Wetland Hydrodynamics Using Interferometric Synthetic Aperture Radar, Remote Sensing, and Modeling

Home , Chari River, Logone River, Volumetric flow rate

WETLAND HYDRODYNAMICS USING INTERFEROMETRIC SYNTHETIC APERTURE RADAR, REMOTE SENSING, AND MODELING

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

Hahn Chul Jung, M. S.

Graduate Program in Geological Sciences

The Ohio State University

2011

Dissertation Committee:

Dr. Douglas Alsdorf, Advisor

Dr. Ralph R.B. von Frese

Dr. Kenneth C. Jezek

Dr. C.K. Shum

Hahn Chul Jung

2011

ABSTRACT

The wetlands of low-land rivers and lakes are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. The complexity of floodwater flows has not been well captured because flood waters move laterally across wetlands and this movement is not bounded like that of typical channel flow. The importance of these issues is exemplified by wetland loss in the Lake Chad

Basin, which has been accelerated due primarily to natural and anthropogenic processes.

This loss makes an impact on the magnitude of flooding in the basin and threatens the ecosystems. In my research, I study three wetlands: the Amazon, Congo, and Logone wetlands. The three wetlands are different in size and location, but all are associated with rivers. These are representative of riparian tropical, swamp tropical and inland Saharan wetlands, respectively. First, interferometric coherence variations in JERS-1 (Japanese

Earth Resources Satellite) L-band SAR (Synthetic Aperture Radar) data are analyzed at three central Amazon sites. Lake Balbina consists mostly of upland forests and inundated trunks of dead, leafless trees as opposed to Cabaliana and Solimões-Purús which are dominated by flooded forests. Balbina has higher coherence values than either Cabaliana or Solimões-Purús likely because the dead, leafless trees support strong double-bounce returns. Flooded and nonflooded wetland coherence varies with the season whereas terre-

ii firme and open water do not have similarly evident seasonal variations. Second, interferometric processing of JERS-1 SAR data from the central portions of both the

Amazon and Congo Basins provides centimeter-scale measurements of water level change (h/t). Despite being large, low-relief, tropical river systems, the floodplains and wetlands of the Amazon and Congo Basins show markedly different surface water flow hydraulics. Amazon patterns of h/t are well defined with clear boundaries whereas the

Congo patterns are not well defined and have diffuse boundaries. Amazon floodplain channels, lakes and pans are well interconnected, whereas the Congo wetlands are expanses with few boundaries or flow routes. Third, flood inundation maps in Logone floodplain, Lake Chad Basin are generated from 33 multi-temporal Landsat Enhanced

Thematic Mapper Plus (ETM+) images. The maximum flooding extent in the study area increases up to ~5.8K km2 in late October 2008. Coefficients of determination between flooding extents and water height variations are greater than 0.91 with 4 to 36 days in phase lag. Fourth, the spatial and temporal distribution of water level and storage changes are quantified in the central Congo wetland using spaceborne data and the LISFLOOD-

FP hydrodynamic model. This model provides 1-D diffusive channel flow and 2-D dynamic floodplain flow. The model shows meter scale water level changes on the main stem Congo River and in its tributaries (e.g. Ubangi, Sangha, Likouala-aux-Herbes, and

Likouala Rivers) at 500-meters/pixel spatial resolution. In this dissertation, my research improves the characterization of wetland surface water dynamics by making inter- comparisons of the three wetlands.

iii

DEDICATION

This document is dedicated to my family.

ACKNOWLEDGMENTS

―For the earth will be filled with the knowledge of the glory of the LORD, as the waters cover the sea.‖ (Habakkuk 2:14)

I am grateful to Dr. Douglas Alsdorf, my advisor, who helped me and encouraged me during my challenging yet fruitful years of Ph. D. studies. Without him, this piece of dissertation was impossible to complete. Especially, his time and energy for this study are unforgettable. I am also thankful to three other members of the dissertation committee,

Dr. Ralph R.B. von Frese, Dr. Kenneth C. Jezek, and Dr. C. K. Shum for thoughtful reviews and comments of this dissertation. I would like to thank my research group members and the following graduate students and researchers at OSU for their friendship and useful discussions during my study: Hyongki Lee, Michael Durand, Kostas

Andreadis, Mark Moritz, Jeremiah Lant, Brian Kiel, James Hamski, Natalie Johnson, Dai

Yamazaki, Yeosang Yoon, Sooyeun Ahn, Changki Hong, Sangho Baek, Yushin Ahn, and

Jinwoo Kim. I would like to thank brothers and sisters in Korean Church of Columbus who have shared joy and suffering together during all the years of my graduate studies.

My utmost respect, love, and appreciation go to my parents for their complete support, patience, and prayers during the past 35 years of living. I especially thank my wife,

Jiwon, and 1 year old daughter, Serene. I dedicate this dissertation to my family for their love and their boundless faith in me.

My dissertation research was funded by grants from NASA Earth and Space

Science Fellowship Program (09-Earth09F-197) and the Korea Science and Engineering

Foundation Grant (No.C00131). Additional funding was provided by NASA’s Terrestrial

Hydrology Program and the Ohio State University’s Climate, Water, and Carbon program. Thanks to the School of Earth Sciences for providing me with a Graduate

Teaching Assistantship in spring 2009 and spring 2010 quarters. On a formal note, the provisions of JERS-1 SAR data from JAXA, ALOS PALSAR data from ASF, Landsat

ETM+ data from USGS EROS, ENVISAT radar altimetry data from ESA/ESRIN, and river gauge data from the Lake Chad Basin Commission for this study are gratefully acknowledged.

