... 103 4.6. Conclusion ...... 105 4.7. Acknowledgement ...... 106 4.8. References ...... 106 Chapter 5 Synthesis and Future Directions ...... 112 5.1. Objective 1: To assess the utility of hyperspectral remote sensing data for modelling Arctic PVC using field spectra...... 112 5.2. Objective 2: To model Arctic PVC at landscape scales using field digital images and high- resolution satellite data...... 114 5.3. Objective 3: To develop models of fAPAR based on relationships between fAPAR and high spatial resolution satellite derived VIs...... 115 5.4. Conclusion ...... 116 5.5. References ...... 117 Appendix A Sabine Peninsula Site ...... 119 Appendix B Cape Bounty Arctic Watershed Observatory (CBAWO) Site ...... 123 Appendix C Apex River Watershed (ARW) Site ...... 131
ix
List of Figures
Figure 1-1. Three study sites located in the Canadian Arctic: (1) Sabine Peninsula (7627’N, 10833’W), Melville Island, NU; (2) Cape Bounty Arctic Watershed Observatory (7524’N, 10930’W), Melville Island, NU; and (3) Apex River Watershed (63°45’N, 68°30’W), Baffin Island, NU...... 6
Figure 2-1. Study area on the Sabine Peninsula, Melville Island, Nunavut, Canada (A-B). The WorldView- 2 unsupervised classification in panel C was adapted from Allux et al., (2012). Sampling regions (i.e., ellipses: ~2.5 km × 1 km) spanning at least two geological units were identified. Plots (i.e., dots: 20 m × 20 m) exhibiting homogeneous vegetation types were randomly located within each sampling region and validated in the field...... 21
Figure 2-2. PVC and spectral reflectance sampling design. For each homogeneous plot (~ 20 m x 20 m, A), PVC and spectra were measured within a 1 m x 1 m quadrat located at the center of the plot (i.e., green square in A). For PVC measurements, a 1 m x 1 m quadrat with 10 cm grids was used (B). For spectral reflectance measurements, fifteen measurements were collected at five different locations within the 1 m x 1 m quadrat: three in the center of the quadrat and three in each corner of the quadrat (C). The circles in C represent the field of view (FOV) of the ASD spectroradiometer...... 24
Figure 2-3. PVC by plant functional group for each vegetation type. Vegetation types are organized along a moisture gradient (i.e., polar semi-desert to wet sedge/moss). A) PVC3D results (i.e., PVC calculated using all contact points); B) PVC2D results (i.e., PVC calculated using first contact points only). N represents the number of plots sampled for each vegetation type...... 34
Figure 2-4. Comparison between PVC3D (y-axis) and PVC2D (x-axis) for each plant functional group (i.e., Green Moss, Green Forb, Green Graminoid/Sedge) and Total Green Vegetation (i.e., aggregated PVC). These plots include the 1:1 line...... 36
Figure 2-5. The regression for the difference in total PVC between the two methods (i.e., total PVC3D - PVC2D) and canopy height. Polar Semi-Desert (PD); Dry Mesic Tundra (DMT); Mesic Tundra (MT); Wet Mesic Tundra (WMT); Wet Sedge/Moss (WSM)...... 37
Figure 2-6. Mean (± 1 std. dev.) field spectra for each vegetation type: Polar Semi-Desert (PD); Dry Mesic Tundra (DMT); Mesic Tundra (MT); Wet Mesic Tundra (WMT); Wet Sedge/Moss (WSM), as measured at Sabine Peninsula, Melville Island...... 41
Figure 2-7. Correlations (R2) between green PVC and WorldView-3 derived VIs: NDSI (A) and RDSI (B). The x- and y-axes are the bands of WorldView-3, respectively. For each map, the lower-right diagonal area 2 2 represents the R between PVC2D and VIs; the upper-left diagonal area represents the R between PVC3D and VIs. The R2 > 0.70 values were labelled directly on each map...... 43
Figure 2-8. Correlations (R2) between PVC and Hyperion first-order derivatives (A: correlations with PVC2D; B: correlations with PVC3D). Green dots represent the correlation between PVC and the first derivatives of reflectance (i.e., dR) while red dots represent the correlation between PVC and the first derivatives of the reciprocal of reflectance (i.e., d(1/R)). Thresholds are identified for R2 = 0.6 and R2 = 0.7...... 44
x
Figure 2-9. Correlations (R2) between Hyperion-derived NDSIs/RDSIs and green PVC. For each correlation map, the x and y axes are the wavelengths of Hyperion, respectively; the lower-right diagonal 2 2 area represents the R between PVC2D and NDSI/RDSI; the upper-left area represents the R between PVC3D and NDSI/RDSI; the regions with high R2 (> 0.7) are numbered and the highest R2 position in each region is marked by a black square. The tables associated with each correlation map list the optimal band combinations and R2 values for each region...... 46
Figure 3-1. Cape Bounty Arctic Watershed Observatory (CBAWO) (panel a-b) and three sampling sites (300 m x 300 m) (panel c) on Melville Island, Nunavut, Canada. In panel b, the WorldView-2 image of Cape Bounty, acquired on July 12th, 2012, is displayed as a colour infrared composite with the near infrared, red and green channels display as red, green and blue (RGB) respectively. Panel c shows detailed satellite images (left) and classification maps (right) for the three sites...... 62
Figure 3-2. Sampling scheme for collecting digital images at the CBAWO. The WorldView-2 imagery (panel a) of the 300 m x 300 m site was classified into five land covers (panel b). Within each 6 m x 6 m plot, several 0.5 m x 0.25 m quadrats were randomly located (panel c). Colour infrared (NGB) and normal colour (RGB) images were collected for each of these quadrats (panel d)...... 66
Figure 3-3. Examples of the image classification and simulated point-frame method for three vegetation types (i.e., PD (polar semi-desert), MT (mesic tundra) and WS (wet sedge)). The first column shows the field GNDVI digital images of three vegetation communities and the second column shows the classification results. The white circles superposed on the image were used to simulate the point-frame method: i.e., the vegetation class within each circle was visually identified...... 69
Figure 3-4. Comparison of PVC between the OBIA and simulated point-frame methods. Polar semi-desert: Squares; Mesic tundra: Circles; Wet sedge: Triangles. The x- and y-axes are the OBIA derived PVC and simulated point-frame PVC, respectively...... 72
Figure 3-5. PVC and NDVI comparisons between early (July 9-11, 2014) and late (July 25-28, 2014) growing season for three vegetation communities. Plant functional groups on the X-axis are: forbs (FB), green graminoids/sedges (GG), senesced graminoids/sedges (SG), willow (WL), green moss (GM), senesced moss (SM), bare soil (BS), shadow (SH), green vegetation (GV=FB+GG+WL+GM), non-green vegetation (NGV=SG+SM) and non-vegetation (NV=BS+SH)...... 74
2 2 Figure 3-6. The PVC map of CBAWO. The regression model: PVC=(13.41×NDSINIR1_Yellow+1.26) (R = 0.77, RMSE = 9.99) was applied to the WorldView-2 derived NDSINIR1_Yellow image...... 77
Figure 4-1. The Apex River Watershed (ARW) study area (panel A) and five sample plant communities at the ARW: P1 - prostrate dwarf shrub, herb tundra; P2 - prostrate/hemi-prostrate, dwarf shrub tundra; G2 - graminoids, prostrate dwarf shrub; G3 - non-tussock sedge, dwarf shrub, moss tundra; and W1 - sedge/grass, moss wetlands...... 90
Figure 4-2. Field sampling design. (A) WorldView-2 false color image of the ARW (R: band 7 (near infrared); G: band 5 (red); B: band 3 (green); sample sites are identified by green dots; the red dot represents the locations of six permanent plots); (B) Field transect sampling design: each sampling site consists of
xi paired field transects; and five 6 m × 6 m plots are located along each transect at intervals of 20 m; (C) Within each plot, field measurements were made for four 0.6 m × 0.6 m quadrats located at the center of quadrants...... 92
Figure 4-3. Seasonal changes in the green PVC of different plant species and fAPAR for six permeant plots (i.e., the red dot in Figure 4-1) sampled in 2015. The location for these permanent plots is identified in Figure 4-2...... 98
Figure 4-4. Comparison of PVC derived by the image classification (PVCImage) and point-frame methods (PVC2D: PVC of the top layer derived from the point-frame method; PVC3D: PVC of all layers derived from the point-frame method). Dashed line represents the 1:1 correspondence. MAE (mean absolute error): N | yi xi |/ N ...... 100 i1
Figure 4-5. Comparison of LAI/fAPAR between MODIS and WorldView-2. Dash line is 1:1 line. MAE
(mean absolute error): ...... 104
Figure 4-6. MODIS LAI/fAPAR seasonal variation in 2015...... 105
xii
List of Tables
Table 2-1. Number of sampling plots and mean NDVI (from ground spectra) for each vegetation type. .. 23
Table 2-2. Specifications for simulating Hyperion and WorldView-3 spectral bands...... 26
Table 2-3. Hyperspectral vegetation indices used in previous studies...... 28
Table 3-1. Statistical information of three sampling sites at Cape Bounty (Normalized Difference Vegetation Index: NDVI) ...... 64
Table 3-2. Specifications and program settings of the Canon digital near-infrared camera...... 65
Table 3-3. Linear regression results for models of PVC (root squared) using NDSI ((Rx-Ry)/(Rx+Ry)). Only the band combinations with R2 > 0.7 are displayed. The models are grouped according to band x...... 75
Table 4-1. Major species of each plant functional type...... 94
Table 4-2. Plot-scale regression relationships between fAPAR, PVC2D, PVC3D and NDVI...... 101
Table 4-3. Satellite-scale regression relationships between fAPAR, PVC3D and NDSI ((Rx - Ry)/ (Rx + Ry)). 2 For the fAPAR-NDSI relationship, the band combinations with R > 0.6 are presented. For the PVC3D- 2 NDSI relationships, the band combinations with R > 0.7 are displayed. The model results are grouped according to band x...... 103
Table A-1. Plot information of the Sabine Peninsula site. PD: Polar semi-desert; DMT: Dry Mesic Tundra; MT: Mesic Tundra; WMT: Wet Mesic Tundra; WSM: Wet Sedge/Moss ...... 119
Table A-2. PVC2D and PVC3D data used in Figures 2-3 and 2-4. Grey: Dry mesic tundra; Blue: Mesic tundra; Orange: Polar semi-desert; Green: Wet mesic tundra; Brown: Wet sedge/moss; Italics font: PVC2D; Non- Italics font: PVC3D. BS: bare soil; GF: green forb; DF: dead forb; GG: green graminoids; DG: dead graminoids; LC: lichens; GM: green moss; SM: senesced moss. H: canopy height...... 121
Table B-1. Plot Information (UTM-12N projection, WGS-84 datum)...... 123
Table B-2. PVC (%) data used for Figure 3-4. Orange: Mesic tundra; Brown: Polar semi-desert; Green: Wet sedge; Italics font: PVC derived from OBIA; Non-Italics font: PVC derived from point-frame method. FB: Forb; GG: Green Graminoids; SG: Senesced Graminoids; WL: Willows; GM: Green Moss; SM: Senesced Moss; SL: Soil...... 127
Table B-3. PVC (%) data and WorldView-2 spectra used for Table 3-3; Yellow: PVC data acquired on July 9, 2014; Blue: PVC data acquired July 10, 2014; Green: PVC data acquired July 11, 2014; Brown: PVC data acquired July 12, 2014; WorldView-2 spectra acquired on July 12, 2014...... 129
xiii
Table C-1. Plot Information (UTM-19N projection, WGS-84 datum)...... 131
Table C-2. 1-PARReflected/PARIncident; PVCImg: PVC derived from image classification; Number of contacts with green vegetation in 2D and 3D cases (PVC2D=Contacts/25; PVC3D=Contacts/25, each quadrat has 5 x 5 = 25 grids); Height: canopy height. Field NDVI (some NDVI images are not usable due to under- exposure). Q1-Q4: Quadrat number. fAPAR=(1-PARReflected/PARIncident) x PVC. T1AP1: Transect 1A Plot 1...... 134
xiv
List of Abbreviations
ARW Apex River Watershed
CAVM Circumpolar Arctic Vegetation Map
CBAWO Cape Bounty Arctic Watershed Observatory
DMT Dry Mesic Tundra fAPAR fraction of Absorbed Photosynthetically Active Radiation
ISODATA Iterative Self-Organized Data Analysis Technique
ITEX International Tundra EXperiment
LAI Leaf Area Index
MODIS MOderate Resolution Imaging Spectroradiometer
MT Mesic Tundra
NDVI Normalized Difference Vegetation Index
NDSI Normalized Difference Spectral Index
OBIA Object-Based Image Analysis
PAR Photosynthetically Active Radiation
PD Polar semi-Desert
PVC Percent Vegetation Cover
PVC2D Two-dimensional PVC
PVC3D Three-dimensional PVC
R2 Coefficients of Determination
RDSI Reciprocal Difference Spectral Index
RMSE Root Mean Squared Error
VI Vegetation Index
WMT Wet Mesic Tundra
WSM Wet Sedge/Moss
xv
Chapter 1
Introduction
1.1. Arctic Vegetation and Environmental Change
Arctic ecosystems are characterized by low air and soil temperatures, limited precipitation, permafrost, a short growing season, and low vegetation productivity (Stow et al., 2004). Hence, it follows that Arctic vegetation patterns are closely related to environmental site conditions (Chapin et al., 1995; Walker et al.,
2006; Elmendorf et al., 2012). Based on warming experiments in moist tussock tundra near Toolik, Alaska,
Chapin et al. (1995) investigated the response of Arctic vegetation to solar irradiance, temperature and soil nutrients. The authors found: (1) increased soil nutrients (e.g., nitrogen and phosphorus) tended to increase biomass production of deciduous shrubs (with an associated decline in evergreen shrubs and nonvascular plants); (2) increased air temperature tended to increase shrub production and reduce nonvascular plant production; and (3) reduced solar irradiance tended to reduce the biomass of all growth forms. Walker et al. (2006) used meta-analysis on plant community measurements collected across 11 Arctic tundra locations and found that warming treatments resulted in an increase in the height and cover of deciduous shrubs and graminoids, and a decreased in moss cover and species diversity. In a more recent study, Elmendorf et al.
(2012) analyzed plot-level data collected between 1980 and 2010 for 158 plant communities spread across
46 Arctic locations. Some common changes in Arctic vegetation on decadal scales were summarized as follows: (1) the height and abundance of vascular plants (e.g., shrubs, graminoids and forbs) showed an increasing trend, whereas the cover of mosses and lichens decreased; (2) the biological response of Arctic vegetation to warming exhibited spatial heterogeneity: the increases in shrub abundance primarily occurred in study locations that had a high ambient temperature, whereas the increases in graminoids predominantly occurred in colder areas; and (3) in addition to temperature, other factors such as soil moisture and nutrients, growing season length, hydrology, precipitation and grazing intensity were also important drivers of Arctic vegetation change. In addition, these changes observed in Arctic warming experiments were in line with the repeated field photographs taken in past decades. By comparing photographs taken from 24 flight lines
over the past 50 years in northern Alaska, Tape et al. (2006) found that the shrub expansion took place in three ways: (1) the boundaries of shrub patches expanded; (2) patches filled in; and (3) individual shrub cover increased.
1.2. Remote Sensing of Arctic Vegetation
Compared to field surveys, remote sensing offers several advantages for large-scale assessment and monitoring of Arctic vegetation. Field surveys are spatially limited in that they can only provide observations at a finite number of points/locations. Hence, exhaustive field surveys are generally not feasible, nor desirable in practice, especially for remote and inaccessible Arctic environments. In addition, the short growing season in the Arctic complicates timely field sampling. In contrast, the acquisition of remote sensing data is more cost effective to provide observations over large areas when supported by a limited field campaign. Current remote sensing of Arctic vegetation studies can be grouped into three broad categories: (1) vegetation classification at different spatial scales; (2) spatial and temporal trend analyses; and (3) modelling biophysical variables. The following sections provide a summary of the current research in these areas.
1.2.1. Arctic Vegetation Classification
In general, Arctic vegetation classification is based on spectral distinctions between different vegetation communities (Stow et al., 2004). Many studies have found that Arctic vegetation possesses unique spectral characteristics (Riedel et al., 2005b; Ulrich et al., 2009; Buchhorn et al., 2013). Ulrich et al. (2009) compared the field spectra of fifteen typical periglacial vegetation communities collected in the western
Lena Delta area, NE Siberia, Russia. They found that Arctic vegetation spectra changed along a moisture gradient: (1) moist surfaces that were dominated by green grasses/shrubs exhibited typical vegetation spectral signatures, including green reflectance peaks, steep red-edge slopes, and high near-infrared (NIR) reflectance; and (2) dry surfaces that were dominated by large amounts of senesced vegetation and dry soil exhibited characteristic spectral signatures of these surfaces, including flat reflectance curves in the visible spectral region, shallow red-edge slopes, dampened water absorption features and apparent dry matter
2 absorption features at shortwave infrared (SWIR) wavelengths. These distinctive spectral characteristics were also reported by studies conducted at other Arctic sites, which laid the foundation for mapping Arctic vegetation based on multi-spectral satellite imagery (Riedel et al., 2005a; Buchhorn et al., 2013; Bratsch et al., 2016; Davidson et al., 2016).
By integrating advanced very high resolution radiometer (AVHRR) satellite imagery with topography, hydrology, vegetation, surficial geology and soil maps, Walker et al. (2005) produced the first Pan-Arctic vegetation classification map. On this 1-km resolution circumpolar Arctic vegetation map (CAVM), Arctic vegetation was classified into five broad biome categories and 15 physiognomic units based on plant growth forms. This map has been widely used for investigating Arctic vegetation responses to climate warming at large scales (Raynolds et al., 2008; Bhatt et al., 2010; Epstein et al., 2012; Pearson et al., 2013). However, it is insufficient for characterizing fine-grain spatial variation of Arctic vegetation types which tend to be extremely heterogeneous, variability that is not captured within the 1 km resolution pixels of AVHRR data
(Stow et al., 2004). High- (e.g., IKONOS: 4 m) and medium-spatial resolution (e.g., Landsat: 30 m) satellite data have shown potential for discerning complex patterns of Arctic vegetation (Rees et al., 2003; Stow et al., 2004; Olthof et al., 2009; Schneider et al., 2009; Atkinson and Treitz, 2012; Davidson et al., 2016). By applying unsupervised classification algorithms to Landsat imagery, Olthof et al. (2009) created a 30 m resolution Canadian Arctic land cover map which was suitable for land use planning, wildlife habitat assessment and climate impact assessment. Atkinson and Treitz (2012) combined environmental variables with IKONOS imagery and derived a 4 m resolution classification map suitable for landscape scale studies.
It is also worth noting that some recent studies are attempting to fuse multi-resource remote sensing data to improve classification (Ullmann et al., 2014; Fraser et al., 2016). For instance, vegetation heights derived from airborne light detection and ranging (LiDAR) data were combined with true-color images to classify
Arctic shrubs (Fraser et al., 2016). Ullmann et al. (2014) found that the different synthetic aperture radar
(SAR) backscattering characteristics of Arctic vegetation could improve classification.
3
1.2.2. Spatial and Temporal Dynamics of Arctic Vegetation
Studies of Arctic vegetation dynamics mainly focus on trend analyses of vegetation indices (VIs) time series derived from historical satellite imagery such as Landsat, AVHRR and moderate resolution imaging spectroradiometer (MODIS) (Jia et al., 2003; 2009; Myneni et al., 1997; Olthof and Pouliot, 2010). In most studies, the normalized difference vegetation index (NDVI) is used as an indicator of the photosynthetic activity of Arctic vegetation. The increase in NDVI has been reported in many Arctic/sub-Arctic regions:
(1) a 10% increase in NDVI for the northern hemisphere, with the greatest change occurring between 45N and 70N from 1981 and 1991 (Myneni et al., 1997); and (2) 16.9% and 11.2% increases in peak NDVI for northern Alaska (between 1981 and 2001) and the Canadian Arctic, respectively (Jia et al., 2003, 2009). In addition to NDVI, some researchers have incorporated other satellite products into their analyses. For instance, Raynolds et al. (2008) found a positive correlation between the AVHRR-derived NDVI and
Summer Warmth Index (SWI, defined as the accumulated land-surface temperature during the growing season) (R2 = 0.58): a 5C increase in SWI corresponded to an increase in NDVI of 0.07 for the Arctic overall. Bhatt et al. (2010) analyzed the satellite products of sea ice concentration, surface temperature and
NDVI for the Pan-Arctic coastal areas between 1982 and 2008 and found that the decline in Arctic coastal sea ice was linked to the warming of terrestrial surfaces and the associated increase in tundra productivity.
1.2.3. Arctic Biophysical Variable Estimation
Arctic biophysical estimation is based on the statistical relationships between optical VIs and biophysical variables such as leaf area index (LAI) (Williams, 2005), biomass (Hope et al., 1993; Raynolds et al., 2006;
Epstein et al., 2012; Johansen and Tømmervik, 2014; Sweet et al., 2015), green percent vegetation cover
(PVC) (Laidler et al., 2008; Blok et al., 2011; Atkinson and Treitz, 2013), fraction of absorbed photosynthetically active radiation (fAPAR) (Huemmrich et al., 2010, 2013; Tagesson et al., 2012) and CO2 flux (la Puma et al., 2007; Emmerton et al., 2015). The single value of NDVI (ranging from -1 and 1) is derived from the low reflectance in the red wavelength region attributed to chlorophyll absorption and the high reflectance in the near-infrared wavelength region caused by scattering from leaf cellular structures
4
(Jensen, 2005). Laidler et al. (2008) found that the NDVI derived from high-resolution satellite imagery
(i.e., IKONOS: 4 m) had a stronger correlation with PVC than the NDVI derived from medium-resolution satellite imagery (i.e., Landsat ETM+: 30 m). The stronger correlation was attributed to the enhanced capacity of high-resolution satellite imagery to stratify heterogeneous Arctic land surfaces into relatively homogeneous areas (Hope et al., 1993; Laidler et al., 2008). Atkinson and Treitz (2013) and Walker et al.
(2012) found that some caution needs to be exercised when extending regression equations calibrated for one site to another site, particularly over large distances. In addition, nonlinear relationships between NDVI and biophysical variables have been reported by some authors (Williams, 2005; Blok et al., 2011; Sweet et al., 2014). Blok et al. (2011) found that NDVI started to saturate when shrub cover exceeded 40%, indicating that NDVI may not be optimally suited to detecting changes in the Arctic tundra for scenarios of high biomass. It should also be mentioned that NDVI does not always correlate well with Arctic biophysical variables. Emmerton et al. (2015) demonstrated that NDVI performed poorly in estimating the CO2 flux for a polar semi-desert. The poor correlation was mainly because: (1) the faint seasonal changes in plant growth/coverage of polar semi-desert could not be detected using optical remote sensing; and (2) the CO2 flux variation of polar semi-desert caused by other factors (e.g., soil moisture) was not captured by NDVI.
1.3. Research Issues
The overall objective of this study is to model Arctic biophysical variables using remote sensing data.
Specifically, two important biophysical variables including PVC and fAPAR are examined. PVC is defined as the percentage of the ground surface covered by green vegetation (Purevdorj et al., 1998). It has been widely used as an ecosystem variable for characterizing and monitoring Arctic vegetation (Sturm et al.,
2001; Walker et al., 2006; Elmendorf et al., 2012; Tape et al., 2012; Frost and Epstein, 2014). Further, PVC has been found to be closely related to other key remote sensing derivatives such as NDVI (Purevdorj et al., 1998; Stow et al., 2004; Laidler et al., 2008; Atkinson and Treitz, 2013), biomass (Krebs et al., 2003;
Chen et al., 2009, 2013) and fAPAR (Schubert et al., 2010; Tagesson et al., 2012; Huemmrich et al., 2013). fAPAR refers to the portion of photosynthetically active radiation absorbed by vegetation canopies in the
5
400 – 700 nm spectral wavelength region (Jensen, 2005). It is an important parameter in describing the exchange of energy and mass fluxes between the surface and atmosphere in climate, hydrological and ecological models (Running et al., 2004; GCOS, 2006; King et al., 2011). For the research reported in this dissertation, a series of field studies were conducted at three Canadian Arctic sites: (1) Sabine Peninsula
(7627’N, 10833’W), Melville Island, NU; (2) Cape Bounty Arctic Watershed Observatory (7524’N,
10930’W), Melville Island, NU; and (3) Apex River Watershed (63°45’N, 68°30’W), Baffin Island, NU
(Figure 1-1). Three major research objectives were addressed as separate manuscripts (i.e., Chapters 2 through 4).
Figure 1-1. Three study sites located in the Canadian Arctic: (1) Sabine Peninsula (7627’N, 10833’W), Melville Island, NU; (2) Cape Bounty Arctic Watershed Observatory (7524’N, 10930’W), Melville Island, NU; and (3) Apex River Watershed (63°45’N, 68°30’W), Baffin Island, NU.
1.3.1. Research Objective 1
To assess the utility of hyperspectral remote sensing data for modelling Arctic PVC using field spectra.
Remote sensing of Arctic vegetation currently has two major limitations. First, most studies are based on multispectral satellite data (Stow et al., 2004; Laidler et al., 2008; Kushida et al., 2009; Atkinson and Treitz,
6
2013). Compared to multispectral remote sensing, hyperspectral remote sensing provides the contiguous spectrum for the land surface which has demonstrated great utility in many environments such as croplands, grasslands and forests (le Maire et al., 2004; Hernández-Clemente et al., 2012; Thenkabail et al., 2013;
Kalacska et al., 2015). However, the availability of hyperspectral data for sites in the High Arctic is limited by the low frequency of data coverage at high latitudes. With the forthcoming hyperspectral satellite missions such as the Environmental Mapping and Analysis Program (EnMAP) and the Hyperspectral
Environment and Resource Observer (HERO) (Rogge et al., 2014), a routine monitoring of Arctic tundra through hyperspectral satellite remote sensing will be feasible.
Second, the spectral bands used in previous studies are constrained within the visible-NIR wavelength range
(i.e., 450-1400 nm) (Stow et al., 2004; Buchhorn et al., 2013; Bratsch et al., 2016; Davidson et al., 2016).
Few studies have explored the potential of SWIR spectral bands (i.e., 1400-2500 nm) for modelling Arctic biophysical variables. Recent studies have indicated that the strong SWIR cellulose absorption features that have been identified in Arctic vegetation spectra are related to the substantial presence of senesced vegetation (Ulrich et al., 2009). Further exploration of SWIR spectral bands could therefore enhance our understanding of Arctic biophysical variable estimation and plant community dynamics given the predominance of senesced vegetation in some vegetation types (Buchhorn et al., 2013; Bratsch et al., 2016;
Davidson et al; 2016).
Chapter 2 presents an assessment of the utility of hyperspectral data for modelling Arctic green PVC within the 450 –2500 nm wavelength range. The analyses are based on field spectra collected on the Sabine
Peninsula. Specifically, the objectives were to: (1) characterize the PVC and spectral properties of five vegetation types distributed along the moisture gradient for Sabine Peninsula; (2) identify the optimal hyperspectral bands for estimating green PVC; and (3) evaluate the performance of multispectral broadband and hyperspectral narrowband VIs for predicting green PVC. The methods and results presented in Chapter
2 provide a guide on future studies aiming to model other Arctic biophysical variables through hyperspectral remote sensing data.
7
1.3.2. Research Objective 2
To model Arctic PVC at landscape scales using field digital images and high-resolution satellite data.
This objective addresses a challenge in Arctic vegetation research regarding how to obtain field PVC effectively and efficiently. The point-frame method has been used to measure biodiversity at fine scales and assess change in tundra vegetation across the tundra biome. Currently, it is probably the most widely adopted means for measuring Arctic PVC (Molau and Mølgaard, 1996; Bonham, 2013). This method has been found to be time-consuming and cannot be used to extrapolate over large areas (Chen et al., 2010).
Given the short growing season and adverse weather often encountered at High Arctic study sites, it is not practical to obtain field PVC for many sites using the point-frame method. Recent studies have demonstrated that image-based methods offer great promise for obtaining Arctic field PVC efficiently
(Chen et al., 2010; Beamish et al., 2016; Fraser et al., 2016). Basically, nadir-view digital images are taken in the field and analyzed in the laboratory to obtain PVC by using various classification methods (Luscier et al., 2006; Chen et al., 2010; Fraser et al., 2016). However, its application in Arctic tundra is limited and requires further exploration.
In Chapter 3, the image-based method is tested to determine how well it captures the seasonal PVC changes in Arctic vegetation. It involves collecting field digital images for three vegetation types (i.e., wet sedge, mesic tundra and polar semi-desert) at the Cape Bounty Arctic Watershed Observatory (CBAWO). First, an object-based image analysis (OBIA) approach, which considers both the spectral and geometric characteristics of surfaces, was adopted to classify field digital images. Second, PVC estimated from images collected at the beginning and at the end of the growing season were compared for each vegetation type.
Furthermore, the relationships between field-derived PVC and high-resolution satellite derived VIs was explored. The optimal predictive PVC-VI model was applied to generate landscape scale PVC maps for the study area.
8
1.3.3. Research Objective 3
To develop models of fAPAR based on relationships between fAPAR and high spatial resolution satellite derived VIs.
The point-frame method is likely the most commonly applied method for measuring Arctic PVC at plot scales (Molau and Mølgaard, 1996; Bonham, 2013). However, due to its time-consuming and laborious nature, many recent studies have focused on deriving PVC from field digital images (Luscier et al., 2006;
Chen et al., 2010; Liu and Treitz, 2016). However, the image classification method has seldom been compared to the traditional point-frame method for Arctic tundra, with the exceptions being Chen et al.
(2010) and Beamish et al. (2016). The discrepancy of the PVC derived from the two methods is still relatively unknown.
fAPAR is an important parameter in describing the exchange of energy and mass fluxes (e.g., CO2) between the land surface and atmosphere (Liu et al., 2013) and its relationship with NDVI has been evaluated for many environments (Goward and Huemmrich, 1992; Myneni and Williams, 1994; Gitelson et al., 2002;
Fensholt and Sandholt, 2003). However, very few studies have tested the fAPAR-NDVI relationship for
Arctic vegetation. Arctic vegetation types differ from other vegetated landscapes (e.g., forests, croplands and grasslands) given their low stature and presence of senesced vegetation. This makes the fAPAR measurement and calculation of Arctic vegetation different from other vegetation types (Huemmrich et al.,
2010; Schubert et al., 2010). Meanwhile, MODIS LAI/fAPAR is an operational global-scale product and has been validated for various land covers such as evergreen needle forests (Jensen et al., 2011; Serbin et al., 2013), evergreen deciduous forests (Wang et al., 2004; Aragao et al., 2005; De Kauwe et al., 2011), semi-arid savannah (Privette et al., 2002; Tian et al., 2002; Fensholt et al., 2004; Hill et al., 2006) and grassland (Pasolli et al., 2015). However, its validation for Arctic tundra has been seldom investigated due to limited field measurements (Huemmrich et al., 2010; Tagesson et al., 2012).
As a result, Chapter 4 focuses on exploring the spatial and temporal patterns of PVC/fAPAR and their correlations with VIs at the Apex River Watershed (ARW). The specific objectives are to: (1) assess the
9 performance of the image classification method in estimating PVC; (2) investigate the seasonal patterns of
PVC/fAPAR for different vegetation types; (3) explore the PVC/fAPAR – VI relationships at plot and fine- resolution satellite scales; and (4) validate MODIS LAI/fAPAR products for Arctic tundra by using field measured PVC/fAPAR and high-resolution satellite imagery.
Chapters 5 summarizes the ideas, results and conclusions presented in the manuscripts in the context of an overall research agenda. Their implications for modelling Arctic biophysical variables are discussed in the context of an overall monitoring strategy. Further research ideas stemming from this dissertation are also presented.
1.4. References
Aragao, L.E.O.C., Shimabukuro, Y.E., Espirito-Santo, F.D.B., Williams, M., 2005. Spatial validation of the collection 4 MODIS LAI product in eastern Amazonia. IEEE Transactions on Geoscience and Remote Sensing 43(11), 2526–2534. doi:10.1109/TGRS.2005.856632 Atkinson, D.M., Treitz, P.M., 2012. Arctic ecological classifications derived from vegetation community and satellite spectral data. Remote Sensing 4(12), 3948–3971. doi:10.3390/rs4123948 Atkinson, D.M., Treitz, P.M., 2013. Modeling biophysical variables across an Arctic latitudinal gradient using high spatial resolution remote sensing data. Arctic, Antarctic, and Alpine Research 45(2), 161– 178. doi:10.1657/1938-4246-45.2.161 Beamish, A.L., Nijland, W., Edwards, M., Coops, N.C., Henry, G.H.R., 2016. Phenology and vegetation change measurements from true colour digital photography in High Arctic tundra. Arctic Science 2(2), 33–49. doi:10.1139/as-2014-0003 Bhatt, U.S., Walker, D.A., Raynolds, M.K., Comiso, J.C., Epstein, H.E., Jia, G., Gens, R., Pinzon, J.E., Tucker, C.J., Tweedie, C.E., Webber, P.J., 2010. Circumpolar Arctic tundra vegetation change is linked to sea ice decline. Earth Interactions 14(8), 1–20. doi:10.1175/2010EI315.1 Blok, D., Schaepman-Strub, G., Bartholomeus, H., Heijmans, M.M.P.D., Maximov, T.C., Berendse, F., 2011. The response of Arctic vegetation to the summer climate: relation between shrub cover, NDVI, surface albedo and temperature. Environmental Research Letters 6(3), 35502. doi:10.1088/1748- 9326/6/3/035502 Bonham, C.D., 2013. Measurements for terrestrial vegetation (Second Edition). John Wiley & Sons, West Sussex. Bratsch, S.N., Epstein, H.E., Buchhorn, M., Walker, D.A., 2016. Differentiating among four Arctic tundra plant communities at Ivotuk, Alaska using field spectroscopy. Remote Sensing 8(1), 1-17. doi:10.3390/rs8010051 Buchhorn, M., Walker, D.A., Heim, B., Raynolds, M.K., Epstein, H.E., Schwieder, M., 2013. Ground- based hyperspectral characterization of Alaska tundra vegetation along environmental gradients. Remote Sensing 5(8), 3971–4005. doi:10.3390/rs5083971
10
Chapin, F.S., Shaver, G.R., Giblin, A.E., Nadelhoffer, K.J., Laundre, J.A., 1995. Responses of Arctic tundra to experimental and observed changes in climate. Ecology 76(3), 694–711. doi:10.2307/1939337 Chen, W., Wang, J., Li, J., Zhang, Y., Zhang, Y., Zhou, F., Koehler, K., Leblanc, S., Fraser, R., Olthof, I., 2009. Relating biomass and leaf area index to non-destructive measurements in order to monitor changes in Arctic vegetation. Arctic 62(3), 281–294. Chen, W., Zorn, P., Chen, Z., Latifovic, R., Zhang, Y., Li, J., Quirouette, J., Olthof, I., Fraser, R., Mclennan, D., Poitevin, J., Stewart, H.M., Sharma, R., 2013. Propagation of errors associated with scaling foliage biomass from field measurements to remote sensing data over a northern Canadian national park. Remote Sensing of Environment 130, 205–218. doi:10.1016/j.rse.2012.11.012 Chen, Z., Chen, W., Leblanc, S.G., Henry, G.H.R., 2010. Digital photograph analysis for measuring percent plant cover in the Arctic. Arctic 63(3), 315–326. Davidson, S., Santos, M., Sloan, V., Watts, J., Phoenix, G., Oechel, W., Zona, D., 2016. Mapping Arctic tundra vegetation communities using field spectroscopy and multispectral satellite data in North Alaska, USA. Remote Sensing 8(12), 978. doi:10.3390/rs8120978 De Kauwe, M.G., Disney, M.I., Quaife, T., Lewis, P., Williams, M., 2011. An assessment of the MODIS collection 5 leaf area index product for a region of mixed coniferous forest. Remote Sensing of Environment 115(2), 767–780. doi:10.1016/j.rse.2010.11.004 Elmendorf, S.C., Henry, G.H.R., Hollister, R.D., Björk, R.G., Boulanger-Lapointe, N., Cooper, E.J., Cornelissen, J.H.C., Day, T.A., Dorrepaal, E., Elumeeva, T.G., Gill, M., Gould, W.A., Harte, J., Hik, D.S., Hofgaard, A., Johnson, D.R., Johnstone, J.F., Jónsdóttir, I.S., Jorgenson, J.C., Klanderud, K., Klein, J.A., Koh, S., Kudo, G., Lara, M., Lévesque, E., Magnússon, B., May, J.L., Mercado-Dı´az, J.A., Michelsen, A., Molau, U., Myers-Smith, I.H., Oberbauer, S.F., Onipchenko, V.G., Rixen, C., Martin Schmidt, N., Shaver, G.R., Spasojevic, M.J., Þórhallsdóttir, Þ.E., Tolvanen, A., Troxler, T., Tweedie, C.E., Villareal, S., Wahren, C.-H., Walker, X., Webber, P.J., Welker, J.M., Wipf, S., 2012. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2(6), 453–457. doi:10.1038/nclimate1465 Emmerton, C.A., St. Louis, V.L., Humphreys, E.R., Gamon, J.A., Barker, J.D., Pastorello, G.Z., 2015. Net ecosystem exchange of CO 2 with rapidly changing high Arctic landscapes. Global Change Biology n/a-n/a. doi:10.1111/gcb.13064 Epstein, H.E., Calef, M.P., Walker, M.D., Chapin, F.S., Starfields, A.M., 2004. Detecting changes in arctic tundra plant communities in response to warming over decadal time scales. Global Change Biology 10(8), 1325–1334. doi:10.1111/j.1529-8817.2003.00810.x Fensholt, R., Sandholt, I., 2003. Derivation of a shortwave infrared water stress index from MODIS near- and shortwave infrared data in a semiarid environment. Remote Sensing of Environment 87(1), 111– 121. doi:10.1016/j.rse.2003.07.002 Fensholt, R., Sandholt, I., Rasmussen, M.S., 2004. Evaluation of MODIS LAI, fAPAR and the relation between fAPAR and NDVI in a semi-arid environment using in situ measurements. Remote Sensing of Environment 91(3–4), 490–507. doi:10.1016/j.rse.2004.04.009 Fraser, R.H., Olthof, I., Lantz, T.C., Schmitt, C., 2016. UAV photogrammetry for mapping vegetation in the low-Arctic. Arctic Science 2(3), 79–102. doi:10.1139/as-2016-0008 Frost, G. V, Epstein, H.E., 2014. Tall shrub and tree expansion in Siberian tundra ecotones since the 1960s. Global change biology 20(4), 1264–1277. doi:10.1111/gcb.12406 GCOS, 2006. Systematic observation requirements for satellite-based products for climate. GCOS.