―Do your best to present yourself to God as one approved, a workman who does not need to be ashamed and who correctly handles the word of truth.‖ (2 Timothy 2:15)

VITA

Jun. 1975 ...... Born, Philadelphia, PA

Feb. 1998 ...... B.S. Geology, Yonsei University, Seoul, Korea

Aug. 2003 ...... M.S. Remote Sensing, Yonsei University, Seoul, Korea

Jun. 2005 – Aug. 2009 ...... Graduate Research Assistant, The Ohio State University

Sept. 2005 – Aug. 2007 ...... Korea Science and Engineering Foundation (KOSEF) Scholarship

Mar. 2009 – May 2010 ...... Graduate Teaching Assistant, The Ohio State University

May 2010 ...... Spieker Book Award (Distinguished Senior Ph.D. Student), School of Earth Sciences, The Ohio State University

Sept. 2009 - Present ...... NASA Earth and Space Science Fellowship (NESSF)

PUBLICATIONS

Peer-reviewed Articles Jung, H. C., J. Hamski, M. Durand, D. Alsdorf, F. Hossain, H. Lee, A. K. M. A. Hossain, K. Hasan, A. S. Khan, and A. K. M. Z. Hoque, 2010, Characterization of complex fluvial systems via remote sensing of spatial and temporal water level variations, Earth Surface Processes and Landform, 35, 294-304.

vii

Jung, H. C., and D. Alsdorf, 2010, Repeat-pass multi-temporal interferometric SAR coherence variations with Amazon floodplain and lake habitats, International Journal of Remote Sensing, 31, 881-901. Lee, H., M. Durand, H. C. Jung, D. Alsdorf, C. K. Shum, and Y. Sheng, 2010, Characterization of surface water storage changes in Arctic lakes using simulated SWOT measurements, International Journal of Remote Sensing, 31, 3931-3953. Jung, H. C., S. Kim, H. Jung, K. D. Min, and J. Won, 2007, Satellite observation of coal mining subsidence by persistent scatterer analysis, Engineering Geology, 92, 1-13. Jung, H. C., S. Kim, B. C. Kim, K. D. Min, and J. Won, 2004,Observation of the ground subsidence in the abandoned Gaeun coal mining area using JERS-1 SAR, Korea Society of Economic and Environmental Geology, 37(5), 509-519.

Selected Conference Proceeding and Abstracts Jung, H. C., D. E. Alsdorf, H. Lee, M. Trigg, and T. Fewtrel, 2010, Hydrogeomorphic flood classification and hydrodynamic modeling of the Congo interfluvial wetlands, AGU Fall Meeting, San Francisco, CA, December 13-17, 2010. Lee, H., D. E. Alsdorf, H. C. Jung, C. K. Shum, J. Duan, J. Guo, and K. Andreadis, 2010, Characterization of terrestrial water dynamics in the Congo Basin using GRACE and satellite radar altimetry, AGU Fall Meeting, San Francisco, CA, December 13-17, 2010. Wilson, M., M. Durand, D. Alsdorf, and H. C. Jung, 2010, Swath altimetry measurements of the mainstem amazon river: measurement errors and hydraulic implications, Ocean Surface Topography Science Team (OSTST) Meeting, Lisbon, Portugal, October 18-20, 2010. Kim, J., J. Won, H. Lee, C. Shum, S. Calmant, A. E. Souza, and H. C. Jung, 2010, River velocity estimation from ENVISAT ASAR observations, Eos Trans. AGU, 91(26), Jt. Assem. Suppl., Abstract U24A-04. Jung, H. C., D. Alsdorf, H. Lee, M. Wilson, E. Beighley, M. Durand, C.K. Shum, J. Kim, and K. Andreadis, 2010, Hydrodynamic modeling of the Congo wetlands using LISFLOOD and satellite based measurements, EGU General Assembly 2010, Geophysical Research Abstracts, Vol. 12, EGU2010-6202. Lee, H., D. Alsdorf, J. Duan, M. Durand, J. Guo, H. C. Jung, L. Schaller, and C. Shum, 2009, Terrestrial water dynamics in the Congo basin using satellite radar altimetry and GRACE, Eos Trans. AGU, 90(52), Fall Meet. Suppl., Abstract H51F-0824. Schaller, L, M. Durand, D. Alsdorf, H. C. Jung, and H. Lee, 2009, Discharge of the Congo River Estimated from Satellite Measurements, Eos Trans. AGU, 90(52), Fall Meet. Suppl., Abstract H51F-0823. Jung, H. C., D. Alsdorf, and H. Lee, 2008, A comparison of Congo and Amazon wetland hydraulics from repeat-pass interferometric SAR satellite measurements, Eos Trans. AGU, 89(53), Fall Meet. Suppl., Abstract H43G-1107. Jung, H. C., and D. Alsdorf, 2008, Repeat-pass interferometric SAR measurements of seasonal changes in Congo flood water elevations, Ocean Science Meeting, Orlando, FL, March 2-7, 2008. viii

Jung, H. C., and D. Alsdorf, 2006, Repeat-pass multi-temporal interferometric SAR coherence variations with Amazon floodplain and lake habitats, Eos Trans. AGU, 87(52), Fall Meet. Suppl., Abstract H23A-1459. Jung, H. C., N. Johnson, and D. Alsdorf, 2005, Interferometric SAR coherence variations with Amazon floodplain and lake vegetation, Eos Trans. AGU, 86(52), Fall Meet. Suppl., Abstract H21D-1376. Jung, H. C., and K. D. Min, 2005, Observing coal mining subsidence from JERS-1 permanent scatterer analysis, Geoscience and Remote Sensing Symposium, IGARSS '05. Proceedings, 4578-4581.

FIELDS OF STUDY

Major Field: Geological Sciences Studies in: 1. Satellite Hydrology 2. Interferometric Synthetic Aperture Radar (SAR) Processing 3. Hydrodynamics 4. Floodplain Dynamics 5. Satellite Mission Orbit Design

TABLE OF CONTENTS

Page

ABSTRACT ...... ii

DEDICATION ...... iv

ACKNOWLEDGMENTS ...... v

VITA…...... vii

TABLE OF CONTENTS ...... x

LIST OF TABLES ...... xiii

LIST OF FIGURES ...... xiv

Chapter 1: Introduction ...... 1 1.1 Statement of Problem ...... 1 1.2 Background and Relevance to Previous Work ...... 3 1.2.1 Interferometric SAR Observations in Wetlands ...... 3 1.2.2 Remote Sensing for Flood Inundation Mapping ...... 8 1.2.3 Hydrodynamic Modeling ...... 10

Chapter 2: Repeat-pass Multi-temporal Interferometric SAR Coherence Variations with Amazon Floodplain and Lake Habitats ...... 14 2.1 Introduction ...... 14 2.2 Study Area and Classification ...... 16 2.2.1 Study Location ...... 16 2.2.2 Classification Scheme ...... 20 x

2.3 SAR Data and Processing ...... 25 2.3.1 Interferometric Processing ...... 25 2.3.2 Coherence Variations ...... 28 2.4 Results & Discussions ...... 30 2.4.1 Coherence Variations with Perpendicular Baselines ...... 30 2.4.2 Coherence Variations with Temporal Baseline ...... 37 2.4.3 Coherence Variations within High- and Low-Water Seasons ...... 43 2.5 Conclusions ...... 47

Chapter 3: A Comparison of Congo and Amazon Wetland Hydraulics from Repeat-pass Interferometric Satellite Measurements ...... 49 3.1 Introduction ...... 49 3.2 Study Areas and Interferometric SAR Data ...... 53 3.3 Flow Hydraulics...... 58 3.4 Conclusions ...... 63

Chapter 4: Flood Inundation Mapping in the Logone Floodplain from Multi-temporal Landsat ETM+ Imagery ...... 64 4.1 Introduction ...... 64 4.2 Study Area ...... 67 4.3 Materials and Methods ...... 69 4.3.1 Landsat ETM+ Data ...... 69 4.3.2 ENVISAT Altimetry Data ...... 74 4.3.3 Ground-based Data...... 75 4.4 Results and Discussion ...... 75 4.4.1 Flooding Extent ...... 75 4.4.2 Water Height Variation ...... 80 4.4.3 Correlation between Flooding Extent and Water Height Variation ...... 81 4.4.4 Logone Floodplain Dynamics ...... 87 4.5 Conclusions ...... 89

Chapter 5: Hydrodynamic Modeling of the Congo Wetlands Using LISFLOD-FP and Spaceborne Data ...... 91 5.1 Introduction ...... 91 5.2 Methods ...... 92 5.2.1 LISFLOOD-FP ...... 92 5.2.2 River Parameters ...... 94 5.2.3 Boundary Conditions...... 97 5.2.4 Floodplain Bare Ground Elevation ...... 99 xi

5.3 Results ...... 102 5.3.1 Hydrodynamics Models...... 102 5.3.2 Comparisons of the Models with Altimetry and SAR Interferometry ...... 105 5.4 Discussions and Conclusions ...... 109

Chapter 6: Summary and Conclusions ...... 110

REFERENCES ...... 115

xii

LIST OF TABLES

Table Page

Table 2.1: Upland and wetland areas and their percentages in Balbina, Cabaliana, and Solimões-Purús as mapped from high- and low-water seasons. Diagonal downs (\) indicate shortage of data to compute statistics of the classes...... 24

Table 2.2: The temporal and spatial parameters of acquired SAR images...... 27

Table 2.3: The mean plus and minus the standard deviation of coherences of terre-firme, open water, flooded herbaceous, flooded woodland, flooded forest, and nonflooded forest in Figure 2.4...... 34

Table 2.4: Absolute z values of Mann-Whitney Statistical Test (T: terre-firme, O: open water, FH: flooded herbaceous, FW: flooded woodland, FF: flooded forest, and NF: nonflooded forest). z values after below 2.33 in the study indicate that the corresponding two classes are statistically identical...... 35

Table 3.1: Description of Satellite Data...... 54

Table 4.1: Summary of Landsat ETM+ dataset and classification results...... 71

Table 4.2: Summary statistics for regression models in Figure 4.4...... 85

Table 4.3: The estimation of flow rates for flooding extents in the regression models with time shifting...... 86

xiii

LIST OF FIGURES

Figure Page

Figure 1.1: (Left) Amplitude composite, (middle) polynomially-flattened wrapped interferogram, and (right) coherence images for a large floodplain lake immediately adjacent to the Solimões-Amazon River (source: Alsdorf et al., 2001b)...... 6

Figure 1.2: Schematic figures showing the contributions of radar backscattering over (a) forests and (b) marshes due to canopy surface backscattering, canopy volume backscattering, specular scattering, and double-bounce backscattering (source: Lu and Kwoun, 2008)...... 7

Figure 2.1: Location map of the Amazon Basin study areas. A marks Balbina (JERS-1 path-row 414-303), B is Cabaliana (416-306), and C indicates Solimões-Purús (417-307). Blue indicates rivers and lakes that do not completely drain, light blue are annually flooded areas that drain, and green is terre-firme or upland areas that never flood...... 18

Figure 2.2: Classification maps for high and low water seasons: (a) Balbina, (b) Cabaliana, and (c) Solimões-Purús (source: Hess et al., 2003)...... 23

Figure 2.3: Physical and temporal baseline distributions for each study location. Temporal baselines are noted on the x-axis of all three plots whereas physical baselines are on the y-axis...... 26

Figure 2.4: Mean coherence values of various vegetation-hydrologic habitat classes (i.e. terre-firme, open water, flooded herbaceous, flooded woodland, flooded forest, and nonflooded forest) compared to interferometric perpendicular baselines for each study location. Note that open water is consistently separated from the other classes...... 31

xiv

order polynomials. Annual periodicity is found in the flooded and nonflooded habitats but absent in the open-water and terre-firme classes...... 38

Figure 2.6: Power spectrums of figure 2.5 mean coherence values calculated after subtraction of polynomial trends. Clear, annual periodicity is always found in flooded classes and never in open water...... 40

Figure 2.7: Variation in coherence values with temporal and physical interferometric baselines for (a) Balbina, (b) Cabaliana and (c) Solimões-Purús. Point locations are noted in figure 2 and krigging is used to contour the coherence values...... 42

Figure 2.8: Mean coherence values for interferometric pairs where both dates occur during the high-water season. Coherence values of various vegetation- hydrologic habitat classes (i.e. terre-firme, open water, flooded, and nonflooded) are compared to interferometric temporal baselines for each study location. Dots are the habitat mean coherence and circles are the average of these mean coherence values...... 45