11
Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms for remote estimation of vegetation fraction. Remote Sensing of Environment 80(1), 76–87. doi:10.1016/S0034- 4257(01)00289-9 Goward, S.N., Huemmrich, K.F., 1992. Vegetation canopy PAR absorptance and the normalized difference vegetation index: An assessment using the SAIL model. Remote Sensing of Environment 39(2), 119– 140. doi:10.1016/0034-4257(92)90131-3 Hernández-Clemente, R., Navarro-Cerrillo, R.M., Zarco-Tejada, P.J., 2012. Carotenoid content estimation in a heterogeneous conifer forest using narrow-band indices and PROSPECT+DART simulations. Remote Sensing of Environment 127, 298–315. doi:10.1016/j.rse.2012.09.014 Hill, M.J., Senarath, U., Lee, A., Zeppel, M., Nightingale, J.M., Williams, R. (Dick) J., McVicar, T.R., 2006. Assessment of the MODIS LAI product for Australian ecosystems. Remote Sensing of Environment 101(4), 495–518. doi:10.1016/j.rse.2006.01.010 Hope, A.S., Kimball, J.S., Stow, D.A., 1993. The relationship between tussock tundra spectral reflectance properties and biomass and vegetation composition. International Journal of Remote Sensing 14(10), 1861–1874. doi:10.1080/01431169308954008 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Oberbauer, S.F., Kinoshita, G., Houston, S., Kuchy, A., Hollister, R.D., Kwon, H., Mano, M., Harazono, Y., Webber, P.J., Oechel, W.C., 2010. Remote sensing of tundra gross ecosystem productivity and light use efficiency under varying temperature and moisture conditions. Remote Sensing of Environment 114(3), 481–489. doi:http://dx.doi.org/10.1016/j.rse.2009.10.003 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Campbell, P.K.E., Landis, D.R., Middleton, E.M., 2013. Arctic tundra vegetation functional types based on photosynthetic physiology and optical properties. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6, 265–275. doi:10.1109/JSTARS.2013.2253446 Jensen, J.R., 2005. Introductory digital image processing: a remote sensing perspective (Third Edition). Prentice Hall, Upper Saddle River, N.J. Jensen, J.L.R., Humes, K.S., Hudak, A.T., Vierling, L.A., Delmelle, E., 2011. Evaluation of the MODIS LAI product using independent lidar-derived LAI: A case study in mixed conifer forest. Remote Sensing of Environment 115(12), 3625–3639. doi:10.1016/j.rse.2011.08.023 Jia, G.J., Epstein, H.E., Walker, D.A., 2003. Greening of Arctic Alaska, 1981-2001. Geophysical Research Letters 30(20), n/a-n/a. doi:10.1029/2003GL018268 Jia, G.J., Epstein, H.E., Walker, D.A., 2009. Vegetation greening in the Canadian arctic related to decadal warming. Journal of Environmental Monitoring 11(12), 2231. doi:10.1039/b911677j Johansen, B., Tømmervik, H., 2014. The relationship between phytomass, NDVI and vegetation communities on Svalbard. International Journal of Applied Earth Observation and Geoinformation 27, 20–30. doi:10.1016/j.jag.2013.07.001 Kalacska, M., Lalonde, M., Moore, T.R., 2015. Estimation of foliar chlorophyll and nitrogen content in an ombrotrophic bog from hyperspectral data: Scaling from leaf to image. Remote Sensing of Environment 169, 270–279. doi:10.1016/j.rse.2015.08.012 King, D.A., Turner, D.P., Ritts, W.D., 2011. Parameterization of a diagnostic carbon cycle model for continental scale application. Remote Sensing of Environment 115(7), 1653–1664. doi:http://dx.doi.org/10.1016/j.rse.2011.02.024
12
Krebs, C.J., Danell, K., Angerbjörn, A., Agrell, J., Berteaux, D., Bråthen, K.A., Danell, Ö., Erlinge, S., Fedorov, V., Fredga, K., Hjältén, J., Högstedt, G., Jónsdóttir, I.S., Kenney, A.J., Kjellén, N., Nordin, T., Roininen, H., Svensson, M., Tannerfeldt, M., Wiklund, C., 2003. Terrestrial trophic dynamics in the Canadian Arctic. Canadian Journal of Zoology 81(5), 827–843. doi:10.1139/z03-061 Laidler, G.J., Treitz, P.M., Atkinson, D.M., 2008. Remote sensing of Arctic vegetation: relations between the NDVI, spatial resolution and vegetation cover on Boothia Peninsula, Nunavut. Arctic 61(1), 1–13. la Puma, I.P., Philippi, T.E., Oberbauer, S.F., 2007. Relating NDVI to ecosystem CO2 exchange patterns in response to season length and soil warming manipulations in arctic Alaska. Remote Sensing of Environment 109(2), 225–236. doi:10.1016/j.rse.2007.01.001 Molau, U., Mølgaard, P., 1996. ITEX Manual (Second Edition). Copenhagen. Myneni, R.B., Williams, D.L., 1994. On the relationship between FAPAR and NDVI. Remote Sensing of Environment 49(3), 200–211. doi:10.1016/0034-4257(94)90016-7 Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386(6626), 698–702. doi:10.1038/386698a0 Olthof, I., Latifovic, R., Pouliot, D., 2009. Development of a circa 2000 land cover map of northern Canada at 30 m resolution from Landsat. Canadian Journal of Remote Sensing 35(2), 152–165. doi:10.5589/m09-007 Olthof, I., Pouliot, D., 2010. Treeline vegetation composition and change in Canada’s western Subarctic from AVHRR and canopy reflectance modeling. Remote Sensing of Environment 114(4), 805–815. doi:10.1016/j.rse.2009.11.017 Pasolli, L., Asam, S., Castelli, M., Bruzzone, L., Wohlfahrt, G., Zebisch, M., Notarnicola, C., 2015. Retrieval of Leaf Area Index in mountain grasslands in the Alps from MODIS satellite imagery. Remote Sensing of Environment 165, 159–174. doi:CO10.1016/j.rse.2015.04.027 Pearson, R.G., Phillips, S.J., Loranty, M.M., Beck, P.S.A., Damoulas, T., Knight, S.J., Goetz, S.J., 2013. Shifts in Arctic vegetation and associated feedbacks under climate change. Nature Climate Change 3(7), 673–677. doi:10.1038/nclimate1858 Privette, J.., Myneni, R.., Knyazikhin, Y., Mukelabai, M., Roberts, G., Tian, Y., Wang, Y., Leblanc, S.., 2002. Early spatial and temporal validation of MODIS LAI product in the Southern Africa Kalahari. Remote Sensing of Environment 83(1–2), 232–243. doi:10.1016/S0034-4257(02)00075-5 Purevdorj, T., Tateishi, R., Ishiyama, T., Honda, Y., 1998. Relationships between percent vegetation cover and vegetation indices. International Journal of Remote Sensing 19(February 2015), 3519–3535. doi:10.1080/014311698213795 Raynolds, M.K., Comiso, J.C., Walker, D.A., Verbyla, D., 2008. Relationship between satellite-derived land surface temperatures, arctic vegetation types, and NDVI. Remote Sensing of Environment 112(4), 1884–1894. doi:10.1016/j.rse.2007.09.008 Raynolds, M.K., Walker, D.A., Epstein, H.E., Pinzon, J.E., Tucker, C.J., 2012. A new estimate of tundra- biome phytomass from trans-Arctic field data and AVHRR NDVI. Remote Sensing Letters 3(5), 403– 411. doi:10.1080/01431161.2011.609188 Rees, W.G., Williams, M., Vitebsky, P., 2003. Mapping land cover change in a reindeer herding area of the Russian Arctic using Landsat TM and ETM+ imagery and indigenous knowledge. Remote Sensing of Environment 85(4), 441–452. doi:10.1016/S0034-4257(03)00037-3
13
Riedel, S.M., Epstein, H.E., Walker, D.A., 2005a. Biotic controls over spectral reflectance of Arctic tundra vegetation. International Journal of Remote Sensing 26(11), 2391–2405. doi:10.1080/01431160512331337754 Riedel, S.M., Epstein, H.E., Walker, D.A., Richardson, D.L., Calef, M.P., Edwards, E., Moody, A., 2005b. Spatial and temporal heterogeneity of vegetation properties among four tundra plant communities at Ivotuk, Alaska, U.S.A. Arctic, Antarctic, and Alpine Research 37(1), 25–33. doi:10.1657/1523- 0430(2005)037[0025:sathov]2.0.co;2 Rogge, D., Rivard, B., Segl, K., Grant, B., Feng, J., 2014. Mapping of NiCu-PGE ore hosting ultramafic rocks using airborne and simulated EnMAP hyperspectral imagery, Nunavik, Canada. Remote Sensing of Environment 152, 302–317. doi:10.1016/j.rse.2014.06.024 Running, S.W., Nemani, R.R., Heinsch, F.A., Zhao, M., Reeves, M., Hashimoto, H., 2004. A continuous satellite-derived measure of global terrestrial primary production. Bioscience 54(6), 547–560. Schneider, J., Grosse, G., Wagner, D., 2009. Land cover classification of tundra environments in the Arctic Lena Delta based on Landsat 7 ETM + data and its application for upscaling of methane emissions. Remote Sensing of Environment 113(2), 380–391. doi:10.1016/j.rse.2008.10.013 Schubert, P., Lund, M., Ström, L., Eklundh, L., 2010. Impact of nutrients on peatland GPP estimations using MODIS time series data. Remote Sensing of Environment 114(10), 2137–2145. doi:10.1016/j.rse.2010.04.018 Serbin, S.P., Ahl, D.E., Gower, S.T., 2013. Spatial and temporal validation of the MODIS LAI and FPAR products across a boreal forest wildfire chronosequence. Remote Sensing of Environment 133, 71–84. doi:10.1016/j.rse.2013.01.022 Stow, D.A., Hope, A., McGuire, D., Verbyla, D., Gamon, J., Huemmrich, F., Houston, S., Racine, C., Sturm, M., Tape, K., Hinzman, L., Yoshikawa, K., Tweedie, C., Noyle, B., Silapaswan, C., Douglas, D., Griffith, B., Jia, G., Epstein, H., Walker, D., Daeschner, S., Petersen, A., Zhou, L., Myneni, R., 2004. Remote sensing of vegetation and land-cover change in Arctic tundra ecosystems. Remote Sensing of Environment 89(3), 281–308. doi:10.1016/j.rse.2003.10.018 Sturm, M., Racine, C., Tape, K., 2001. Climate change: increasing shrub abundance in the Arctic. Nature 411(6837), 546–7. doi:10.1038/35079180 Sweet, S.K., Asmus, A., Rich, M.E., Wingfield, J., Gough, L., Boelman, N.T., 2015. NDVI as a predictor of canopy arthropod biomass in the Alaskan Arctic tundra. Ecological Applications 25(3), 779–790. doi:10.1890/14-0632.1 Tagesson, T., Mastepanov, M., Tamstorf, M.P., Eklundh, L., Schubert, P., Ekberg, A., Sigsgaard, C., Christensen, T.R., Ström, L., 2012. High-resolution satellite data reveal an increase in peak growing season gross primary production in a high-Arctic wet tundra ecosystem 1992–2008. International Journal of Applied Earth Observation and Geoinformation 18, 407–416. doi:10.1016/j.jag.2012.03.016 Tape, K.D., Sturm, M., Racine, C., 2006. The evidence for shrub expansion in Northern Alaska and the Pan-Arctic. Global Change Biology 12(4), 686–702. doi:10.1111/j.1365-2486.2006.01128.x Thenkabail, P.S., Mariotto, I., Gumma, M.K., Middleton, E.M., Landis, D.R., Huemmrich, K.F., 2013. Selection of hyperspectral narrowbands (HNBs) and composition of hyperspectral twoband vegetation indices (HVIs) for biophysical characterization and discrimination of crop types using field reflectance and Hyperion/EO-1 Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6(2), 427–439. doi:10.1109/JSTARS.2013.2252601
14
Tian, Y., Woodcock, C.E., Wang, Y., Privette, J., Shabanov, N., Zhou, L., Zhang, Y., Buermann, W., Dong, J., Veikkanen, B., 2002. Multiscale analysis and validation of the MODIS LAI productI. Uncertainty assessment. Remote Sensing of Environment 83(3), 414–430. doi:10.1016/S0034-4257(02)00047-0 Ullmann, T., Schmitt, A., Roth, A., Duffe, J., Dech, S., Hubberten, H.-W., Baumhauer, R., 2014. Land cover characterization and classification of Arctic tundra environments by means of polarized synthetic aperture X- and C-Band Radar (PolSAR) and Landsat 8 multispectral Imagery — Richards Island, Canada. Remote Sensing 6(9), 8565–8593. doi:10.3390/rs6098565 Ulrich, M., Grosse, G., Chabrillat, S., Schirrmeister, L., 2009. Spectral characterization of periglacial surfaces and geomorphological units in the Arctic Lena Delta using field spectrometry and remote sensing. Remote Sensing of Environment 113(6), 1220–1235. doi:http://dx.doi.org/10.1016/j.rse.2009.02.009 Walker, M.D., Wahren, C.H., Hollister, R.D., Henry, G.H.R., Ahlquist, L.E., Alatalo, J.M., Bret-Harte, M.S., Calef, M.P., Callaghan, T. V., Carroll, A.B., Epstein, H.E., Jonsdottir, I.S., Klein, J.A., Magnusson, B., Molau, U., Oberbauer, S.F., Rewa, S.P., Robinson, C.H., Shaver, G.R., Suding, K.N., Thompson, C.C., Tolvanen, A., Totland, O., Turner, P.L., Tweedie, C.E., Webber, P.J., Wookey, P.A., 2006. Plant community responses to experimental warming across the tundra biome. Proceedings of the National Academy of Sciences 103(5), 1342–1346. doi:10.1073/pnas.0503198103 Wang, Y., Woodcock, C.E., Buermann, W., Stenberg, P., Voipio, P., Smolander, H., Häme, T., Tian, Y., Hu, J., Knyazikhin, Y., Myneni, R.B., 2004. Evaluation of the MODIS LAI algorithm at a coniferous forest site in Finland. Remote Sensing of Environment 91(1), 114–127. doi:10.1016/j.rse.2004.02.007 Williams, M., 2005. Optical instruments for measuring leaf area index in low vegetation: application in Arctic ecosystems. Ecological Applications 15(4), 1462–1470. doi:10.1890/03-5354
15
Chapter 2
Examining spectral reflectance related to Arctic percent vegetation cover: Implications for
hyperspectral remote sensing of Arctic tundra
2.1. Abstract
In this study, the utility of hyperspectral remote sensing data for estimating green percent vegetation cover
(PVC) was investigated for a study site in the Canadian High Arctic. A field study was conducted in 2011 on Sabine Peninsula (76°27’ N, 108°33’ W), Melville Island, Nunavut, Canada to collect field spectra and
PVC for five vegetation types: polar semi-desert (PD), dry mesic tundra (DMT), mesic tundra (MT), wet mesic tundra (WMT) and wet sedge/moss (WSM). Based on field spectra, two types of 2-band hyperspectral (i.e., Hyperion) and multispectral (i.e., WorldView-3) vegetation indices (VIs) were derived using all possible band combinations. Optimal spectral bands were identified based on their correlations with green PVC. In addition, VIs designed for other landscapes were examined for their ability to estimate green PVC in an Arctic environment. The results indicate that PVC and spectral features for Arctic vegetation types were related to soil moisture content: (1) vegetation types with dry to intermediate soil moisture (e.g., PD, DMT and MT) had large amounts of bare soil and exhibited spectral properties similar to bare soil; and (2) vegetation types with high moisture content (e.g., WMT and WSM) exhibited spectra similar to senesced vegetation given the substantial proportion of senesced vegetation in these vegetation types. The optimal Hyperion spectral bands for estimating green PVC were located at the absorption features observed in Arctic vegetation spectra, including 681.20 nm (leaf chlorophyll absorption); 721.90 nm and 732.07 nm (along the red-edge slope); 1174.77 nm and 1184.87 nm (leaf water absorption); and
1447.14 nm, 1457.23 nm, 2072.65 nm and 2102.94 nm (leaf cellulose and lignin absorption). Narrowband
VIs exhibited a stronger correlation with green PVC than broadband VIs due to the finer spectral features sampled by hyperspectral data. Further, VIs designed to estimate leaf pigment and dry matter content (e.g., lignin and cellulose) showed strong correlations with green PVC.
16
2.2. Introduction
It has been estimated that temperatures in the Arctic will exceed global mean warming by 40% (IPCC,
2013). Hence, it is not surprising that some of the strongest signals of climate change have already been observed at high latitudes (Barber et al., 2008). This amplified warming will have widespread and diverse impacts on Arctic vegetation types (Walker et al., 2006; Post et al., 2009; Myers-Smith et al., 2011); i.e., plant growth will increase (Myneni et al., 1997) and differentially affect species abundance, thereby changing community boundaries, species composition and overall ecosystem processes (Wookey et al.,
2009). In addition to air (and soil) temperature, there are a number of factors that serve as controls on vegetation growth in the High Arctic, including soil moisture (Laidler et al., 2008; Atkinson and Treitz,
2013), available nutrients (Shaver and Chapin, 1980; van Wijk et al., 2005), topography (Evans et al., 1989), soil type (Walker et al., 2011) and permafrost disturbance (Rudy et al., 2013). The spatial variability of these environmental controls contributes to a diverse and heterogeneous vegetation cover.
Long-term measurements of percent vegetation cover (PVC) and species abundance are required to monitor the impact of a warming climate on Arctic vegetation (Stow et al., 2004; Hudson and Henry, 2010;
Elmendorf et al., 2012; Tape et al., 2012; Atkinson and Treitz, 2013; Davidson et al., 2016; Stewart et al.,
2016). The point-frame technique represents a traditional field method for collecting PVC data (Molau and
Mølgaard, 1996; Hudson and Henry, 2010; Elmendorf et al., 2012; Atkinson and Treitz, 2013). Although this technique provides fine-scale quantitative data on plant species richness, canopy height and PVC, it is extremely time-consuming since the plant identification of all vegetation layers is required (Chen et al.,
2010). In addition, the changes in tundra vegetation detected by the point-frame method could not be mapped or extrapolated across areas that have not been measured.
Satellite remote sensing provides an opportunity to monitor Arctic vegetation at a variety of spatial and temporal scales (Stow et al., 2004). While there have been many studies examining biophysical variables at high latitudes using satellite remote sensing, they have been predominantly limited to coarse (i.e., 1-8 km2) (Myneni et al., 1997; Walker et al., 2005; Jia et al., 2009; Zeng et al., 2011; Raynolds et al., 2012) and
17 to a lesser extent, intermediate spatial resolutions (Olthof et al., 2009; Fraser et al., 2011; Johansen and
Tømmervik, 2014; Ju and Masek, 2016; Nitze and Grosse, 2016). In these studies, the broadband normalized difference vegetation index (NDVI) has been applied to characterize Arctic biophysical variables such as green PVC, biomass and carbon flux (Hope et al., 1993; Epstein et al., 2004; Stow et al.,
2004; Laidler et al., 2008; Tagesson et al., 2012; Atkinson and Treitz, 2013). However, this relationship is complicated for Arctic tundra by the optical properties of non-vascular plants/organisms (e.g., mosses, lichens, cyanobacteria), non-photosynthetic components of vegetation (vascular and non-vascular) (e.g., woody stems, leaf litter, senesced vegetation), leaf and plant architecture/structure/organization, illumination conditions and sensor viewing geometry, and soil/substrate reflectance (and organic content)
(Curran, 1989; Asner, 1998; Stow et al., 2004; Feret et al., 2008; Juszak et al., 2014).
Compared to multispectral (i.e., broadband) remote sensing, hyperspectral (i.e., narrowband) remote sensing provides numerous spectral bands which have demonstrated great utility in Arctic research. For instance, satellite/airborne hyperspectral data are capable of capturing subtle changes in mineral absorption features and mapping Arctic lithological units (Harris et al., 2005; Bedini, 2009; Gleeson et al., 2010;
Leverington, 2010). With regards to hyperspectral remote sensing of Arctic vegetation, most studies have been based on field spectra (Hope et al., 1993; Riedel et al., 2005a, 2005b; Laidler et al., 2008; Ulrich et al., 2009; Kushida et al., 2009, 2015; Buchhorn et al., 2013; Huemmrich et al., 2013; Bratsch et al., 2016;
Davidson et al., 2016). These studies can be grouped into two broad categories: (1) vegetation classification; and (2) biophysical variable estimation.
Studies of hyperspectral classification of Arctic vegetation have largely focused on characterizing spectral features of the tundra and investigating spectral separability among different vegetation types (Ulrich et al.,
2009; Buchhorn et al., 2013; Huemmrich et al., 2013; Davidson et al., 2016). Generally, Arctic vegetation spectra were found to be a mixed signal of vascular plants (green or non-green), non-vascular plants (e.g., lichens and mosses) and soil background (dry or wet) (Ulrich et al., 2009). For some vegetation types with dry to intermediate soil moisture, spectral features common to green vegetation, such as the green
18 reflectance peak (around 550 nm), steep red-edge slope (690-720 nm) and the high near-infrared reflectance plateau (720-1300 nm) were not observed (Ulrich et al., 2009; Buchhorn et al., 2013; Bratsch et al., 2016;
Davidson et al., 2016). Classification results from these studies have suggested that hyperspectral data may be useful for Arctic vegetation mapping (Buchhorn et al., 2013; Bratsch et al., 2016; Davidson et al., 2016).
By using the sparse partial least squares regression, Bratsch et al. (2016) identified several optimal hyperspectral bands for classifying four Alaskan Arctic vegetation types. These optimal spectral bands were located at important spectral features (e.g., pigment absorptions at blue and red bands, red-edge slopes and near-infrared bands) and could achieve a high classification accuracy (>80%). One recent study by
Davidson et al. (2016) has also highlighted these spectral bands. In addition, Davidson et al. (2016) found that classification accuracy could be improved by incorporating vegetation indices (VIs) into classification.
With regards to biophysical estimation, most studies have focused on correlating narrowband VIs with field measured biophysical variables (Stow et al., 2004; Riedel et al., 2005a, 2005b; Laidler et al., 2008;
Buchhorn et al., 2013; Huemmrich et al., 2013; Kushida et al., 2015). For instance, NDVI, enhanced VI
(EVI) and soil-adjusted VI (SAVI) have been examined for their ability to estimate green PVC (Laidler et al., 2008; Kushida et al., 2009), biomass (Riedel et al., 2005a, 2005b, Kushida et al., 2009, 2015), and leaf area index (LAI) (Riedel et al., 2005a, 2005b; Williams, 2005; Williams et al., 2008). Huemmrich et al.
(2013) used a 3-band hyperspectral VI as a proxy for the chlorophyll concentration of Arctic vegetation.
Buchhorn et al. (2013) assessed the performance of the NDVI involving all possible two band combinations within the 420-1100 nm wavelength region for estimating shrub biomass and found that narrowband NDVI performed better than broadband NDVI. Recent studies have demonstrated that the absorption depth extracted from continuum-removed spectrum could be used to estimate Arctic biophysical variables with high accuracy (Ulrich et al., 2009; Buchhorn et al., 2013; Bratsch et al., 2016).
To date, hyperspectral remote sensing of Arctic vegetation has not been thoroughly explored. For instance, the spectral bands used in previous studies were limited to the visible-near infrared (VIS-NIR: 450-1300 nm) wavelength region (Kushida et al., 2009; Buchhorn et al., 2013). Studies that apply the shortwave
19 infrared (SWIR: 1400-2500 nm) wavelengths for estimating Arctic biophysical variables are lacking.
Further, spectral VIs designed for landscapes such as croplands, grasslands and forests have seldom been tested for their utility in sparsely vegetated High Arctic tundra vegetation with exposed soil/tills, large quantities of non-vascular plants (i.e., mosses, lichens) or large amounts of senesced vegetation (vascular and non-vascular).
As a result, the overall purpose of this research was to evaluate the performance of satellite hyperspectral data for estimating green PVC within the 450-2500 nm wavelength region. The first objective was to characterize the PVC and spectral properties of five vegetation types distributed along a moisture gradient for a study site in the Canadian High Arctic. The second objective was to identify the optimal spectral bands for estimating green PVC. Finally, the performance of broadband and narrowband VIs for predicting green
PVC was evaluated.
2.3. Study Area
The study area is located on the Sabine Peninsula, Melville Island, Nunavut, Canada (76°27’ N, 108°33’
W) (Figure 2-1). The mean monthly July temperature and precipitation are approximately 4.5 °C and 28.1 mm, respectively, based on 1981 to 2010 Canadian Climate Normals for Resolute Bay, Nunavut
(http://climate.weather.gc.ca/climate_normals/index_e.html). Geologically, this site is located within the
Sverdrup Basin and has four major geological units: Kanguk and Christopher shales, Hassel sandstone and an anhydrite dome (Harrison, 1990). Weathered bedrock generates a surficial material commonly observed in the Kanguk shale and anhydrite dome, and is too coarse to support vascular plants (Harrison, 1990;
Edlund, 1993). Soil pH varies from weakly to moderately acidic (Kanguk) to weakly alkaline (Hassel and
Christopher) to highly alkaline (anhydrite dome) (Harrison, 1990). The active layer depths for this area range from approximately 0.5-1.0 m (Rudy et al., 2013).
The vegetation for this study area was grouped into five vegetation types: (1) polar semi-desert (PD); (2) dry mesic tundra (DMT); (3) mesic tundra (MT); (4) wet mesic tundra (WMT); and (5) wet sedge/moss
(WSM) (Figure 2-1 C). The distribution of vegetation types is determined by topographic and soil moisture
20 gradients (Edlund, 1993): the PD type is generally located on well-drained uplands; the WMT and WSM types generally occur in low-lying areas alongside waterways and in the proximity of permanent snowbanks; and the other vegetation types tend to occur on intermediate moisture sites.
Figure 2-1. Study area on the Sabine Peninsula, Melville Island, Nunavut, Canada (A-B). The WorldView- 2 unsupervised classification in panel C was adapted from Allux et al., (2012). Sampling regions (i.e., ellipses: ~2.5 km × 1 km) spanning at least two geological units were identified. Plots (i.e., dots: 20 m × 20 m) exhibiting homogeneous vegetation types were randomly located within each sampling region and validated in the field.
2.4. Methods
2.4.1. Field Sampling Design
A field campaign was conducted to collect field measurements of PVC and spectral reflectance from 7th
July to 3rd August, 2011. A stratified random sampling strategy was employed to locate sampling plots within the study area (Figure 2-1 C). Firstly, a 2-m spatial resolution unsupervised classification (i.e.,
21 iterative self-organized data analysis technique, ISODATA) with 20 spectral classes was created from a
WorldView-2 image acquired on 12th July, 2010 (Allux et al., 2012). Classes were then labeled and merged manually to create 10 clusters based on comparisons of class mean spectra and georeferenced field photographs collected in 2010 (Allux et al., 2012). Secondly, nine sampling regions (i.e., identified by ellipses in Figure 2-1 C; ~ 2.5 km×1 km) spanning at least two geological units were manually created on the classification. Thirdly, sampling plots (i.e., identified by dots in Figure 2-1 C; 20 m×20 m) were randomly distributed within each region according to the proportion of vegetation types within the region.
To ensure that sample plots would be sufficiently homogeneous and representative of the vegetation types at that location (i.e., 20 m×20 m sample plots consisted of a single vegetation type), plots that were not located within a homogeneous vegetation type were relocated to the nearest homogeneous area of the same vegetation type. It should be noted that in this environment, biophysical variables (e.g., PVC, biomass), vegetation type (i.e., PD, DMT, MT, WMT and WSM) and soil moisture are closely related to NDVI
(Atkinson and Treitz, 2012). Hence, once plot locations were selected, WorldView-2 NDVI histograms were generated and examined to determine whether the selected plots would be representative of the full range of PVC (and vegetation types). Where gaps in the NDVI range were observed, additional plots were added to the analyses to ensure that the full range of PVC (and vegetation types) were represented. Finally, plot location was confirmed in the field and adjusted if necessary to ensure plots fell within a homogeneous vegetation type. In select cases, where unique environments were observed in the field (e.g., bare soil/substrate or permafrost disturbances such as active layer detachments), additional plots were established and sampled.
In total, 47 plots were sampled: 29 plots had PVC and spectral reflectance measurements collected, while
18 plots had PVC measurements only due to adverse weather conditions restricting spectral data collection
(Table 2-1). All field spectra were collected under clear sky conditions and PVC data were collected within two days of the spectral measurements. In the field, sampling plots were located using a handheld Garmin
GPS with an accuracy of approximately 3 m.
22
Table 2-1. Number of sampling plots and mean NDVI (from ground spectra) for each vegetation type.
# of plots with PVC and # of plots with PVC NDVI Land Cover spectra only (Mean ± Std.) Polar semi-desert (PD) 5 3 0.099 ± 0.010 Dry Mesic Tundra (DMT) 4 1 0.214 ± 0.037 Mesic Tundra (MT) 4 4 0.280 ± 0.017 Wet Mesic Tundra (WMT) 9 4 0.371 ± 0.108 Wet Sedge/Moss (WSM) 7 6 0.459 ± 0.112 29 18
2.4.2. Percent Vegetation Cover
The international tundra experiment (ITEX) protocol was adopted for the measurement of PVC (Molau and
Mølgaard, 1996). In the center of each 20 m×20 m plot, a 1 m×1 m quadrat with equidistantly spaced grids
(grid interval = 10 cm) was placed over the vegetation canopy (Figure 2-2). At each grid point, a rod with
1 cm gradations was lowered to measure the height of plants contacting the rod. Species and plant status
(i.e., green or senesced) for each contact were recorded. The rod was lowered until it contacted the ground surface.
Vegetation was first grouped into five plant functional groups: shrubs, graminoids/sedges, mosses, forbs and lichens. Then, two measures of PVC (i.e., two-dimensional (PVC2D) and three-dimensional (PVC3D)) were calculated for each plant functional group and each status (i.e., senesced or green) (Buchhorn et al.,
2013). The PVC2D was calculated as the total number of top-layer contacts of a specific plant functional group divided by the number of frame grids (i.e., 100). Since PVC2D uses the top-of-canopy hits, it has a range of 0% - 100%. The PVC3D was calculated as the total number of all the point-frame contacts with a specific plant functional group divided by the number of frame grids (i.e., 100). PVC3D can be equated to the LAI of a specific functional group (with the exception of mosses and lichens) (Williams, 2005) and can have a value greater than 100% in the case of multilayer canopies. Here, the focus is on the biophysical variable PVC2D given NDVI is more closely related to PVC than LAI (Carlson and Ripley, 1997). The percent cover of all green plant functional groups was aggregated to obtain the total green PVC (i.e., PVCtotal)
23 for the 2D and 3D cases, respectively. In addition, the canopy height (H) was calculated as the average height of all the point-frame contacts of the top layer.
Figure 2-2. PVC and spectral reflectance sampling design. For each homogeneous plot (~ 20 m x 20 m, A), PVC and spectra were measured within a 1 m x 1 m quadrat located at the center of the plot (i.e., green square in A). For PVC measurements, a 1 m x 1 m quadrat with 10 cm grids was used (B). For spectral reflectance measurements, fifteen measurements were collected at five different locations within the 1 m x 1 m quadrat: three in the center of the quadrat and three in each corner of the quadrat (C). The circles in C represent the field of view (FOV) of the ASD spectroradiometer.
2.4.3. Field Spectra
Since satellite hyperspectral data were not available for the study area, field spectra were measured and used to simulate satellite spectra. The instrument used to measure field spectra was an ASD FieldSpec spectroradiometer (ASD Inc., Boulder, Colorado, US) which collects spectral radiance across the VIS-NIR-
SWIR wavelength region (i.e., 350-2500 nm) using one VIS-NIR spectrometer (350-1000 nm, spectral
24 resolution: 3 nm) and two SWIR spectrometers (1000-1770 nm and 1770-2500 nm, spectral resolution: 10-
12 nm).
Field spectra were collected within a 1 m × 1 m quadrat located at the center of each 20 m × 20 m plot
(Figure 2-2 A). To minimize the effect of solar zenith angle (SZA) on nadir spectra, all radiance measurements were collected at, or near solar noon, when changes in SZA are small (i.e., within 2 degrees).
The field of view (FOV) of the fiber-optic cable probe was 18 degrees and the height of the horizontal tripod extension was ~1.5 m, allowing for the radiance measurements of a 0.5 m diameter ground surface
(Figure 2-2 C). To ensure the spectral variation of the 1 m2 quadrat was fully characterized, 15 radiance measurements were collected at different locations within the quadrat: three in the center of the quadrat and three in each corner of the quadrat (Figure 2-2 C). A 25.4 cm × 25.4 cm white reference panel (Labsphere
Inc., North Sutton, New Hampshire, US) was used for reflectance calibration. White panel radiance measurements were collected before and after quadrat radiance measurements to derive calibrated spectral reflectance. Radiance measurements of the quadrat were divided by white panel radiance to derive spectral reflectance. The white panel radiance applied for reflectance calibration was dependent on whether atmospheric conditions changed during the quadrat measurement period. The 15 calibrated reflectance measurements were averaged to obtain a single reflectance spectrum for each plot.
Two representative sensors, Hyperion and WorldView-3, were chosen for simulation. As a hyperspectral sensor, Hyperion provides 242 narrow spectral bands (spectral resolution: ~11 nm) in the 350-2500 nm wavelength region. Due to strong atmospheric water absorption, Hyperion spectral bands within the 1306-
1437 nm, 1790-1992 nm and 2365-2600 nm wavelength regions were not simulated (Thenkabail et al.,
2013). Therefore, a total of 185 Hyperion spectral bands were simulated in this study (Table 2-2). The multispectral satellite WorldView-3 provides eight broad spectral bands in the VIS – NIR and eight in the
SWIR (Table 2-2)
25
Table 2-2. Specifications for Hyperion and WorldView-3 spectral bands.
Hyperion Wavelength region # of bands (spectral resolution: ~ 11 nm) Visible (350 – 690 nm) 33 Red-edge (690 – 720 nm) 4 Near infrared (720 – 1300 nm) 78 Shortwave infrared -a (1400 – 1900 nm) 34 Shortwave infrared -b (2000 – 2500 nm) 36 WorldView-3 Wavelength region Band name (wavelength range) Coastal Blue (400 – 450 nm) Blue (450 – 510 nm) Green (510 – 580 nm) Yellow (585 – 625 nm) Visible – Near infrared (400 – 1040 nm) Red (630 – 690 nm) Red-edge (705 – 745 nm) Near infrared 1 (770 – 895 nm) Near infrared 2 (860 – 1040 nm) SWIR 1 (1195 – 1225 nm) SWIR 2 (1550 – 1590 nm) SWIR 3 (1640 – 1680 nm) SWIR 4 (1710 – 1750 nm) Shortwave infrared (1190 – 2500 nm) SWIR 5 (2145 – 2185 nm) SWIR 6 (2185 – 2225 nm) SWIR 7 (2235 – 2285 nm) SWIR 8 (2295 – 2365 nm)
Broadband WorldView-3 and narrowband Hyperion spectra were simulated by convolving field spectra with the spectral response functions (SRFs) of each sensor (Liang, 2005):
2 R() f (,0 ) R( ) 1 0 2 Eq. (2-1) f (,0 ) 1
where R(0 ) is the simulated spectral reflectance at a central wavelength 0 ; R() is the field spectral
reflectance at wavelength ; f (,0 ) is the SRF at wavelength with a central wavelength ; and
1 2 is the convolution range. Here, the SRFs of WorldView-3 were obtained from Digital Globe
(http://global.digitalglobe.com/sites/default/files/DigitalGlobe_Spectral_Response_1.pdf). For Hyperion, the SRF for each band was simulated using a Gaussian function (Liu et al., 2009):
26
2 (i ) FWHMi f (,i ) exp( 2 ), i ,i 1,...,242 Eq. (2-2) i 2 2ln 2 where f (,i ) is the SRF at wavelength with a central wavelength i ; and F W H M i is the full width at half maximum of band i. The central wavelength and F WH M data were obtained from: https://eo1.usgs.gov/sensors/hyperioncoverage.
For broadband and narrowband spectra, two types of 2-band VIs were calculated: normalized difference spectral index (NDSI) and a reciprocal difference spectral index (RDSI):
R(i ) R( j ) NDSI(i , j ) R(i ) R( j ) Eq. (2-3) 1 1 RDSI (i , j ) (i j) R(i ) R( j )
where R(i ) and R( j ) are the spectral reflectance values of band i and j; i and j are the wavelengths
of band i and j. Considering that NSDI(i , j ) and NSDI( j ,i ) (or RDSI (i , j ) and RDSI ( j ,i ) ) had the same R2 values with green PVC, 120 unique band combinations were obtained for WorldView-3 and 17,020 (i.e., 185x184/2) combinations for Hyperion. For narrowband spectra, the first-order derivatives of reflectance (dR) and the reciprocal transformation of reflectance (d(1/R)) were derived:
R(i1) R(i1) d(R(i )) i1 i1 Eq. (2-4) 1/ R(i1) 1/ R(i1) d(1/ R(i )) (i 2,...,241) i1 i1
In addition to the above VIs, the narrowband VIs used in previous studies were also examined (Table 2-3).
Clearly, many VIs presented in Table 2-3 are of the NDSI type (e.g., NDVI, SIPI, PRI, NDWI, SIWI, NDII,
NDFWC, NDTI and NDMI); or the RDSI type (e.g., RARS and CRI). These VIs were grouped based on their sensitivity to different biochemical components including: chlorophyll/carotenoids, water and dry matter (Table 2-3). Most VIs were designed for vegetation canopies such as croplands, forests and grasslands. However, they have seldom been tested in Arctic environments with sparse vegetation canopies.
27
Table 2-3. Hyperspectral vegetation indices used in previous studies. Columns PVC2D and PVC3D show the R2 results between PVC and VIs.