Figure 2.9: Mean coherence values for interferometric pairs where both dates occur during the low-water season. Coherence values of various vegetation- hydrologic habitat classes (i.e. terre-firme, open water, flooded, and nonflooded) are compared to interferometric temporal baselines for each study location. Dots are the habitat mean coherence and circles are the average of these mean coherence values...... 46

Figure 3.1: The (A) Amazon and (B) Congo study areas are shown using overlays of the low and high water GRFM mosaics. Light blue marks seasonally flooded areas; green is non-flooded areas; dark blue is indicative of areas that always contain water, e.g., river channels. Topex/Poseidon altimetric measurements are marked with yellow and white lines. Red diagonal boxes locate JERS-1 SAR swaths...... 50

Figure 3.2: Water surface heights derived from Topex/Poseidon altimetry. See Figure 1 for locations. X axis numbers refer to months where 1 is January, 2 is February, etc. and y axis numbers refer to orthometric heights with respect to EGM96 geoid model. Dates of interferometric pairs are noted. Lines a1, c1, and 2 correspond to 143, 114, and 18 altimetric points (yellow in Figure 3.1), respectively...... 52

Figure 3.3: Measurements of changes in water level (h/t) superimposed on SRTM elevation maps in (A) Amazon and (B) Congo. Spatial patterns of temporal water level changes are measured from repeat-pass interferometric JERS-1 SAR. Acquisition dates of interferometric pairs are noted in Table 3.1.

Locations without interferometric measurements were not flooded during at least one of the overpasses...... 57

Figure 3.4: Detail of interferometric SAR measurements, SRTM elevations, Landsat color composite images (bands 7, 5, 3 shown in RGB), and JERS-1 amplitude images. Landsat acquisition dates are noted in Table 3.1. The close-up images all have the same spatial scale and have an elevation relief of 20-30m. Black flow arrows are based on continuity with directions pointing toward areas of greater water accumulation at water increasing times in R1 and R2 and indicate evacuation directions at water decreasing times in R3 and R4. In the profiles, the red lines are topography and blue lines are interferometric h/t measurements. Green circles note locations of floodplain channels, yellow ellipses note channels serving as pathways for water flow, and red ellipses note topographic depressions infilled by greater h/t during rising water. In Congo regions R3 and R4, swamp forests, raphia palms, and grass savannas are marked with SF, RP, and GS, respectively...... 60

Figure 4.1: (a) SRTM elevation map of the Lake Chad Basin. Logone floodplain at the south of the Lake Chad is hydrologically linked with two major branches of the Logone and Chari Rivers. Black diagonal box indicates Landsat ETM+ frame used for flood inundation ...... 69

Figure 4.2: Flood inundation maps derived from multi temporal Landsat ETM+ imagery. See text for processing details...... 77

Figure 4.3: Time series of flooding extents and water heights in the study area. Bar graphs in the uppermost panel represent flooding extents calculated from Landsat ETM+ in Fig. 2. The ENVISAT altimetry provides 35-day repeated measurements of water height variation in the second and third upper graphs. These altimetric measurements are linearly interpolated between two successive points. Three local river gauge stations provide daily measurements at the bottom three graphs...... 79

Figure 4.4: Relationships of Landsat-derived flooding extents with water height variations from ENVISAT altimetry and river gauge stations. A second order polynomial regression and time shifting are performed to find the best-fit lines to a set of data points. Left/right graphs are the regression model results before/after time shifting. An increase in R2 after time shifting represents phase lag between flooding extents and water height variations...... 83

Figure 4.5: The detailed spatial pattern and size of the flooding extents in the Logone floodplain during high water. Monthly average flooding extents are calculated based on monthly flood probability maps for the 3-year study period. As red goes gradually through orange to white, it ranges from 100 to 0% in the flood probability. Blue represents open water such as Logone/Chari Rivers and xvi

Lake Maga. The mode of the flooded area at each map is marked with ―X‖ in latitude and longitude...... 88

Figure 5.1: LISFLOOD-FP Hydraulic model area in SRTM DEM (left-upper), water mask (left-bottom), and river networks in GRFM mosaic (right). This area is composed of several river networks: R0-Congo, R1-Giri, R2-Ubangui, R3- Likoula-aux-Herbes, R4-Sangha, R5-Mambili, R6-Likoula (Mossaka), R7- Kouyou, R8-Alima, and R9-Nkeni Rivers. Red dots indicate 10 input discharge locations in the rivers for upstream boundary conditions and green dot indicates an input river height location for downstream boundary condition. Yellow box locates ALOS PALSAR interferometric swath and orange line locates ENVISAT altimetric measurements...... 96

Figure 5.2: The time series of 10 upstream locations (upper and middle) are generated from Hillslopes River Routing (HRR) hydrological model (Beighley et al., 2009). The time series of downstream location are generated from 35 day repeat-pass ENVISAT altimetry (bottom). See the locations in Figure 5.1. ... 98

Figure 5.3: The difference between SRTM and ICESAT elevations of the signal start, centroid, and end are plotted in green, red, and blue, respectively (upper left). ICESAT elevations (GLA14, L1A: Feb-March 2003) in the model area are obtained (upper right). Canopy heights in the SRTM are calculated using ICESAT, MODIS VCF, and SRTM roughness (Carabajal and Harding, 2005, 2006)...... 101

Figure 5.4: The spatial and temporal variations of flood extent and water depth (m) in the study wetland...... 103

Figure 5.5: The profiles of water elevations in the Congo, Ubangui and Sangha Rivers as compared to river bed elevations...... 104

Figure 5.6: ENVISAT altimetry track (upper left; courtesy of Google Imagery), water elevations from ENVISAT altimetry (solid lines) and LISFLOOD-FP hydraulic model (dots) (bottom), and altimetry water levels compared to model water levels (upper right). Water levels during high water (e.g. November in black dots and December in yellow dots) show higher correlation between ENVISAT altimetry and model than during low water...... 106