Index Formula References PVC2D PVC3D R2 R2 Chlorophyll /Carotenoid Concentration
NDVI (Rouse et 0.73 0.74 R R 800 670 al., 1973; (Normalized Difference R800 R670 Tucker, Vegetation Index) 1979) SIPI (Penuelas et 0.61 0.63 R R 800 445 al., 1995) (Structure Insensitive R R Pigment Index) 800 680 MCARI (Daughtry, 0.64 0.73 [(R R ) 2000) (Modified Chlorophyll 700 670 Absorption Reflectance 0.2(R700 R550)]R700 / R670 Index) TCARI (Haboudane 0.68 0.76 3[(R R ) et al., 2002) (Transformed Chlorophyll 700 670 Absorption Reflectance 0.2(R700 R550)R700 / R670] Index) PSSR a-c (Blackburn, 0.67 0.72 R R R 800 , 800 , 800 1998) 0.65 0.73 (Pigment-Specific Simple R R R 0.56 0.62 Ratio) 680 635 470 RARS (Chappelle 0.56 0.64 R 746 et al., 1992) (Ratio Analysis of R Reflectance Spectra) 513
28
Table 2-3. Hyperspectral vegetation indices used in previous studies (Continued). 1 1 (Gitelson et 0.43 0.33 R R al., 2003, 0.51 0.40 515 550 2006) 0.43 0.49 1 1 0.52 0.59 CRI 550, 700 R515 R700 (Carotenoid Concentration Index) 1 1 ( )R770 R515 R550 1 1 ( )R770 R515 R700 R R R R (Hernández- 0.26 0.32 512 531 , 600 531 R R R R Clemente et 0.01 0.01 PRI m1-4 512 531 600 531 al., 2011) 0.01 0.01 R R R R R 0.04 0.01 (Modified Photochemical 670 531 , 600 531 670 Reflectance Index) R670 R531 R600 R531 R670
PRI (Gamon et 0.01 0.02 R R 570 530 al., 1992) (Photochemical R R Reflectance Index) 570 530 PSRI (Merzlyak et 0.59 0.56 R R 680 500 al., 1999) (Plant Senescence R Reflectance Index) 750
R860 (Datt, 1999; 0.60 0.45 Datt, 1998) R550R708 0.13 0.03 0.02 0.02 R Three-band Index 672 R550R708 R R 850 710 R850 R680
29
Table 2-3. Hyperspectral vegetation indices used in previous studies (Continued). (Zarco- 0.73 0.66 Tejada, Miller, CI R675R690 Mohammed, R 2 (Curvature Index) 683 Noland, & Sampson, 2002) (Carter, 0.29 0.37 R695 R695 R605 R710 1994) 0.73 0.76 Carter 1-4 , , , R420 R760 R760 R760 0.69 0.72 0.70 0.78 (le Maire, 0.15 0.34 DDI François, & (R749 R720) (R701 R672) (Double Difference Index) Dufrêne, 2004)
mNDVI R R (Sims and 0.82 0.76 800 680 Gamon, R R 2R (Modified NDVI) 800 680 445 2002) Water Content NDWI (Gao, 1996) 0.08 0.02 R R 862 1239 (Normalized Difference R R Water Index) 862 1239
WI R900 (Penuelas et 0.26 0.11 al., 1997) (Water Index) R970 (Hunter and 0.06 0.18 MSI R1599 Rock, 1989) (Moisture Stress Index) R819
SIWSI (Fensholt 0.12 0.26 R R 858 1640 and (Short Infrared Water R858 R1640 Sandholt, Stress Index) 2003)
SRWI (Zarco- 0.08 0.02 R860 Tejada,
(Simple Ratio Water Rueda, & R1240 Index) Ustin, 2003)
30
Table 2-3. Hyperspectral vegetation indices used in previous studies (Continued). NDII (Hardisky et 0.04 0.15 R R 819 1649 al., 1983) (Normalized Difference R R Infrared Index) 819 1649 NDFWC (Féret et al., 0.45 0.60 R R 1062 1393 2011) (Normalized Difference R R Foliar Water Content) 1062 1393 RIFWC (Féret et al., 0.46 0.62 R 1062 2011) (Ratio Index Foliar Water R Content) 1393 Dry Matter NDTI (van 0.61 0.60 R R 1650 2215 Deventer et (Normalized Difference R R al., 1997) Tillage Index) 1650 2215
NDLI log(1/ R ) log(1/ R ) (Serrano et 0.62 0.71 1510 1680 al., 2002) (Normalized Difference log(1/ R ) log(1/ R ) Lignin Index) 1510 1680 CAI (Nagler et 0.58 0.48 al., 2000) (Cellulose Absorption 0.2(R2000 R2200) R2100 Index) NDMI (Wang et 0.54 0.43 R R 1649 1722 al., 2011) (Normalized Dry Matter R R Index) 1649 1722 DMCI (Romero et 0.51 0.40 R R 2305 1495 al., 2012) (Dry Matter Content R R Index) 2305 1495 R R (Maire et 0.63 0.64 1340 1710 NDLMA al., 2008; 0.64 0.63 R1340 R1710 Féret et al., (Normalized Difference 2011) R R Leaf Mass per Area) 1368 1722 R1368 R1722
31
Table 2-3. Hyperspectral vegetation indices used in previous studies (Continued). RILMA (Féret et al., 0.62 0.60 R 1368 2011) (Ratio Index Leaf Mass R per Area) 1722 LCA (Daughtry et 0.58 0.50 al., 2005) (Lignin-Cellulose 2R2205 (R2165 R2330) Absorption Index)
2.4.4. Regression
Based on the definitions of PVC3D and PVC2D, the difference between PVC3D and PVC2D (i.e.,
PVC PVC3D PVC2D ) is always ≥ 0 and is related to the number of canopy layers, that is, the more layers in the canopy, the larger the P V C . Therefore, a linear relationship was assumed between P V C and canopy height (H) which was used to characterize the number of canopy layers: PVC kH , where k is the slope. In other words, PVC3D could be modelled as:
PVC3D PVC 2D kH Eq. (2-5)
The above equation indicates that PVC3D = PVC2D for single-layer canopies (e.g., the PD or DMT with a single layer) since H = 0, whereas in the case of multi-layer canopies (e.g., WMT or WSM), the PVC3D could be estimated from PVC2D and canopy height (H).
To predict PVC from spectral data, linear regression models were developed between PVC and VIs:
PVC aVI b Eq. (2-6)
For NDSI and RDSI, ‘maps’ of R2 (coefficient of determination) were constructed to help visually identify which spectral band combinations were best suited for predicting green PVC.
32
2.5. Results and Discussion
2.5.1. Percent Vegetation Cover
Based on field measurements, the PVC by plant functional group for each vegetation type is presented in
Figure 2-3. An increase in the moss cover (i.e., the PVC3D of moss) was observed across the moisture gradient (i.e., from dry to wet vegetation types). However, for vegetation types (i.e., WMT and WSM) with increased cover of vascular plants, mosses became less visible and thus were not fully accounted for in
PVC2D. For forbs, an apparent PVC difference (~25%) was only observed between the PD and other vegetation types. Due to the contrasting moisture regime and likely competition between forbs and graminoids/sedges for light and space, the forbs cover of the WSM type was less than that of the WMT type. For the graminoids/sedges cover, only the WSM type showed an apparent difference with other vegetation types.
With regards to total PVC (i.e., the aggregated PVC of all green plant function groups), wet vegetation types exhibited higher PVC than dry types (Figure 2-3). Since PVC3D measured the multi-layer PVC, the increase in total PVC3D from dry types to wet types was much more apparent than that of total PVC2D: the increase in total PVC3D between two neighboring types was ~25%, while the total PVC2D among the MT,
WMT and WSM types was similar (i.e., the difference between vegetation types was less than 5%).
33
Figure 2-3. PVC by plant functional group for each vegetation type. Vegetation types are organized along a moisture gradient (i.e., polar semi-desert to wet sedge/moss). A) PVC3D results (i.e., PVC calculated using all contact points); B) PVC2D results (i.e., PVC calculated using first contact points only). N represents the number of plots sampled for each vegetation type.
34
The comparison between PVC3D and PVC2D for each plant functional group is presented in Figure 2-4. The
PVC (i.e., PVC3D - PVC2D) is indicative of the difference between top-of-canopy hit cover and total-hit cover; characterized by the divergence from the 1:1 lines in Figure 2-4. For instance, since the moss understory layer was effectively ‘masked’ by vascular plants, a large difference was observed between
PVC3D and PVC2D for mosses, especially for the WSM type which was almost fully ‘masked’ by vascular plants (Figure 2-4 A). For forbs and graminoids/sedges, P V C became pronounced when PVC2D exceeded 25% (Figure 2-4 B, C). For total PVC, became apparent when the total PVC2D exceeded
50% (Figure 2-4 D). Large differences in PVC were predominant for types that possessed multiple canopy layers such as WMT and WSM.
The vegetation canopy height for this study site was less than 10 cm (Figure 2-5). A strong correlation (R2
= 0.80) existed between and canopy height (Figure 2-5). This strong correlation indicates that the conventional point-frame method could be simplified if the purpose is to obtain the total PVC3D. Instead of measuring the PVC of all layers, we could simply measure the PVC and height of the top layer. Then, total
PVC3D could be modelled for each vegetation type. Compared with the conventional point-frame method, this method could obtain the field measurements of total PVC3D more efficiently with a comparable accuracy.
35
Figure 2-4. Comparison between PVC3D (y-axis) and PVC2D (x-axis) for each plant functional group (i.e., Green Moss, Green Forb, Green Graminoid/Sedge) and Total Green Vegetation (i.e., aggregated PVC). These plots include the 1:1 line.
36
Figure 2-5. The regression for the difference in total PVC between the two methods (i.e., total PVC3D - PVC2D) and canopy height. Polar Semi-Desert (PD); Dry Mesic Tundra (DMT); Mesic Tundra (MT); Wet Mesic Tundra (WMT); Wet Sedge/Moss (WSM).
2.5.2. Spectral Characteristics of Arctic Vegetation
The simulated Hyperion spectra (based on convolved field spectra) for each vegetation type along with the spectral ranges for each of the WorldView-3 bands are illustrated in Figure 2-6. Since Arctic tundra environments are quite different from other environments such as forests, grasslands and croplands, a detailed description of their spectral characteristics can help interpret the VI results reported in Sections
2.5.3 (‘Broadband VIs’) and 2.5.4 (‘Narrowband VIs’). Here, the spectral range of each region was adopted from Ulrich et al. (2009). It should be noted that the 1000-1300 nm spectral region can be discussed in the context of the shortwave infrared. However, since the spectrum in this region was continuous with that in the 720-1000 nm near infrared region and was discontinuous with that in the 1400-1900 nm shortwave infrared region due to atmospheric effects, the 1000-1300 nm region was grouped into the near infrared region for convenience.
37
In the visible (VIS: 450-690 nm) wavelength region, reflectance was generally lower for wet vegetation types (i.e., WSM, WMT) due to the increase in pigment concentration (e.g., chlorophylls and carotenoids) associated with an increase in PVC (Figure 2-6). Unlike the typical spectrum of green vegetation which has a pronounced reflectance peak at approximately 550 nm and a deep absorption feature at approximately
650 nm, the spectrum of each vegetation type here increased with wavelength. This is interpreted to be due to the presence of a large amount of dry bare soil (for PD and DMT) and senesced vegetation (e.g., senesced sedges, mosses and forbs) (for MT, WMT and WSM) (Figure 2-3). The dampened pigment absorption in the red wavelength region has also been found in previous studies (Ulrich et al., 2009; Buchhorn et al.,
2013; Bratsch et al., 2016; Davidson et al., 2016). In addition, it could be seen that wet vegetation types had a slightly deeper absorption in the red wavelength region than dry vegetation types. Ulrich et al. (2009),
Buchhorn et al. (2013) and Davidson et al., (2016) have demonstrated that continuum-removal analysis could enhance this absorption difference and the extracted absorption depth was useful for separating different Arctic vegetation types.
The red-edge (RE: 690-720 nm) wavelength region is a transition between the red wavelength region (of low reflectance for healthy vegetation) and the NIR wavelength region (of high reflectance for healthy vegetation) (Buchhorn et al., 2013). The RE reflectance is affected by both pigment absorption and leaf cell structure (Buchhorn et al., 2013). RE parameters (e.g., first-derivative inflection point) are therefore often used to estimate chlorophyll content (Curran, 1989; Ulrich et al., 2009) and detect vegetation stress (Zarco-
Tejada et al., 2003). Here, PD exhibited a shallow RE slope while WSM possessed the steepest slope due to the high contrast between red and NIR reflectance. Other vegetation types between these extremes exhibited moderate slopes related to low-to-moderate green PVC (Figure 2-6).
In the NIR (720-1300 nm) wavelength region, all vegetation types showed a low reflectance (<0.4) and did not exhibit the typical NIR reflectance plateau found for green vegetation. The low NIR reflectance could be attributed to low vegetation height structure and low green leaf area of the vegetation, i.e., less multiple backscattering occurred within the vegetation canopy and understory mosses/lichens reduced NIR
38 reflectance (Buchhorn et al., 2013). This effect is especially important in the High Arctic which has lower vegetation stature and less green vegetation than the Low Arctic. Wet and dry vegetation types exhibited distinct spectral trends/shapes. The spectral shape of the WMT and WSM types was similar but with a distinct offset: the reflectance increased sharply from 720 nm to 1140 nm and had an absorption feature at
1140 nm-1240 nm (Figure 2-6). In some studies, this absorption was attributed to leaf water (Ulrich et al.,
2009; Buchhorn et al., 2013). However, since WMT and WSM types possessed large amounts of senesced vegetation, the water absorption feature was not as strong as that observed for green vegetation. It was noteworthy that cellulose/lignin absorption overlapped water absorption in this region (Curran, 1989).
Therefore, it was believed that the absorption feature at 1140 nm-1240 nm might also be affected by cellulose/lignin. The reflectance offset between WMT and WSM was caused by different degrees of multiple scattering within the canopy: WSM had a greater proportion of vascular plants than WMT; thereby enhancing the multiple scattering of NIR reflectance (Figure 2-3) (Curran, 1989; Davidson et al., 2016).
Dry vegetation types such as PD, DMT and MT exhibited a spectral shape similar to soil due to the large proportion of bare soil (Figure 2-3): the reflectance steadily increased from 720 nm to 1300 nm with no visible water absorption features. These vegetation types were characterized by patches of moss, fewer forbs and sedges, thereby having less multiple scatting within the canopy and producing lower NIR reflectance than wet vegetation types (Buchhorn et al., 2013). Previous studies have also highlighted the importance of NIR reflectance for discriminating vegetation types with different structures (e.g., sedge- dominant and moss-dominant) (Ulrich et al., 2009; Buchhorn et al., 2013; Bratsch et al., 2016; Davidson et al., 2016).
The SWIR (1400-2500 nm) wavelength region was divided into two sub-regions: SWIR-a (1400-1900 nm) and SWIR-b: (2000-2500 nm) (Figure 2-6). For green vegetation, water absorption dominated and obscured dry matter (e.g., protein, lignin and cellulous) absorption features in the SWIR region, resulting in an enhanced bell shape (Curran, 1989; Asner, 1998; Ulrich et al., 2009). However, due to the presence of dry bare soil (PD, DMT and MT) or senesced vegetation (WMT and WSM), water absorption within the SWIR
39 regions was not strong. Here, WSM and WMT showed slightly bell-shaped spectra while PD, DMT and
MT showed flat curves (Figure 2-6). Several dry matter absorption features in the SWIR were observed: 1) in the SWIR-a region, lignin absorption features around 1450 and 1700 nm were observed for WMT and
WSM; and 2) in the SWIR-b region, the lower reflectance for DMT, MT, WMT and WSM at 2100 nm and
2350 nm was a function of cellulous/lignin absorption (Curran, 1989; Ulrich et al., 2009). The absorption feature of PD and DMT around 2200 nm is typically associated with clay mineral absorption features common in these soils (Ulrich et al., 2009). Although few studies have focused on exploring this spectral region, SWIR bands can be useful for the study of Arctic tundra, especially High Arctic vegetation given the strong cellulous/lignin absorption caused by senesced vegetation (Ulrich et al., 2009; Davidson et al.,
2016).
40
Figure 2-6. Mean (± 1 std. dev.) field spectra for each vegetation type: Polar Semi-Desert (PD); Dry Mesic Tundra (DMT); Mesic Tundra (MT); Wet Mesic Tundra (WMT); Wet Sedge/Moss (WSM), as measured at Sabine Peninsula, Melville Island.
41
Figure 2-6 (Continued). Mean spectra of all vegetation types. Grey areas represent the spectral range of all WorldView-3 bands - CB (Coastal Blue), B (Blue), G (Green), Y (Yellow), R (Red), RE (Red-edge), NIR1, 2 (Near infrared 1, 2), SWIR 1 – 8 (Shortwave infrared 1 - 8) (see Table 2-2 for wavelength ranges). N represents the number of plots for which mean spectra were derived.
2.5.3. Broadband VIs
The correlations between green PVC and the WorldView-3 derived VIs (i.e., NDSI and RDSI) are presented
2 in Figure 2-7. The NDSIs that were highly correlated (R > 0.7) with PVC2D and PVC3D included the RE and yellow/red, or NIR and yellow/red band combinations. The NDSIs using the NIR and SWIR (e.g.,
NIR2 and SWIR5, 7, 8), or SWIR and SWIR (e.g., SWIR1 or 3 and SWIR2) combinations were strongly correlated with PVC3D only. For RDSI, strong correlations were observed in the SWIR and SWIR combination (e.g., SWIR4 and SWIR5, or SWIR1 and SWIR2, 5, 6, 7, 8), especially for PVC2D. Some
RDSIs using the NIR and SWIR combination (e.g., NIR2 and SWIR5, 7, 8) showed strong correlations with PVC3D only (Figure 2-7).
42
Figure 2-7. Correlations (R2) between green PVC and WorldView-3 derived VIs: NDSI (A) and RDSI (B). The x- and y-axes are the bands of WorldView-3, respectively. For each map, the lower-right diagonal area 2 2 represents the R between PVC2D and VIs; the upper-left diagonal area represents the R between PVC3D and VIs. The R2 > 0.70 values were labelled directly on each map.
2.5.4. Narrowband VIs
The correlations between green PVC and the Hyperion derived first-order derivatives are presented in
Figure 2-8. In essence, the first-order derivative dR was the reflectance difference between two neighboring spectral bands and was often used to detect/enhance some weak spectral features in the original reflectance spectra (Ben-Dor, 1997). It was found that dRs near the red absorption feature (i.e., 650.67 nm and 660.85
2 nm) were sensitive to PVC2D (R = 0.77 and 0.82 respectively) whereas the dRs near the sharp RE change
2 feature (i.e., 711.72 nm) were sensitive to PVC3D (R = 0.70). The reciprocal transformation (1/R) (or log- reciprocal transformation (log(1/R)) was often used to simulate absorbance spectra (Ben-Dor, 1997) and were found to improve the correlations. For instance, the dR within the 1400-1800 nm and 2100-2200 nm regions showed a weak/moderate correlation with PVC (R2 < 0.70). However, after reciprocal transformation, the R2 between PVC and d(1/R) exceeded 0.70.
43
Figure 2-8. Correlations (R2) between PVC and Hyperion first-order derivatives (A: correlations with PVC2D; B: correlations with PVC3D). Green dots represent the correlation between PVC and the first derivatives of reflectance (i.e., dR) while red dots represent the correlation between PVC and the first derivatives of the reciprocal of reflectance (i.e., d(1/R)). Thresholds are identified for R2 = 0.6 and R2 = 0.7.
44
Several broad regions with R2 > 0.7 were found for NDSI and RDSI (Figure 2-9). For NDSI (Figure 2-9
A), these regions were grouped into six categories according to their band combinations: (1) VIS and NIR;
(2) NIR and SWIR-a; (3) NIR and SWIR-b; (4) SWIR-a and SWIR-a; (5) SWIR-a and SWIR-b; and (6)
SWIR-b and SWIR-b. In contrast, the RDSI formulation using VIS and NIR band combinations only showed moderate correlations with PVC (Figure 2-9 B). The tables in Figure 2-9 identify the optimal band combinations (and their positions) for each region.
The strong correlation between green PVC and NDSI(VIS-NIR) is not surprising since the reflectance in the VIS region is mainly controlled by chlorophyll/carotenoids content which is directly related to the percent cover of green/senesced vegetation. In this region, the optimal band combination for estimating
PVC3D was RE and RE (i.e., 721.90 nm and 732.07 nm) whereas the optimal band combination for estimating PVC2D was red and NIR (i.e., 681.20 nm and 872.10 nm). This can be attributed to the PVC3D-
NDSI(red, NIR) relationship where the NDSI reached a maximum of 0.62 when PVC3D exceeded 100%, that is, a further increase in PVC3D could not result in an overall change in NDSI. This was consistent with the saturation reported in the NDVI-LAI/biomass/shrub cover relationship of previous Arctic studies
(Hansen, 1991; Shippert et al., 1995; Williams, 2005; Blok et al., 2011). In contrast, the NDSI(RE, RE) using two RE spectral bands (i.e., 721.90 nm and 732.07 nm) captured the different RE slopes observed for different vegetation types and modelled PVC3D well. The importance of RE bands was also highlighted by previous studies. For instance, Buchhorn et al. (2013) found that the NDSI using two RE spectral bands could produce better separations of the sedge-dominant and shrub-dominant vegetation types than NDVI.
In the NIR and SWIR-a region, the optimal NIR bands (e.g., 1174.77 nm and 1184.87 nm) occurred within the leaf water absorption region: 1140-1240 nm whereas the SWIR band (e.g., 1447.14 nm) occurred around the lignin absorption feature at 1450 nm (Figure 2-9). In the NIR and SWIR-b region, the optimal SWIR bands (e.g., 2072.65 nm, 2102.94 nm, 2123.14 nm, 2335.01 nm) occurred near the cellulous/lignin absorption features: 2100 nm and 2350 nm (Figure 2-9) (Curran, 1989; Ulrich et al., 2009). However, the importance of SWIR bands has seldom been investigated for Arctic vegetation.
45
Figure 2-9. Correlations (R2) between Hyperion-derived NDSIs/RDSIs and green PVC. For each correlation map, the x and y axes are the wavelengths of Hyperion, respectively; the lower-right diagonal area 2 2 represents the R between PVC2D and NDSI/RDSI; the upper-left area represents the R between PVC3D and NDSI/RDSI; the regions with high R2 (> 0.7) are numbered and the highest R2 position in each region is marked by a black square. The tables associated with each correlation map list the optimal band combinations and R2 values for each region.
46
When compared with other types of band combinations, the SWIR-b and SWIR-b band combinations
2 showed a higher R with green PVC (Figure 2-9). For instance, in the PVC2D-NDSI relationship, the optimal
R2 of the SWIR-b and SWIR-b combination was 0.80 whereas the optimal R2 of all other combinations was
2 0.75 or less (for PVC3D-NDSI, the optimal R was 0.84 with all other combinations less than 0.80). This improvement implies that SWIR spectral bands might be important for estimating the biophysical variables of Arctic vegetation dominated by dry senesced vegetation (Ulrich et al., 2009; Davidson et al., 2016).
Despite its importance, the use of SWIR bands has seldom been investigated in previous Arctic studies
(Riedel et al., 2005b; Huemmrich et al., 2010; Buchhorn et al., 2013). This can likely be attributed to the lack of data available at these wavelengths.
Overall, when using the same band combinations, narrowband VIs (Figure 2-9) showed higher optimal R2 values than broadband VIs (Figure 2-7). For instance, in the PVC3D-NDSI(VIS, NIR) relationship, the
2 highest R were 0.75 and 0.79 for broadband and narrowband VIs, respectively. Similarly, in the PVC3D-
RDSI(SWIR-b, SWIR-b) relationship, the R2 improved from 0.77 to 0.84, respectively. The better performance of narrowband VIs could be attributed to the finer spectral details provided by hyperspectral data. For instance, in the PVC3D-NDSI(VIS, NIR) relationship, the optimal bands were two RE bands:
721.90 nm and 732.07 nm (Figure 2-9). These two RE spectral bands could detect the variations in RE slopes (Figure 2-6). However, the spectral range of the WorldView-3 RE band was from 705 nm to 745 nm, which smoothed the different RE slopes observed in Figure 2-6. Another example was that the important cellulose/lignin absorption features around 1450 nm and 2100 nm were too narrow to be captured by the shortwave bands of WorldView-3 (Figure 2-6).
The correlations between green PVC and the VIs used in previous studies are presented in Table 2-3. Within the chlorophyll/carotenoids-sensitive group, most of the VIs having strong correlations with green PVC included the combinations of red and RE/NIR bands (e.g., NDVI, MCARI, TCARI, PSSR a-b, Three-band index 3, Carter 2-3, mNDVI). The water-sensitive VIs did not show strong correlations with green PVC.
These VIs were generally used in previous studies to estimate leaf water content, but may not be suited for
47
Arctic vegetation which had weak water absorption features due to the great proportions of dry bare soil or senesced vegetation. Dry matter-sensitive VIs showed moderate correlations with green PVC, hence they show promise for estimating Arctic vegetation biochemical content. When Arctic vegetation is dominated by dry senesced vegetation, reflectance in the SWIR region is greatly affected by the absorption of dry matter content such as cellulose, lignin and protein and the water absorption influence appears to be minimal
(Ulrich et al., 2009).
2.5.5. Conclusion
Several key findings were presented in this paper. First, PVC of the five High Arctic vegetation types examined in this study was found to be related to available moisture. Dry vegetation types (i.e., polar semi- desert, dry mesic tundra and mesic tundra) possessed greater proportions of bare soil and smaller amounts of green vegetation. In contrast, moist vegetation types (i.e., wet mesic tundra and wet sedge/moss) were dominated by senesced and green vegetation with small amounts of bare soil. Second, the spectra of the five vegetation types were influenced by PVC. Generally, these vegetation types had mixed spectral signals of bare soil, senesced vegetation and green vegetation. Dry vegetation types showed spectral features similar to clay mineral soil while moist vegetation types exhibited spectra similar to senesced vegetation.
None of the vegetation types exhibited the typical spectral features (i.e., curve) of healthy green vegetation.
Third, optimal spectral bands for estimating green PVC were identified based on two types of 2-band VIs.
Among the spectral bands prevalent in the optimal 2-band VIs, several were located at the important absorption features observed in the original spectra such as 681.20 nm (pigment absorption), 721.90 nm and 732.07 nm (along the red-edge slopes), 1174.77 nm and 1184.87 nm (water absorption), and 1447.14 nm, 1457.23 nm, 2072.65 nm and 2102.94 nm (cellulose and lignin absorption). When compared with broadband VIs, narrowband VIs exhibited slightly stronger correlations with green PVC due to the narrow and deeper absorption features sampled by hyperspectral data.
Finally, the narrowband VIs used in previous studies were tested. Most of the VIs with moderate to strong correlations (R2 = 0.6-0.7) were within the pigment-sensitive and dry matter-sensitive VIs group. Water-
48 sensitive VIs showed weak correlations with PVC; this was attributed to greater proportions of dry bare soil or senesced vegetation present in the Arctic vegetation spectra. The dry matter-sensitive VIs designed to estimate leaf lignin/cellulose contents are suggested to be given further consideration for examining
Arctic vegetation types which are often dominated by senesced vegetation (particularly for seasonal studies examining phenology). Although, this study focused on exploring hyperspectral data for estimating green
PVC, it is suggested that future studies examine the PVC of different plant functional groups by using spectral mixture analysis (Song, 2005; Somers et al., 2011) or examine biochemical variables such as chlorophyll, cellulose, and/or lignin content of Arctic vegetation by using partial least square regression
(Asner et al., 2014).
2.6. Acknowledgement
The authors would like to gratefully acknowledge financial support from ArcticNet (Project 1.7: Water
Security and Quality in a Changing Arctic), the Northern Science Training Program (NSTP) (Grant-
306001), the Natural Sciences and Engineering Research Council (NSERC) (Grant-RGPIN-2014-03822) and Queen's University (Grant-379616), Kingston, Canada. Logistical support provided by the Polar
Continental Shelf Project (PCSP) was instrumental in supporting this research. The authors would like to thank Drs. Scott Lamoureux and Melissa Lafreniere for the support of this research on the Sabine Peninsula,
Melville Island, NU. The authors would also like to thank Sarah Allux for her significant contribution to the field data collection.
2.7. References
Allux, S., Budkewitsch, P., Treitz, P., 2012. High-resolution satellite mapping of High Arctic vegetation community composition and cover on the Sabine Peninsula, Melville Island, Nunavut, in: International Polar Year 2012. Montreal, Quebec. Asner, G.P., 1998. Biophysical and biochemical sources of variability in canopy reflectance. Remote Sensing of Environment 64(3), 234–253. doi:10.1016/S0034-4257(98)00014-5 Asner, G.P., Martin, R.E., Carranza-Jiménez, L., Sinca, F., Tupayachi, R., Anderson, C.B., Martinez, P., 2014. Functional and biological diversity of foliar spectra in tree canopies throughout the Andes to Amazon region. The New phytologist. doi:10.1111/nph.12895 Atkinson, D.M., Treitz, P.M., 2012. Arctic ecological classifications derived from vegetation community and satellite spectral data. Remote Sensing 4(12), 3948–3971. doi:10.3390/rs4123948
49
Atkinson, D.M., Treitz, P.M., 2013. Modeling biophysical variables across an Arctic latitudinal gradient using high spatial resolution remote sensing data. Arctic, Antarctic, and Alpine Research 45(2), 161– 178. doi:10.1657/1938-4246-45.2.161 Barber, D.G., Lukovich, J. V, Keogak, J., Baryluk, S., Fortier, L., Henry, G.H.R., 2008. The changing climate of the Arctic. Arctic 61, 7–26. Bedini, E., 2009. Mapping lithology of the Sarfartoq carbonatite complex, southern West Greenland, using HyMap imaging spectrometer data. Remote Sensing of Environment 113(6), 1208–1219. doi:10.1016/j.rse.2009.02.007 Ben-Dor, E., 1997. The reflectance spectra of organic matter in the visible near-infrared and short wave infrared region (400–2500 nm) during a controlled decomposition process. Remote Sensing of Environment 61(1), 1–15. doi:10.1016/S0034-4257(96)00120-4 Blackburn, G.A., 1998. Quantifying chlorophylls and caroteniods at leaf and canopy scales. Remote Sensing of Environment 66(3), 273–285. doi:10.1016/S0034-4257(98)00059-5 Blok, D., Schaepman-Strub, G., Bartholomeus, H., Heijmans, M.M.P.D., Maximov, T.C., Berendse, F., 2011. The response of Arctic vegetation to the summer climate: relation between shrub cover, NDVI, surface albedo and temperature. Environmental Research Letters 6(3), 35502. doi:10.1088/1748- 9326/6/3/035502 Bratsch, S.N., Epstein, H.E., Buchhorn, M., Walker, D.A., 2016. Differentiating among four Arctic tundra plant communities at Ivotuk, Alaska using field spectroscopy. Remote Sensing 8(1). doi:10.3390/rs8010051 Buchhorn, M., Walker, D.A., Heim, B., Raynolds, M.K., Epstein, H.E., Schwieder, M., 2013. Ground- based hyperspectral characterization of Alaska tundra vegetation along environmental gradients. Remote Sensing 5(8), 3971–4005. doi:10.3390/rs5083971 Carlson, T.N., Ripley, D.A., 1997. On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sensing of Environment 62(3), 241–252. doi:10.1016/S0034-4257(97)00104-1 Carter, G.A., 1994. Ratios of leaf reflectances in narrow wavebands as indicators of plant stress. International Journal of Remote Sensing 15(3), 697–703. doi:10.1080/01431169408954109 Chappelle, E.W., Kim, M.S., McMurtrey, J.E., 1992. Ratio analysis of reflectance spectra (RARS): An algorithm for the remote estimation of the concentrations of chlorophyll A, chlorophyll B, and carotenoids in soybean leaves. Remote Sensing of Environment 39(3), 239–247. doi:10.1016/0034- 4257(92)90089-3 Chen, Z., Chen, W., Leblanc, S.G., Henry, G.H.R., 2010. Digital photograph analysis for measuring percent plant cover in the Arctic. Arctic 63(3), 315–326. Curran, P.J., 1989. Remote sensing of foliar chemistry. Remote Sensing of Environment 30(3), 271–278. doi:10.1016/0034-4257(89)90069-2 Datt, B., 1998. Remote sensing of chlorophyll A, chlorophyll B, chlorophyll A+B, and total carotenoid content in eucalyptus leaves. Remote Sensing of Environment 66(2), 111–121. doi:10.1016/S0034- 4257(98)00046-7 Datt, B., 1999. Visible/near infrared reflectance and chlorophyll content in Eucalyptus leaves. International Journal of Remote Sensing 20(14), 2741–2759. doi:10.1080/014311699211778 Daughtry, C., 2000. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance.
50
Remote Sensing of Environment 74(2), 229–239. doi:10.1016/S0034-4257(00)00113-9 Daughtry, C., Hunt, E.R., Doraiswamy, P.C., McMurtrey, J.E., 2005. Remote sensing the spatial distribution of crop residues. Agronomy Journal 97(3), 864–871. doi:10.2134/agronj2003.0291 Davidson, S., Santos, M., Sloan, V., Watts, J., Phoenix, G., Oechel, W., Zona, D., 2016. Mapping Arctic tundra vegetation communities using field spectroscopy and multispectral satellite data in North Alaska, USA. Remote Sensing 8(12), 978. doi:10.3390/rs8120978 Edlund, S.E., 1993. The distribution of plant communities on Melville Island, Arctic Canada, in: Christie, R.L., McMillan, N.J. (Eds.), The Geology of Melville Island, Arctic Canada. Geological Survey of Canada, pp. 247–255. Elmendorf, S.C., Henry, G.H.R., Hollister, R.D., Björk, R.G., Boulanger-Lapointe, N., Cooper, E.J., Cornelissen, J.H.C., Day, T.A., Dorrepaal, E., Elumeeva, T.G., Gill, M., Gould, W.A., Harte, J., Hik, D.S., Hofgaard, A., Johnson, D.R., Johnstone, J.F., Jónsdóttir, I.S., Jorgenson, J.C., Klanderud, K., Klein, J.A., Koh, S., Kudo, G., Lara, M., Lévesque, E., Magnússon, B., May, J.L., Mercado-Dı´az, J.A., Michelsen, A., Molau, U., Myers-Smith, I.H., Oberbauer, S.F., Onipchenko, V.G., Rixen, C., Martin Schmidt, N., Shaver, G.R., Spasojevic, M.J., Þórhallsdóttir, Þ.E., Tolvanen, A., Troxler, T., Tweedie, C.E., Villareal, S., Wahren, C.-H., Walker, X., Webber, P.J., Welker, J.M., Wipf, S., 2012. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2(6), 453–457. doi:10.1038/nclimate1465 Epstein, H.E., Calef, M.P., Walker, M.D., Chapin, F.S., Starfields, A.M., 2004. Detecting changes in arctic tundra plant communities in response to warming over decadal time scales. Global Change Biology 10(8), 1325–1334. doi:10.1111/j.1529-8817.2003.00810.x Evans, B.M., Walker, D.A., Benson, C.S., Nordstrand, E.A., Petersen, G.W., 1989. Spatial interrelationships between terrain , snow distribution and vegetation patterns at an Arctic foothills site in Alaska. Holarctic Ecology 12(3), 270–278. Fensholt, R., Sandholt, I., 2003. Derivation of a shortwave infrared water stress index from MODIS near- and shortwave infrared data in a semiarid environment. Remote Sensing of Environment 87(1), 111– 121. doi:10.1016/j.rse.2003.07.002 Feret, J.-B., François, C., Asner, G.P., Gitelson, A. a., Martin, R.E., Bidel, L.P.R., Ustin, S.L., le Maire, G., Jacquemoud, S., 2008. PROSPECT-4 and 5: Advances in the leaf optical properties model separating photosynthetic pigments. Remote Sensing of Environment 112(6), 3030–3043. doi:10.1016/j.rse.2008.02.012 Féret, J.-B., François, C., Gitelson, A., Asner, G.P., Barry, K.M., Panigada, C., Richardson, A.D., Jacquemoud, S., 2011. Optimizing spectral indices and chemometric analysis of leaf chemical properties using radiative transfer modeling. Remote Sensing of Environment 115(10), 2742–2750. doi:10.1016/j.rse.2011.06.016 Fraser, R.H., Olthof, I., Carrière, M., Deschamps, a, Pouliot, D., 2011. Detecting long-term changes to vegetation in northern Canada using the Landsat satellite image archive. Environmental Research Letters 6(4), 45502. doi:10.1088/1748-9326/6/4/045502 Gamon, J.A., Peñuelas, J., Field, C.B., 1992. A narrow-waveband spectral index that tracks diurnal changes in photosynthetic efficiency. Remote Sensing of Environment 41(1), 35–44. doi:10.1016/0034- 4257(92)90059-S Gao, B.C., 1996. NDWI - A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment 58(3), 257–266. doi:10.1016/S0034-
51
4257(96)00067-3 GCOS, 2006. Systematic observation requirements for satellite-based products for climate. GCOS. Gitelson, A., Gritz, Y., Merzlyak, M., 2003. Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. Journal of plant physiology 160(3), 271–282. doi:10.1078/0176-1617-00887 Gitelson, A., Keydan, G., Merzlyak, M., 2006. Three-band model for noninvasive estimation of chlorophyll, carotenoids, and anthocyanin contents in higher plant leaves. Geophysical Research Letters 33(11), L11402. doi:10.1029/2006GL026457 Gleeson, D.F., Pappalardo, R.T., Grasby, S.E., Anderson, M.S., Beauchamp, B., Castaño, R., Chien, S.A., Doggett, T., Mandrake, L., Wagstaff, K.L., 2010. Characterization of a sulfur-rich Arctic spring site and field analog to Europa using hyperspectral data. Remote sensing of Environment 114(6), 1297– 1311. doi:10.1016/j.rse.2010.01.011 Haboudane, D., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J., Dextraze, L., 2002. Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture. Remote Sensing of Environment 81(2–3), 416–426. doi:10.1016/S0034-4257(02)00018-4 Hansen, B.U., 1991. Monitoring natural vegetation in Southern Greenland using NOAA AVHRR and field measurements. Arctic 44(5). doi:10.14430/arctic1575 Hardisky, M.A., Klemas, V., Smart, R.M., 1983. The influence of soil salinity, growth form, and leaf moisture on the spectral radiance of Spartina Alterniflora canopies. Photogrammetric Engineering and Remote Sensing 49(1), 77–84. Harris, J.R., Rogge, D., Hitchcock, R., Ijewliw, O., Wright, D., 2005. Mapping lithology in Canada’s Arctic: application of hyperspectral data using the minimum noise fraction transformation and matched filtering. Canadian Journal of Earth Sciences 42(12), 2173–2193. Harrison, J.C., 1990. Geology, Melville Island and adjacent smaller islands, Canadian Arctic Archipelago, District of Franklin, Northwest Territories. doi:10.4095/131504 Hernández-Clemente, R., Navarro-Cerrillo, R.M., Suárez, L., Morales, F., Zarco-Tejada, P.J., 2011. Assessing structural effects on PRI for stress detection in conifer forests. Remote Sensing of Environment 115(9), 2360–2375. doi:10.1016/j.rse.2011.04.036 Hope, A.S., Kimball, J.S., Stow, D.A., 1993. The relationship between tussock tundra spectral reflectance properties and biomass and vegetation composition. International Journal of Remote Sensing 14(10), 1861–1874. doi:10.1080/01431169308954008 Hudson, J., Henry, G., 2010. High Arctic plant community resists 15 years of experimental warming. Journal of Ecology 98(5), 1035–1041. Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Oberbauer, S.F., Kinoshita, G., Houston, S., Kuchy, A., Hollister, R.D., Kwon, H., Mano, M., Harazono, Y., Webber, P.J., Oechel, W.C., 2010. Remote sensing of tundra gross ecosystem productivity and light use efficiency under varying temperature and moisture conditions. Remote Sensing of Environment 114(3), 481–489. doi:http://dx.doi.org/10.1016/j.rse.2009.10.003 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Campbell, P.K.E., Landis, D.R., Middleton, E.M., 2013. Arctic tundra vegetation functional types based on photosynthetic physiology and optical properties. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6, 265–275. doi:10.1109/JSTARS.2013.2253446
52
Hunter, E., Rock, B., 1989. Detection of changes in leaf water content using Near- and Middle-Infrared reflectances. Remote Sensing of Environment 30(1), 43–54. doi:10.1016/0034-4257(89)90046-1 IPCC, 2013. Climate Change 2013 - The Physical Science Basis. Cambridge University Press, Cambridge. doi:10.1017/CBO9781107415324 Jia, G.J., Epstein, H.E., Walker, D.A., 2009. Vegetation greening in the canadian arctic related to decadal warming. Journal of Environmental Monitoring 11(12), 2231. doi:10.1039/b911677j Johansen, B., Tømmervik, H., 2014. The relationship between phytomass, NDVI and vegetation communities on Svalbard. International Journal of Applied Earth Observation and Geoinformation 27, 20–30. doi:10.1016/j.jag.2013.07.001 Ju, J., Masek, J.G., 2016. The vegetation greenness trend in Canada and US Alaska from 1984–2012 Landsat data. Remote Sensing of Environment 176, 1–16. doi:10.1016/j.rse.2016.01.001 Juszak, I., Erb, A.M., Maximov, T.C., Schaepman-Strub, G., 2014. Arctic shrub effects on NDVI, summer albedo and soil shading. Remote Sensing of Environment 153, 79–89. doi:10.1016/j.rse.2014.07.021 Kushida, K., Kim, Y., Tsuyuzaki, S., Fukuda, M., 2009. Spectral vegetation indices for estimating shrub cover, green phytomass and leaf turnover in a sedge‐shrub tundra. International Journal of Remote Sensing 30(6), 1651–1658. doi:10.1080/01431160802502632 Kushida, K., Hobara, S., Tsuyuzaki, S., Kim, Y., Watanabe, M., Setiawan, Y., Harada, K., Shaver, G.R., Fukuda, M., 2015. Spectral indices for remote sensing of phytomass, deciduous shrubs, and productivity in Alaskan Arctic tundra. International Journal of Remote Sensing 36(17), 4344–4362. doi:10.1080/01431161.2015.1080878 Laidler, G.J., Treitz, P.M., Atkinson, D.M., 2008. Remote sensing of Arctic vegetation: relations between the NDVI, spatial resolution and vegetation cover on Boothia Peninsula, Nunavut. Arctic 61(1), 1–13. le Maire, G., François, C., Dufrêne, E., 2004. Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements. Remote Sensing of Environment 89(1), 1–28. doi:10.1016/j.rse.2003.09.004 Leverington, D.W., 2010. Discrimination of sedimentary lithologies using Hyperion and Landsat Thematic Mapper data: a case study at Melville Island, Canadian High Arctic. International Journal of Remote Sensing 31(1), 233–260. Liang, S., 2005. Quantitative remote sensing of land surfaces. John Wiley & Sons. Liu, B., Zhang, L., Zhang, X., Zhang, B., Tong, Q., 2009. Simulation of EO-1 Hyperion data from ALI multispectral data based on the spectral reconstruction approach. Sensors 9(4), 3090–3108. doi:10.3390/s90403090 Maire, G. le, François, C., Soudani, K., Berveiller, D., Pontailler, J.-Y., Bréda, N., Genet, H., Davi, H., Dufrêne, E., 2008. Calibration and validation of hyperspectral indices for the estimation of broadleaved forest leaf chlorophyll content, leaf mass per area, leaf area index and leaf canopy biomass. Remote Sensing of Environment 112(10), 3846–3864. doi:10.1016/j.rse.2008.06.005 Merzlyak, M.N., Gitelson, A.A., Chivkunova, O.B., Rakitin, V.Y., 1999. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiologia Plantarum 106(1), 135– 141. doi:10.1034/j.1399-3054.1999.106119.x Molau, U., Mølgaard, P., 1996. ITEX Manual (Second Edition). Copenhagen. Myers-Smith, I.H., Forbes, B.C., Wilmking, M., Hallinger, M., Lantz, T., Blok, D., Tape, K.D., Macias-
53
Fauria, M., Sass-Klaassen, U., Lévesque, E., Boudreau, S., Ropars, P., Hermanutz, L., Trant, A., Collier, L.S., Weijers, S., Rozema, J., Rayback, S.A., Schmidt, N.M., Schaepman-Strub, G., Wipf, S., Rixen, C., Ménard, C.B., Venn, S., Goetz, S., Andreu-Hayles, L., Elmendorf, S., Ravolainen, V., Welker, J., Grogan, P., Epstein, H.E., Hik, D.S., 2011. Shrub expansion in tundra ecosystems: dynamics, impacts and research priorities. Environmental Research Letters 6(4), 45509. doi:10.1088/1748-9326/6/4/045509 Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386(6626), 698–702. doi:10.1038/386698a0 Nagler, P.L., Daughtry, C.S.T., Goward, S.N., 2000. Plant Litter and Soil Reflectance. Remote Sensing of Environment 71(2), 207–215. doi:10.1016/S0034-4257(99)00082-6 Nitze, I., Grosse, G., 2016. Detection of landscape dynamics in the Arctic Lena Delta with temporally dense Landsat time-series stacks. Remote Sensing of Environment 181, 27–41. doi:10.1016/j.rse.2016.03.038 Olthof, I., Latifovic, R., Pouliot, D., 2009. Development of a circa 2000 land cover map of northern Canada at 30 m resolution from Landsat. Canadian Journal of Remote Sensing 35(2), 152–165. doi:10.5589/m09-007 Penuelas, J., Baret, F., Filella, I., 1995. Semi-empirical indices to assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31(2), 221–230. Penuelas, J., Pinol, J., Ogaya, R., Filella, I., 1997. Estimation of plant water concentration by the reflectance Water Index WI (R900/R970). International Journal of Remote Sensing 18(13), 2869–2875. doi:10.1080/014311697217396 Post, E., Forchhammer, M.C., Bret-Harte, M.S., Callaghan, T. V., Christensen, T.R., Elberling, B., Fox, A.D., Gilg, O., Hik, D.S., Hoye, T.T., Ims, R.A., Jeppesen, E., Klein, D.R., Madsen, J., McGuire, A.D., Rysgaard, S., Schindler, D.E., Stirling, I., Tamstorf, M.P., Tyler, N.J.C., van der Wal, R., Welker, J., Wookey, P.A., Schmidt, N.M., Aastrup, P., 2009. Ecological dynamics across the Arctic associated with recent climate change. Science 325(5946), 1355–1358. doi:10.1126/science.1173113 Raynolds, M.K., Walker, D.A., Epstein, H.E., Pinzon, J.E., Tucker, C.J., 2012. A new estimate of tundra- biome phytomass from trans-Arctic field data and AVHRR NDVI. Remote Sensing Letters 3(5), 403– 411. doi:10.1080/01431161.2011.609188 Riedel, S.M., Epstein, H.E., Walker, D.A., 2005a. Biotic controls over spectral reflectance of Arctic tundra vegetation. International Journal of Remote Sensing 26(11), 2391–2405. doi:10.1080/01431160512331337754 Riedel, S.M., Epstein, H.E., Walker, D.A., Richardson, D.L., Calef, M.P., Edwards, E., Moody, A., 2005b. Spatial and temporal heterogeneity of vegetation properties among four tundra plant communities at Ivotuk, Alaska, U.S.A. Arctic, Antarctic, and Alpine Research 37(1), 25–33. doi:10.1657/1523- 0430(2005)037[0025:sathov]2.0.co;2 Romero, A., Aguado, I., Yebra, M., 2012. Estimation of dry matter content in leaves using normalized indexes and PROSPECT model inversion. International Journal of Remote Sensing 33(2), 396–414. doi:10.1080/01431161.2010.532819 Rouse, J.W., Haas, J.A., Schell, J.A., Deering, D.W., 1973. Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation, in: Third ERTS Symposium. Greenbelt, Maryland, pp. 309–317. Rudy, A.C.A., Lamoureux, S.F., Treitz, P., Collingwood, A., 2013. Identifying permafrost slope
54
disturbance using multi-temporal optical satellite images and change detection techniques. Cold Regions Science and Technology 88, 37–49. doi:10.1016/j.coldregions.2012.12.008 Serrano, L., Peñuelas, J., Ustin, S.L., 2002. Remote sensing of nitrogen and lignin in Mediterranean vegetation from AVIRIS data. Remote Sensing of Environment 81(2–3), 355–364. doi:10.1016/S0034-4257(02)00011-1 Shaver, G.R., Chapin, F.S.I., 1980. Response to fertilization by various plant growth forms in an Alaskan tundra: nutrient accumulation and growth. Ecology 61(3), 662–675. Shippert, M.M., Walker, D.A., Auerbach, N.A., Lewis, B.E., 1995. Biomass and leaf-area index maps derived from SPOT images for Toolik Lake and Imnavait Creek areas, Alaska. Polar Record 31(177), 147. doi:10.1017/S0032247400013644 Sims, D.A., Gamon, J.A., 2002. Relationships between leaf pigment content and spectral reflectance across a wide range of species, leaf structures and developmental stages. Remote Sensing of Environment 81(2–3), 337–354. doi:10.1016/S0034-4257(02)00010-X Somers, B., Asner, G.P., Tits, L., Coppin, P., 2011. Endmember variability in Spectral Mixture Analysis: A review. Remote Sensing of Environment. doi:10.1016/j.rse.2011.03.003 Song, C., 2005. Spectral mixture analysis for subpixel vegetation fractions in the urban environment: How to incorporate endmember variability? Remote Sensing of Environment 95(2), 248–263. doi:10.1016/j.rse.2005.01.002 Stewart, L., Alsos, I.G., Bay, C., Breen, A.L., Brochmann, C., Boulanger-Lapointe, N., Broennimann, O., Bültmann, H., Bøcher, P.K., Damgaard, C., Daniëls, F.J.A., Ehrich, D., Eidesen, P.B., Guisan, A., Jónsdóttir, I.S., Lenoir, J., le Roux, P.C., Lévesque, E., Luoto, M., Nabe-Nielsen, J., Schönswetter, P., Tribsch, A., Tveraabak, L.U., Virtanen, R., Walker, D.A., Westergaard, K.B., Yoccoz, N.G., Svenning, J.-C., Wisz, M., Schmidt, N.M., Pellissier, L., 2016. The regional species richness and genetic diversity of Arctic vegetation reflect both past glaciations and current climate. Global Ecology and Biogeography 430–442. doi:10.1111/geb.12424 Stow, D.A., Hope, A., McGuire, D., Verbyla, D., Gamon, J., Huemmrich, F., Houston, S., Racine, C., Sturm, M., Tape, K., Hinzman, L., Yoshikawa, K., Tweedie, C., Noyle, B., Silapaswan, C., Douglas, D., Griffith, B., Jia, G., Epstein, H., Walker, D., Daeschner, S., Petersen, A., Zhou, L., Myneni, R., 2004. Remote sensing of vegetation and land-cover change in Arctic tundra ecosystems. Remote Sensing of Environment 89(3), 281–308. doi:10.1016/j.rse.2003.10.018 Tagesson, T., Mastepanov, M., Tamstorf, M.P., Eklundh, L., Schubert, P., Ekberg, A., Sigsgaard, C., Christensen, T.R., Ström, L., 2012. High-resolution satellite data reveal an increase in peak growing season gross primary production in a high-Arctic wet tundra ecosystem 1992–2008. International Journal of Applied Earth Observation and Geoinformation 18, 407–416. doi:10.1016/j.jag.2012.03.016 Tape, K.D., Hallinger, M., Welker, J.M., Ruess, R.W., 2012. Landscape heterogeneity of shrub expansion in Arctic Alaska. Ecosystems 15(5), 711–724. doi:10.1007/s10021-012-9540-4 Thenkabail, P.S., Mariotto, I., Gumma, M.K., Middleton, E.M., Landis, D.R., Huemmrich, K.F., 2013. Selection of hyperspectral narrowbands (HNBs) and composition of hyperspectral twoband vegetation indices (HVIs) for biophysical characterization and discrimination of crop types using field reflectance and Hyperion/EO-1 Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6(2), 427–439. doi:10.1109/JSTARS.2013.2252601 Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote
55
Sensing of Environment 8(2), 127–150. doi:10.1016/0034-4257(79)90013-0 Ulrich, M., Grosse, G., Chabrillat, S., Schirrmeister, L., 2009. Spectral characterization of periglacial surfaces and geomorphological units in the Arctic Lena Delta using field spectrometry and remote sensing. Remote Sensing of Environment 113(6), 1220–1235. doi:http://dx.doi.org/10.1016/j.rse.2009.02.009 van Deventer, A.P., Ward, A.D., Gowda, P.H., Lyon, J.G., 1997. Using Thematic Mapper data to identify contrasting soil plains and tillage practices. Photogrammetric Engineering and Remote Sensing 63(1), 87–93. van Wijk, M.T., Williams, M., Shaver, G.R., 2005. Tight coupling between leaf area index and foliage N content in arctic plant communities. Oecologia 142(3), 421–7. doi:10.1007/s00442-004-1733-x Walker, D.A., Raynolds, M.K., Daniëls, F.J.A., Einarsson, E., Elvebakk, A., Gould, W.A., Katenin, A.E., Kholod, S.S., Markon, C.J., Melnikov, E.S., Moskalenko, N.G., Talbot, S.S., Yurtsev, B.A., 2005. The Circumpolar Arctic vegetation map. Journal of Vegetation Science 16(3), 267–282. doi:10.1111/j.1654-1103.2005.tb02365.x Walker, M.D., Wahren, C.H., Hollister, R.D., Henry, G.H.R., Ahlquist, L.E., Alatalo, J.M., Bret-Harte, M.S., Calef, M.P., Callaghan, T. V., Carroll, A.B., Epstein, H.E., Jonsdottir, I.S., Klein, J.A., Magnusson, B., Molau, U., Oberbauer, S.F., Rewa, S.P., Robinson, C.H., Shaver, G.R., Suding, K.N., Thompson, C.C., Tolvanen, A., Totland, O., Turner, P.L., Tweedie, C.E., Webber, P.J., Wookey, P.A., 2006. Plant community responses to experimental warming across the tundra biome. Proceedings of the National Academy of Sciences 103(5), 1342–1346. doi:10.1073/pnas.0503198103 Wang, L., Hunt, E.R., Qu, J.J., Hao, X., Daughtry, C.S.T., 2011. Towards estimation of canopy foliar biomass with spectral reflectance measurements. Remote Sensing of Environment 115(3), 836–840. doi:10.1016/j.rse.2010.11.011 Williams, M., 2005. Optical instruments for measuring leaf area index in low vegetation: application in Arctic ecosystems. Ecological Applications 15(4), 1462–1470. doi:10.1890/03-5354 Williams, M., Bell, R., Spadavecchia, L., Street, L.E., van Wijk, M.T., 2008. Upscaling leaf area index in an Arctic landscape through multiscale observations. Global Change Biology 14(7), 1517–1530. doi:10.1111/j.1365-2486.2008.01590.x Wookey, P.A., Aerts, R., Bardgett, R.D., Baptist, F., Brathen, K.A., Cornelissen, J.H.C., Gough, L., Hartley, I.P., Hopkins, D.W., Lavorel, S., Shaver, G.R., 2009. Ecosystem feedbacks and cascade processes: understanding their role in the responses of Arctic and alpine ecosystems to environmental change. Global Change Biology 15(5), 1153–1172. doi:10.1111/j.1365-2486.2008.01801.x Zarco-Tejada, P., Miller, J.R., Mohammed, G.H., Noland, T.L., Sampson, P.H., 2002. Vegetation stress detection through chlorophyll + estimation and fluorescence effects on hyperspectral imagery. Journal of Environment Quality 31(5), 1433–1441. doi:10.2134/jeq2002.1433 Zarco-Tejada, P., Pushnik, J., Dobrowski, S., Ustin, S., 2003a. Steady-state chlorophyll a fluorescence detection from canopy derivative reflectance and double-peak red-edge effects. Remote Sensing of Environment 84(2), 283–294. doi:10.1016/S0034-4257(02)00113-X Zarco-Tejada, P., Rueda, C., Ustin, S., 2003b. Water content estimation in vegetation with MODIS reflectance data and model inversion methods. Remote Sensing of Environment 85(1), 109–124. doi:10.1016/S0034-4257(02)00197-9 Zeng, H., Jia, G., Epstein, H., 2011. Recent changes in phenology over the northern high latitudes detected from multi-satellite data. Environmental Research Letters 6(4), 45508.
56
Chapter 3
Modelling High Arctic Percent Vegetation Cover Using Field Digital Images and High
Resolution Satellite Data
3.1. Abstract
In this study, field digital images were used to examine the spatial-temporal patterns of Percent Vegetation
Cover (PVC) of three vegetation types (i.e., polar semi-desert, mesic tundra and wet sedge) at the Cape
Bounty Arctic Watershed Observatory (CBAWO, 7524’N, 10930’W), Melville Island, NU. In 2014, field
NDVI images were taken through the growing season for each vegetation type. An Object-Based Image
Analysis (OBIA) approach was applied to classify NDVI images to obtain the percent cover of different plant functional groups (i.e., forbs, graminoids/sedges and mosses). Green Normalized Difference
Vegetation Index (GNDVI: (RNIR–RGreen)/(RNIR+RGreen)) using green and NIR bands was calculated from field NDVI images. The analyses comparing different vegetation types confirmed: (1) the polar semi-desert exhibited the lowest PVC with a large proportion of bare soil/rock cover; (2) the mesic tundra consisted of approximately 60% mosses; and (3) the wet sedge consisted almost exclusively of graminoids and sedges.
As expected, the green PVC and GNDVI, derived from field digital images, increased during the summer growing season for each vegetation type: i.e., ~5% (0.01) for polar semi-desert; ~10% (0.04) for mesic tundra; and ~12% (0.03) for wet sedge, respectively. Furthermore, field green PVC was found to be strongly correlated with WorldView-2 derived Normalized Difference Spectral Indices (NDSI: (Rx-Ry)/(Rx+Ry)), where Rx was the reflectance of the red edge (724.1 nm) or near infrared (832.9 nm and 949.3 nm) bands;
2 Ry was the reflectance of the yellow (607.7 nm) or red (658.8 nm) bands with R ’s ranging from 0.74 to
0.81. NDSIs that incorporated the yellow band (607.7 nm) performed slightly better than the NDSIs without, indicating that this band may be more useful for investigating Arctic vegetation that often includes large proportions of senesced vegetation throughout the growing season.
57
3.2. Introduction
Percent vegetation cover (PVC) is an important biophysical variable and a key indicator of ecosystem health and productivity. Long-term records of PVC have been used to examine Arctic vegetation dynamics (e.g., shrub expansion) (Sturm et al., 2001; Elmendorf et al., 2012; Tape et al., 2012; Frost and Epstein, 2014). It is also an important factor that controls the surface energy balance; e.g., increased shrub cover results in a decrease in albedo and an increase in evapotranspiration and surface roughness, which greatly affects the surface energy balance (Chapin, 2005; Subin, 2012; Pearson et al., 2013; Juszak et al., 2014). Further, green
PVC has been found to be closely related to other important biophysical variables such as Leaf Area Index
(LAI) (Chen et al., 2009), biomass (Chen et al., 2009; Hudson and Henry, 2009; Edwards and Henry 2016), carbon flux (Huemmrich et al., 2010; Sharp et al., 2013; Sweet et al., 2015), fraction of absorbed
Photosynthetically Active Radiation (fAPAR) (Huemmrich et al., 2010; Gamon et al., 2013) and Vegetation
Indices (VIs) (Hope et al., 1993; Laidler et al., 2008; Kushida et al., 2009; Atkinson and Treitz, 2013). In many climate and land-surface models, green PVC and LAI are input for modelling the amount of photosynthetic biomass (Gutman and Ignatov, 1998; Zeng et al., 2000; Brovkin et al., 2013; Case et al.,
2014).
Among the conventional field techniques for measuring PVC, the point-frame method is probably the most widely adopted for low vegetation communities (Molau and Mølgaard, 1996; Bonham, 2013). Here, a level rectangular quadrat (e.g., 1 m x 1 m) consisting of a gridded network of intersecting strings/wires (e.g., at
10 cm = 100 grid points) is placed above the vegetated surface. At each grid, a pin is lowered until it encounters vegetation, whereby the plant species, status (senesced or green) and the canopy height are recorded. The PVC for each species is then derived from the total number of contacts with vegetation divided by the total number of grid points (i.e., 100). Compared with the visual estimate method (Barbour et al., 1980; Chen et al., 2010; Bonham, 2013), this method is considered to be objective and as a result has been adopted by many as the standard protocol for measuring percent cover of Arctic vegetation (Hope et al., 1993; Molau and Mølgaard, 1996; Wahren et al., 2005; Laidler et al., 2008; Elmendorf et al., 2012;
58
Atkinson and Treitz, 2013; Sharp et al., 2013; Sweet et al., 2015). However, this method is time-consuming and laborious (Chen et al., 2010). Further, a dense grid is required to maximize the number of species recorded, thereby requiring additional time and resources to complete (Chen et al., 2010).
Recently, true colour (i.e., RGB) digital images, collected for field plots, have proven to be useful and effective for estimating PVC in various landscapes such as grasslands (Purevdorj et al., 1998; Seefeldt and
Booth, 2006), croplands (Gitelson, 2013), semi-arid/arid environments (Laliberte et al., 2007; Laliberte,
2010; Zhang et al., 2012) and Arctic tundra (Chen et al., 2010; Beamish et al., 2016). By applying various thresholds/rules to pixels’ RGB values, green and non-green vegetation can be distinguished (Purevdorj et al., 1998; White et al., 2000; Gitelson et al., 2002; Liu et al., 2012). Classification techniques, such as k- means and maximum likelihood, have also been used to extract PVC from digital images (Vanha-Majamaa et al., 2000; Zhou and Robson, 2001; Mirik and Ansley, 2012). Since these methods consider the reflectance characteristics of cover types, they are most suited to generating broad categories, such as green (i.e., photosynthetic) versus non-green (i.e., non-photosynthetic) vegetation. When differentiating different plant species or functional groups with similar reflectance characteristics, the geometric information of vegetation can be incorporated into the classification (Luscier et al., 2006; Laliberte et al., 2007; Chen et al., 2010; Laliberte, 2010). For instance, Chen et al. (2010) segmented digital images into polygons/objects and used the length/width ratio of polygons to differentiate grasses from forbs since the length/width ratio of grasses is larger than that of forbs. Luscier et al. (2006) also used objects’ length/width ratios and RGB colours as the k-NN classification features for differentiating forbs, grasses and shrubs.
Investigating the relationship between green PVC and VIs is important for mapping green PVC at satellite scales. Generally, strong correlations have been reported in related studies: e.g., Hope et al. (1993), Laidler et al. (2008), Kushida et al. (2009) and Atkinson and Treitz (2013). Hope et al.'s (1993) results indicated that vegetation communities with different soil moisture conditions and vegetation composition may have different regression relationships with Normalized Difference Vegetation Index (NDVI). Laidler et al.
(2008) found that the correlation between NDVI and PVC derived from high-resolution satellite images
59
(i.e., 4 m IKONOS) was greater than those derived from medium-resolution images (i.e., 30 m Landsat).
This may be due to the enhanced capacity to stratify heterogeneous Arctic land surfaces into relatively homogeneous areas using high-resolution satellite data, thereby strengthening the correlation between
NDVI and biophysical variables (Hope et al., 1993; Laidler et al., 2008). Atkinson and Treitz (2013) examined the linear PVC-NDVI relationship along latitudinal gradients. The statistical tests indicated that there was no significant difference between the regression slopes for two disparate sites, while the intercepts were significantly different, i.e., the regression equations for the two sites were parallel. This parallelism suggests that NDVI can be used as an indicator for monitoring the changes in PVC since the response of
NDVI to PVC appears to be equal at different sites; however, some caution needs to be exercised in applying the regression equation of one site to estimate the PVC of another site due to the different intercepts observed (Atkinson and Treitz, 2013). Other VIs such as the Soil Adjusted Vegetation Index
(SAVI) (Huete, 1988) and Modified SAVI (MSAVI) (Qi et al., 1994) were developed to minimize the influence of soil brightness on NDVI for low PVC. These have also been tested in the Arctic but do not appear to improve the PVC-VI correlation when compared to NDVI (Laidler et al., 2008; Kushida et al.,
2009). However, Beamish et al. (2016) used the Greenness Excess Index (GEI) to successfully track phenology and track changes in biomass at the plot scale for their study on the central east coast of
Ellesmere Island, Nunavut.
Although image-based classification is an effective way to derive PVC, to the author’s knowledge, it has not been widely tested in Arctic environments, particularly in the High Arctic. Further, due to the time- consuming nature of the point-frame method for conducting landscape or watershed-scale studies, most studies focus on measuring the PVC at the peak of growing season rather than examining seasonal vegetation change. Hence, the goal of this research is to model PVC for a High Arctic study site using field digital images collected throughout the growing season and high spatial resolution satellite images. The specific objectives are to: (1) characterize the seasonal changes in PVC and NDVI for a study site in the
60
Canadian High Arctic using field near infrared-green-blue (NGB) digital images, and (2) ‘scale up’ green
PVC to satellite scales using the PVC derived from field NGB images and high-resolution satellite images.
3.3. Study Area
The Cape Bounty Arctic Watershed Observatory (CBAWO) (75.4 °N, 109.5 °E) is located on the southern coast of Melville Island, Nunavut, Canada and covers approximately 150 km2 (Figure 3-1). The terrain of this area is undulating with gradual slopes (above sea level elevation: 5 m to 125 m). The study site is underlain by sedimentary rocks of the Devonian Weatherall and Hecla Bay formations (Hodgson et al.,
1984). Climatically, the growing season is short (i.e., late June to early August) (Environment and Climate
Change Canada) and cool with low stratus cloud and fog being common during the growing season
(Atkinson and Treitz, 2013). The mean monthly July temperature in 2014 was 3.7 °C and the total precipitation for July was 70.2 mm.
Previous studies at the CBAWO have classified vegetation into three broad vegetation types based on topographic and soil moisture conditions: Polar semi-Desert (PD), Mesic Tundra (MT) and Wet Sedge (WS)
(Edlund, 1993; Atkinson and Treitz, 2013). The PD generally occurs on drained uplands and consists of willow (e.g., Salix arctica Pall.) and mountain avens (Dryas intergrifolia Vahl.), forbs (e.g., Papaver cornwallisense D. Love, Saxifraga oppositifolia L., Saxifraga hyperborean R. Br., Saxifraga tricuspidata
Rothb.), rushes (e.g., Luzula confusa Lindeberg., Luzula confusa Spreng.), grasses (e.g., Dupontia fisheri
R. Br., Phippsia algida Sol. R. Br., Poa abbreviata R. Br.), mosses and lichens with large patches of bare soil and rock. The MT tends to occur on intermediate moisture sites and exhibits a thick moss layer with patches of exposed bare soil and rock. The WS possess a thick mat of grasses (e.g., Alopecurus alpinus,
Phippsia algida), sedges (e.g., Carex aquatilis var. stans, Eriophorum triste, E. scheuchzeri) and mosses, occurring in low-lying areas alongside waterways or downslope of semi-permanent snow fields.
61
Figure 3-1. Cape Bounty Arctic Watershed Observatory (CBAWO) (panel a-b) and three sampling sites (300 m x 300 m) (panel c) on Melville Island, Nunavut, Canada. In panel b, the WorldView-2 image of Cape Bounty, acquired on July 12th, 2012, is displayed as a colour infrared composite with the near infrared, red and green channels display as red, green and blue (RGB) respectively. Panel c shows detailed satellite images (left) and classification maps (right) for the three sites.
62
3.4. Methods
3.4.1. Field Sampling Design
In July 2014, a field campaign was undertaken at three sites at the CBAWO to examine vegetation properties (i.e., PVC and VIs). Each sampling site was 300 m×300 m (i.e., identified as squares in Figure
3-1, panel b). Within each site, several 6 m × 6 m plots (designed to correspond to 3 by 3 pixel windows of
WorldView-2 data), were located using a stratified random sampling procedure described as follows.