Figure 5.7: Water level changes (cm) for 92 days from December 18 to September 17, 2007 are calculated from ALOS PALSAR differential interferograms (upper left) and model (upper right). The differential interferogram is unwrapped to compare with model-derived water level change along a black profile at Easting 282 (bottom)...... 108

xvii

Chapter 1: Introduction

1.1 Statement of Problem

The wetlands of low-land rivers and lakes are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. Wetlands, lakes, and rivers cover five to eight million square kilometers globally, blanketing up to six percent of the earth’s land surface (Matthews, 1993). Water flow through wetlands controls a number of processes including changes in stored water, biogeochemical cycling, sediment delivery, and nutrient exchange. Water storage in floodplains is also a key, governing parameter in continental-scale hydrologic models (e.g. Vorosmarty et al.,

1989, Richey et al., 1989). Monitoring discharge in the main channels of rivers and upland tributaries as well as storage changes in floodplain lakes is necessary for understanding flooding hazards, methane production, sediment transport, and nutrient exchange. The complexity of floodwater flows has not well been captured because flood waters move laterally across wetlands and this movement is not bounded like that of typical channel flow. Floods are two-dimensional. Water flow across wetlands is more complex than implied by 1D, point-based measurements. Flow paths and water sources are not fixed in space and time, but rather vary with flood water elevations. 1

Anthropogenic forcings such as deforestation in the Amazon and Congo Basins and the impoundment of waters in the Lake Chad Basin have made an impact on the water cycle, magnitude of flooding, and ecosystems in these three basins. The three wetlands are different in size and location, but all are associated with rivers. These are representative of riparian tropical, swamp tropical and inland Saharan wetlands, respectively. The Amazon Basin, with an area of ~6.0M km2 and containing the largest tropical rainforest in the world, contributes 15% to 20% of the global river discharge

(Junk, 1993) to the oceans (annually averaged discharge of 200,000 m3/s). The floodplains of the Amazon River and its tributaries are built from overbank deposits with an overall annually inundated area of about 750,000 km2 (Melack and Forsberg, 2001).

The floodplain along just the mainstem Amazon River in Brazil covers at least 92,000 km2 (Sippel et al., 1992). The Congo River, with the second-largest discharge and basin area of any rivers, has not had the same attention in terms of research compared to the

Amazon (annually averaged discharge is ~40,600 m3/s and area is 3.5M km2, Laraque et al., 2001). The Lake Chad Basin covers an area of 2.5M km2 (Crétaux and Birkett, 2006;

Coz et al., 2009). The Logone River is a major tributary of the Chari River. The

Logone/Chari River waters flow into the Lake Chad Basin from the south and gradually move northwards, supplying permanent open water and seasonally inundating the marsh regions. Approximately 90% of the Lake Chad’s water stems from the Logone/Chari

River system. These rivers have their origins in Cameroon and the Central African

Republic, respectively. The remaining 10% stems from other tributaries and local precipitation (FEWS, 1997). West African wetlands including the Logone floodplain

2 have been extensively damaged or destroyed by the construction of major agricultural and hydro-electric infrastructure over the last thirty years.

This study significantly adds to our understanding of wetland hydraulic knowledge and provides an opportunity to investigate the impacts of deforestation, drought, and flood hazards on surface water stores and fluxes. The overall scientific questions to address are as follows: How do the riparian tropical, swamp tropical and inland Saharan wetlands fill and empty? Can we discern differences in flow hydraulics and their storage changes using combinations of satellites and models? Are these differences significant and thus indicative of potentially different fluxes and storages of sediments, nutrients, and carbon?

1.2 Background and Relevance to Previous Work

1.2.1 Interferometric SAR Observations in Wetlands

Interferometric processing of SAR (Synthetic Aperture Radar) data has already been used to map the centimeter-scale changes in topography related to earthquakes, volcanic activity, and groundwater flux (e.g., Massonnet et al., 1993; Zebker et al., 1994a;

Wicks et al., 1998; Hoffman et al., 2001), to map the velocity field of flowing glaciers

(e.g., Goldstein et al., 1993), and to map the atmospheric water vapor change resulting from storms (e.g. Hanssen et al., 1999). The processing method required two SAR image acquisitions from identical (or nearly identical) viewing geometries before and after the

3 displacement phenomenon; co-registration of the two images to a sub-pixel accuracy; and subtraction of the complex phase and amplitude values at each SAR image pixel. The value of the resulting interferometric phase at each pixel varies between –pi and +pi and is primarily a function of the distance between the radar antenna positions during acquisition (i.e. the platform baseline), topographic relief, surface displacement, and the degree of correlation between the individual scattering elements that comprise each pixel location (i.e. coherence; Rosen et al., 1996).

The performance of a radar interferometer system depends on the radar instrument parameters, the orbit parameters, and the errors induced by the data processing and post-processing operations. Analyses of noise expected in ERS-1 interferometric data collected over Alaska and the southwestern United States indicate that maps with relative errors less than 5 m RMS are possible in some regions (Zebker et al., 1994b). For the repeat-pass implementation in particular, temporal decorrelation constitutes an important error source in the operation of topographic mapping radar. Temporal decorrelation contributions to the height errors may be limited to 1.5 and 2.6 m for the forested and lava areas, respectively, if suitable attention is given to experiment design (Zebker and

Villasenor, 1992). Height error as a function of phase error for topographic analysis is given by Zebker and Goldstein (1986) or Rodriguez and Martin (1992). Spatial and temporal changes of 20% in relative humidity lead to 10 cm errors in deformation products, and perhaps 100 m of error in derived topographic maps for those pass pairs with unfavorable baseline geometries (Zebker et al., 1997). The phase signature is related to the atmosphere is likely because of several points: (1) the ratio of the phase artifact

4 level in the images is the wavelength r ratio, (2) the phase irregularities are unrelated to surface features, so that they are probably not surface scattering changes, (3) they are different from day to day, whereas surface-induced changes should be spatially correlated in different data sets, and (4) the effect is most pronounced at lower elevations, where the ray path through the atmosphere is longest and thus most sensitive to variation in an absolute sense (Zebker et al., 1997).

As SAR transmits radar pulses at an off-nadir look angle, a smooth open-water surface causes most of the radar energy to reflect away from the radar sensor, resulting in little energy being returned back to the SAR receiver. When the open-water surface is rough and turbulent, part of the radar energy can be scattered back to the sensor; however, the SAR signals over open water are not coherent if two radar images are acquired at different times. Thus, it has been generally accepted that InSAR is an inappropriate tool to use in studying changes in the water level of open water. However, Alsdorf et al.

(2000, 2001a, 2001b) found that interferometric analysis of L-band (i.e. wavelength of

∼24 cm) Shuttle Imaging Radar-C and Japanese Earth Resources Satellite (JERS-1)

SAR imagery can yield centimeter-scale measurements of water-level changes throughout inundated floodplain vegetation. Their work confirmed that scattering elements for L-band radar consist primarily of the water surface and vegetation trunks, which allows double-bounce backscattering returns. The margins of most floodplain lakes include emergent shrubs that provide substantial double-bounce radar returns in

Figure 1.1. The profiles of phase and coherence values show stage decrease across

5 inundated vegetation during one day separating the SIR-C acquisitions. The shorelines of this lake are well defined. Thus, the phase shows a sharp transition from land to water.

The radar backscattering over flooded vegetation consists of contributions from the interactions of radar waves with the canopy surface, canopy volume, and water surface. Based on the canopy backscattering model for continuous tree canopies developed by Sun (1990), the total radar backscattering over wetlands can be approximated as the incoherent summation of contributions from the following: 1) 6 canopy surface backscattering; 2) canopy volume backscattering that includes backscattering from multiple path interactions of canopy and water; and 3) double- bounce trunk-water backscattering (Figure 1.2). The relative contributions from surface backscattering, volume backscattering, and double-bounce backscattering are controlled primarily by vegetation type, vegetation leaf on/off condition, canopy closure, and other environmental factors (Hess et al., 1995).

Chapter 2 analyzes interferometric coherence variations in JERS-1 (Japanese

Earth Resources Satellite) L-band SAR (Synthetic Aperture Radar) data at three central

Amazon sites. The findings suggest that repeat-pass interferometric coherence of flooded habitats is capable of showing the annual periodicity of the Amazon flood wave.

Chapter 3 carries out interferometric processing of JERS-1 SAR data from the central portions of both the Amazon and Congo Basins to provide centimeter-scale measurements of water level change (h/t). Despite being large, low-relief, tropical river systems, the floodplains and wetlands of the Amazon and Congo Basins show markedly different surface water flow hydraulics.

1.2.2 Remote Sensing for Flood Inundation Mapping

Satellite remote sensing has been implemented in classifications of wetlands and their separation from other land cover classes (Ozesmi and Bauer, 2002). Remote sensing is expected to provide powerful techniques to determine flood inundation areas. Flood detection is one of the classical themes of remote sensing, and many studies have been undertaken to map spatial and temporal changes of flood inundation areas, study flood dynamics and behaviors, and assess flood damage in urban areas (Islam et al., 2010).

Satellite-derived flood inundation maps are invaluable to state or national agencies for disaster monitoring and relief efforts (Smith, 1997). Flood inundation maps that are tied to U. S. Geological Survey (USGS) real-time stream gage data and National Weather

Service (NWS) flood forecast sites enable officials to make timely operational and public

8 safety decisions during floods. Because floods are the leading cause of natural-disaster losses, and because disasters associated with flooding can be reduced with proper preventative measures, development of a USGS National flood inundation mapping science initiative is critical to meeting the USGS science strategy goals for the National

Hazards, Risk, and Resilience Assessment Program. A powerful new tool for flood response and mitigation is digital geospatial flood inundation mapping that shows floodwater extent and depth on the land surface (USGS, 2010).

The application of satellite imagery for flood mapping began with the use of

Landsat Thematic Mapper (TM) and Multi-Spectral Scanner (MSS), the Satellite Pour l’Observation de la Terre (SPOT), the Advanced Very High Resolution Radiometer

(AVHRR), Advanced Spaceborne Thermal Emission and Reflection Radiometer

(ASTER), and Moderate-Resolution Imaging Spectroradiometer (MODIS) (Khan et al,

2010). The visible and infrared sensors have been used to determine the extent of water bodies using simple classification procedures (Jain et al., 2005). These studies have relied on the water bodies having a unique spectral response in this range of electromagnetic radiation when compared to the surrounding landscape. Many studies using density slicing of Landsat MSS band 7 (e.g. Bennett, 1987) and TM bands 4 and 7 (e.g. Wang,

2002; Frazier and Page, 2000) have been reported. In addition, satellite radar remote sensing systems have been used to study wetlands (e.g. Hess et al., 1990; Kasischke and

Bourgeau-Chavez, 1997; Townsend and Walsh, 1998; Kushwaha et al., 2000). Satellite radar data is available from ERS-1, launched by the European Space Agency in 1991,

RADARSAT, launched by the Canadian Space agency in 1995, and JERS-1 and ALOS

PALSAR, launched by the National Space Development Agency of Japan in 1992 and

2006, respectively. Radar has advantages for remote sensing for two reasons: 1) radar systems can collect data at any time of day or night and under almost any weather conditions; 2) radar reflections (backscatter) provide different information than optical sensors (Ozesmi and Bauer, 2002). The microwave measurements from space enable the frequent monitoring of flood inundation at a large scale. However, this is not a feasible approach for certain studies because of the limitations of the recurrence period, the performance of the pointing device, and in particular the high cost of data acquisition

(Sakamoto et al., 2007).

Chapter 4 deals with flood inundation mapping in the Logone floodplain using 33 multi-temporal Landsat Enhanced Thematic Mapper Plus (ETM+) images. Yearly flooding in the Logone floodplain has a direct impact on agricultural, pastoral, and fishery systems in the Lake Chad Basin. The study provides strong correlation between the flooding extents and water height variations in both the floodplain and the river based on a polynomial regression model.

1.2.3 Hydrodynamic Modeling

Hydraulic models can be classified according to the number of dimensions in which they represent the spatial domain and flow processes (Hunter et al., 2007). Though flow in compound channels is known to be fully three-dimensional, 1D and 2D models predominantly have been implemented to study the flow dynamics. Complex three-

10 dimensional approaches may also be unnecessary for many scales of compound channel flow with the limitation of computational feasibility and the problems of accurately representing the water free surface, high-order turbulence, and transient flood shorelines

(Lane et al., 1999; Booker et al., 2001; Morvan et al., 2002, Wilson et al., 2003).