First, the Worldview-2 image data, acquired on July 12th, 2012, were classified into five vegetation communities for each sampling site using the k-means algorithm (Figure 3-1, panel c). These five vegetation classes, defined along a moisture gradient from dry to wet, were PD, MT1, MT2, WS1 and WS2. Unlike previous studies (Atkinson and Treitz, 2013), the MT and WS were divided into two sub-classes according to their soil moisture in order to better characterize vegetation properties along topographic and soil moisture gradients. Given the dependence on available moisture, the NDVI of MT1 represents a midrange between that of PD and MT2 (NDVI: MT2 3-1, panel c). At Site 2, the PD is located on the northeastern uplands. Compared with other sites, the PD of Site 2 has a greater spatial variability based on the coefficient of variation of NDVI. Therefore, more samples were required to characterize the spatial heterogeneity of this site. At Site 3, part of the PD is located on the northeastern uplands. A long and narrow active layer detachment occurs in Site 3 and has similar spectral responses as the PD. 63 Second, based on the classification, 6 m × 6 m plots were randomly selected within each vegetation type for each sampling site (Figure 3-1, panel b). The number of plots of each vegetation type was computed as: 2 N (Z /(r)) , where Z 1.96 (p-value: 0.05 ); and are the standard deviation and mean values respectively of WorldView-2 NDVI for each vegetation type within each site; and r is the relative error (10%) (Jensen, 2005). The sample size can be derived at a 95% confidence level within a relative error of 10% of the mean NDVI for each vegetation type of each sampling site. The geographical coordinates of each plot’s center were registered using a handheld Trimble GeoExplorer II GPS with a horizontal accuracy of approximately 1.0 m (after averaging). Table 3-1. Statistical information of three sampling sites at Cape Bounty (Green Normalized Difference Vegetation Index: GNDVI) Site 1 (74.91125 °N, 109.5567 °E) Class Percentage (% GNDVI # of plots cover) Mean Std. (36) Polar Semi-Desert 13.1 0.154 0.037 22 Mesic Tundra 1 26.9 0.268 0.027 4 Mesic Tundra 2 23.5 0.362 0.027 3 Wet Sedge 1 20.7 0.343 0.028 3 Wet Sedge 2 15.8 0.382 0.039 4 Site 2 (74.9115 °N, 109.5800 °E) Class Percentage (%) GNDVI # of plots Mean Std. (66) Polar Semi-Desert 16.8 0.098 0.034 47 Mesic Tundra 1 16.8 0.208 0.029 8 Mesic Tundra 2 27.1 0.294 0.022 4 Wet Sedge 1 26.7 0.360 0.024 3 Wet Sedge 2 12.6 0.421 0.040 4 Site 3 (74.9146 °N, 109.5941 °E) Class Percentage (%) GNDVI # of plots Mean Std. (38) Polar Semi-Desert 13.5 0.150 0.037 24 Mesic Tundra 1 16.9 0.244 0.027 5 Mesic Tundra 2 29.6 0.298 0.028 3 Wet Sedge 1 26.1 0.345 0.022 3 Wet Sedge 2 13.9 0.412 0.027 3 When taking measurements in the field, each plot was evenly divided into quadrants (i.e., 3 m × 3 m) and several 0.5 m × 0.25 m quadrats were located randomly within each quadrant without overlapping (Figure 64 3-2). According to the spatial heterogeneity of different land covers, different numbers of quadrats were photographed in each plot using a Canon digital near-infrared camera (Maxmax Inc., Carlstadt, NJ). For PD, at least 8 quadrats were sampled; for MT, 4-8 quadrats; and for WS, 2-4 quadrats. The PVC and NDVI values derived from the images of each of the quadrats were averaged to obtain plot-level estimates. Field images were collected at the beginning (i.e., Day188-191) and end of the field season (Day 204-209). The specifications of the Canon digital camera are summarized in Table 3-2. This camera captures digital images in the near-infrared, green and blue (NGB) regions of the spectrum, which were used to calculate ‘green’ NDVI (i.e., GNDVI: (RNIR–RGreen)/(RNIR+RGreen)). When collecting images, the camera was positioned ~1.5 m above the ground (and levelled) to capture images of a 0.5 m × 0.25 m quadrat. The manual exposure program was adopted and the “exposure time” parameter was adjusted to ensure zero exposure bias (Sakamoto et al., 2012). Digital images were recorded in both raw (i.e., CR2) and JPEG formats. All the images were acquired between 10:00 and 14:30 local time, with the top of the camera oriented towards the sun, to minimize the influence of solar zenith and azimuth angle’s respectively. In addition to the NGB images, normal colour (RGB) digital images were also collected (Figure 3-2, panel d). Table 3-2. Specifications and program settings of the Canon digital near-infrared camera. Camera Model Canon EOS REBEL T4i Channels: Spectral bands R: Near-infrared; G: Green; B: Blue Image Width, Height 5184×3456 pixels Format JPEG & CR2 image Shooting parameters Exposure program Manual F-stop 2.2 Exposure time* (in seconds) 1/320 – 1/4000 ISO speed 100 Camera height ~1.5 m * Exposure time was adjusted to ensure zero exposure bias. 65 Figure 3-2. Sampling scheme for collecting digital images at the CBAWO. The WorldView-2 imagery (panel a) of the 300 m x 300 m site was classified into five land covers (panel b). Within each 6 m x 6 m plot, several 0.5 m x 0.25 m quadrats were randomly located (panel c). Colour infrared (NGB) and normal colour (RGB) images were collected for each of these quadrats (panel d). 3.4.2. Percent Vegetation Cover In this study, the estimation of PVC was based on image classification. Similar to previous studies (Atkinson and Treitz, 2013), different plant species were grouped according to their plant functional groups: graminoids/sedges, forbs, mosses, and willows. Non-vegetated surfaces were classified into two classes: bare soil/rock and shadows. OBIA was applied to classify the digital images collected in the field. Unlike traditional pixel-based classifiers, which simply rely on spectral information, OBIA considers both the spectral and geometric characteristics of objects/surfaces. This approach is particularly useful for growth form identification. For instance, in pixel-based classification, it is hard to distinguish green graminoids from forbs due to their spectral similarity (i.e., chlorophyll has similar spectral responses in the visible and 66 near infrared for green vegetation). However, in OBIA, these two plant functional groups can be distinguished according to their geometric characteristics (e.g., the length/width ratio): graminoids have long, narrow leaves while forbs often exhibit round, broad leaves (i.e., the length/width ratio for graminoids is much larger than that for forbs). Before image segmentation, the NGB images were transformed into hue-saturation-value (HSV) images by using the “RGB to HSV (USGS Munsell)” tool in ENVI. Instead of using RGB, the Munsell transformation characterizes colours based on Hue (0 and 360 = blue, 120 = green and 240 = red), saturation (purity of colours) and value (brightness of colours). HSV reduced the inter-correlations of RGB bands and made the setting of colour-related thresholds much easier (Chen et al., 2010). The HSV images were imported into eCognition (Trimble, Sunnyvale, CA, USA) and the image pixels within quadrats were grouped into image objects by using the multi-resolution segmentation algorithm (Benz et al., 2004). There are three user- defined segmentation parameters in the multi-resolution segmentation algorithm: hscale, hshape and hcompactness. The parameter hscale determines the maximum allowable heterogeneity of image objects, i.e., larger scale values generally yield larger objects (Evans and Costa, 2013). The hshape parameter (range: 0-1) describes the extent to which shape affects the segmentation compared to colour. Meanwhile, the hcompactness parameter (range: 0-1) characterizes the degree of smoothness of object boarders. Generally, it is difficult to standardize these values because of the differences in brightness/complexity of images. However, for images taken on the same day within the same plot, these values were similar. In this study, the values of these parameters were determined in a trial and error fashion: hscale = 50, 100 and 150 (dependent on the complexity of images), hshape = 0.1 and hcompactness = 0.2. These settings proved successful for segmenting individual leaves into separate polygons. After segmentation, object samples of each plant functional group were manually selected. Based on these object samples, histograms of the object features, including HSV values and the length/width ratio, were created for each plant functional group using the “sample editor” tool. Using these histograms, different thresholds were assigned to the length/width ratio and spectral values of objects in order to classify image 67 objects (Figure 3-3). First, a threshold was assigned to hue to separate two broad categories: green vegetation and non-green vegetation. Since green vegetation was red on the NGB images, the hue threshold was approximately 240. Second, within the green vegetation category, a threshold (range: 3-5, depending on segmentation parameters) was assigned to the length/width ratio to separate graminoids/sedges from other vegetation. Compared to graminoid/sedge, the length/width ratios of forbs, willows and mosses were much smaller (usually less than 3). Third, objects defining willows and forbs were manually identified quickly due to their small coverages. The same procedure was applied to the non-green vegetation category to separate graminoids/sedges and mosses. The PVC of each plant functional group was calculated as the pixel proportion of each class. Green (i.e., including green graminoid/sedge, moss, willow and forb) and non-green PVCs were also calculated. Furthermore, the estimates of PVC derived from image classification and the traditional point-frame method were compared. Since there were no direct point-frame measurements collected in the field, a simulation was conducted by superimposing a grid (20×10) on field digital images (Figure 3-3). This results in an effective sampling grid of 2.5×2.5 cm. The cover class located at each grid point was visually identified and used to estimate PVC (by functional group) for the quadrat. 68 Figure 3-3. Examples of the image classification and simulated point-frame method for three vegetation types (i.e., PD (polar semi-desert), MT (mesic tundra) and WS (wet sedge)). The first column shows the field GNDVI digital images of three vegetation communities and the second column shows the classification results. The white circles superposed on the image were used to simulate the point-frame method: i.e., the vegetation class within each circle was visually identified. 69 3.4.3. Vegetation Indices An eight-band (Coastal: 427 nm, Blue: 478.3 nm, Green: 545.8 nm, Yellow: 607.7 nm, Red: 658.8 nm, Red Edge: 724.1 nm, NIR1: 832.9 nm and NIR2: 949.3 nm) WorldView-2 satellite image was acquired for the study area on July 10, 2014. The preprocessing of these WorldView-2 data included: 1) geometric rectification (using 10 ground control points); and 2) atmospheric correction (applying the FLAASH atmospheric correction package in ENVI 5.0). The input parameters for the atmospheric correction included: 1) the mid-latitude winter atmospheric model; 2) the maritime aerosol model; and 3) scene visibility of 40 km. Although NDVI is the most commonly applied VI, additional VIs were examined given the additional channels available from WorldView-2. Here, a range of 2-band VIs were tested for predicting green PVC. 2-band Normalized Difference Spectral Indices (NDSI) were calculated by systematically combining all WorldView-2 bands as follows: NDSIx_y=(Rx-Ry)/(Rx-Ry), where Rx and Ry are the WorldView-2 reflectance values at bands x and y. For field NDVI images, Rx is the near-infrared band reflectance and Ry is the green reflectance. 3.4.4. PVC-VI Regression Models Linear regression models were developed to examine the performance of NDSI in predicting green PVC. Green PVC (the dependent variable) was transformed (using a square root transformation) to satisfy the assumption of normality, whereas NDSI was the independent variable (Atkinson and Treitz, 2013). Due to the limited number of sample plots, a ten-fold cross validation was performed to evaluate the performance of the regression models. The coefficient of determination (R2) and root mean squared error (RMSE) of the ten-fold cross validations were used to assess model accuracy. 70 3.5. Results and Discussion 3.5.1. Percent Vegetation Cover Estimation – A Comparison of Methods Visually, it was found that the OBIA achieved high accuracy for classifying different species (Figure 3-3). On the segmented NGB images, the objects of graminoids/sedges were long and narrow; the forbs or willows were typically round; and mosses were generally irregular. The contrasts in object shapes together with the colour information was helpful for discriminating plant functional groups/species. A comparison of estimates of PVC derived from OBIA and simulated point-frame methods for each plant functional group is presented in Figure 3-4. PVC values derived from both methods exhibit good correlation (R2 > 0.7), indicating that the simulated point-frame method can serve as an alternative to OBIA to derive PVC quickly. For forbs and willows, PVC was underestimated by OBIA and over-estimated by the simulated point-frame method. The PVC of green moss was also overestimated by OBIA, especially for the WS. Although the point-frame method was simulated and compared with OBIA in this study, it should be noted that the simulated method still differs from the point-frame method. For instance, the point-frame method characterizes the three-dimensional (3D) canopy structure: the number of layers of the canopy and the height of each layer above the ground (characterized by the number and height of leaves that come into contact with the pin at each grid). The 3D PVC also plays an important role in remote sensing. For instance, NIR reflectance can be greatly enhanced by the multiple-scattering process within a canopy (Jensen, 2005). In comparison, the simulated point-frame method only describes how much vegetation cover can be ‘seen’ by the sensor and cannot provide any structural information of the canopy (i.e., a two-dimensional representation of the canopy, however low that canopy may be). Therefore, when comparing the PVC derived by traditional point-frame method with that derived by image classification, the PD class is expected to have a small PVC difference given a canopy close to the ground surface. Meanwhile, the MT and WS communities should have a larger difference given their more complex and multi-layered canopies. Therefore, the point-frame method has a greater capacity for characterizing the PVC of multi-layered 71 canopies (e.g., the moss understory and multi-layer vascular plants) (Laidler et al., 2008; Atkinson and Treitz, 2013; Edwards and Henry 2016). The importance of the canopy structure of Arctic vegetation has been emphasized in many studies. For instance, increases in NDVI have been attributed to increases in height and cover of Arctic vegetation (Verbyla, 2008; Forbes et al., 2010; Juszak et al., 2014). Together with PVC, changes in canopy height can result in changes in surface albedo and roughness, thereby affecting the surface energy budget (Walker et al., 2006). Some studies have reported that vegetation volume, i.e., PVC × canopy height, had stronger correlations with NDVI or other biophysical variables (e.g., biomass, leaf area index) than PVC (Chen et al., 2009; Gregory, 2011; Greaves et al., 2015). Figure 3-4. Comparison of PVC between the OBIA and simulated point-frame methods. Polar semi-desert: Squares; Mesic tundra: Circles; Wet sedge: Triangles. The x- and y-axes are the OBIA derived PVC and simulated point-frame PVC, respectively. 72 3.5.2. Temporal Patterns of Percent Cover A comparison of PVC between the early (July 9-11, 2014) and late (July 25-28, 2014) growing season for each vegetation type is presented in Figure 3-5. The three vegetation types exhibited different spatial patterns of PVC. The PD had the lowest PVC (i.e., NGV + GV in Figure 3-5, typically 20% - 30%), with a large proportion of bare soil/rock cover. For the MT, PVC was approximately 80%, dominated by moss (~ 60%). The WS had the highest PVC (~ 100%), dominated by graminoids/sedges (~ 80%). Not surprisingly, these vegetation communities had significantly different PVC and NDVI values: the WS with the largest PVC and NDVI values; the MT with moderate values; and the PD the lowest values. In terms of temporal patterns, an increase in green PVC (i.e., GV in Figure 3-5) was observed during the month of July in each vegetation type. For the WS, the increase in green PVC was mainly attributed to the increase in green graminoids/sedges. In contrast, the increase in green moss was considered negligible. For the MT, however, the contribution of green moss to green PVC change was larger than graminoids/sedges (10% versus 1%). Compared with other communities, the increase in PVC for the green PD was less pronounced (smaller than 5%). The absolute differences in average NDVI between late and early season for each vegetation type were 0.03 (WS), 0.04 (MT) and 0.01 (PD) respectively, indicative of the ‘greening’ of vegetation observed in the field over the growing season. A paired t-test conducted on PVC and NDVI indicated that the temporal increase was only significant for the WS and MT communities. 73 Figure 3-5. PVC and NDVI comparisons between early (July 9-11, 2014) and late (July 25-28, 2014) growing season for three vegetation communities. Plant functional groups on the X-axis are: forbs (FB), green graminoids/sedges (GG), senesced graminoids/sedges (SG), willow (WL), green moss (GM), senesced moss (SM), bare soil (BS), shadow (SH), green vegetation (GV=FB+GG+WL+GM), non-green vegetation (NGV=SG+SM) and non-vegetation (NV=BS+SH). 3.5.3. PVC-NDSI relationships The relationships between PVC and various NDSIs are summarized in Table 3-3. In general, the NDSIs that are strongly correlated with PVC (R2 > 0.7) incorporated the red edge (or near infrared) + visible band combinations. The field-measured GNDVI exhibited stronger correlations with PVC than the WorldView- 2 derived NDSIs, likely due to reduced atmospheric effects (i.e., path radiance). The strong correlations further confirm that VIs can be used as an indicator of PVC changes in Arctic environments (Buus-Hinkler et al., 2006; Hudson & Henry, 2009; Leblanc et al., 2014; Olthof & Fraser, 2007; Puma et al., 2007; Raynolds et al., 2006; Tremblay et al., 2012; Edwards and Henry 2016; Beamish et al. 2016). Interestingly, within each NDSI group, the NDSI with the yellow band, which was designed to detect the “yellowness” of vegetation, showed a slightly higher R2 value than the NDSI without the yellow band 74 (Table 3-3). It is difficult to conclude that the yellow band may provide additional information for modelling percent vegetation cover. However, given that senesced (i.e., brown/yellow) vegetation is prevalent in these vegetation types (Figure 3-5), the spectral information embedded in the yellow channel may be very useful for distinguishing vegetation types with different proportions of senesced vegetation. The yellow band, as reported in previous studies at low- to mid-latitudes, was found to improve the accuracy of tree species identification (Gong et al., 1997; Pu, 2009; Cho et al., 2012), land-cover classification (Marchisio et al., 2010; Ghosh and Joshi, 2014; Lane et al., 2014) and biochemical content estimation (Zengeya et al., 2013). Meanwhile, Nelson et al. (2013) found that some lichen (e.g., Usnea spp.) species have a much higher reflectance in the yellow portion of the electromagnetic spectrum than other vegetation, and therefore can be used to detect lichen abundance using remote sensing data. At the very least, this observation requires further study in the context of Arctic vegetation. Table 3-3. Linear regression results for models of PVC using NDSI ((Rx-Ry)/(Rx+Ry)). Only the band combinations with R2 > 0.7 are displayed. The models are grouped according to band x. Band x Band y Equation R2 RMSE Field NGB Image-Based Models NIR Green 19.61x -0.12 0.81 7.98 Green 17.26x + 1.50 0.71 11.02 Red Edge Yellow 19.58x + 1.58 0.76 9.83 Red 21.55 x + 1.18 0.75 10.02 Green 12.59 x + 1.18 0.74 10.32 WorldView-2 Based Models Yellow 13.41 x + 1.26 0.77 9.99 NIR1 Red 14.37 x + 0.96 0.76 9.71 Red Edge 37.18 x + 0.83 0.74 10.31 Green 11.67 x + 0.96 0.74 10.15 NIR2 Yellow 12.21 x + 1.05 0.75 9.82 Red 12.86 x + 0.87 0.74 10.14 The PVC map of the CBAWO derived from the WorldView-2 data is presented in Figure 3-6. Due to the limited tasking window for a short summer season and the sky conditions encountered in 2014, the WorldView-2 satellite data were only acquired during the early growing season. It is clear that the majority 75 of the CBAWO has a low PVC (~0% -10%; PD). The MT and WS communities have medium PVC (~20% - 50%), distributed along topographic and moisture gradients. In this study, the PVC of each plant functional group was not estimated separately. It is expected that the PVC of each plant functional group will demonstrate weaker correlations with VIs than total PVC (Atkinson and Treitz, 2013). However, examining PVC according to functional groups may be important. For instance, moss has a distinct photosynthetic capacity compared to vascular plants (Martin and Adamson, 2001; Huemmrich et al., 2010). Hence, for modelling various functional properties (i.e., CO2 flux), it would be advantageous to examine vascular and non-vascular plants separately (Huemmrich et al., 2010). To estimate the PVC for different plant functional groups, algorithms, such as regression trees (Olthof and Fraser, 2007; Selkowitz, 2010), random forest (Gessner et al., 2013) and spectral mixture analysis (Somers et al., 2011; Song, 2005; Thorp et al., 2013) are suggested. 76 2 2 Figure 3-6. The PVC map of CBAWO. The regression model: PVC=(13.41×NDSINIR1_Yellow+1.26) (R = 0.77, RMSE = 9.99) was applied to the WorldView-2 derived NDSINIR1_Yellow image. 3.6. Conclusion PVC plays an important role in classifying Arctic vegetation and monitoring change in Arctic vegetation condition. The conventional point-frame method for measuring PVC is time-consuming, costly, and limited in terms of generating a large sample size. Here, classification of field digital images was adopted to investigate spatial and temporal patterns of PVC for a study site in the Canadian High Arctic. OBIA was performed to classify NGB data collected in the field. Based on these classifications, the PVC of different plant functional groups such as forbs, graminoids and mosses was calculated. The OBIA method was compared with the simulated point-frame method which measured the top of the canopy to estimate PVC (by superimposing artificial grids on the images). Results indicate that the PVC derived from the two 77 methods are highly correlated. Compared to the traditional point-frame method, the image-based method possesses many advantages. For instance, it does not require much time in the field. Digital photographs can be taken quickly in the field, making it possible to take more PVC samples during the short growing season in the Arctic. On average, the image classification (i.e., the OBIA in this study) required 15 to 45 minutes (the PD requires the least time while WS requires the most time due to their different vegetation complexities). However, in the field, the traditional point-frame method provides detailed information about the species composition, abundance and canopy height but requires 1 to 1.5 hours to collect these measurements for conversion to PVC. The results demonstrate that different vegetation communities exhibited different spatial and temporal patterns of PVC. For instance, and as expected, the PD exhibited the lowest PVC, with a large proportion of bare soil/rock cover. Meanwhile, the MT was dominated by moss (~60%), and WS was ~100% covered by graminoids/sedges (green and senesced). Each vegetation type exhibited slight to modest increases in total green PVC and NDVI over the summer growing season. The increase in PVC for the WS was mainly attributed to the increase in green graminoids/sedges, whereas for MT, the increase was mainly due to green moss. The increase in PVC for the PD was least pronounced due to the lower abundance of green vegetation. In order to study the impact of climate warming on Arctic vegetation types, it will be important to determine the effects on the various functional groups in order to predict the overall impact on productivity, biodiversity and carbon fluxes at watershed scales (Edwards and Henry, 2016). To upscale the field image derived PVC to satellite scales, linear regression models were built between PVC and the WorldView-2 image derived NDSIs. In general, PVC was strongly correlated with these NDSIs (R2 = 0.74-0.81). The NDSIs that incorporated the yellow band performed slightly better than the NDSIs without the yellow band. This may result from the relatively high proportion of senesced (brown/yellow) vegetation in each vegetation type at this study site; therefore, the yellow band may be more useful for modelling biophysical properties of High Arctic vegetation. This is an important consideration that should be tested further in these environments, particularly now that these spectral 78 channels are available from high spatial resolution remote sensing satellites. Finally, the spectral bands used in this study were limited to the visible-NIR wavelength range. It is expected that the shortwave spectral data of WorldView-3, which are sensitive to water content, will further improve the performance in estimating PVC for these vegetation types in the Canadian Arctic. 3.7. Acknowledgement The authors would like to gratefully acknowledge financial support from ArcticNet, the Northern Science Training Program (NSTP), the Natural Sciences and Engineering Research Council (NSERC) and Queen’s University, Kingston, Canada. Logistical support provided by the Polar Continental Shelf Project (PCSP) was instrumental in supporting this research. The authors would also like to thank Drs. Scott Lamoureux and Melissa Lafreniere for the support of this research at the Cape Bounty Arctic Watershed Observatory (CBAWO). 3.8. References Atkinson, D.M., Treitz, P.M., 2013. Modeling biophysical variables across an Arctic latitudinal gradient using high spatial resolution remote sensing data. Arctic, Antarctic, and Alpine Research 45(2), 161– 178. doi:10.1657/1938-4246-45.2.161 Barbour, M.G., Burk, J.H., Pitts, W.D., 1980. Terrestrial plant ecology. Benjamin/Cummings., Menlo Park, California. Beamish, A.L., Nijland, W., Edwards, M., Coops, N.C., Henry, G.H.R., 2016. Phenology and vegetation change measurements from true colour digital photography in High Arctic tundra. Arctic Science 2(2), 33–49. doi:10.1139/as-2014-0003 Benz, U.C., Hofmann, P., Willhauck, G., Lingenfelder, I., Heynen, M., 2004. Multi-resolution, object- oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry and Remote Sensing 58(3–4), 239–258. doi:10.1016/j.isprsjprs.2003.10.002 Buchhorn, M., Walker, D.A., Heim, B., Raynolds, M.K., Epstein, H.E., Schwieder, M., 2013. Ground- based hyperspectral characterization of Alaska tundra vegetation along environmental gradients. Remote Sensing 5(8), 3971–4005. doi:10.3390/rs5083971 Buus-Hinkler, J., Hansen, B.U., Tamstorf, M.P., Pedersen, S.B., 2006. Snow-vegetation relations in a High Arctic ecosystem: Inter-annual variability inferred from new monitoring and modeling concepts. Remote Sensing of Environment 105(3), 237–247. doi:10.1016/j.rse.2006.06.016 Case, J.L., LaFontaine, F.J., Bell, J.R., Jedlovec, G.J., Kumar, S. V., Peters-Lidard, C.D., 2014. A real-time MODIS vegetation product for land surface and numerical weather prediction models. IEEE Transactions on Geoscience and Remote Sensing 52(3), 1772–1786. doi:10.1109/TGRS.2013.2255059 Chapin, F.S., 2005. Role of land-surface changes in Arctic summer warming. Science 310(5748), 657–660. 79 doi:10.1126/science.1117368 Chen, W., Wang, J., Li, J., Zhang, Y., Zhang, Y., Zhou, F., Koehler, K., Leblanc, S., Fraser, R., Olthof, I., 2009. Relating biomass and leaf area index to non-destructive measurements in order to monitor changes in Arctic vegetation. Arctic 62(3), 281–294. Chen, Z., Chen, W., Leblanc, S.G., Henry, G.H.R., 2010. Digital photograph analysis for measuring percent plant cover in the Arctic. Arctic 63(3), 315–326. Cho, M.A., Mathieu, R., Asner, G.P., Naidoo, L., van Aardt, J., Ramoelo, A., Debba, P., Wessels, K., Main, R., Smit, I.P.J., Erasmus, B., 2012. Mapping tree species composition in South African savannas using an integrated airborne spectral and LiDAR system. Remote Sensing of Environment 125, 214–226. doi:10.1016/j.rse.2012.07.010 Edlund, S.E., 1993. The distribution of plant communities on Melville Island, Arctic Canada, in: Christie, R.L., McMillan, N.J. (Eds.), The Geology of Melville Island, Arctic Canada. Geological Survey of Canada, pp. 247–255. Edwards, M., Henry, G.H.R., 2016. The effects of long-term experimental warming on the structure of three High Arctic plant communities, Journal of Vegetation Science 27, 904-913. Elmendorf, S.C., Henry, G.H.R., Hollister, R.D., Björk, R.G., Boulanger-Lapointe, N., Cooper, E.J., Cornelissen, J.H.C., Day, T.A., Dorrepaal, E., Elumeeva, T.G., Gill, M., Gould, W.A., Harte, J., Hik, D.S., Hofgaard, A., Johnson, D.R., Johnstone, J.F., Jónsdóttir, I.S., Jorgenson, J.C., Klanderud, K., Klein, J.A., Koh, S., Kudo, G., Lara, M., Lévesque, E., Magnússon, B., May, J.L., Mercado-Dı´az, J.A., Michelsen, A., Molau, U., Myers-Smith, I.H., Oberbauer, S.F., Onipchenko, V.G., Rixen, C., Martin Schmidt, N., Shaver, G.R., Spasojevic, M.J., Þórhallsdóttir, Þ.E., Tolvanen, A., Troxler, T., Tweedie, C.E., Villareal, S., Wahren, C.-H., Walker, X., Webber, P.J., Welker, J.M., Wipf, S., 2012. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2(6), 453–457. doi:10.1038/nclimate1465 Evans, T.L., Costa, M., 2013. Landcover classification of the Lower Nhecolândia subregion of the Brazilian Pantanal Wetlands using ALOS/PALSAR, RADARSAT-2 and ENVISAT/ASAR imagery. Remote Sensing of Environment 128, 118–137. doi:10.1016/j.rse.2012.09.022 Forbes, B.C., Fauria, M.M., Zetterberg, P., 2010. Russian Arctic warming and “greening” are closely tracked by tundra shrub willows. Global Change Biology 16(5), 1542–1554. doi:10.1111/j.1365- 2486.2009.02047.x Frost, G. V, Epstein, H.E., 2014. Tall shrub and tree expansion in Siberian tundra ecotones since the 1960s. Global change biology 20(4), 1264–1277. doi:10.1111/gcb.12406 Gamon, J.A., Huemmrich, K.F., Stone, R.S., Tweedie, C.E., 2013. Spatial and temporal variation in primary productivity ( NDVI ) of coastal Alaskan tundra : Decreased vegetation growth following earlier snowmelt. Remote Sensing of Environment 129, 144–153. doi:10.1016/j.rse.2012.10.030 Gessner, U., Machwitz, M., Conrad, C., Dech, S., 2013. Estimating the fractional cover of growth forms and bare surface in savannas. A multi-resolution approach based on regression tree ensembles. Remote Sensing of Environment 129, 90–102. doi:10.1016/j.rse.2012.10.026 Ghosh, A., Joshi, P.K., 2014. A comparison of selected classification algorithms for mapping bamboo patches in lower Gangetic plains using very high resolution WorldView 2 imagery. International Journal of Applied Earth Observation and Geoinformation 26, 298–311. doi:10.1016/j.jag.2013.08.011 Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms for remote estimation of vegetation fraction. Remote Sensing of Environment 80(1), 76–87. doi:10.1016/S0034- 4257(01)00289-9 80 Gitelson, A. a., 2013. Remote estimation of crop fractional vegetation cover: the use of noise equivalent as an indicator of performance of vegetation indices. International Journal of Remote Sensing 34(17), 6054–6066. doi:10.1080/01431161.2013.793868 Gong, P., Pu, R., Yu, B., 1997. Conifer species recognition: An exploratory analysis of in situ hyperspectral data. Remote Sensing of Environment. doi:10.1016/S0034-4257(97)00094-1 Greaves, H.E., Vierling, L.A., Eitel, J.U.H., Boelman, N.T., Magney, T.S., Prager, C.M., Griffin, K.L., 2015. Estimating aboveground biomass and leaf area of low-stature Arctic shrubs with terrestrial LiDAR. Remote Sensing of Environment 164, 26–35. doi:10.1016/j.rse.2015.02.023 Gregory, F.M., 2011. Biophysical remote sensing and terrestrial CO2 exchange at Cape Bounty, Melville Island. Department of Geography. Queen’s University, Kingston, ON. Gutman, G., Ignatov, A., 1998. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. International Journal of Remote Sensing 19(8), 1533– 1543. doi:10.1080/014311698215333 Hodgson, D.A., Vincent, J.-S., Fyles, J.G., 1984. Quaternary geology of central Melville Island, Northwest Territories. doi:10.4095/119784 Hope, A.S., Kimball, J.S., Stow, D.A., 1993. The relationship between tussock tundra spectral reflectance properties and biomass and vegetation composition. International Journal of Remote Sensing 14(10), 1861–1874. doi:10.1080/01431169308954008 Hudson, J.M.G., Henry, G.H.R., 2009. Increased plant biomass in a high arctic heath community from 1981 to 2008. Ecology 90, 2657–2663. doi:10.1890/09-0102.1 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Oberbauer, S.F., Kinoshita, G., Houston, S., Kuchy, A., Hollister, R.D., Kwon, H., Mano, M., Harazono, Y., Webber, P.J., Oechel, W.C., 2010. Remote sensing of tundra gross ecosystem productivity and light use efficiency under varying temperature and moisture conditions. Remote Sensing of Environment 114(3), 481–489. doi:http://dx.doi.org/10.1016/j.rse.2009.10.003 Huete, A.., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment 25, 295–309. doi:10.1016/0034-4257(88)90106-X Jensen, J.R., 2005. Introductory digital image processing: a remote sensing perspective (Third Edition). Prentice Hall, Upper Saddle River, N.J. Juszak, I., Erb, A.M., Maximov, T.C., Schaepman-Strub, G., 2014. Arctic shrub effects on NDVI, summer albedo and soil shading. Remote Sensing of Environment 153, 79–89. doi:10.1016/j.rse.2014.07.021 Kushida, K., Kim, Y., Tsuyuzaki, S., Fukuda, M., 2009. Spectral vegetation indices for estimating shrub cover, green phytomass and leaf turnover in a sedge‐shrub tundra. International Journal of Remote Sensing 30(6), 1651–1658. doi:10.1080/01431160802502632 Laidler, G.J., Treitz, P.M., Atkinson, D.M., 2008. Remote sensing of Arctic vegetation: relations between the NDVI, spatial resolution and vegetation cover on Boothia Peninsula, Nunavut. Arctic 61(1), 1–13. la Puma, I.P., Philippi, T.E., Oberbauer, S.F., 2007. Relating NDVI to ecosystem CO2 exchange patterns in response to season length and soil warming manipulations in arctic Alaska. Remote Sensing of Environment 109(2), 225–236. doi:10.1016/j.rse.2007.01.001 Laliberte, A.S., Rango, A., Herrick, J.E., Fredrickson, E.L., Burkett, L., 2007. An object-based image analysis approach for determining fractional cover of senescent and green vegetation with digital plot photography. Journal of Arid Environments 69, 1–14. doi:10.1016/j.jaridenv.2006.08.016 Lane, C., Liu, H., Autrey, B., Anenkhonov, O., Chepinoga, V., Wu, Q., 2014. Improved wetland classification using eight-band high resolution satellite imagery and a hybrid approach. Remote 81 Sensing 6(12), 12187–12216. doi:10.3390/rs61212187 Leblanc, S.G., Chen, W., Maloley, M., 2014. NDVI digital camera for monitoring arctic vegetation, in: IEEE International Geoscience & Remote Sensing Symposium and 35th Canadian Symposium on Remote Sensing. Quebec City. Liu, Y., Mu, X., Wang, H., Yan, G., 2012. A novel method for extracting green fractional vegetation cover from digital images. Journal of Vegetation Science 23(3), 406–418. doi:10.1111/j.1654- 1103.2011.01373.x Luscier, J.D., Thompson, W.L., Wilson, J.M., Gorham, B.E., Dragut, L.D., 2006. Using digital photographs and object-based image analysis to estimate percent ground cover in vegetation plots. Frontiers in Ecology and the Environment 4, 408–413. doi:http://dx.doi.org/10.1890/1540- 9295(2006)4[408:UDPAOI]2.0.CO;2 Marchisio, G., Pacifici, F., Padwick, C., 2010. On the relative predictive value of the new spectral bands in the WorldWiew-2 sensor, in: 2010 IEEE International Geoscience and Remote Sensing Symposium. IEEE, pp. 2723–2726. doi:10.1109/IGARSS.2010.5649771 Martin, C.E., Adamson, V.J., 2001. Photosynthetic capacity of mosses relative to vascular plants. journal of Bryology 23(January), 319–323. Mirik, M., Ansley, R.J., 2012. Comparison of ground-measured and image-classified mesquite (Prosopis glandulosa) canopy cover. Rangeland Ecology & Management. doi:10.2111/REM-D-11-00073.1 Molau, U., Mølgaard, P., 1996. ITEX Manual (Second Edition). Copenhagen. Nelson, P.R., Roland, C., Macander, M.J., McCune, B., 2013. Detecting continuous lichen abundance for mapping winter caribou forage at landscape spatial scales. Remote Sensing of Environment 137, 43– 54. doi:10.1016/j.rse.2013.05.026 Olthof, I., Fraser, R.H., 2007. Mapping northern land cover fractions using Landsat ETM+. Remote Sensing of Environment 107(3), 496–509. doi:http://dx.doi.org/10.1016/j.rse.2006.10.009 Pearson, R.G., Phillips, S.J., Loranty, M.M., Beck, P.S. a., Damoulas, T., Knight, S.J., Goetz, S.J., 2013. Shifts in Arctic vegetation and associated feedbacks under climate change. Nature Climate Change 3(7), 673–677. doi:10.1038/nclimate1858 Pu, R., 2009. Broadleaf species recognition with in situ hyperspectral data. International Journal of Remote Sensing. doi:10.1080/01431160802555820 Purevdorj, T., Tateishi, R., Ishiyama, T., Honda, Y., 1998. Relationships between percent vegetation cover and vegetation indices. International Journal of Remote Sensing 19(February 2015), 3519–3535. doi:10.1080/014311698213795 Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified soil adjusted vegetation index. Remote Sensing of Environment 48(2), 119–126. doi:10.1016/0034-4257(94)90134-1 Raynolds, M.K., Walker, D.A., Maier, H.A., 2006. NDVI patterns and phytomass distribution in the circumpolar Arctic. Remote Sensing of Environment 102(3–4), 271–281. doi:http://dx.doi.org/10.1016/j.rse.2006.02.016 Sakamoto, T., Gitelson, A. a., Nguy-Robertson, A.L., Arkebauer, T.J., Wardlow, B.D., Suyker, A.E., Verma, S.B., Shibayama, M., 2012. An alternative method using digital cameras for continuous monitoring of crop status. Agricultural and Forest Meteorology 154–155, 113–126. doi:10.1016/j.agrformet.2011.10.014 Selkowitz, D.J., 2010. A comparison of multi-spectral, multi-angular, and multi-temporal remote sensing datasets for fractional shrub canopy mapping in Arctic Alaska. Remote Sensing of Environment 114(7), 1338–1352. doi:http://dx.doi.org/10.1016/j.rse.2010.01.012 82 Sharp, E.D., Sullivan, P.F., Steltzer, H., Csank, A.Z., Welker, J.M., 2013. Complex carbon cycle responses to multi-level warming and supplemental summer rain in the high Arctic. Global Change Biology 19(6), 1780–1792. doi:10.1111/gcb.12149 Somers, B., Asner, G.P., Tits, L., Coppin, P., 2011. Endmember variability in Spectral Mixture Analysis: A review. Remote Sensing of Environment. doi:10.1016/j.rse.2011.03.003 Song, C., 2005. Spectral mixture analysis for subpixel vegetation fractions in the urban environment: How to incorporate endmember variability? Remote Sensing of Environment 95(2), 248–263. doi:10.1016/j.rse.2005.01.002 Stow, D.A., Hope, A., McGuire, D., Verbyla, D., Gamon, J., Huemmrich, F., Houston, S., Racine, C., Sturm, M., Tape, K., Hinzman, L., Yoshikawa, K., Tweedie, C., Noyle, B., Silapaswan, C., Douglas, D., Griffith, B., Jia, G., Epstein, H., Walker, D., Daeschner, S., Petersen, A., Zhou, L., Myneni, R., 2004. Remote sensing of vegetation and land-cover change in Arctic tundra ecosystems. Remote Sensing of Environment 89(3), 281–308. doi:10.1016/j.rse.2003.10.018 Sturm, M., Racine, C., Tape, K., 2001. Climate change: increasing shrub abundance in the Arctic. Nature 411(6837), 546–7. doi:10.1038/35079180 Sweet, S.K., Griffin, K.L., Steltzer, H., Gough, L., Boelman, N.T., 2014. Greater deciduous shrub abundance extends tundra peak season and increases modeled net CO2 uptake. Global Change Biology n/a-n/a. doi:10.1111/gcb.12852 Tape, K.D., Hallinger, M., Welker, J.M., Ruess, R.W., 2012. Landscape heterogeneity of shrub expansion in Arctic Alaska. Ecosystems 15(5), 711–724. doi:10.1007/s10021-012-9540-4 Thorp, K.R., French, A.N., Rango, A., 2013. Effect of image spatial and spectral characteristics on mapping semi-arid rangeland vegetation using multiple endmember spectral mixture analysis (MESMA). Remote Sensing of Environment 132, 120–130. doi:10.1016/j.rse.2013.01.008 Tremblay, B., Lévesque, E., Boudreau, S., 2012. Recent expansion of erect shrubs in the Low Arctic: evidence from Eastern Nunavik. Environmental Research Letters 7(3), 35501. doi:10.1088/1748- 9326/7/3/035501 Vanha-Majamaa, I., Salemaa, M., Tuominen, S., Mikkola, K., 2000. Digitized photographs in vegetation analysis - a comparison of cover estimates. Applied Vegetation Science 3, 89–94. doi:10.2307/1478922 Verbyla, D., 2008. The greening and browning of Alaska based on 1982–2003 satellite data. Global Ecology and Biogeography 17(4), 547–555. doi:10.1111/j.1466-8238.2008.00396.x Wahren, C.H.A., Walker, M.D., Bret-Harte, M.S., 2005. Vegetation responses in Alaskan arctic tundra after 8 years of a summer warming and winter snow manipulation experiment. Global Change Biology 11, 537–552. doi:10.1111/j.1365-2486.2005.00927.x Walker, D.A., Kuss, P., Epstein, H.E., Kade, A.N., Vonlanthen, C.M., Raynolds, M.K., Daniëls, F.J.A., 2011. Vegetation of zonal patterned-ground ecosystems along the North America Arctic bioclimate gradient. Applied Vegetation Science 14(4), 440–463. doi:10.1111/j.1654-109X.2011.01149.x Walker, D.A., Raynolds, M.K., Daniëls, F.J.A., Einarsson, E., Elvebakk, A., Gould, W.A., Katenin, A.E., Kholod, S.S., Markon, C.J., Melnikov, E.S., Moskalenko, N.G., Talbot, S.S., Yurtsev, B.A., 2005. The Circumpolar Arctic vegetation map. Journal of Vegetation Science 16(3), 267–282. doi:10.1111/j.1654-1103.2005.tb02365.x Walker, M.D., Wahren, C.H., Hollister, R.D., Henry, G.H.R., Ahlquist, L.E., Alatalo, J.M., Bret-Harte, M.S., Calef, M.P., Callaghan, T. V., Carroll, A.B., Epstein, H.E., Jonsdottir, I.S., Klein, J.A., Magnusson, B., Molau, U., Oberbauer, S.F., Rewa, S.P., Robinson, C.H., Shaver, G.R., Suding, K.N., Thompson, C.C., Tolvanen, A., Totland, O., Turner, P.L., Tweedie, C.E., Webber, P.J., Wookey, P.A., 83 2006. Plant community responses to experimental warming across the tundra biome. Proceedings of the National Academy of Sciences 103(5), 1342–1346. doi:10.1073/pnas.0503198103 White, M.A., Asner, G.P., Nemani, R.R., Privette, J.L., Running, S.W., 2000. Measuring fractional cover and leaf area index in arid ecosystems: Digital camera, radiation transmittance, and laser altimetry methods. Remote Sensing of Environment 74, 45–57. doi:10.1016/S0034-4257(00)00119-X Zeng, X., Dickinson, R.E., Walker, A., Shaikh, M., DeFries, R.S., Qi, J., 2000. Derivation and evaluation of global 1-km fractional vegetation cover data for land modeling. Journal of Applied Meteorology 39(6), 826–839. doi:10.1175/1520-0450(2000)039<0826:DAEOGK>2.0.CO;2 Zengeya, F.M., Mutanga, O., Murwira, A., 2013. Linking remotely sensed forage quality estimates from WorldView-2 multispectral data with cattle distribution in a savanna landscape. International Journal of Applied Earth Observation and Geoinformation 21, 513–524. doi:10.1016/j.jag.2012.07.008 Zhang, X., Liao, C., Li, J., Sun, Q., 2012. Fractional vegetation cover estimation in arid and semi-arid environments using: HJ-1 satellite hyperspectral data. International Journal of Applied Earth Observation and Geoinformation 21, 506–512. doi:10.1016/j.jag.2012.07.003 Zhou, Q., Robson, M., 2001. Automated rangeland vegetation cover and density estimation using ground digital images and a spectral-contextual classifier. International Journal of Remote Sensing 22(17), 3457–3470. doi:10.1080/01431160010004504 84 Chapter 4 Remote Sensing of Percent Vegetation Cover and fAPAR on Baffin Island, Nunavut, Canada 4.1. Abstract Percent vegetation cover (PVC) and the fraction of absorbed photosynthetically active radiation (fAPAR) are important functional variables for assessing Arctic vegetation density and vigor. In this study, field measures of PVC, fAPAR and normalized difference vegetation index (NDVI) were collected from July 5 to August 8, 2015 (i.e., summer season) along a moisture gradient in the Apex River Watershed (ARW) (63°45’N, 68°30’W), Baffin Island, Nunavut, Canada. Two field methods for estimating PVC (i.e., the point-frame and image classification methods) were examined and it was determined that the image classification approach provided a suitable alternative to the point-frame method, more specifically for detecting changes in Arctic PVC. Meanwhile, fAPAR was measured based on the method presented by Tagesson et al. (2012): fAPAR (1 PARreflectected / PARincomin g ) PVC . The spatial and temporal patterns of PVC and fAPAR in the context of NDVI derived from remote sensing data were examined. For this site, vegetation types exhibited contrasting spatial and temporal patterns of PVC as a result of differing moisture regimes: (1) vegetation types with saturated soils (e.g., sedge dominated) exhibited a continuous increase in PVC throughout the growing season; (2) mesic vegetation types with moderate soil moisture (e.g., mosses dominated with a mixture of sedges and shrubs) exhibited an increase in PVC from July to early August followed by a decline in mid-August; and (3) semi-desert vegetation types that tended to dry early in the growing season (e.g., dominated by prostrate dwarf shrubs, herb tundra and large areas of bare soil and rock) exhibited little variation in PVC. Field measures of PVC and fAPAR demonstrated strong relationships to field NDVI data and vegetation indices (VIs) derived from 2m-resolution WorldView-2 data, thereby providing further evidence that VIs are suited for modelling PVC/fAPAR of Arctic vegetation. 85 4.2. Introduction Arctic temperatures are increasing at almost twice the global average (IPCC, 2013) and as a result, are having widespread and diverse impacts on Arctic vegetation (Myneni et al., 1997; Jia et al., 2003, 2009; Epstein et al., 2004; Walker et al., 2006; Elmendorf et al., 2012; Stewart et al., 2016). For instance, Elmendorf et al. (2012) collected plot-level data from a large number of re-measurement studies between 1980 and 2010. Based on 158 plant communities spread across 46 Arctic locations, the authors found that the height and abundance of vascular plants (e.g., shrubs, graminoids and forbs) increased over time, whereas the cover of mosses and lichens decreased. Tape et al. (2006) compared aerial photographs collected over the past 50 years, and found substantial shrub expansion in northern Alaska. Increasingly, satellite spectral vegetation indices (VIs) have demonstrated a significant ‘greening’ trend in many Arctic/sub-Arctic regions (Myneni et al., 1997; Stow et al., 2004; Raynolds et al., 2006; Emmerton et al., 2015; Gonsamo and Chen, 2016; Ju and Masek, 2016; Zhang et al., 2017). For instance, Myneni et al. (1997) reported a 10% increase in the normalized difference vegetation index (NDVI) for the northern hemisphere from 1981 to 1991, with the greatest change occurring between 45N and 70N. Jia et al. (2003, 2009) reported increases in NDVI of 16.9% for northern Alaska (1981 to 2001) and 11.2% for the Canadian Arctic (1982 to 2003). Given the range of spatial resolutions available from different platforms and sensors, satellite remote sensing offers an efficient and systematic method for assessing and monitoring Arctic vegetation across scales (i.e., local to continental). Due to the short growing season and the remote nature of the Arctic, field surveys can only be conducted at a finite number of points/locations, rendering extrapolation over large areas difficult (Stow et al., 2004). However, by integrating field measurements with satellite remote sensing data, it is possible to model biophysical variables over large spatial scales and monitor changes in these variables over time. There have been many studies reporting on the strong statistical relations between satellite derived spectral indices and Arctic biophysical variables, including leaf area index (LAI) (Williams, 2005), biomass (Hope et al., 1993; Raynolds et al., 2006; Epstein et al., 2012; Johansen and Tømmervik, 86 2014; Sweet et al., 2015), percent vegetation cover (PVC) (Laidler et al., 2008; Blok et al., 2011; Atkinson and Treitz, 2013; Liu and Treitz, 2016; Liu et al., 2017), fraction of absorbed photosynthetically active radiation (fAPAR) (Huemmrich et al., 2010, 2013; Tagesson et al., 2012) and carbon dioxide (CO2) flux (la Puma et al., 2007; Emmerton et al., 2015). Among these variables, PVC and fAPAR are important from a vegetation functional perspective. PVC describes the percentage of ground area covered by vegetation. It has been widely used for the long-term monitoring of Arctic vegetation (Sturm et al., 2001; Walker et al., 2006; Elmendorf et al., 2012; Tape et al., 2012; Frost and Epstein, 2014). The point-frame method is likely the most commonly applied method for measuring Arctic PVC (Molau and Mølgaard, 1996; Bonham, 2013) albeit time-consuming and laborious (Chen et al., 2010). Recent studies have demonstrated that field digital images offer great promise for monitoring Arctic vegetation efficiently (Chen et al., 2010; Beamish et al., 2016; Edwars and Henry, 2016; Fraser et al., 2016; Liu and Treitz, 2016). For instance, Anderson et al. (2016) and Beamish et al. (2016) found that the greenness index derived from field normal color images was effective at characterizing the seasonal phenology of Arctic vegetation. Other studies have demonstrated that the image classification method is capable of estimating PVC of different plant functional groups/species (Luscier et al., 2006; Chen et al., 2010; Liu and Treitz, 2016). However, this method has seldom been compared to the traditional point- frame method for estimating PVC for Arctic vegetation types (Chen et al., 2010; Beamish et al., 2016; Edwards and Henry 2016). Alternatively, fAPAR is an important variable for describing the exchange of energy and mass fluxes (e.g., CO2) between the land surface and atmosphere (Myneni and Williams, 1994). Although fAPAR has a strong correlation with NDVI in many environments (Goward and Huemmrich, 1992; Myneni and Williams, 1994; Fensholt et al., 2004), very few studies have collected fAPAR or examined the fAPAR-VI relationship for Arctic tundra (Huemmrich et al., 2010, 2013; Tagesson et al., 2012; Juszak et al., 2017). Arctic vegetation types are quite different from more southern vegetation types in that they typically exhibit low stature (often less than 10 cm) and include substantial portions of exposed soil, senesced vegetation and a range of non- 87 vascular plants (e.g., mosses, cyanobacteria and lichens). These characteristics may influence the field measurements of fAPAR and its remote sensing estimation. For instance, the low stature of Arctic vegetation often hinders measurement of below-canopy PAR. In some studies, the fAPAR measurements were based on the PAR measured above the canopy (Huemmrich et al., 2010, 2013; Tagesson et al., 2012). Further, large proportions of senesced vegetation present in Arctic vegetation types have been found to strongly influence their spectral properties; thereby impacting the ability to model biophysical variables using remote sensing data (Buchhorn et al., 2013; Davidson et al., 2016; Liu et al., 2017). MODerate resolution Imaging Spectroradiometer (MODIS) LAI/fAPAR is an operational global-scale product. Its primary retrieval algorithm is based on a look-up-table (LUT) which links LAI/fAPAR with MODIS bidirectional reflectance factors at red and NIR bands (Knyazikhin et al., 1998; Myneni et al., 2002). A 3D canopy radiative transfer model describing the interaction of photons with vegetation canopies is used to generate the LUT (Knyazikhin et al., 1998). When the primary retrieval algorithm fails, an empirical relationship between LAI/fAPAR and MODIS NDVI (also called the backup algorithm), is applied to estimate LAI/fAPAR, both of which are biome dependent. The global vegetation land cover is divided into six biomes (i.e., grasses and cereal crops; shrubs; broadleaf crops; savannah; broadleaf forests; and needle forests) and the primary retrieval/backup algorithms are developed for each biome. The MODIS LAI/fAPAR product has been validated for various land cover types such as evergreen needle forests (Jensen et al., 2011; Serbin et al., 2013), evergreen deciduous forests (Wang et al., 2004; Aragao et al., 2005; De Kauwe et al., 2011), semi-arid savannah (Privette et al., 2002; Tian et al., 2002; Fensholt et al., 2004; Hill et al., 2006), and grasslands (Pasolli et al., 2015). However, this product has not been validated for vegetation types in the Arctic given the lack of field studies to provide suitable field measurements. Hence, the overall goal of this study was to explore the spatial and temporal patterns of PVC and fAPAR and their relationships to VIs at a study site in the Canadian Arctic. The specific objectives were to: (1) assess the performance of the image classification method for estimating PVC; (2) investigate the seasonal patterns of PVC/fAPAR for different vegetation types; (3) explore the PVC/fAPAR-VI relationship at plot 88 and fine-resolution satellite scales; and (4) validate MODIS LAI/fAPAR products for Arctic tundra using field-measured PVC/fAPAR and high spatial resolution satellite data. 4.3. Study Area This study, as part of a larger integrated watershed study examining water quality, was conducted in the Apex River Watershed (ARW) (63°45’ N, 68°30’ W; ~58 km2) located on the southern coast of Baffin Island, Nunavut, Canada (Figure 4-1). The terrain is undulating with elevations ranging from 50-300 m above sea level, with the highest elevations occurring in the northern reaches of the study area (Hodgson, 2003). Geologically, the ARW is underlain by early Holocene and late Wisconsinan till on the uplands with coarse-grained glaciofluvial deposits and Holocene marine sediments in localized lowlands (Hodgson, 2003). During the summer season, the active layer depth is approximately 1.5-2.5 m. The growing season normally starts in late June and ends in late August. In 2015, the mean monthly July air temperature was 8.2 °C and the total July precipitation was 82.3 mm (Environment Canada, 2015). Based on the 1971-2000 climate normal station data at Mould Bay, Northwest Territories, the mean July air temperature and total July precipitation were 4.0 °C and 13.5 mm respectively (Environment Canada, 2015). Vegetation types for this area can be grouped into five broad categories (after Walker et al., 2005): i.e., one dry vegetation type (P1 - prostrate dwarf shrub, herb tundra); three mesic vegetation types (P2 - prostrate/hemi-prostrate dwarf shrub tundra; G2 - graminoids, prostrate dwarf shrub; G3 - non-tussock sedge, dwarf shrub, moss tundra) and one wet vegetation type (W1 - sedge/grass, moss wetland) (Figure 4- 1). P1 is generally located on uplands with dry soil conditions and often exhibits continuous areas of exposed bare soil/rock with patches of vegetation such as moss, lichen and Arctic willow (Olthof et al., 2009). P2, G2 and G3 tend to occur in areas of intermediate slopes with moderate soil moisture regimes. The dominant species of P2 are low shrubs (e.g., Cassiope tetragona (Arctic white heather), Rhododendron tomenrosum subsp. decumbens (Labrador tea) and Vaccinium vitis-idaea subsp. minor (mountain cranberry)) and tend to occur over granite-gneiss bedrock. G2 occurs in moist to dry tundra areas with a large diversity of plants including sedge, rushes and some prostrate dwarf shrubs (e.g., Salix herbacea 89 (Snow-bed willow) and Salix reticulata (Net-vein willow)). G3 exhibits a continuous mat of vegetation with a mixture of graminoids and dwarf shrubs (e.g., Actostaphylos alpina (Black bearberry), Actostaphylos rubra (Red bearberry), Vaccinium uliginosum (Blueberry), and Empetrum nigrum subsp. hermaphroditum (Crowberry)). Finally, W1 represents the wettest vegetation type and tends to occur in areas with saturated soils, sometimes with standing surface water. W1 exhibits continuous areas of sedges/grasses with a thick layer of moss understory. Figure 4-1. The Apex River Watershed (ARW) study area (panel A) and five sample plant communities at the ARW: P1 - prostrate dwarf shrub, herb tundra; P2 - prostrate/hemi-prostrate, dwarf shrub tundra; G2 - graminoids, prostrate dwarf shrub; G3 - non-tussock sedge, dwarf shrub, moss tundra; and W1 - sedge/grass, moss wetlands. 4.4. Methods 4.4.1. Field Sampling Design Based on a 2m-resolution WorldView-2 NDVI image acquired in 2014, seven sample sites (i.e., the green dots) were chosen to provide a range of highly vegetated (i.e., with high NDVI) to non-vegetated areas (i.e., 90 with low NDVI), where field data (i.e., PVC, fAPAR and NDVI) were collected. Each site consisted of a pair of 100 m transects oriented across different slope angles to ensure that sufficient data were collected for vegetation types across different environmental gradients (Figure 4-2 B). Five plots were located along each transect at an interval of 20 m (Figure 4-2 B). The plot size was 6 m × 6 m to correspond to a 3 × 3 pixel window of WorldView-2 multispectral data. Each plot was then divided into four 3 × 3 m quadrants (i.e., northeast, southeast, southwest and northwest). Field measurements were collected within a 0.6 m × 0.6 m quadrat placed at the center of each quadrant and then averaged for the plot (Figure 4-2 C). Transect sampling was conducted from July 5 to August 8, 2015. The first-iteration of sampling occurred between July 5 to July 20 and the second iteration from July 21 to August 8. For each iteration, one transect was selected from each site for sampling. Since transect plots were chosen to provide a wide range of fAPAR and PVC over the study area, they were sampled only once during the growing season. In addition to transect plots, six permanent sample plots (i.e., the red dot in Figure 4-1) with different vegetation cover types were sampled repeatedly every one or two weeks to obtain seasonal changes in vegetation (i.e., phenology). 91 Figure 4-2. Field sampling design. (A) WorldView-2 false color image of the ARW (R: band 7 (near infrared); G: band 5 (red); B: band 3 (green); sample sites are identified by green dots; the red dot represents the locations of six permanent plots); (B) Field transect sampling design: each sampling site consists of paired field transects; and five 6 m × 6 m plots are located along each transect at intervals of 20 m; (C) Within each plot, field measurements were made for four 0.6 m × 0.6 m quadrats located at the center of quadrants. 4.4.2. Percent Vegetation Cover The point-frame method was used to measure field PVC (once for transect plots and more than once for permanent plots) (Molau and Mølgaard, 1996). In the field, a 0.6 m × 0.6 m quadrat with equidistantly spaced grids (grid interval = 10 cm, 36 points per quadrat) was placed at the center of each quadrant of a plot (Figure 4-2 C). At each grid point, a needle with 1 cm gradations was lowered, and each time the needle 92 encountered vegetation, the plant species, plant status (i.e., green or senesced) and above ground heights of each contact were recorded. For multilayered vegetation canopies, multiple contacts may occur at a single grid point. The major plant species are presented in Table 4-1. Two- and three-dimensional representations of PVC (i.e., PVC2D and PVC3D) were derived from the point- frame data (Buchhorn et al., 2013; Liu et al., 2017). PVC2D was based on the number of the top layer contacts divided by the number of frame grids (i.e., 25), whereas PVC3D was based on the number of contacts of all layers divided by the number of frame grids. According to their definitions, PVC2D has a value ranging from 0-100% while PVC3D can be related to the LAI of a specific plant species (with the exception of mosses and lichens since they do not have ‘leaves’) and can have a value that exceeds 100% in the case of multilayer canopies (Molau and Mølgaard, 1996; Atkinson and Treitz, 2013; Buchhorn et al., 2013; Liu et al., 2017). Since one of the objectives of this study was to compare the PVC derived from the point-frame and image classification methods, field normal color images were classified to estimate PVC (Chen et al., 2010; Liu and Treitz, 2016; Beamish et al., 2016). Digital images were first segmented into many objects (i.e., polygons) by using a multi-resolution segmentation algorithm and three user-defined segmentation parameters in eCognition (Benz et al., 2004). Values for hscale (which describes the maximum allowable heterogeneity of polygons), hshape (which characterizes the extent to which shape information affects the segmentation compared to color information), and hcompactness (which estimates the smoothness of object boarders), were determined in a trial and error manner: hscale = 50, 100 and 150 (depending on the complexity of vegetation), hshape = 0.1 and hcompactness = 0.2. Then, samples of green and non-green vegetation objects were manually selected as training data for a nearest neighbor (NN) classification. Finally, PVC was calculated based on this binary classification. 93 Table 4-1 Major species of each plant functional type. Plant Functional Type Major Species Graminoids/Sedges Carex lugens, Carex scirpoidea Mosses Sphagnum spp. Forbs/Herbs Oxytropis maydelliana (Yellow Oxytrope), Pedicularis flammea ( Flame Lousewort), Pendicularis hirsuta ( Hairy Lousewort), Braya glabella Richardson ssp. glabella (Smooth Northern-Rockcress) Shrubs Arctous alpina, Arctous rubra (Alpine Bearberry, Red Bearberry), Vaccinium uliginosum (Blueberry), Empetrum nigrum subsp. hermaphroditum (Crowberry), Vaccinium vitis-idaea subsp. minor (Mountain Craneberry), Dryas integrifolia (Mountain Avens), Salix herbacea (Snow-bed Willow), Salix arctica (Arctic Willow), Salix reticulata (Net-vein Willow), Rhododendron tomenrosum subsp. decumbens (Labrador Tea), Cassiope tetragona (White Heather), Dispensia lapponica (Lapland Pincushion), Rhododendron lapponicum (Lanpland Rosebay) Lichens Flavocetravia cucullata 4.4.3. The Fraction of Absorbed Photosynthetically Active Radiation Due to the low stature of High Arctic vegetation, it is impossible to place the PAR sensor below the canopy to measure the transmitted and reflected PAR. Therefore, PAR was measured above the canopy only. For transect plots, a 0.5 m MQ-306 PARmeter (6 quantum sensors, Apogee Instrument Inc., Logan, USA) was horizontally held ~30 cm above the canopy with the sensors looking upward to measure incoming PAR (i.e., PARIncoming) and then downward to measure reflected PAR (i.e., PARReflected) (Huemmrich et al., 2010; Schubert et al., 2010; Tagesson et al., 2012). For the six permanent plots, incoming and reflected PAR was measured with hemispherical QSO-S PAR sensors (Decagon Device Inc., Pullman, USA) mounted approximately 1 m above the ground. Incident and reflected PAR were sampled at 5-minute intervals and recorded by a Decagon Em50 data logger (Decagon Devices Inc., Pullman, WA, USA). While Arctic tundra possess senesced vegetation (Bratsch et al., 2016; Davidson et al., 2016; Liu and Treitz, 2016; Liu et al., 2017), the PAR intercepted by non-photosynthetic vegetation is not used for photosynthesis (Gitelson et al., 2014). Therefore, field PVC (i.e., the PVC2D described in Section 3.2) was used to adjust fAPAR (Huemmrich et al., 2010; Schubert et al., 2010; Tagesson et al., 2012; Gitelson et al., 2014): 94 PAR Reflected fAPAR (1 ) PVC Eq. (4-1) PAR Incoming 4.4.4. Normalized Difference Vegetation Index An agricultural digital camera (ADC) (TetraCam Inc., Chatsworth, CA, USA) was used to collect ground NDVI digital data (Edwards and Henry, 2016). This camera provides the digital number (DN) values of the near-infrared (NIR), red and green channels. In the field, a white panel was placed on the ground and imaged by the ADC for reflectance calibration (ADC User’s Guide, 2011). Then, the ADC was positioned approximately 1.5 m above the ground to capture the NDVI digital data for each quadrat (Figure 4-2 C). In the PixelWrench2 software, a region of interest (ROI) was manually created from the calibration panel data to obtain the (NIR/Red)Incoming ratio (i.e., the ratio of incoming NIR radiance to red radiance) (ADC User’s Guide, 2011). Then, this ratio was applied as an offset to calculate NDVI for each quadrat (ADC User’s Guide, 2011): (NIR / Red)Reflected (NIR / Red)Incoming NDVI Eq. (4-2) (NIR / Red) Reflected (NIR / Red)Incoming where (NIR/Red)Reflected was the ratio of reflected NIR radiance to red radiance derived from NDVI digital images. Linear regression models were derived between field NDVI and PVC/fAPAR to explore their relationship. 4.4.5. Satellite Image Processing WorldView-2 satellite data were acquired for the ARW on July 21, 2015. This image consists of eight spectral bands: Coastal: 400-450 nm, Blue: 450-510 nm, Green: 510-580 nm, Yellow: 585-625 nm, Red: 630-690 nm, Red Edge: 705-745 nm, NIR1: 770-895 nm and NIR2: 860-1040 nm. WorldView-2 data were radiometrically corrected by using the FLAASH tool in ENVI 5.0 (input parameters: sub-Arctic summer atmosphere model; maritime aerosol model; 40 km visibility). Then a range of 2-band normalized difference spectral indices (NDSI) using all possible spectral band combinations were tested for predicting PVC3D/fAPAR: 95 Rx Ry NDSIx _ y Eq. (4-3) Rx Ry where Rx and Ry are the WorldView-2 reflectance values at bands x and y. These NDSIs were linearly regressed with field measured PVC/fAPAR for examining their ability to predict PVC/fAPAR. The regressions with the highest R2 values were applied to derive fine-scale maps of PVC/fAPAR. Since another objective was to validate MODIS LAI/fAPAR, the most recent MODIS LAI/fAPAR Collection 6 products (i.e., MCD15A3) were acquired for this study. Using Terra- and Aqua-MODIS reflectance as the algorithm inputs, this product has a spatial resolution of 500 m and a temporal resolution of 4 days. The MCD15A3 product (tile number: H14V02) of the ARW area was downloaded for the period of June 1 to September 22, 2015 (day of year (DOY): 153-265) from Earth Explorer (https://earthexplorer.usgs.gov/). The 2m resolution WorldView-2 PVC3D/fAPAR map was aggregated to 500 m spatial resolution and compared to the MODIS LAI/fAPAR product (July 20, 2015) on a pixel-by- pixel basis. In addition, MODIS LAI/fAPAR was averaged over the ARW area to obtain seasonal time- series to investigate the seasonal phenology. 4.5. Results and Discussion 4.5.1. Temporal Patterns of PVC and fAPAR The seasonal changes in PVC of different plant species and fAPAR at the six permanent plots (i.e., the red dot in Figure 4-1) are presented in Figure 4-3 (Note: Plots 5 and 6 had no PVC measurements on July 5, 2015). As evident here, each plot was dominated by different plant species and exhibited unique seasonal trends. Plot 1 was dominated by sedges/grasses and mosses, with a minimal presence of shrubs (e.g., Vaccinium uliginosum (blueberry), Braya glabella Richardson ssp. glabella (smooth northern-rockcress) and Salix reticulata (net-vein willows)). The total PVC (i.e., the PVC of all green vegetation) exhibited a continuously increasing trend during the growing season. The increase in the total PVC of Plot 1 was 20% (from 12% on July 5 to 32% on August 8), most of which was due to the increase in sedges/grasses (10%) and moss (7%). The overall trend in the total PVC was similar for Plot 2 and Plot 3. Plot 2 was dominated 96 by mosses with a mixture of shrubs (e.g., Salix reticulata (net-vein willows)) and sedges. The total PVC of Plot 2 increased rapidly from 19% (July 5) to 29% (August 3) and then decreased to 22% (August 8) due to the drying of mosses observed in the field. Plot 3 contained a mix of mosses, sedges and a variety of shrubs. The total PVC increased from 8% (July 5) to 21% (August 3) and then decreased rapidly to 7% (August 8). Plot 4 had large patches of bare soil with small amounts of shrubs (~10%) with the total PVC remaining consistent during the growing season. Plot 5 was dominated by Cassiope tetragona (white heather) with small amounts of mosses and willows. From July 20 to August 3, the white heather started to green up and PVC increased 13%. After flowering, white heather started to senesce, resulting in a decrease of 15% from August 3 to August 8. Plot 6 was dominated by mosses and shrubs with an overall change in total PVC of only 2% during the study period. The seasonal trend of fAPAR was consistent with that of total PVC (Figure 4-3). This is not surprising given that total PVC was used to adjust the calculation of fAPAR. It should be noted that the ratio of PARReflected to PARIncoming was found to be very small (~0.05) and was almost constant with a small variation throughout the growing season. This phenomenon was also reported by Schusbert et al. (2010). It indicates that the variation in fAPAR is mainly determined by PVC. In one recent study, by simulating a 3D radiative transfer model at small scales, Juszak et al. (2017) also confirmed that fAPAR was almost completely controlled by PVC. Species composition and their PVC were found to be driven by soil moisture in this study. For instance, the continuous increase in PVC observed in Plot 1 was related to the saturated moisture conditions throughout the growing season. In contrast, Plot 4 was located on an upland site with available soil moisture decreasing quickly early in the growing season. Due to moisture limitations, there was no obvious increase in total PVC. For other plots (e.g., Plots 2, 3 and 6) with intermediate soil moisture, mosses greened up early in the growing season and then underwent senescence. This was consistent with many previous studies suggesting that soil moisture may be the primary determining factor affecting Arctic vegetation growth (Chapin et al., 1995; Stow et al., 2004; Gamon et al., 2013; Huemmrich et al., 2013; Beamish et al., 2016; Liu and Treitz, 97 2016; Liu et al., 2017). Huemmrich et al. (2010) extend this observation to suggest that changes to species composition caused by soil moisture change over time will likely impact landscape level carbon exchange. Figure 4-3. Seasonal changes in the green PVC of different plant species and fAPAR for six permeant plots (i.e., the red dot in Figure 4-1) sampled in 2015. The location for these permanent plots is identified in Figure 4-2. 4.5.2. Comparison of PVC between the point-frame and image classification methods 2 High coefficients of determination (i.e., R > 0.80) were observed between the green PVC derived from the image classification (PVCImage) and point-frame methods (PVC2D; PVC3D) (Figure 4-4). As expected, it was observed that PVC2D had a stronger relationship with PVCImage than PVC3D, given that PVC2D and PVCImage 98 characterize the PVC from a synoptic perspective above the canopy, while PVC3D also incorporates the PVC below the top canopy layer. Overall, the PVC derived by the image classification method was systematically smaller than that derived by the point-frame method; i.e., a positive intercept was observed in each of the regressions. The image classification method underestimated PVC by 9.67% for the PVC2D- PVCImage relationship, and 12.14% for the PVC3D-PVCImage relationship. Potential explanations for this discrepancy include: (1) some plants such as Labrador tea, white heather and Lapland rosebay exhibited a dark green color and may have been classified as black soil in the image data; and/or (2) some plants (e.g., sedges and mosses) that were in the shadows of taller plants may have been classified as shadows in the image data but identified visually in the field as plants. The difference between PVC2D and PVC3D was also evident based on the slopes of the two regressions. The PVC2D-PVCImage relationship had a slope of 1.17 (close to 1), while the slope of the PVC3D-PVCImage relationship was 2.12 since the PVC below the top layer was included in the calculation of PVC3D. Although a systematic bias was observed between PVC2D and PVCImage (i.e., regression intercept = 9.67), the close- to-1 slope of the regression (i.e., equal to 1.17) indicates that the change in PVCImage was similar to the change in PVC2D. Therefore, the image classification method could serve as an efficient alternative to detect changes in Arctic PVC2D (Chen et al., 2010; Liu and Treitz, 2016; Beamish et al. 2016; Edwards and Henry 2016). However, it is suggest that calibration between these two methods is necessary when the purpose is to estimate PVC2D/PVC3D from PVCImage, i.e., samples should be collected across a variety of vegetation types to develop a calibration model between PVC2D/PVC3D and PVCImage. PVC2D and PVC3D are important measures for assessing and monitoring Arctic vegetation. For instance, PVC2D has been used to adjust the field measured fAPAR of canopies that exhibit significant amounts of senesced vegetation (Huemmrich et al., 2010; Schubert et al., 2010; Tagesson et al., 2012). PVC3D has the capacity to characterize the changes/differences in the PVC below the top layer (Buchhorn et al., 2013). For instance, Liu et al. (2017) found that the Wet Mesic Tundra vegetation type dominated by mosses and forbs showed similar PVC2D to the Wet Sedge/Moss vegetation type dominated by sedges/mosses. However, 99 due to the multiple layers of sedges present in Wet Sedge/Moss, the PVC3D difference between these two vegetation types was ~25%. This difference in PVC3D resulted in a higher NIR reflectance exhibited in Wet Sedge/Moss than that in Wet Mesic Tundra (Liu et al., 2017). Recent studies have indicated that this spectral difference could be used to differentiate different Arctic vegetation types (Richardson et al., 2009; Ulrich et al., 2009; Buchhorn et al., 2013; Beamish et al. 2016; Bratsch et al., 2016; Davidson et al., 2016; Liu et al., 2017). Figure 4-4. Comparison of PVC derived by the image classification (PVCImage) and point-frame methods (PVC2D: PVC of the top layer derived from the point-frame method; PVC3D: PVC of all layers derived from the point-frame method). Dashed line represents the 1:1 correspondence. MAE (mean absolute error): N | yi xi |/ N . i1 4.5.3. PVC/fAPAR-VI relationships 2 At the plot scale, there was a strong correspondence between field-measured PVC/fAPAR and NDVI (R > 0.70) (Table 4-2). This strong relationship was consistent with previous field studies in the Arctic (Huemmrich et al., 2010, 2013; Schubert et al., 2010; Tagesson et al., 2012; Beamish et al., 2016; Edwards and Henry 2016; Liu and Treitz, 2016; Liu et al., 2017). These results have further confirmed that NDVI 100 can serve as a surrogate for characterizing PVC/fAPAR of Arctic vegetation. In the fAPAR-NDVI relationship, when fAPAR = 0 (i.e., there was no photosynthetic vegetation), NDVI had a value of 0.05. This value was smaller than that found by previous studies (Stow et al., 2003; Huemmrich et al., 2010; Schubert et al., 2010; Tagesson et al., 2012). The main reason may be that the vegetation types measured in previous studies were dense canopies with a relatively high biomass (dominant plants: heath or sedge meadows) while the fAPAR was sampled for a range of highly vegetated (in an Arctic sense) to non- vegetated areas in this study. Table 4-2. Plot-scale regression relationships between fAPAR, PVC2D, PVC3D and NDVI. fAPAR = 1.85 × NDVI - 0.10 R2 = 0.74 RMSE = 0.07 2 PVC2D = 1.95 × NDVI - 0.08 R = 0.74 RMSE = 0.08 2 PVC3D = 3.83 × NDVI - 0.27 R = 0.83 RMSE = 0.12 Results of the PVC3D/fAPAR-NDSI relationship at the fine satellite scale (i.e., 6 m) are presented in Table 2 2 4-3. The NDSIs with moderate/strong correlations (R > 0.60 for fAPAR and R > 0.70 for PVC3D) were grouped into four categories: (1) green + NIR; (2) yellow + NIR; (3) red + NIR; and (4) red-edge + NIR. Within each category, the R2 and RMSE values of different band combinations were quite similar, i.e., no band combination showed a better performance than others. For each band combination, the correlation between PVC3D and NDSI was stronger than the correlation between fAPAR and NDSI. To explain this discrepancy, we suggest that the NIR is more sensitive to the changes in PVC3D, especially for the vegetation types with multiple layers. Generally, an increase in PVC3D would result in more multiple scattering within the canopy, thereby increasing NIR reflectance (Davidson et al., 2016; Liu et al., 2017). However, the increase in PVC3D did not necessarily mean an increase in PVC2D that was the controlling factor of fAPAR. As stated in Section 4.2, two different vegetation types may show similar PVC2D but quite different PVC3D (Liu et al., 2017). 