The most popular approaches to modelling fluvial hydraulics at the reach scale have been one-dimensional finite difference solutions of the full St. Venant equations

(see for example Fread, 1984; Samuels, 1990; Ervine and MacLeod, 1999) such as

MIKE11 (DHI Water and Environment, 2001), ISIS (HR Wallingford), FLUCOMP

(Samuels and Gray, 1982) and HEC-RAS (USACE, 2001). Such schemes describe the river channel and floodplain as a series of cross sections perpendicular to the flow direction and are thus well suited to parameterization using traditional field surveying methods (Bates et al., 1992). Numerical solution of the controlling equations for prescribed inflow and outflow boundary conditions then enables the cross section averaged velocity and water depth at each location to be calculated. Although 1D codes are computationally very efficient, they suffer from a number of drawbacks when applied to floodplain flows (Hunter et al., 2007). These include the inability to simulate lateral diffusion of the flood wave, the discretization of topography as cross sections rather than as a surface and the subjectivity of cross-section location and orientation. All of these fundamental constraints can be overcome with two-dimensional codes and numerous classes of 2D schemes have been developed in response.

Since first proposed by Zanobetti et al. (1970) methods to predict floodplain inundation using storage cell approaches have become justifiably popular. Two-

11 dimensional finite difference and finite element models have been developed (see for example Feldhaus et al., 1992; Bates et al., 1992; Bates et al., 1995). These provide a higher order representation of river hydraulics through a full solution of the 2D St.

Venant equations that is more consistent with known processes, includes a continuous representation of topography and requires no secondary processing step to determine the flood inundation. Two-dimensional models are best employed in conjunction with a

DEM of the channel and floodplain surface which, in conjunction with suitable inflow and outflow boundary conditions, allows the water depth and depth-averaged velocity to be computed at each computational node at each time step (Bates et al., 1992). For reasons of computational cost, full 2D codes are not a currently viable solution here and this has lead a number of researchers to develop coupled 1D/2D codes which combine the simplicity of 1D channel routing approaches with simpler methods of treating floodplain flow that make use of improved topographic data (Bates et al., 1995).

LISFLOOD-FP is one such 1D/2D model and was originally developed by Bates and De Roo (2000) in the PC-Raster dynamic modelling language. Subsequently, the code was re-coded in C++ in order to improve computational efficiency and allow application to larger domains (Horritt and Bates, 2001) or multiple realisations of the same problem (Aronica et al., 2002). Inputs include a floodplain DEM, bathymetric depths, channel widths, channel roughness, and upstream flow boundary conditions.

Chapter 5 quantifies the spatial and temporal distribution of water level and storage changes in the central Congo wetlands using spaceborne data and the

LISFLOOD-FP hydrodynamic model. This model provides 1-D diffusive channel flow

12 and 2-D dynamic floodplain flow. The model results are compared with ALOS PALSAR repeat pass interferometric SAR measurements.

Chapter 2: Repeat-pass Multi-temporal Interferometric SAR Coherence Variations with Amazon Floodplain and Lake Habitats

2.1 Introduction

Wetlands, lakes, and rivers cover five to eight million square kilometers globally, blanketing up to six percent of the earth’s land surface (Matthews and Fung, 1987;

Matthews, 1993; Mitchell, 1990). The flow of water through these environments is a control on both biogeochemical and sediment fluxes. Water storage in floodplains is also a key, governing parameter in continental-scale hydrologic models (e.g. Coe, 1998;

Vorosmarty et al., 1989; Richey et al., 1989). Monitoring discharge in the main channels of rivers and upland tributaries as well as storage changes in floodplain lakes is necessary for understanding flooding hazards, methane production, sediment transport, and nutrient exchange. An understanding of the flooding dynamics and hydrologic exchange between rivers and related floodplains relies on measurements of water levels recorded at gauging stations along a main channel. For nearly all wetlands, however, the lack of floodplain stage recording devices results in poorly constrained estimates of floodplain water storage.

Given the vast size and remote location of large tropical basins such as the

Amazon and Congo, satellite observations remain a viable approach to constraining and validating basin scale hydrologic models. For example, modeling efforts have begun to rely on remotely sensed observations that either directly record water surface elevations using satellite radar altimetry (e.g. Birkett 1998; Koblinsky et al., 1993; Maheu et al.,

2003; Leon et al., 2006) or infer stage and discharge from relationships between main channel gauge data and remotely sensed inundated area (e.g. Sippel et al., 1998; Smith

1997; Smith et al., 1995, 1996; Vorosmarty et al., 1996). Interferometric synthetic aperture radar (SAR) has recently been demonstrated to measure water level changes with time (dh/dt) (Alsdorf et al. 2000, 2001a, 2001b; Lu et al., 2005; Kim et al., 2005) and has been coupled with model based understanding of storage changes (Alsdorf 2003) and flow hydraulics (Alsdorf et al., 2005).

Previous investigations have used interferometric SAR for forest mapping (Askne et al., 1997; Engdahl and Hyyppa, 1997), forest change detection (Wegmuller et al.,

1995, 2000), and flood water studies (Alsdorf et al., 2000). Numerous studies have demonstrated the value of SAR amplitude (radar backscatter) for delineation of wetland ecosystems (e.g. Hess et al., 2003; Harris and Digby, 1986, Hess and Melack 1994,

Richards et al., 1987) and especially flooding beneath the forest canopy (Hess et al. 1990,

Wang et al., 1995). Recently, interferometric SAR phase coherence was found to be more effective than radar backscatter for differentiating willow-alder (broad leaf tree), spruce

(needle leaf tree), ice, and open water (Hall-Atkinson and Smith, 2001). The phase coherence between repeat-pass SAR observations of a forest region was investigated to

15 estimate the growing–stock volume (Eriksson et al., 2003; Luckman et al., 2000).

However, few studies focus on interferometric SAR coherence variations with vegetation and with seasonal water fluctuations, particularly related to the utility of using the related phase for measuring dh/dt.