101 Compared to the plot-scale, the relationship observed at the satellite scale was weaker (Table 4-3). This may be attributed to differences in spectral bandwidth, atmospheric effects (i.e., path radiance), geometric errors in the remote sensing data and/or scale differences. For example, the plot scale PVC/fAPAR-NDVI relationship was based on the measurements of PVC/fAPAR and NDVI measured within the same quadrats. However, when developing satellite scale relationships, PVC/fAPAR measurements were based on the average of quadrats (i.e., 0.6 m × 0.6 m) while NDSI was a measure at the pixel scale (i.e., 6 m × 6 m) thereby covering a much larger area than the quadrat measures (Tagesson et al., 2012). Since Arctic tundra is extremely heterogeneous even at small scales (Juszak et al., 2017), this discrepancy in scale may affect the PVC/fAPAR-NDSI relationship. The potential for scale to impact the relationships between biophysical variables and satellite VIs have also been reported for other Arctic studies (i.e., Hope et al., 1993; Stow et al., 2004; Laidler et al., 2008). Compared to medium (e.g., 30 m Landsat imagery) and coarse resolution (e.g., 500m or 1000 m MODIS imagery) satellite data, high spatial resolution satellite data may capture fine-grained heterogeneity and exhibit stronger relationships to Arctic biophysical variables. As a result, more studies are focusing on using high spatial resolution imagery to map Arctic biophysical variables at local/regional scales (Tagesson et al., 2012; Siewert et al., 2015; Fraser et al., 2016; Greaves et al., 2016; Liu and Treitz, 2016). 102 Table 4-3. Satellite-scale regression relationships between fAPAR, PVC3D and NDSI ((Rx - Ry)/ (Rx + Ry)). 2 For the fAPAR-NDSI relationship, the band combinations with R > 0.6 are presented. For the PVC3D- 2 NDSI relationships, the band combinations with R > 0.7 are displayed. The model results are grouped according to band x. Band Combination fAPAR-NDSI PVC3D-NDSI Band x Band y R2 (RMSE) R2 (RMSE) Red-Edge 0.65 (0.12) 0.70 (0.20) Green NIR1 0.66 (0.12) 0.70 (0.20) NIR2 0.65 (0.12) 0.70 (0.20) Red-Edge 0.64 (0.12) 0.70 (0.20) Yellow NIR1 0.65 (0.12) 0.71 (0.20) NIR2 0.64 (0.12) 0.71 (0.20) Red-Edge 0.66 (0.12) 0.73 (0.19) Red NIR1 0.66 (0.12) 0.73 (0.19) NIR2 0.65 (0.12) 0.73 (0.20) Red-Edge NIR1 0.64 (0.12) - 4.5.4. Comparison with MODIS LAI/fAPAR product Figure 4-5 presents the comparison of LAI and fAPAR as derived for WorldView-2 and MODIS. These two datasets showed moderate coefficients of determination (R2 = 0.68 and 0.62 respectively). Generally, the PVC3D/fAPAR derived from WorldView-2 was higher than MODIS LAI/fAPAR. For LAI, the average difference (i.e., mean absolute error in Figure 4-5) was ~ 0.06 m2/m2. For fAPAR, it was ~0.08. This difference was much larger when MODIS LAI/fAPAR values were small (i.e., for areas with lower PVC). It should be noted that although the PVC of moss was included in the measures of PVC3D in this study, it did not contribute to LAI since moss does not possess leaves (Williams, 2005). However, moss cover has a substantial contribution to satellite signals (Bratsch et al., 2016; Davidson et al., 2016; Liu et al., 2017). Excluding moss cover from PVC3D will dampen the correspondence between PVC3D and vegetation indices, thereby affecting the estimation accuracy of PVC3D. For fAPAR, we suggest that the systematic 103 overestimation of WorldView-2 data is due to the PAR absorption by the soil background that has not been excluded from the fAPAR calculation. Measuring fAPAR for low Arctic vegetation is challenging since it is difficult to place PAR sensors below the canopy (Tagesson et al., 2012). For most Arctic studies for which PAR was measured below the canopy, the canopy height exceeded 10 cm, sometimes reaching 1-2m for tall shrubs (Huemmrich et al., 2010; Juszak et al., 2017). These conditions were not available at our study site. Figure 4-5. Comparison of LAI/fAPAR between MODIS and WorldView-2. Dash line is 1:1 line. MAE N (mean absolute error): | yi xi |/ N . i1 The MODIS LAI/fAPAR time series for the ARW from DOY161-265 (i.e., June 10 to September 22, 2015) is presented in Figure 4-6. Here, there was a clear seasonal trend for LAI and fAPAR. As expected, LAI was minimal at the beginning of the growing season (DOY: 161, June 10). It then increased rapidly to approximately 0.30 (i.e., DOY 205, July 23). During the peak of the growing season (DOY: 205-229, July 23-August 16), the LAI value remained within the range of 0.30-0.40 and then decreased rapidly to 0.15. The seasonal trend of fAPAR was consistent with that of LAI, with a value of 0.23 during the peak of the 104 season. The seasonal trajectory of MODIS LAI/fAPAR was consistent with other studies investigating seasonal trends of Arctic NDVI (Schubert et al., 2010; Tagesson et al., 2012; Gonsamo and Chen, 2016). These trends are important as they can be used to extract phenology parameters for Arctic vegetation (e.g., start, end and length of growing season) (Narasimhan and Stow, 2010; Anderson et al., 2016; Beamish et al., 2016; Gonsamo and Chen, 2016). Figure 4-6. MODIS LAI/fAPAR seasonal variation in 2015. 4.6. Conclusion In this study, field PVC, fAPAR and NDVI were measured and analyzed along a moisture gradient in the Apex River Watershed (ARW) (63°45’N, 68°30’W), Baffin Island, Nunavut, Canada. The results of seasonal changes in PVC at six plots indicated that different vegetation types had different temporal patterns of PVC change and these patterns appear to be a function of available soil moisture. Although a systematic PVC difference was observed between the point-frame and image classification methods, the close-to-1 slope of the regression indicates that the image classification method can be applied as an efficient alternative to detect change in PVC. At both plot and satellite scales, field measured PVC/fAPAR demonstrated moderate to strong relationships with VIs. Comparison of high-resolution PVC3D/fAPAR data with MODIS LAI/fAPAR products showed that high-resolution data overestimated PVC3D/fAPAR. The overestimation of LAI was attributed to moss cover being included in the calculation of PVC3D, and the systematic overestimation of fAPAR to PAR being absorbed by soil background (a phenomenon that could not be excluded from the fAPAR calculation). 105 4.7. Acknowledgement The authors would like to acknowledge financial support from the Networks of Centres of Excellence of Canada (i.e., ArcticNet), the Northern Science Training Program (NTSP) – Polar Knowledge Canada, the Natural Sciences and Engineering Research Council (NSERC) and Queen’s University, Kingston, Canada. We gratefully acknowledge logistical support provided by the Nunavut Research Institute (NRI). The authors would also like to thank our field assistant Rebecca Edwards for her significant contribution to the field data collection. The authors would also like to thank Drs. Melissa Lafrenière and Scott Lamoureux for their ongoing support of this research at the Apex River Watershed. 4.8. References Agricultural Digital Camera User’s Guide, 2011. TETRACAM Inc., Chatsworth, USA. Anderson, H., Nilsen, L., Tømmervik, H., Karlsen, S., Nagai, S., Cooper, E., 2016. Using ordinary digital cameras in place of near-infrared sensors to derive vegetation indices for phenology studies of high Arctic vegetation. Remote Sensing 8(10), 847. doi:10.3390/rs8100847 Aragao, L.E.O.C., Shimabukuro, Y.E., Espirito-Santo, F.D.B., Williams, M., 2005. Spatial validation of the collection 4 MODIS LAI product in eastern Amazonia. IEEE Transactions on Geoscience and Remote Sensing 43(11), 2526–2534. doi:10.1109/TGRS.2005.856632 Atkinson, D.M., Treitz, P.M., 2013. Modeling biophysical variables across an Arctic latitudinal gradient using high spatial resolution remote sensing data. Arctic, Antarctic, and Alpine Research 45(2), 161– 178. doi:10.1657/1938-4246-45.2.161 Beamish, A.L., Nijland, W., Edwards, M., Coops, N.C., Henry, G.H.R., 2016. Phenology and vegetation change measurements from true colour digital photography in High Arctic tundra. Arctic Science 2(2), 33–49. doi:10.1139/as-2014-0003 Benz, U.C., Hofmann, P., Willhauck, G., Lingenfelder, I., Heynen, M., 2004. Multi-resolution, object- oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry and Remote Sensing 58(3–4), 239–258. doi:10.1016/j.isprsjprs.2003.10.002 Blok, D., Schaepman-Strub, G., Bartholomeus, H., Heijmans, M.M.P.D., Maximov, T.C., Berendse, F., 2011. The response of Arctic vegetation to the summer climate: relation between shrub cover, NDVI, surface albedo and temperature. Environmental Research Letters 6(3), 35502. doi:10.1088/1748- 9326/6/3/035502 Bonham, C.D., 2013. Measurements for terrestrial vegetation (Second Edition). John Wiley & Sons, West Sussex. Bratsch, S.N., Epstein, H.E., Buchhorn, M., Walker, D.A., 2016. Differentiating among four Arctic tundra plant communities at Ivotuk, Alaska using field spectroscopy. Remote Sensing 8(1). doi:10.3390/rs8010051 Buchhorn, M., Walker, D.A., Heim, B., Raynolds, M.K., Epstein, H.E., Schwieder, M., 2013. Ground- 106 based hyperspectral characterization of Alaska tundra vegetation along environmental gradients. Remote Sensing 5(8), 3971–4005. doi:10.3390/rs5083971 Chapin, F.S., Shaver, G.R., Giblin, A.E., Nadelhoffer, K.J., Laundre, J.A., 1995. Responses of Arctic tundra to experimental and observed changes in climate. Ecology 76(3), 694–711. doi:10.2307/1939337 Chen, Z., Chen, W., Leblanc, S.G., Henry, G.H.R., 2010. Digital photograph analysis for measuring percent plant cover in the Arctic. Arctic 63(3), 315–326. Davidson, S., Santos, M., Sloan, V., Watts, J., Phoenix, G., Oechel, W., Zona, D., 2016. Mapping Arctic tundra vegetation communities using field spectroscopy and multispectral satellite data in North Alaska, USA. Remote Sensing 8(12), 978. doi:10.3390/rs8120978 De Kauwe, M.G., Disney, M.I., Quaife, T., Lewis, P., Williams, M., 2011. An assessment of the MODIS collection 5 leaf area index product for a region of mixed coniferous forest. Remote Sensing of Environment 115(2), 767–780. doi:10.1016/j.rse.2010.11.004 Edwards, M., Henry, G.H.R., 2016. The effects of long-term experimental warming on the structure of three High Arctic plant communities, Journal of Vegetation Science 27, 904-913. Elmendorf, S.C., Henry, G.H.R., Hollister, R.D., Björk, R.G., Boulanger-Lapointe, N., Cooper, E.J., Cornelissen, J.H.C., Day, T.A., Dorrepaal, E., Elumeeva, T.G., Gill, M., Gould, W.A., Harte, J., Hik, D.S., Hofgaard, A., Johnson, D.R., Johnstone, J.F., Jónsdóttir, I.S., Jorgenson, J.C., Klanderud, K., Klein, J.A., Koh, S., Kudo, G., Lara, M., Lévesque, E., Magnússon, B., May, J.L., Mercado-Dı´az, J.A., Michelsen, A., Molau, U., Myers-Smith, I.H., Oberbauer, S.F., Onipchenko, V.G., Rixen, C., Martin Schmidt, N., Shaver, G.R., Spasojevic, M.J., Þórhallsdóttir, Þ.E., Tolvanen, A., Troxler, T., Tweedie, C.E., Villareal, S., Wahren, C.-H., Walker, X., Webber, P.J., Welker, J.M., Wipf, S., 2012. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2(6), 453–457. doi:10.1038/nclimate1465 Emmerton, C.A., St. Louis, V.L., Humphreys, E.R., Gamon, J.A., Barker, J.D., Pastorello, G.Z., 2015. Net ecosystem exchange of CO2 with rapidly changing high Arctic landscapes. Global Change Biology n/a-n/a. doi:10.1111/gcb.13064 Environment Canada, 2015. Climate normals. Epstein, H.E., Calef, M.P., Walker, M.D., Chapin, F.S., Starfields, A.M., 2004. Detecting changes in arctic tundra plant communities in response to warming over decadal time scales. Global Change Biology 10(8), 1325–1334. doi:10.1111/j.1529-8817.2003.00810.x Epstein, H.E., Raynolds, M.K., Walker, D. a, Bhatt, U.S., Tucker, C.J., Pinzon, J.E., 2012. Dynamics of aboveground phytomass of the circumpolar Arctic tundra during the past three decades. Environmental Research Letters 7(1), 15506. doi:10.1088/1748-9326/7/1/015506 Fensholt, R., Sandholt, I., 2003. Derivation of a shortwave infrared water stress index from MODIS near- and shortwave infrared data in a semiarid environment. Remote Sensing of Environment 87(1), 111– 121. doi:10.1016/j.rse.2003.07.002 Fensholt, R., Sandholt, I., Rasmussen, M.S., 2004. Evaluation of MODIS LAI, fAPAR and the relation between fAPAR and NDVI in a semi-arid environment using in situ measurements. Remote Sensing of Environment 91(3–4), 490–507. doi:10.1016/j.rse.2004.04.009 Fraser, R.H., Olthof, I., Carrière, M., Deschamps, A., Pouliot, D., 2011. Detecting long-term changes to vegetation in northern Canada using the Landsat satellite image archive. Environmental Research Letters 6(4), 45502. doi:10.1088/1748-9326/6/4/045502 107 Fraser, R.H., Olthof, I., Lantz, T.C., Schmitt, C., 2016. UAV photogrammetry for mapping vegetation in the low-Arctic. Arctic Science 2(3), 79–102. doi:10.1139/as-2016-0008 Frost, G. V, Epstein, H.E., 2014. Tall shrub and tree expansion in Siberian tundra ecotones since the 1960s. Global change biology 20(4), 1264–1277. doi:10.1111/gcb.12406 Gamon, J.A., Huemmrich, K.F., Stone, R.S., Tweedie, C.E., 2013. Spatial and temporal variation in primary productivity ( NDVI ) of coastal Alaskan tundra : Decreased vegetation growth following earlier snowmelt. Remote Sensing of Environment 129, 144–153. doi:10.1016/j.rse.2012.10.030 Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms for remote estimation of vegetation fraction. Remote Sensing of Environment 80(1), 76–87. doi:10.1016/S0034- 4257(01)00289-9 Gitelson, A.A., Peng, Y., Huemmrich, K.F., 2014. Relationship between fraction of radiation absorbed by photosynthesizing maize and soybean canopies and NDVI from remotely sensed data taken at close range and from MODIS 250m resolution data. Remote Sensing of Environment 147, 108–120. doi:10.1016/j.rse.2014.02.014 Gonsamo, A., Chen, J.M., 2016. Circumpolar vegetation dynamics product for global change study. Remote Sensing of Environment 182, 13–26. doi:10.1016/j.rse.2016.04.022 Goward, S.N., Huemmrich, K.F., 1992. Vegetation canopy PAR absorptance and the normalized difference vegetation index: An assessment using the SAIL model. Remote Sensing of Environment 39(2), 119– 140. doi:10.1016/0034-4257(92)90131-3 Greaves, H.E., Vierling, L.A., Eitel, J.U.H., Boelman, N.T., Magney, T.S., Prager, C.M., Griffin, K.L., 2016. High-resolution mapping of aboveground shrub biomass in Arctic tundra using airborne lidar and imagery. Remote Sensing of Environment 184, 361–373. doi:10.1016/j.rse.2016.07.026 Hill, M.J., Senarath, U., Lee, A., Zeppel, M., Nightingale, J.M., Williams, R. (Dick) J., McVicar, T.R., 2006. Assessment of the MODIS LAI product for Australian ecosystems. Remote Sensing of Environment 101(4), 495–518. doi:10.1016/j.rse.2006.01.010 Hodgson, D.A., 2003. Surficial geology, Frobisher Bay, Baffin Island, Nunavut. doi:10.4095/214808 Hope, A.S., Kimball, J.S., Stow, D.A., 1993. The relationship between tussock tundra spectral reflectance properties and biomass and vegetation composition. International Journal of Remote Sensing 14(10), 1861–1874. doi:10.1080/01431169308954008 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Campbell, P.K.E., Landis, D.R., Middleton, E.M., 2013. Arctic tundra vegetation functional types based on photosynthetic physiology and optical properties. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6, 265–275. doi:10.1109/JSTARS.2013.2253446 Huemmrich, K.F., Gamon, J.A., Tweedie, C.E., Oberbauer, S.F., Kinoshita, G., Houston, S., Kuchy, A., Hollister, R.D., Kwon, H., Mano, M., Harazono, Y., Webber, P.J., Oechel, W.C., 2010. Remote sensing of tundra gross ecosystem productivity and light use efficiency under varying temperature and moisture conditions. Remote Sensing of Environment 114(3), 481–489. doi:http://dx.doi.org/10.1016/j.rse.2009.10.003 IPCC, 2013. Climate Change 2013 - The Physical Science Basis. Cambridge University Press, Cambridge. doi:10.1017/CBO9781107415324 Jensen, J.L.R., Humes, K.S., Hudak, A.T., Vierling, L.A., Delmelle, E., 2011. Evaluation of the MODIS LAI product using independent lidar-derived LAI: A case study in mixed conifer forest. Remote 108 Sensing of Environment 115(12), 3625–3639. doi:10.1016/j.rse.2011.08.023 Jia, G.J., Epstein, H.E., Walker, D.A., 2003. Greening of Arctic Alaska, 1981-2001. Geophysical Research Letters 30(20), n/a-n/a. doi:10.1029/2003GL018268 Jia, G.J., Epstein, H.E., Walker, D.A., 2009. Vegetation greening in the canadian arctic related to decadal warming. Journal of Environmental Monitoring 11(12), 2231. doi:10.1039/b911677j Johansen, B., Tømmervik, H., 2014. The relationship between phytomass, NDVI and vegetation communities on Svalbard. International Journal of Applied Earth Observation and Geoinformation 27, 20–30. doi:10.1016/j.jag.2013.07.001 Ju, J., Masek, J.G., 2016. The vegetation greenness trend in Canada and US Alaska from 1984–2012 Landsat data. Remote Sensing of Environment 176, 1–16. doi:10.1016/j.rse.2016.01.001 Juszak, I., Erb, A.M., Maximov, T.C., Schaepman-Strub, G., 2014. Arctic shrub effects on NDVI, summer albedo and soil shading. Remote Sensing of Environment 153, 79–89. doi:10.1016/j.rse.2014.07.021 Juszak, I., Iturrate-Garcia, M., Gastellu-Etchegorry, J.-P., Schaepman, M.E., Maximov, T.C., Schaepman- Strub, G., 2017. Drivers of shortwave radiation fluxes in Arctic tundra across scales. Remote Sensing of Environment 193, 86–102. doi:10.1016/j.rse.2017.02.017 Knyazikhin, Y., Martonchik, J.V., Myneni, R.B., Diner, D.J., Running, S.W., 1998. Synergistic algorithm for estimating vegetation canopy leaf area index and fraction of absorbed photosynthetically active radiation from MODIS and MISR data. Journal of Geophysical Research 103, 32257–32274. la Puma, I.P., Philippi, T.E., Oberbauer, S.F., 2007. Relating NDVI to ecosystem CO2 exchange patterns in response to season length and soil warming manipulations in arctic Alaska. Remote Sensing of Environment 109(2), 225–236. doi:10.1016/j.rse.2007.01.001 Laidler, G.J., Treitz, P.M., Atkinson, D.M., 2008. Remote sensing of Arctic vegetation: relations between the NDVI, spatial resolution and vegetation cover on Boothia Peninsula, Nunavut. Arctic 61(1), 1–13. Liu, N., Treitz, P., 2016. Modelling high arctic percent vegetation cover using field digital images and high resolution satellite data. International Journal of Applied Earth Observation and Geoinformation 52, 445–456. Liu, N., Budkewitsch, P., Treitz, P., 2017. Examining spectral reflectance features related to Arctic percent vegetation cover: Implications for hyperspectral remote sensing of Arctic tundra. Remote Sensing of Environment 192, 58–72. doi:10.1016/j.rse.2017.02.002 Luscier, J.D., Thompson, W.L., Wilson, J.M., Gorham, B.E., Dragut, L.D., 2006. Using digital photographs and object-based image analysis to estimate percent ground cover in vegetation plots. Frontiers in Ecology and the Environment 4, 408–413. doi:http://dx.doi.org/10.1890/1540- 9295(2006)4[408:UDPAOI]2.0.CO;2 Molau, U., Mølgaard, P., 1996. ITEX Manual (Second Edition). Copenhagen. Myneni, R.B., Williams, D.L., 1994. On the relationship between FAPAR and NDVI. Remote Sensing of Environment 49(3), 200–211. doi:10.1016/0034-4257(94)90016-7 Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386(6626), 698–702. doi:10.1038/386698a0 Myneni, R.., Hoffmana, S., Knyazikhina, Y., Privetteb, J.., Glassyc, J., Tiana, Y., Wanga, Y., Songa, X., Zhanga, Y., Smitha, G.., Lotscha, A., Friedla, M., Morisetteb, J.., Votavac, P., Nemanic, R.., Runningc, S.., 2002. Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS 109 data. Remote Sensing of Environment 83(1–2), 214–231. Narasimhan, R., Stow, D., 2010. Daily MODIS products for analyzing early season vegetation dynamics across the North Slope of Alaska. Remote Sensing of Environment 114(6), 1251–1262. doi:10.1016/j.rse.2010.01.017 Olthof, I., Latifovic, R., Pouliot, D., 2009. Development of a circa 2000 land cover map of northern Canada at 30 m resolution from Landsat. Canadian Journal of Remote Sensing 35(2), 152–165. doi:10.5589/m09-007 Pasolli, L., Asam, S., Castelli, M., Bruzzone, L., Wohlfahrt, G., Zebisch, M., Notarnicola, C., 2015. Retrieval of Leaf Area Index in mountain grasslands in the Alps from MODIS satellite imagery. Remote Sensing of Environment 165, 159–174. doi:10.1016/j.rse.2015.04.027 Privette, J.., Myneni, R.., Knyazikhin, Y., Mukelabai, M., Roberts, G., Tian, Y., Wang, Y., Leblanc, S.., 2002. Early spatial and temporal validation of MODIS LAI product in the Southern Africa Kalahari. Remote Sensing of Environment 83(1–2), 232–243. doi:10.1016/S0034-4257(02)00075-5 Raynolds, M.K., Walker, D.A., Maier, H.A., 2006. NDVI patterns and phytomass distribution in the circumpolar Arctic. Remote Sensing of Environment 102(3–4), 271–281. doi:http://dx.doi.org/10.1016/j.rse.2006.02.016 Richardson, A.D., Braswell, B.H., Hollinger, D.Y., Jenkins, J.P., Ollinger, S. V, 2009. Near-surface remote sensing of spatial and temporal variation in canopy phenology. Ecological Applications 19, 1417– 1428. doi:http://dx.doi.org.proxy.queensu.ca/10.1890/08-2022.1 Schubert, P., Lund, M., Ström, L., Eklundh, L., 2010. Impact of nutrients on peatland GPP estimations using MODIS time series data. Remote Sensing of Environment 114(10), 2137–2145. doi:10.1016/j.rse.2010.04.018 Serbin, S.P., Ahl, D.E., Gower, S.T., 2013. Spatial and temporal validation of the MODIS LAI and FPAR products across a boreal forest wildfire chronosequence. Remote Sensing of Environment 133, 71–84. doi:10.1016/j.rse.2013.01.022 Siewert, M.B., Hanisch, J., Weiss, N., Kuhry, P., Maximov, T.C., Hugelius, G., 2015. Comparing carbon storage of Siberian tundra and taiga permafrost ecosystems at very high spatial resolution. Journal of Geophysical Research: Biogeosciences 120(10), 1973–1994. doi:10.1002/2015JG002999 Stewart, L., Alsos, I.G., Bay, C., Breen, A.L., Brochmann, C., Boulanger-Lapointe, N., Broennimann, O., Bültmann, H., Bøcher, P.K., Damgaard, C., Daniëls, F.J.A., Ehrich, D., Eidesen, P.B., Guisan, A., Jónsdóttir, I.S., Lenoir, J., le Roux, P.C., Lévesque, E., Luoto, M., Nabe-Nielsen, J., Schönswetter, P., Tribsch, A., Tveraabak, L.U., Virtanen, R., Walker, D.A., Westergaard, K.B., Yoccoz, N.G., Svenning, J.-C., Wisz, M., Schmidt, N.M., Pellissier, L., 2016. The regional species richness and genetic diversity of Arctic vegetation reflect both past glaciations and current climate. Global Ecology and Biogeography 430–442. doi:10.1111/geb.12424 Stow, D., Daeschner, S., Hope, A., Douglas, D., Petersen, A., Myneni, R., Zhou, L., Oechel, W., 2003. Variability of the Seasonally Integrated Normalized Difference Vegetation Index Across the North Slope of Alaska in the 1990s. International Journal of Remote Sensing 24(5), 1111–1117. doi:10.1080/0143116021000020144 Stow, D.A., Hope, A., McGuire, D., Verbyla, D., Gamon, J., Huemmrich, F., Houston, S., Racine, C., Sturm, M., Tape, K., Hinzman, L., Yoshikawa, K., Tweedie, C., Noyle, B., Silapaswan, C., Douglas, D., Griffith, B., Jia, G., Epstein, H., Walker, D., Daeschner, S., Petersen, A., Zhou, L., Myneni, R., 2004. Remote sensing of vegetation and land-cover change in Arctic tundra ecosystems. Remote Sensing of 110 Environment 89(3), 281–308. doi:10.1016/j.rse.2003.10.018 Sturm, M., Racine, C., Tape, K., 2001. Climate change: increasing shrub abundance in the Arctic. Nature 411(6837), 546–7. doi:10.1038/35079180 Tagesson, T., Mastepanov, M., Tamstorf, M.P., Eklundh, L., Schubert, P., Ekberg, A., Sigsgaard, C., Christensen, T.R., Ström, L., 2012. High-resolution satellite data reveal an increase in peak growing season gross primary production in a high-Arctic wet tundra ecosystem 1992–2008. International Journal of Applied Earth Observation and Geoinformation 18, 407–416. doi:10.1016/j.jag.2012.03.016 Tape, K.D., Sturm, M., Racine, C., 2006. The evidence for shrub expansion in Northern Alaska and the Pan-Arctic. Global Change Biology 12(4), 686–702. doi:10.1111/j.1365-2486.2006.01128.x Tape, K.D., Hallinger, M., Welker, J.M., Ruess, R.W., 2012. Landscape heterogeneity of shrub expansion in Arctic Alaska. Ecosystems 15(5), 711–724. doi:10.1007/s10021-012-9540-4 Tian, Y., Woodcock, C.E., Wang, Y., Privette, J., Shabanov, N., Zhou, L., Zhang, Y., Buermann, W., Dong, J., Veikkanen, B., 2002. Multiscale analysis and validation of the MODIS LAI productI. Uncertainty assessment. Remote Sensing of Environment 83(3), 414–430. doi:10.1016/S0034-4257(02)00047-0 Ulrich, M., Grosse, G., Chabrillat, S., Schirrmeister, L., 2009. Spectral characterization of periglacial surfaces and geomorphological units in the Arctic Lena Delta using field spectrometry and remote sensing. Remote Sensing of Environment 113(6), 1220–1235. doi:http://dx.doi.org/10.1016/j.rse.2009.02.009 Walker, D.A., Raynolds, M.K., Daniëls, F.J.A., Einarsson, E., Elvebakk, A., Gould, W.A., Katenin, A.E., Kholod, S.S., Markon, C.J., Melnikov, E.S., Moskalenko, N.G., Talbot, S.S., Yurtsev, B.A., 2005. The Circumpolar Arctic vegetation map. Journal of Vegetation Science 16(3), 267–282. doi:10.1111/j.1654-1103.2005.tb02365.x Walker, M.D., Wahren, C.H., Hollister, R.D., Henry, G.H.R., Ahlquist, L.E., Alatalo, J.M., Bret-Harte, M.S., Calef, M.P., Callaghan, T. V., Carroll, A.B., Epstein, H.E., Jonsdottir, I.S., Klein, J.A., Magnusson, B., Molau, U., Oberbauer, S.F., Rewa, S.P., Robinson, C.H., Shaver, G.R., Suding, K.N., Thompson, C.C., Tolvanen, A., Totland, O., Turner, P.L., Tweedie, C.E., Webber, P.J., Wookey, P.A., 2006. Plant community responses to experimental warming across the tundra biome. Proceedings of the National Academy of Sciences 103(5), 1342–1346. doi:10.1073/pnas.0503198103 Wang, Y., Woodcock, C.E., Buermann, W., Stenberg, P., Voipio, P., Smolander, H., Häme, T., Tian, Y., Hu, J., Knyazikhin, Y., Myneni, R.B., 2004. Evaluation of the MODIS LAI algorithm at a coniferous forest site in Finland. Remote Sensing of Environment 91(1), 114–127. doi:10.1016/j.rse.2004.02.007 Williams, M., 2005. Optical instruments for measuring leaf area index in low vegetation: application in Arctic ecosystems. Ecological Applications 15(4), 1462–1470. doi:10.1890/03-5354 Zhang, Y., Song, C., Band, L.E., Sun, G., Li, J., 2017. Reanalysis of global terrestrial vegetation trends from MODIS products: Browning or greening? Remote Sensing of Environment 191, 145–155. doi:10.1016/j.rse.2016.12.018 111 Chapter 5 Synthesis and Future Directions The Arctic is warming at a rapid rate and vegetation is being affected in a widespread and diverse manner (Myneni et al., 1997; Jia et al., 2003, 2009; Epstein et al., 2004; Walker et al., 2006; Elmendorf et al., 2012; IPCC, 2013; Stewart et al., 2016). However, due to the short growing season and the remote nature of the Arctic, it is logistically challenging to conduct field surveys for more than a small number of locations in order to capture and study these impacts and changes. Alternatively, satellite remote sensing provides an efficient way to assess and monitor Arctic vegetation, and subsequent changes over large areas (Stow et al., 2004). The overall objective of this dissertation was to model Arctic biophysical variables using remote sensing data. A series of field measurements were conducted at three Canadian Arctic sites, including the Sabine Peninsula, the Cape Bounty Arctic Watershed Observatory (CBAWO) and the Apex River Watershed (ARW), to measure percent vegetation cover (PVC) and the fraction of absorbed photosynthetically active radiation (fAPAR). These three sites represent a latitudinal gradient, a surrogate for a ‘climate gradient’ that can be used to examine differences in vegetation biophysical variables given their contrasting temperature and moisture regimes. The field measurements collected at these sites were then modelled based on relationships to spectral indices and extended over space using satellite data in Chapters 2-4. Chapter 2 focused on exploring the capacity of high spectral resolution (i.e., hyperspectral) remote sensing data for modelling Arctic PVC. Chapters 3 and 4 focused on modelling Arctic PVC and fAPAR using high spatial resolution multispectral satellite data, respectively. 5.1. Objective 1: To assess the utility of hyperspectral remote sensing data for modelling Arctic PVC using field spectra. Hyperspectral data exhibited a better performance for characterizing Arctic vegetation and for modelling PVC than multispectral data; a result attributed to the capacity for hyperspectral data to sample fine spectral features. Results from Chapter 2 demonstrated that vegetation in the High Arctic did not exhibit the typical 112 spectral characteristics of green vegetation such as the reflectance peak in the green wavelengths, the deep pigment absorption feature in the red, or the water absorption features in the near- and shortwave-infrared regions (Figure 2-3). Instead, its spectral characteristics were similar to bare soil or senesced vegetation. Strong dry matter absorption features, especially lignin and cellulose absorption, were evident at shortwave infrared wavelengths. Based on field PVC measurements, this was due to the large proportion of senesced vegetation present in these vegetation types. The optimal hyperspectral bands for estimating PVC were found to include important absorption features observed in Arctic vegetation spectra (Figure 2-6). These spectral bands included 681.20 nm (leaf pigment absorption); 721.90 nm and 732.07 nm (along the red-edge slope); 1174.77 nm and 1184.87 nm (leaf water absorption); and 1447.14 nm, 1457.23 nm, 2072.65 nm and 2102.94 nm (leaf cellulose and lignin absorption) (Figure 2-9). Compared to visible and near infrared wavelengths, shortwave infrared wavelengths showed a better performance in predicting PVC. VIs using shortwave infrared spectral bands had a higher R2 with PVC than VIs only using the visible and near infrared spectral bands (Figure 2-9). This was mainly due to the strong dry matter absorption (i.e., lignin and cellulose absorption) in shortwave bands. It is suggested that shortwave infrared spectral data could be used for the study of Arctic tundra with large amounts of senesced vegetation. The approach proposed and optimal spectral bands found in this study can be directly applied to map Arctic biochemical variables (e.g., chlorophyll/carotenoid concentrations, water, dry matter and nitrogen content) over large scales by using satellite remote sensing data. This can aid our understanding of the physiological processes of Arctic vegetation such as photosynthesis, respiration and transpiration, as well as their physiological and biochemical responses to climate warming. The forthcoming hyperspectral satellite missions such as the Environmental Mapping and Analysis Program (EnMAP, 30-m spatial resolution and 244 spectral bands from 420 to 2450 nm) and the Hyperspectral Environment and Resource Observer 113 (HERO, 30-m spatial resolution and 240 spectral bands from 420 and 2450 nm) (Rogge et al., 2014) make the routine mapping of Arctic vegetation biochemicals feasible. 5.2. Objective 2: To model Arctic PVC at landscape scales using field digital images and high- resolution satellite data. Field digital image classification was an efficient way to derive field measures of Arctic PVC. The object- based image analysis (OBIA) approach used in Chapter 3 was successful in distinguishing Arctic plant growth forms with different shapes (e.g., graminoids and forbs). Compared to pixel-based classification methods (e.g., K-means), the OBIA considered both the spectral and geometric characteristics of objects/surfaces, which improved classification accuracy. In addition, the OBIA was found to be more efficient than pixel-based methods due to the less classification samples (i.e., the number of segmented polygons was much less than that of image pixels). At the plot scale, the green normalized difference vegetation index (i.e., GNDVI; (RNIR − RGreen)/(RNIR + 2 RGreen)), derived from field digital images, was found to be a strong predictor of PVC (R = 0.81) (Table 3- 3). At the satellite level, WorldView-2 derived normalized difference spectral indices (NDSI; (Rx− Ry)/(Rx + Ry)), where Rx is the reflectance of the red edge (724.1 nm) or near infrared (832.9 nm and 949.3 nm) bands; Ry is the reflectance of the yellow (607.7 nm) or red (658.8 nm) bands, also exhibited strong correlations with field PVC with R2’s ranging from 0.74 to 0.81 (Table 3-3). Specifically, NDSIs that incorporated the yellow band (607.7 nm) had a slightly higher R2 than the NDSIs without. It is suggested that the yellow band may be more useful for investigating Arctic vegetation with large proportions of senesced vegetation. It should be noted that the point-frame method and image classification method have their own advantages and disadvantages. The point-frame method can obtain Arctic vegetation biodiversity. However, its laborious nature makes plots spatially limited. Therefore, measurements are often made at fine scales (e.g., 6 m in this study). Due to the high spatial heterogeneity of Arctic tundra, PVC changes detected at fine scales may not be representative of those at coarse scales (Phoenix and Bjerke, 2016; Epstein et al. 2016). 114 In contrast, taking field digital images is relatively faster, and therefore allows for greater spatial coverage and more frequent sampling over the growing season. However, it should be calibrated before being compared to the point-frame method. The strong correlation between these two methods in Chapters 3 and 4 suggests a timely and efficient way to obtain point-frame PVC measurements at coarser scales (e.g., 30-m Landsat and 500-m MODIS). First, calibration models should be built by using the PVC measurements derived by two methods within fine- scale plots. Then, field digital images can be taken within coarser-scale plots. Based on calibration models and the PVC derived from field images, we can generate point-frame PVC at coarser scales. This measurement can help us better understand how Artic PVC changes at coarse scales. More importantly, it can be directly correlated with remote sensing VIs at coarse scales. 5.3. Objective 3: To develop models of fAPAR based on relationships between fAPAR and high spatial resolution satellite derived VIs. For Arctic vegetation, PVC was a determining factor for the variation in fAPAR (Tagesson et al. 2012). The ratio of PARReflected to PARIncoming was very small (< 0.05) and showed little change throughout the growing season. In other words, PVC can serve as a surrogate for characterizing fAPAR. In addition, it should be noted that, since the PAR measurements below the canopy were not available, the fraction of PAR absorbed by soil background were not excluded. Therefore, the field measured fAPAR was expected to be higher in this study. This possibly explains why field-based fAPAR was systematically higher than the MODIS fAPAR product when up-scaled to satellite scales. 2 At the plot scale, a strong correspondence between field fAPAR and NDVI was observed (R > 0.74) (Table 4-2). Similar to the PVC-NDSI relationship (Table 3-3), the fAPAR-NDSI relationship at the satellite scale was weaker than that at the plot scale with R2 ranging from 0.64 to 0.66 (Table 4-3). One reason was attributed to atmosphere effects. Atmosphere scattering is additive to remote sensing signals while atmosphere absorption is multiplicative. These two effects could dampen the reflectance contrast between different spectral bands, thereby affecting the correlations between biophysical variables and VIs (Song, 115 2005). The scale discrepancy between plots and satellite pixels could also provide some explanation to this observation. The plot-scale relationship was based on the measurements of biophysical variables and VIs made within the same quadrats. However, when building satellite-scale relationships, biophysical measurements were based on the average of quadrats within plots while VIs were derived at the pixel scale which covered larger areas than quadrats. Since Arctic tundra is extremely heterogeneous even at small scales (Juszak et al., 2017), this discrepancy in spatial scales may affect the relationship between biophysical variables and VIs. Measuring fAPAR for Arctic vegetation is still difficult due to their low stature. Since no PAR measurements were made below canopies, the contribution of soil absorption is still unknown in this study. In the future, radiative transfer simulation models are suggested to estimate ‘true’ fAPAR for Arctic vegetation (Fensholt et al., 2004; Juszak et al., 2014). However, it requires the measurement of more parameters to describe the transfer of photons within canopies. In a study by Fensholt et al. (2004), canopy albedo, soil background albedo (after vegetation being removed) and canopy leaf area index (LAI) were used as the inputs into a simplified radiative transfer model to calculate fAPAR. This method may be feasible for estimating Arctic fAPAR. Canopy and soil background albedo can be measured by PAR sensors in the way described in Chapter 4 while canopy LAI can be obtained by the point-frame method used in Chapters 2 and 4. 5.4. Conclusion Mapping the spatial distribution of Arctic biophysical variables is an important step towards understanding Arctic vegetation processes (e.g., energy exchanges) over different spatial scales. Results from Chapters 2- 4 provide strong evidence that high spectral and spatial resolution remote sensing data have great promise for modelling Arctic biophysical variables. Although only two variables (i.e., PVC and fAPAR) were modelled in this study, it is anticipated that the fine spectral and spatial information provided by these remote sensing data would be useful for modelling other biophysical/biochemical variables. Furthermore, the modelling methods and results presented in this research provided a guide for future Arctic studies. 116 5.5. References Atkinson, D.M., Treitz, P.M., 2013. Modeling biophysical variables across an Arctic latitudinal gradient using high spatial resolution remote sensing data. Arctic, Antarctic, and Alpine Research 45(2), 161– 178. doi:10.1657/1938-4246-45.2.161 Bratsch, S.N., Epstein, H.E., Buchhorn, M., Walker, D.A., 2016. Differentiating among four Arctic tundra plant communities at Ivotuk, Alaska using field spectroscopy. Remote Sensing 8(1). doi:10.3390/rs8010051 Buchhorn, M., Walker, D.A., Heim, B., Raynolds, M.K., Epstein, H.E., Schwieder, M., 2013. Ground- based hyperspectral characterization of Alaska tundra vegetation along environmental gradients. Remote Sensing 5(8), 3971–4005. doi:10.3390/rs5083971 Elmendorf, S.C., Henry, G.H.R., Hollister, R.D., Björk, R.G., Boulanger-Lapointe, N., Cooper, E.J., Cornelissen, J.H.C., Day, T.A., Dorrepaal, E., Elumeeva, T.G., Gill, M., Gould, W.A., Harte, J., Hik, D.S., Hofgaard, A., Johnson, D.R., Johnstone, J.F., Jónsdóttir, I.S., Jorgenson, J.C., Klanderud, K., Klein, J.A., Koh, S., Kudo, G., Lara, M., Lévesque, E., Magnússon, B., May, J.L., Mercado-Dı´az, J.A., Michelsen, A., Molau, U., Myers-Smith, I.H., Oberbauer, S.F., Onipchenko, V.G., Rixen, C., Martin Schmidt, N., Shaver, G.R., Spasojevic, M.J., Þórhallsdóttir, Þ.E., Tolvanen, A., Troxler, T., Tweedie, C.E., Villareal, S., Wahren, C.-H., Walker, X., Webber, P.J., Welker, J.M., Wipf, S., 2012. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2(6), 453–457. doi:10.1038/nclimate1465 Epstein, H.E., Calef, M.P., Walker, M.D., Chapin, F.S., Starfields, A.M., 2004. Detecting changes in arctic tundra plant communities in response to warming over decadal time scales. Global Change Biology 10(8), 1325–1334. doi:10.1111/j.1529-8817.2003.00810.x Epstein, H.E., Bhatt, U.S., Raynolds, M.K., Walker, D.A., Forbes, B.C., Macias-Fauria, M., Loranty, M., Phoenix, G., Bjerke, J., 2016. Tundra Greenness [in Arctic Report Card 2016], http://www.arctic.noaa.gov/Report-Card. Fensholt, R., Sandholt, I., & Rasmussen, M. S. (2004). Evaluation of MODIS LAI, FPAR and the relation between FPAR and NDVI in a semi-arid environment using in situ measurements. Remote Sensing of Environment, 91(3–4), 490–507. doi:http://dx.doi.org/10.1016/j.rse.2004.04.009 Jia, G.J., Epstein, H.E., Walker, D.A., 2009. Vegetation greening in the canadian arctic related to decadal warming. Journal of Environmental Monitoring 11(12), 2231. doi:10.1039/b911677j Jia, G.J., Epstein, H.E., Walker, D.A., 2003. Greening of Arctic Alaska, 1981-2001. Geophysical Research Letters 30(20), n/a-n/a. doi:10.1029/2003GL018268 Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386(6626), 698–702. doi:10.1038/386698a0 Phoenix, G.K. Bjerke, J.W., 2016. Arctic browning: extreme events and trends reversing arctic greeing, Global Change Biology 22, 2960-2962. doi: 10.1111/gcb.13261 Song, C., 2005. Spectral mixture analysis for subpixel vegetation fractions in the urban environment: How to incorporate endmember variability? Remote Sensing of Environment 95(2), 248–263. doi:10.1016/j.rse.2005.01.002 Stow, D.A., Hope, A., McGuire, D., Verbyla, D., Gamon, J., Huemmrich, F., Houston, S., Racine, C., Sturm, M., Tape, K., Hinzman, L., Yoshikawa, K., Tweedie, C., Noyle, B., Silapaswan, C., Douglas, D., Griffith, B., Jia, G., Epstein, H., Walker, D., Daeschner, S., Petersen, A., Zhou, L., Myneni, R., 2004. 117 Remote sensing of vegetation and land-cover change in Arctic tundra ecosystems. Remote Sensing of Environment 89(3), 281–308. doi:10.1016/j.rse.2003.10.018 Tagesson, T., Mastepanov, M., Tamstorf, M.P., Eklundh, L., Schubert, P., Ekberg, A., Sigsgaard, C., Christensen, T.R., Ström, L., 2012. High-resolution satellite data reveal an increase in peak growing season gross primary production in a high-Arctic wet tundra ecosystem 1992–2008. International Journal of Applied Earth Observation and Geoinformation 18, 407–416. doi:10.1016/j.jag.2012.03.016 Walker, M.D., Wahren, C.H., Hollister, R.D., Henry, G.H.R., Ahlquist, L.E., Alatalo, J.M., Bret-Harte, M.S., Calef, M.P., Callaghan, T. V., Carroll, A.B., Epstein, H.E., Jonsdottir, I.S., Klein, J.A., Magnusson, B., Molau, U., Oberbauer, S.F., Rewa, S.P., Robinson, C.H., Shaver, G.R., Suding, K.N., Thompson, C.C., Tolvanen, A., Totland, O., Turner, P.L., Tweedie, C.E., Webber, P.J., Wookey, P.A., 2006. Plant community responses to experimental warming across the tundra biome. Proceedings of the National Academy of Sciences 103(5), 1342–1346. doi:10.1073/pnas.0503198103 118 Appendix A Sabine Peninsula Site Table A- 1. Plot information of the Sabine Peninsula site. PD: Polar semi-desert; DMT: Dry Mesic Tundra; MT: Mesic Tundra; WMT: Wet Mesic Tundra; WSM: Wet Sedge/Moss Plot Class Date Latitude Longitude Kk-3-5 DMT 7/26/2011 76.5397 -108.99 Kk-2-7 DMT 7/9/2011 76.6044 -108.89 Kk-4-4 DMT 7/16/2011 76.5683 -108.86 Kh-3-9 DMT 7/8/2011 76.497 -108.91 Kh-2-6 DMT 7/13/2011 76.5069 -108.79 Kk-1-6 MT 7/18/2011 76.5869 -109.02 Kh-3-5 MT 7/8/2011 76.4901 -108.9 Kh-2-2 MT 7/14/2011 76.5127 -108.78 Kc-2-4 MT 7/10/2011 76.4812 -108.79 Kc-2-7 MT 7/10/2011 76.4853 -108.79 Kk-2-11 PD 7/9/2011 76.609 -108.89 Kk-3-6 PD 7/26/2011 76.54 -108.97 Kk-3-7 PD 7/26/2011 76.5439 -109.01 Kh-2-7 PD 7/14/2011 76.5124 -108.77 Kc-1-10 PD 7/17/2011 76.4584 -108.9 Kk-4-1 WMT 7/16/2011 76.57 -108.87 X-8 WMT 7/18/2011 76.6027 -108.99 Kk-1-5 WMT 7/18/2011 76.5832 -108.98 Kh-3-2 WMT 7/8/2011 76.4935 -108.91 Kh-2-1 WMT 7/14/2011 76.505 -108.8 Kc-2-2 WMT 7/10/2011 76.4838 -108.8 Kc-1-3 WMT 7/17/2011 76.4593 -108.92 Kc-1-4 WMT 7/17/2011 76.4635 -108.86 X-10 WSM 7/9/2011 76.5935 -108.86 Kk-4-11 WSM 7/16/2011 76.575 -108.86 Kk-2-5 WSM 7/19/2011 76.608 -108.86 Kk-2-6 WSM 7/19/2011 76.6101 -108.87 Kk-4-3 WSM 7/16/2011 76.5728 -108.85 Kh-2-3 WSM 7/14/2011 76.5096 -108.81 Kc-1-12 WSM 7/17/2011 76.4571 -108.89 Kk-2-9 DMT 12/07/2011 76.6033 -108.87 Kk-1-9 DMT 30/07/2011 76.5914 -109.02 Kc-2-8 DMT 27/07/2011 76.4784 -108.79 Kc-1-6 MT 31/07/2011 76.4636 -108.92 Kh-1-11 PD 28/07/2011 76.5289 -108.87 Kc-2-10 PD 27/07/2011 76.4789 -108.8 119 Kk-1-12 WMT 30/07/2011 76.5873 -109 X-3 WMT 31/07/2011 76.4628 -108.83 Kh-1-1 WMT 28/07/2011 76.5407 -108.82 Kc-2-3 WMT 27/07/2011 76.4668 -108.79 Kc-1-13 WSM 31/07/2011 76.457 -108.89 Kk-1-8 WSM 30/07/2011 76.5918 -109.01 Kh-1-7 WSM 28/07/2011 76.5274 -108.83 Kh-1-8 WSM 28/07/2011 76.5301 -108.84 X-1 WSM 30/07/2011 76.5948 -108.98 New02 PD 30/07/2011 76.5502 -108.81 New03Act WSM 04/08/2011 76.5405 -108.64 New04Act DMT 03/08/2011 76.5431 -108.63 New01 WSM 30/07/2011 76.5851 -108.95 120 Table A- 2. PVC2D and PVC3D data used in Figures 2-3 and 2-4. Grey: Dry mesic tundra; Blue: Mesic tundra; Orange: Polar semi-desert; Green: Wet mesic tundra; Brown: Wet sedge/moss; Italics font: PVC2D; Non-Italics font: PVC3D. BS: bare soil; GF: green forb; DF: dead forb; GG: green graminoids; DG: dead graminoids; LC: lichens; GM: green moss; SM: senesced moss. H: canopy height. Plot BS GF DF GG DG LC GM SM BS GF DF GG DG LC GM SM H Kc-2-8 53 30 14 1 2 0 0 0 95 41 28 1 3 0 0 0 1.64 Kh-2-6 68 7 1 0 0 12 3 9 69 7 1 0 0 12 4 9 0.06 Kk-1-9 52 20 12 3 5 0 7 1 75 24 16 3 5 1 20 1 0.585 Kk-2-7 39 7 15 7 12 3 12 5 52 7 17 8 13 3 18 6 0.495 Kk-2-9 49 13 16 4 13 1 4 0 57 16 23 5 13 1 8 0 0.59 Kk-3-5 50 27 18 2 0 1 2 0 80 39 30 4 1 1 7 0 1.065 Kk-4-4 52 12 31 0 0 0 3 2 68 16 44 0 0 0 10 2 0.745 Kc-2-4 42 25 7 0 0 0 26 0 42 31 12 0 0 0 26 0 0.17 Kc-2-7 22 17 7 8 10 6 23 7 28 19 8 9 11 7 36 7 0.465 Kh-2-2 18 13 14 17 12 10 13 3 27 14 19 31 17 14 30 4 0.995 Kh-3-5 46 17 0 0 0 3 29 5 46 39 0 0 0 4 36 5 0.3 Kc-1-10 87 9 4 0 0 0 0 0 95 12 5 0 0 0 0 0 0.145 Kc-2-10 72 0 0 22 6 0 0 0 93 0 0 24 9 0 0 0 0.32 Kh-1-11 100 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 Kh-2-7 89 2 0 2 5 1 0 1 93 2 0 2 5 1 0 1 0.035 Kk-2-11 100 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 Kk-3-6 91 7 2 0 0 0 0 0 100 7 4 0 0 0 0 0 0.205 Kk-3-7 74 13 12 0 0 0 1 0 98 13 12 0 0 0 1 0 2.501 New-2 70 0 0 0 0 9 16 5 87 0 0 0 0 11 19 13 0.48 New-4 62 0 0 9 6 18 4 1 80 0 0 19 16 26 5 1 0.78 Kc-1-3 15 26 21 6 3 4 19 6 23 34 29 8 5 6 47 17 1.515 Kc-1-4 4 29 15 24 12 5 6 5 14 38 28 44 24 13 52 22 2.805 Kc-2-2 9 13 7 13 15 5 28 10 19 21 11 17 21 6 57 25 1.09 Kc-2-3 23 22 11 15 18 4 7 0 61 28 18 20 37 7 25 1 2.155 Kh-2-1 18 12 15 2 8 3 21 21 20 13 15 3 8 3 23 25 0.15 Kh-3-2 22 15 3 0 0 0 55 5 22 29 3 0 0 0 62 8 0.28 Kk-1-12 22 42 15 11 2 1 7 0 50 61 28 12 4 4 44 2 1.86 Kk-1-5 23 31 9 6 0 5 26 0 36 38 14 9 1 6 48 0 0.905 Kk-2-5 2 42 39 3 3 1 10 0 8 69 113 5 9 4 78 0 3.701 Kk-4-1 6 12 22 0 0 22 24 14 6 16 28 1 0 32 41 17 0.855 New-1 1 33 10 5 22 10 4 15 1 49 17 12 28 21 6 50 1.555 X-3 2 11 19 5 4 3 44 12 6 15 21 6 7 4 79 18 1.015 X-8 2 31 1 2 4 7 53 0 2 34 2 2 4 9 70 0 0.265 Kc-1-12 0 0 0 76 19 0 5 0 1 0 0 137 86 0 75 0 4.39 Kc-1-13 0 4 2 58 33 0 2 1 4 15 2 97 106 3 93 5 6.44 Kh-1-1 17 6 3 6 5 19 43 1 31 7 4 9 12 25 66 2 0.83 Kh-1-7 10 17 18 10 3 16 25 1 18 26 22 19 3 25 59 4 1.18 Kh-1-8 15 18 13 10 3 11 27 3 35 23 13 16 3 17 51 4 1.45 121 Kh-2-3 5 9 5 12 9 15 37 8 9 14 7 14 11 18 50 10 0.715 Kh-3-8 46 16 0 0 0 6 16 16 46 18 0 0 0 6 16 16 0.02 Kk-1-8 34 30 13 3 1 1 18 0 54 40 17 4 1 2 39 3 0.955 Kk-2-6 3 24 36 10 19 1 6 1 10 53 59 19 26 4 27 10 2.13 Kk-4-11 4 71 0 3 1 0 21 0 0 77 28 5 1 0 74 0 1.985 Kk-4-3 15 11 20 12 15 6 20 1 26 15 27 15 18 9 37 3 0.925 New-3 0 14 4 49 30 2 1 0 23 24 13 79 81 8 80 0 3.295 X-10 0 30 35 2 6 0 20 7 2 37 50 4 8 3 52 9 1.255 122 Appendix B Cape Bounty Arctic Watershed Observatory (CBAWO) Site Table B- 1. Plot Information (UTM-12N projection, WGS-84 datum). Site Plot # Map X (m) Map Y (m) 1 31 542040 8314224 1 36 542012 8314090 1 37 542038 8314120 1 38 542042 8314244 1 41 542018 8314208 1 42 541998 8314224 1 43 541970 8314246 1 44 541916 8314108 1 44 541918 8314110 1 45 541940 8314236 1 45 541938 8314216 1 46 541894 8314218 1 46 541894 8314218 1 47 541894 8314128 1 47 541894 8314130 1 48 541960 8314186 1 49 541826 8314158 1 49 541822 8314154 1 50 541820 8314238 1 50 541822 8314238 1 51 541806 8314196 1 51 541806 8314196 1 52 541824 8314104 1 53 541792 8314154 1 53 541794 8314154 1 54 541804 8314232 1 55 541896 8314078 1 56 541862 8314112 2 01 541358 8314258 2 04 541326 8314278 2 05 541344 8314264 2 06 541378 8314324 2 06N1 541382 8314348 2 06N2 541392 8314360 2 07 541388 8314272 2 08 541276 8314332 2 09 541350 8314316 123 2 10 541394 8314310 2 11 541324 8314322 2 12 541360 8314314 2 14 541230 8314346 2 15 541326 8314350 2 16 541402 8314294 2 20 541380 8314282 2 22 541358 8314304 2 23 541294 8314296 2 24 541300 8314346 2 25 541394 8314324 2 26 541252 8314346 2 28 541384 8314310 2 29 541342 8314242 2 30 541350 8314344 2 32 541376 8314274 2 33 541260 8314364 2 34 541298 8314370 2 35 541364 8314340 2 37 541394 8314288 2 38 541278 8314302 2 39 541278 8314370 2 43 541326 8314298 2 44 541230 8314328 2 46 541276 8314344 2 49 541290 8314240 2 50 541302 8314262 2 55 541292 8314272 2 56 541270 8314264 2 59 541224 8314226 2 60 541286 8314282 2 63 541216 8314178 2 66 541230 8314210 2 67 541218 8314186 2 68 541140 8314130 2 68 541140 8314130 2 69 541148 8314346 2 70 541146 8314374 2 73 541228 8314090 2 77 541250 8314222 2 79 541222 8314148 2 79 541282 8314126 2 79 541222 8314148 124 2 80 541280 8314128 2 80 541308 8314138 2 80 541280 8314128 2 81 541216 8314178 2 81 541342 8314134 2 81 541216 8314178 2 83 541390 8314108 2 83 541404 8314082 2 83 541390 8314108 2 84 541312 8314102 2 84 541392 8314106 2 84 541312 8314102 2 85 541270 8314114 2 85 541270 8314114 3 02 540822 8314548 3 12 540888 8314652 3 14 540834 8314598 3 15 540846 8314608 3 16 540950 8314626 3 18 540994 8314662 3 19 540986 8314676 3 21 540970 8314672 3 22 540988 8314628 3 26 540962 8314614 3 27 540978 8314654 3 29 540982 8314690 3 30 540944 8314614 3 31 540888 8314620 3 32 540902 8314602 3 34 540924 8314650 3 35 540816 8314606 3 36 540762 8314648 3 42 540822 8314536 3 43 540786 8314516 3 44 540742 8314508 3 45 540782 8314688 3 46 540834 8314690 3 47 540794 8314636 3 50 540916 8314584 3 50 540908 8314578 3 52 540972 8314466 3 52 540978 8314472 3 53 540932 8314464 125 3 53 540938 8314468 3 54 540958 8314492 3 54 540956 8314486 3 55 540868 8314480 3 55 540876 8314482 3 N1 540862 8314568 3 N2 540754 8314474 126 Table B- 2. PVC (%) data used for Figure 3-4. Orange: Mesic tundra; Brown: Polar semi-desert; Green: Wet sedge; Italics font: PVC derived from OBIA; Non-Italics font: PVC derived from point- frame method. FB: Forb; GG: Green Graminoids; SG: Senesced Graminoids; WL: Willows; GM: Green Moss; SM: Senesced Moss; SL: Soil. FB GG SG WL GM SM SL FB GG SG WL GM SM SL 2.22 0.00 16.76 0.00 37.15 39.64 4.22 3.29 0.66 12.50 0.00 34.87 40.13 8.55 0.38 0.00 2.58 0.00 30.54 66.50 0.00 0.69 0.00 6.21 0.00 31.03 61.38 0.69 0.14 0.00 4.53 0.00 4.23 90.78 0.31 2.14 0.71 5.71 0.00 10.71 80.00 0.71 0.48 0.59 4.54 0.00 0.00 33.71 60.67 0.74 0.00 5.19 0.00 0.00 39.26 54.81 2.37 5.42 24.42 0.00 27.65 40.14 0.00 1.88 11.88 21.88 0.00 33.13 30.00 0.00 0.88 5.82 17.86 0.00 2.30 27.87 45.26 1.42 9.93 20.57 0.71 6.38 10.64 50.35 4.20 6.50 48.05 0.00 1.98 32.69 6.58 5.26 6.77 33.83 0.00 2.26 43.61 8.27 1.73 7.82 18.18 0.00 3.91 39.59 28.78 1.39 5.56 28.47 0.00 13.19 19.44 31.94 1.82 8.07 22.82 0.00 27.21 24.33 15.74 2.52 7.55 23.90 0.00 27.67 16.35 22.01 0.14 9.38 44.11 2.48 0.00 0.00 43.90 1.43 8.57 50.71 2.86 0.00 0.00 36.43 0.36 15.95 46.50 0.00 31.55 0.00 0.00 0.74 14.71 48.53 0.00 28.68 7.35 0.00 0.00 19.81 19.13 0.00 44.77 16.28 0.00 0.00 12.42 26.14 0.00 43.14 18.30 0.00 1.34 0.00 8.47 0.00 1.98 67.80 20.41 2.94 3.68 15.44 0.00 8.09 27.94 41.91 1.17 0.00 13.69 0.00 61.86 17.59 5.69 2.14 3.57 7.86 0.00 43.57 39.29 3.57 0.53 0.00 0.53 0.00 1.58 0.00 97.35 0.63 0.00 0.63 0.00 0.00 4.40 94.34 6.93 0.00 6.46 0.00 7.21 19.00 60.41 12.41 3.65 5.84 0.00 3.65 43.07 31.39 0.32 0.00 14.48 0.00 15.06 0.00 70.14 1.88 1.25 9.38 0.00 8.75 5.63 73.13 0.00 0.00 15.03 0.00 50.54 12.82 21.61 1.31 5.88 7.19 0.65 38.56 27.45 18.95 1.99 0.00 81.02 0.00 0.00 4.57 12.42 2.01 7.38 53.02 0.00 5.37 0.67 31.54 0.00 0.00 0.00 0.00 0.00 40.06 59.94 0.00 0.00 0.65 0.00 0.00 44.52 54.84 0.00 0.00 0.00 0.00 0.00 7.30 92.70 0.00 0.00 0.00 0.00 0.68 13.51 85.81 0.00 0.00 0.00 0.00 0.00 28.55 71.45 0.00 0.00 0.00 0.00 0.00 32.87 67.13 0.00 0.00 0.00 0.00 0.00 9.22 90.78 0.00 0.00 0.00 0.00 0.00 10.06 89.94 0.00 0.00 0.00 0.00 0.00 2.83 97.17 0.00 0.00 0.00 0.00 0.00 7.80 92.20 0.00 0.00 1.81 8.88 0.00 5.48 83.82 0.00 0.71 7.14 13.57 2.14 7.14 69.29 0.00 0.00 0.00 0.00 0.00 0.00 100.00 2.50 1.25 1.25 0.00 0.00 0.00 95.00 0.00 0.00 0.00 0.00 0.00 75.62 24.38 0.66 0.00 1.32 0.00 7.95 69.54 20.53 0.96 0.00 0.00 0.00 0.00 74.43 24.61 0.71 0.00 2.14 0.00 2.14 60.71 34.29 2.54 0.00 1.94 0.00 0.00 56.29 39.23 4.40 2.52 2.52 0.00 1.26 61.01 28.30 0.92 0.13 0.00 0.00 0.00 4.64 94.31 2.16 0.00 0.00 0.00 0.00 4.32 93.53 0.54 0.17 0.26 0.00 0.00 0.00 99.04 0.71 0.00 0.00 0.00 0.00 0.00 99.29 0.00 0.19 0.62 0.00 0.00 8.57 90.63 0.71 0.71 0.00 0.00 0.00 12.86 85.71 1.01 0.23 0.68 0.00 0.00 0.00 98.08 0.64 0.64 1.28 0.00 0.00 1.28 96.15 0.00 0.24 2.53 0.00 2.61 69.50 25.12 0.00 0.00 4.46 0.00 13.38 50.32 31.85 20.99 0.53 4.39 0.00 0.00 3.42 70.68 30.00 1.43 5.71 0.00 1.43 2.86 58.57 2.22 0.90 7.60 0.00 0.54 0.00 88.74 5.00 2.86 5.00 0.00 0.00 22.14 65.00 0.00 1.45 30.46 0.00 2.61 18.66 46.83 3.92 13.73 19.61 1.31 9.15 18.95 33.33 3.58 1.51 5.63 0.00 19.12 14.54 55.62 4.61 2.63 5.26 0.00 23.68 17.11 46.71 1.17 1.64 3.84 0.00 0.00 0.00 93.34 2.88 2.88 9.35 0.00 0.00 0.00 84.89 1.18 1.82 5.60 0.00 1.67 0.00 89.72 0.63 1.88 6.25 0.00 1.25 3.75 86.25 127 0.81 3.05 2.04 0.00 4.36 69.83 19.90 0.63 2.50 3.75 0.00 9.38 61.88 21.88 2.65 4.01 3.64 0.00 0.00 0.00 89.70 2.53 1.90 6.96 0.00 0.63 3.80 84.18 0.00 4.45 8.06 0.00 0.00 0.00 87.50 0.00 8.13 13.13 0.00 0.00 0.00 78.75 0.55 11.11 47.92 1.37 0.00 0.00 37.66 1.25 7.50 56.88 3.13 0.00 0.00 31.25 0.00 1.59 73.59 0.00 20.83 0.00 0.00 0.00 4.96 78.01 0.00 14.18 0.00 0.00 0.00 4.48 22.96 3.28 0.00 0.00 69.28 1.25 6.25 20.63 6.25 2.50 7.50 55.63 0.53 4.55 22.77 2.40 13.97 55.78 0.00 2.17 7.25 26.09 1.45 17.39 45.65 0.00 0.00 4.79 73.32 0.00 18.47 0.00 0.00 0.00 5.52 76.55 0.00 16.55 0.00 0.00 0.00 5.78 82.55 0.00 4.02 0.00 0.00 0.00 6.25 85.00 0.00 4.38 0.00 0.00 0.00 7.00 6.57 9.48 61.46 0.00 15.50 0.00 8.75 12.50 8.75 53.13 0.63 16.25 0.00 7.61 66.94 0.00 24.07 0.00 0.00 0.00 5.73 68.79 0.00 22.29 0.00 0.00 0.00 8.31 60.42 0.00 26.23 0.00 0.00 0.00 8.75 63.13 0.00 21.88 5.00 0.00 0.00 9.01 34.93 0.00 56.06 0.00 0.00 0.00 8.86 39.87 0.00 46.20 0.00 0.00 0.05 9.81 69.53 0.46 9.43 0.00 0.00 1.86 13.66 70.19 0.00 2.48 0.00 0.00 0.00 10.70 63.94 0.00 22.68 0.00 0.00 0.00 11.49 73.65 0.00 14.19 0.00 0.00 0.00 11.19 61.61 0.00 23.75 0.00 0.00 0.00 13.57 55.71 0.00 25.00 0.00 1.43 0.00 11.22 86.07 0.00 0.00 0.00 0.00 0.00 15.72 84.28 0.00 0.00 0.00 0.00 0.00 11.71 69.34 0.00 14.35 0.00 0.00 0.00 6.49 79.87 0.00 12.99 0.00 0.00 0.00 11.88 57.70 0.00 28.23 0.00 0.00 0.00 6.12 67.35 0.00 23.81 0.00 0.00 0.00 12.26 65.16 0.00 19.52 0.00 0.00 0.75 10.53 77.44 0.00 9.77 0.00 0.00 0.00 12.33 52.39 0.00 35.28 0.00 0.00 0.00 12.50 51.88 0.00 35.00 0.00 0.00 0.00 12.48 26.86 6.79 30.45 0.00 23.42 0.00 13.13 33.75 10.00 15.00 1.88 26.25 0.00 12.88 63.40 0.00 18.76 0.00 0.00 0.00 13.75 70.00 0.00 11.88 0.00 0.00 0.00 13.09 56.60 0.00 26.98 0.00 0.00 0.00 12.58 57.86 0.00 28.93 0.00 0.00 0.00 13.56 69.73 0.00 13.51 0.00 0.00 0.00 20.00 65.63 0.00 10.63 0.00 0.00 0.00 13.86 42.16 0.00 43.98 0.00 0.00 0.00 13.21 48.43 0.00 33.33 0.00 0.00 0.00 14.35 51.27 0.00 34.39 0.00 0.00 0.00 17.11 60.53 0.00 17.11 0.00 0.00 0.00 15.23 51.20 0.00 31.48 0.00 0.00 0.00 13.07 55.56 0.00 30.72 0.00 0.00 0.00 16.14 52.27 0.00 31.59 0.00 0.00 0.71 13.57 61.43 0.00 22.86 0.00 0.00 0.04 16.61 52.41 0.00 24.00 0.00 0.00 0.00 17.50 60.00 0.00 22.50 0.00 0.00 0.00 17.90 53.66 0.00 20.57 0.00 0.00 0.00 18.24 64.78 0.00 9.43 0.00 0.00 1.50 17.95 61.70 0.00 16.55 0.00 0.00 0.00 15.58 55.19 1.30 26.62 0.00 0.00 0.00 22.65 53.80 0.00 16.47 0.00 0.00 0.00 30.26 61.18 0.00 8.55 0.00 0.00 0.00 23.22 46.05 0.00 28.29 0.00 0.00 0.00 17.61 63.52 0.00 16.35 0.00 0.00 0.00 26.55 67.94 0.00 0.00 0.00 0.00 0.00 20.63 75.00 0.00 0.63 0.00 0.00 0.00 26.67 54.82 0.00 9.46 0.00 0.00 0.00 24.20 66.24 0.00 0.00 0.00 0.00 0.00 27.24 43.46 0.00 19.25 0.00 0.00 0.00 20.00 71.43 0.00 6.43 0.00 0.00 0.00 27.51 50.75 0.00 16.95 0.00 0.00 0.00 29.38 53.75 0.00 14.38 0.63 0.00 0.00 27.92 46.01 0.00 18.10 0.00 0.00 0.00 33.33 58.82 0.00 0.00 0.00 0.00 0.00 29.94 65.21 0.00 0.00 0.00 0.00 0.00 24.53 70.44 0.00 0.00 0.00 0.00 0.00 30.77 40.80 0.00 26.05 0.00 0.00 0.00 23.33 57.33 0.00 18.67 0.00 0.00 0.00 34.71 48.98 0.00 7.75 0.00 0.00 0.00 14.67 68.67 0.00 10.67 0.00 2.00 0.00 36.05 54.49 0.00 0.00 0.00 0.00 0.00 45.83 46.53 0.00 1.39 0.00 0.00 128 Table B- 3. PVC (%) data and WorldView-2 spectra used for Table 3-3; Yellow: PVC data acquired on July 9, 2014; Blue: PVC data acquired July 10, 2014; Green: PVC data acquired July 11, 2014; Brown: PVC data acquired July 12, 2014; WorldView-2 spectra acquired on July 12, 2014. Site Plot # PVC Coastal Blue Green Yellow Red Red Edge NIR1 NIR2 1 31 26.87 0.12 0.12 0.12 0.12 0.11 0.17 0.21 0.24 1 36 13.53 0.15 0.17 0.18 0.19 0.19 0.24 0.28 0.29 1 37 9.63 0.14 0.15 0.15 0.15 0.15 0.20 0.24 0.26 1 38 30.73 0.12 0.12 0.11 0.11 0.10 0.14 0.18 0.21 1 42 42.37 0.11 0.11 0.11 0.11 0.11 0.16 0.20 0.21 1 43 45.16 0.11 0.10 0.09 0.09 0.08 0.13 0.17 0.19 1 45 37.64 0.12 0.13 0.14 0.15 0.16 0.23 0.29 0.32 1 47 26.96 0.14 0.16 0.18 0.20 0.21 0.30 0.38 0.42 1 48 19.66 0.13 0.13 0.14 0.15 0.15 0.22 0.29 0.31 1 49 23.75 0.14 0.15 0.16 0.17 0.17 0.24 0.30 0.34 1 50 29.18 0.12 0.13 0.15 0.17 0.18 0.28 0.37 0.41 1 51 37.84 0.13 0.15 0.17 0.20 0.21 0.29 0.38 0.41 1 53 38.85 0.14 0.15 0.17 0.19 0.20 0.29 0.35 0.39 1 54 51.51 0.12 0.13 0.15 0.16 0.17 0.25 0.32 0.35 1 55 12.88 0.14 0.15 0.18 0.20 0.21 0.32 0.43 0.47 1 56 25.26 0.14 0.16 0.18 0.20 0.21 0.29 0.38 0.41 2 9 0.27 0.16 0.18 0.19 0.19 0.18 0.19 0.20 0.22 2 10 7.61 0.17 0.19 0.20 0.21 0.20 0.21 0.22 0.23 2 11 0.05 0.17 0.19 0.20 0.21 0.20 0.20 0.20 0.20 2 12 2.53 0.17 0.19 0.19 0.19 0.19 0.20 0.21 0.22 2 14 3.77 0.15 0.16 0.17 0.17 0.17 0.18 0.20 0.22 2 15 2.49 0.15 0.17 0.17 0.18 0.17 0.18 0.19 0.19 2 22 0.58 0.16 0.18 0.18 0.19 0.18 0.19 0.20 0.22 2 24 5.93 0.14 0.15 0.16 0.16 0.15 0.16 0.15 0.16 2 26 2.90 0.16 0.18 0.19 0.20 0.20 0.21 0.23 0.23 2 28 2.79 0.18 0.20 0.22 0.23 0.23 0.23 0.24 0.24 2 34 0.21 0.15 0.17 0.19 0.20 0.19 0.21 0.23 0.24 2 4 7.63 0.16 0.17 0.17 0.17 0.16 0.19 0.20 0.22 2 5 1.28 0.14 0.15 0.15 0.15 0.14 0.16 0.17 0.19 2 7 5.35 0.17 0.19 0.19 0.20 0.20 0.21 0.22 0.23 2 20 1.98 0.17 0.18 0.19 0.19 0.19 0.20 0.22 0.24 2 23 3.07 0.15 0.17 0.18 0.18 0.17 0.19 0.21 0.22 2 29 5.39 0.13 0.14 0.13 0.13 0.12 0.13 0.13 0.14 2 79 25.93 0.12 0.12 0.11 0.11 0.11 0.18 0.25 0.29 2 80 21.22 0.13 0.13 0.14 0.14 0.15 0.24 0.33 0.35 2 81 20.54 0.13 0.14 0.15 0.16 0.16 0.27 0.36 0.41 2 83 42.43 0.14 0.14 0.14 0.15 0.14 0.22 0.28 0.31 2 84 23.08 0.13 0.14 0.15 0.17 0.17 0.27 0.34 0.38 3 2 34.27 0.15 0.16 0.17 0.19 0.18 0.25 0.30 0.33 129 3 16 3.80 0.18 0.20 0.22 0.22 0.21 0.23 0.24 0.25 3 21 7.72 0.18 0.21 0.24 0.25 0.24 0.26 0.28 0.27 3 26 4.52 0.16 0.18 0.18 0.19 0.18 0.20 0.23 0.25 3 27 6.75 0.17 0.19 0.21 0.22 0.22 0.25 0.28 0.28 3 31 12.41 0.13 0.13 0.12 0.12 0.12 0.16 0.20 0.23 3 35 6.01 0.18 0.21 0.24 0.25 0.25 0.29 0.32 0.33 3 36 28.13 0.14 0.15 0.15 0.16 0.16 0.21 0.25 0.28 3 42 29.00 0.15 0.17 0.18 0.19 0.18 0.25 0.30 0.33 3 43 24.48 0.15 0.16 0.17 0.18 0.18 0.23 0.28 0.31 3 44 31.04 0.14 0.15 0.15 0.16 0.16 0.21 0.26 0.28 3 45 26.17 0.12 0.11 0.10 0.10 0.10 0.14 0.19 0.23 3 46 44.85 0.12 0.13 0.12 0.12 0.12 0.19 0.25 0.28 3 47 24.58 0.13 0.13 0.13 0.13 0.12 0.18 0.23 0.26 3 50 27.95 0.12 0.12 0.11 0.11 0.11 0.17 0.22 0.26 3 52 43.90 0.13 0.14 0.14 0.15 0.14 0.23 0.30 0.34 3 53 37.05 0.12 0.13 0.13 0.13 0.13 0.22 0.29 0.32 3 54 55.08 0.12 0.12 0.12 0.12 0.12 0.20 0.26 0.28 3 55 46.83 0.14 0.16 0.16 0.18 0.18 0.26 0.33 0.37 130 Appendix C Apex River Watershed (ARW) Site Table C- 1. Plot Information (UTM-19N projection, WGS-84 datum). Site Date X (m) Y(m) Transect1A Plot1 7/23/2015 527241.88 7067664.34 Transect1A Plot2 7/23/2015 527263.46 7067667.60 Transect1A Plot3 7/23/2015 527282.71 7067669.44 Transect1A Plot4 7/23/2015 527299.89 7067677.13 Transect1A Plot5 7/23/2015 527319.60 7067676.68 Transect1B Plot1 7/5/2015 527347.78 7067697.16 Transect1B Plot2 7/5/2015 527377.82 7067684.05 Transect1B Plot3 7/9/2015 527394.15 7067684.49 Transect1B Plot4 7/9/2015 527420.26 7067679.36 Transect1B Plot5 7/8/2015 527446.31 7067674.42 Transect2A Plot1 7/22/2015 527477.99 7069745.68 Transect2A Plot2 7/22/2015 527457.93 7069743.77 Transect2A Plot 3 7/22/2015 527440.74 7069738.21 Transect2A Plot4 7/22/2015 527420.78 7069738.06 Transect2A Plot 5 7/22/2015 527395.53 7069731.7 Transect2B Plot1 7/12/2015 527305.42 7069864.07 Transect2B Plot2 7/12/2015 527320.54 7069847.26 Transect2B Plot3 7/12/2015 527337.75 7069831.87 Transect2B Plot4 7/14/2015 527355.76 7069812.81 Transect2B Plot5 7/14/2015 527370.71 7069797.11 Transect3A Plot1 7/25/2015 526464.04 7070649.74 Transect3A Plot2 7/25/2015 526443.41 7070645.73 Transect3A Plot3 7/25/2015 526421.69 7070643 Transect3A Plot4 7/25/2015 526402.59 7070641.34 Transect3A Plot5 7/25/2015 526385.29 7070641.55 Transect3B Plot 1 7/10/2015 526506.84 7070670.52 Transect3B Plot2 7/10/2015 526525.83 7070662.23 Transect3B Plot3 7/11/2015 526548.03 7070644.98 Transect3B Plot4 7/11/2015 526570.02 7070629.47 Transect3B Plot5 7/11/2015 526587.11 7070612.39 Transect4A Plot1 7/26/2015 527037.15 7070726.73 Transect4A Plot2 7/26/2015 527054.43 7070719.64 Transect4A Plot3 7/26/2015 527071.15 7070712.76 Transect4A Plot4 7/26/2015 527095.21 7070703.26 Transect4A Plot5 7/26/2015 527118.1 7070688.88 131 Transect4B Plot1 7/16/2015 527080.65 7070682.14 Transect4B Plot2 7/16/2015 527090.34 7070693.81 Transect4B Plot3 7/16/2015 527097.07 7070701.85 Transect4B Plot4 7/16/2015 527107.79 7070711.82 Transect4B Plot5 7/16/2015 527127.34 7070748.96 Transect5A Plot1 7/27/2015 525083.58 7073082.87 Transect5A Plot2 7/27/2015 525075.65 7073102.56 Transect5A Plot3 7/27/2015 525061.92 7073123.89 Transect5A Plot4 7/27/2015 525036.23 7073145.23 Transect5A Plot5 7/27/2015 525017.63 7073186.08 Transect5B Plot1 7/17/2015 525117.17 7072814.07 Transect5B Plot2 7/17/2015 525136.23 7072826.48 Transect5B Plot3 7/17/2015 525146.64 7072841.2 Transect5B Plot4 7/17/2015 525177.67 7072879.84 Transect5B Plot5 7/17/2015 525209.8 7072900.26 Transect6A Plot1 7/31/2015 523730.06 7073479.59 Transect6A Plot2 7/31/2015 523745.51 7073465.34 Transect6A Plot3 7/31/2015 523761.98 7073454.59 Transect6A Plot4 7/31/2015 523706.8 7073520.8 Transect6A Plot5 7/31/2015 523716.98 7073502.15 Transect6B Plot1 7/18/2015 523655.57 7073580.53 Transect6B Plot2 7/18/2015 523671.97 7073559.31 Transect6B Plot3 7/18/2015 523686.37 7073547.79 Transect6B Plot4 7/18/2015 523706.8 7073520.8 Transect6B Plot5 7/18/2015 523716.98 7073502.15 Transect7A Plot1 7/20/2015 523918.09 7076098.67 Transect7A Plot2 7/20/2015 523936.52 7076113.11 Transect7A Plot3 7/20/2015 523954.95 7076127.55 Transect7A Plot4 7/20/2015 523980.96 7076146.6 Transect7A Plot5 7/20/2015 524003.14 7076161.87 Transect7B Plot1 7/20/2015 524068.6 7076187.64 Transect7B Plot2 7/20/2015 524083.28 7076171.82 Transect7B Plot3 7/20/2015 524095.86 7076158.63 Transect7B Plot4 7/20/2015 524143.96 7076119.47 Transect7B Plot5 7/20/2015 524143.96 7076119.47 PAR1 7/5/2015 526755.54 7070074.26 PAR1 7/21/2015 526755.54 7070074.26 PAR1 8/3/2015 526755.54 7070074.26 PAR1 8/8/2015 526755.54 7070074.26 PAR2 7/5/2015 526823.78 7069956.68 PAR2 7/20/2015 526823.78 7069956.68 PAR2 8/3/2015 526823.78 7069956.68 PAR2 8/8/2015 526823.78 7069956.68 132 PAR3 7/5/2015 526876.12 7069910.53 PAR3 7/20/2015 526876.12 7069910.53 PAR3 8/3/2015 526876.12 7069910.53 PAR3 8/8/2015 526876.12 7069910.53 PAR4 7/5/2015 526900.93 7069861.34 PAR4 7/20/2015 526900.93 7069861.34 PAR4 8/3/2015 526900.93 7069861.34 PAR4 8/8/2015 526900.93 7069861.34 PAR 5 7/20/2015 526910.48 7069851.15 PAR 5 8/3/2015 526910.48 7069851.15 PAR 5 8/8/2015 526910.48 7069851.15 PAR6 7/20/2015 526921.36 7069834.37 PAR6 8/3/2015 526921.36 7069834.37 PAR6 8/8/2015 526921.36 7069834.37 133 Table C- 2. 1-PARReflected/PARIncident; PVCImg: PVC derived from image classification; Number of contacts with green vegetation in 2D and 3D cases (PVC2D=Contacts/25; PVC3D=Contacts/25, each quadrat has 5 x 5 = 25 grids); Height: canopy height. Field NDVI (some NDVI images are not usable due to under-exposure). Q1-Q4: Quadrat number. fAPAR=(1-PARReflected/PARIncident) x PVC. T1AP1: Transect 1A Plot 1. 1-PARRef/PARinc PVC_Img 2D 3D Height (cm) Field_NDVI Site Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 T1A P1 0.94 0.94 0.95 0.92 70.80 67.61 59.74 44.29 21 21 19 10 28 30 47 26 2.44 2.28 4.32 2.5 0.428 0.508 0.374 0.29 T1A P2 0.93 0.95 0.94 0.93 24.46 24.84 50.95 21.58 12 17 20 11 12 25 34 18 1.14 1.92 1.54 0.82 0.235 0.271 0.373 0.214 T1A P3 0.95 0.94 0.94 0.94 48.13 38.61 37.44 28.04 18 15 15 15 20 16 29 21 1.24 1.38 2.36 1.22 0.386 0.305 0.339 0.307 T1A P4 0.95 0.94 0.94 0.94 51.73 37.73 61.12 51.48 22 13 22 18 25 22 53 29 1.42 1.06 1.82 1.02 0.39 0.286 0.361 0.407 T1A P5 0.93 0.93 0.93 0.94 56.56 61.80 45.12 41.63 16 15 16 18 31 26 32 31 2.56 2.18 2.28 2.3 0.355 0.313 0.431 0.397 T1B P1 0.93 0.94 0.93 0.94 22.41 43.13 21.26 27.38 5 12 5 10 20 32 20 13 3.04 2.24 2.16 0.68 0.289 0.28 0.17 0.274 T1B P2 0.95 0.94 0.95 0.94 45.40 31.89 43.15 21.46 14 10 17 6 23 14 34 25 3.14 0.68 2.82 1.8 0.202 0.328 0.358 0.29 T1B P3 0.95 0.94 0.96 0.94 25.91 42.42 18.40 34.51 11 16 6 6 19 26 19 14 0.9 0.4 1.78 6.24 0.287 0.376 0.171 0.235 T1B P4 0.96 0.93 0.94 0.95 29.97 41.30 40.24 19.03 15 17 13 4 20 33 25 16 1.22 1.78 1.54 2.6 0.337 0.342 0.333 0.297 T1B P5 0.94 0.94 0.94 0.92 16.25 43.25 23.32 22.57 7 20 6 5 14 32 20 23 1.46 1.56 3.58 5.42 0.14 0.312 0.23 0.202 T2A P1 0.94 0.94 0.93 0.94 40.52 18.02 26.64 40.74 20 8 13 18 26 8 14 25 1.06 0.16 0.64 1.3 T2A P2 0.93 0.95 0.95 0.94 6.51 30.66 38.77 37.58 4 14 14 15 4 16 28 23 0.02 0.74 0.62 1.32 T2A P 3 0.95 0.92 0.93 0.00 41.63 15.85 22.35 0.00 16 10 7 0 29 12 10 0 2.26 0.98 0.22 0 T2A P4 0.95 0.00 0.94 0.95 56.67 0.00 43.16 48.96 20 0 16 20 27 0 19 48 1.12 0 1.88 1.52 T2A P 5 0.94 0.95 0.94 0.95 57.91 53.89 32.38 46.77 14 13 18 22 20 15 26 38 1.02 0.88 1.5 1.96 0.352 0.387 0.39 0.419 T2B P1 0.95 0.90 0.93 0.94 34.50 5.11 4.56 13.04 9 3 10 7 13 4 10 10 2.22 0.12 0.12 1.54 0.072 0.142 0.242 T2B P2 0.94 0.94 0.95 0.93 30.54 30.74 58.67 19.22 11 8 19 12 16 22 32 15 1.52 2.02 1.44 1.1 0.258 0.247 0.322 0.225 T2B P3 0.94 0.93 0.00 0.93 35.51 25.23 0.00 35.47 15 6 0 11 22 6 0 20 2.14 0.38 0 1.7 0.238 0.255 0.002 0.169 T2B P4 0.93 0.94 0.94 0.95 15 16 17 15 21 27 37 18 1.88 0.84 1.36 0.86 T2B P5 0.95 0.94 0.95 0.94 10 18 19 9 14 21 22 19 1.56 1.44 1.48 3.08 T3A P1 0.93 0.92 0.93 0.92 25.36 7.38 15.66 1.65 10 7 8 2 11 9 9 2 0.48 0.36 0.52 0.04 0.233 0.278 0.162 0.11 T3A P2 0.93 0.93 0.93 0.92 15.69 17.94 49.31 13.39 8 12 14 8 9 16 20 15 0.82 0.88 1.42 1 0.277 0.217 0.286 0.151 T3A P3 0.92 0.92 0.92 0.92 11.62 10.23 16.31 8.95 4 12 10 5 5 16 16 7 1.12 1.34 1.46 0.84 0.119 0.17 0.284 0.233 T3A P4 0.93 0.94 0.94 0.94 23.14 57.55 42.47 43.42 14 17 16 17 17 21 20 21 0.66 1.42 1.14 0.86 0.242 0.327 0.338 0.32 T3A P5 0.93 0.93 0.92 0.93 23.48 12.76 5.37 34.24 9 10 0 11 11 11 0 12 0.9 0.82 0.02 0.5 0.25 0.197 0.055 0.281 T3B P 1 0.93 0.94 0.93 0.94 11.44 5.50 2.89 37.76 2 4 3 14 5 6 3 22 0.42 0.26 0.14 1.12 0.217 0.267 0.158 0.284 T3B P2 0.93 0.94 0.94 0.93 22.64 14.18 35.97 4.97 3 2 15 1 13 8 20 1 -0.04 1.02 0.48 0 0.157 0.202 0.085 0.241 T3B P3 0.96 0.95 0.93 0.93 51.00 30.72 13.22 11.91 10 12 6 11 22 18 9 13 3 1.84 1 0.46 0.308 0.237 0.128 0.132 T3B P4 0.93 0.93 0.93 0.93 13.84 9.19 26.02 8.00 3 2 9 1 3 3 16 1 0.08 0.16 2.08 0.28 0.154 0.085 0.137 0.227 T3B P5 0.97 0.95 0.96 0.93 24.23 17.23 18.34 57.57 4 1 10 10 5 2 34 27 0.52 2.08 3.58 2.22 0.316 0.177 0.27 0.296 T4A P1 0.94 0.93 0.93 0.93 36.14 33.15 22.13 15.25 13 7 5 9 33 18 25 26 2.6 3.72 3.72 2.98 0.308 T4A P2 0.91 0.94 0.93 0.94 30.30 19.24 17.20 9.94 T4A P3 0.94 0.00 0.93 0.93 19.35 2.16 29.52 19.35 11 0 11 12 31 0 38 28 4.58 0 6.08 5.22 T4A P4 0.92 0.93 0.92 0.91 13.51 14.13 20.64 11.55 T4A P5 0.92 0.94 0.94 0.93 16.23 14.47 13.20 20.45 11 7 2 10 19 16 20 35 8.5 2.66 6.08 7.24 T4B P1 0.95 0.94 0.93 0.93 25.69 16.06 17.14 13.88 6 6 6 6 20 19 19 19 5.84 8.82 11.78 13.16 T4B P2 0.94 0.95 0.92 0.94 12.14 46.56 32.38 4.95 T4B P3 0.93 0.93 0.94 0.93 12.56 6.04 15.26 6.34 T4B P4 0.92 0.91 0.91 0.94 3 3 3 3 17 16 23 19 5.68 9.26 5.48 3.72 T4B P5 0.95 0.94 0.93 0.94 T5A P1 0.92 0.94 0.00 0.94 7.18 13.00 0.00 25.27 1 7 0 11 2 8 0 20 0.64 0.74 0.24 1.94 0.188 T5A P2 0.92 0.92 0.93 0.00 11.11 30.54 7.60 0.62 0.142 0.252 0.058 0.057 T5A P3 0.92 0.91 0.92 0.94 13.53 8.08 16.06 29.22 4 3 13 13 4 5 18 19 0.42 0.46 1.12 2.22 0.158 0.15 0.218 0.265 T5A P4 0.94 0.94 0.95 0.93 15.86 5.64 5.96 16.63 0.195 0.14 0.223 0.22 T5A P5 0.91 0.93 0.93 0.93 9.55 10.12 10.72 7.89 7 12 7 5 8 12 7 8 0.36 0.54 0.22 0.38 0.196 0.232 0.11 0.074 T5B P1 0.93 0.92 0.93 0.92 2.76 6.81 10.19 8.15 0 6 2 8 1 6 2 8 0.24 0.06 0.18 0.04 T5B P2 0.93 0.93 0.93 0.93 6.95 8.58 8.39 8.66 T5B P3 0.94 0.93 0.92 0.90 8.93 26.88 9.08 12.80 10 13 1 8 19 24 1 8 2.16 1.16 0.02 0.32 T5B P4 0.93 0.93 0.93 0.94 12.10 5.95 12.19 0.00 T5B P5 0.91 0.94 0.93 0.93 9.68 14.18 20.29 3.49 13 8 8 2 15 8 16 2 0.76 0.24 1.38 0.3 T6A P1 0.00 0.83 0.94 0.00 0.45 6.19 28.09 0.00 1 4 13 0 1 8 15 0 0.06 0.54 0.9 0 0.024 0.085 0.027 0.03 T6A P2 0.94 0.92 0.94 0.93 3.15 25.94 12.92 22.34 4 7 4 8 4 12 6 11 0.16 0.68 0.2 0.2 0.086 0.2 0.16 0.172 T6A P3 0.93 0.93 0.93 0.93 35.33 40.25 26.23 19.87 11 16 10 8 20 21 15 15 1.78 0.42 0.44 1.02 0.278 0.258 0.248 0.198 T6A P4 0.93 0.93 0.93 0.94 9.19 14.50 13.88 32.95 0.182 0.197 0.229 0.279 T6A P5 0.92 0.94 0.93 0.95 20.32 32.26 6.10 13.21 0.193 0.253 0.194 0.247 T6B P1 0.91 0.92 0.93 0.93 6.82 13.92 13.85 2.89 5 9 4 2 6 13 6 2 0.1 0.66 0.68 0.04 135 T6B P2 0.92 0.92 0.92 0.93 23.21 20.48 13.90 23.38 T6B P3 0.93 0.96 0.94 0.95 44.71 20.04 33.63 54.26 21 5 19 23 24 7 21 35 0.5 0.8 0.74 0.9 T6B P4 0.91 0.92 0.91 0.93 19.19 9.43 7.28 14.94 T6B P5 0.92 0.91 0.90 0.93 8.84 9.18 7.04 13.52 5 6 4 7 5 7 7 8 0.04 0.28 1.08 0.04 T7A P1 0.00 0.91 0.92 0.92 0.00 5.66 16.77 3.57 0 3 10 5 0 3 10 5 0 0.82 0.66 1.62 T7A P2 0.92 0.92 0.00 0.91 9.43 0.96 0.00 11.79 T7A P3 0.00 0.00 0.93 0.00 0.00 0.00 2.38 0.00 0 0 2 0 0 0 2 0 0 0 0.1 0 T7A P4 0.93 0.94 0.94 0.92 7.76 10.98 1.47 3.45 9 7 2 2 9 9 2 2 2.16 1.64 0.34 0.12 T7A P5 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 T7B P1 0.00 0.93 0.94 0.93 0.22 10.74 5.40 6.20 0 10 3 6 1 11 3 8 0.74 1.04 0.36 1.2 T7B P2 0.92 0.00 0.00 0.92 2.09 0.00 0.00 6.41 T7B P3 0.94 0.00 0.00 0.93 8.14 0.00 0.23 13.44 6 0 0 14 8 0 0 15 0.78 0 0 1.42 T7B P4 0.00 0.92 0.92 0.00 0.06 5.12 2.35 0.14 T7B P5 0.00 0.00 0.93 0.93 0.21 0.16 2.96 5.08 1 1 7 3 1 1 8 4 0.16 0.24 1.16 0.5 136