I demonstrate the relationships of interferometric SAR coherence with physical and temporal baselines as measured across various Amazon floodplain habitats (e.g. terre-firme, open water, flooded and nonflooded forest, etc.). We use repeat-pass JERS-1

SAR data, which is L-band and has a HH polarization. Because the Amazon has a strong, seasonal flood wave, we also show the relationships of coherence with seasonal flooding.

2.2 Study Area and Classification

2.2.1 Study Location

The central Amazon floodplain contains the confluence of the Amazon and Purús rivers (upstream of the confluence with the Negro River, the Amazon River is referred to as the Solimões River, whereas the combined Amazon-Solimões River is referred to as the Amazon mainstem). The Amazon Basin contains about 750,000 km2 of annually inundated area (Melack and Forsberg, 2001). The alluvial floodplains of the Amazon

River and major tributaries in Brazil are believed to cover over 300,000 km2 (Klinge et al., 1990). The alluvial deposits along just the mainstem Amazon River in Brazil cover approximately 92,000 km2 (Sippel et al., 1992). The floodplain can be divided into the

16 varzea, which is flooded by sediment- and nutrient-rich water (white water), and igapo, which is flooded by sediment- and nutrient-poor water (black water) (Sioli, 1968). The floodplains of these large lowland rivers are ~20 to ~50 km wide, with low topographic relief ranging from 20 to 30 m above mean sea level. Terre-firme uplands, which are never flooded, are developed in river terrace sediments and border the floodplains with elevations ranging from 30 to 50+ m above mean sea level. Figure 2.1 shows the location and geographical coordinates of three study areas.

Lake Balbina is a man-made reservoir created to supply hydroelectric power to the city of Manaus. The reservoir is located on the Uatuma River and drains a 19,100 km2 basin of mostly upland topography where the relief extends from 30 m to 200 m in elevation (Fearnside, 1989). The lake includes a cluster of ~1,500 islands separated by submerged, shallow valleys within a flooded water-surface area of 2,400 km2 (Melack and Wang, 1998). Prior to dam closure on October 1, 1987, the annually averaged flow on the river was about 450 m3/s. Water depths in the full reservoir average 7.4 m whereas the average water level fluctuations have a range of about 3 m/yr. Because the vegetation was not cleared before filling, the lake consists mostly of forest and inundated trunks of dead, leafless trees. The other study areas are the Cabaliana floodplain on the Solimões

River and the confluence of the Purús and Solimões Rivers. The annual rise and fall of the Solimões River averages about 10 m on this reach and inundates large areas of floodplain (i.e. varzea). Amazonian varzea forests have stand densities comparable to upland (terre-firme) forests, but tend to have lower species diversity (Campbell et al.,

1992). The Purús River drains the sediments of the sub-Andean trough and of the central plain. It is a Southern tributary of the Solimões River with an intermediate composition between black and white water (Hedges et al., 1986). Floodplain area between Itapeau and Manacapuru along the Solimões River is about 12,000 km2 (Alsdorf, 2003).

2.2.2 Classification Scheme

Hess et al. (2003) conducted dual-season mapping of wetland inundation and vegetation for the central Amazon basin under both low-water and high-water conditions at 3 arc-second resolution. Mosaics of JERS-1 SAR images were created as part of the

Global Rain Forest Mapping Project (GRFM, Rosenqvist et al., 2000) and four validation overflights for Amazon mosaics (VOAM) surveys were conducted. For GRFM, the entire

Amazon Basin was acquired in a series of orbital passes during the generally low flood season of the Amazon River in September to December 1995. The same area was covered again in May to August 1996, during the high flood period. A pixel-based classifier was used by Hess et al. (2003) to map wetland vegetation and flooding states based on SAR backscattering coefficients of two-season class combinations. The two initial VOAM were flown in 1995 and 1996 during the GRFM imaging periods for the central Amazon and were limited to areas within 600 km of Manaus. The aerial video graphic surveys covered regions of Balbina, Cabaliana, and Solimões-Purús. In order to expand the ground data set to a more extensive region, follow-up surveys were flown in

1997 and 1999. VOAM95 and VOAM97 surveys are timed to low-water whereas

VOAM96 and VOAM99 correspond to high-water stages of the Amazon River. Based on the VOAM surveys, accuracy for flooded and nonflooded forest classes ranged from 78% to 91%, with lower accuracy (63-65%) for flooded herbaceous vegetation (Hess et al.,

2003). Because VOAM97 and VOAM99 surveys were performed in all three study areas,

20 the classification scheme used in this research has higher accuracy compared to the regional accuracy.

The Amazon floodplain was classified by Hess et al. (2003) in terms of both inundation state (flooded or nonflooded) and vegetation covers (non-vegetated, herbaceous, shrub, woodland, or forest) at the time of imaging. The five vegetation classes correspond to physiognomic classes of the National Vegetation Classification

Standard (NVCS, Federal Geographic Data Committee, 1997). Herbaceous is defined as non-woody plants as compared to woody plants of shrub, woodland, and forest. Shrub is dominated with individuals or clumps, woodland is dominated by trees with crowns (i.e. open tree canopy), and forest is dominated by trees with interlocking crowns (i.e. closed tree canopy). Nine hydrologic-vegetative categories include: terre-firme, open water, flooded-herbaceous, flooded-shrub, flooded-woodland, flooded-forest, nonflooded- herbaceous (or bare soil), nonflooded-shrub, and nonflooded-forest (Figure 2.2).

Table 2.1 presents areas and their percentages of the entire interferometric JERS-1 frame for these hydrologic-vegetation habitats as mapped from high- and low-water seasons in Balbina, Cabaliana, and Solimões-Purús. Balbina has 72% terre-firme and shows little change in the area of flooded (range from 22% to 20%) and nonflooded (3% to 8%) classes from high to low water seasons, respectively. In contrast, Cabaliana and

Solimões-Purús indicate significant change in the area of flooded (range from 37% to

12% and 60% to 10%) and nonflooded (7% to 37% and 8% to 59%) classes from high to low water seasons, respectively. Flooded shrub area disappears during the low water

21 season whereas none of the nonflooded herbaceous and nonflooded shrub classes exist during the high water season.

Figure 2.2: Classification maps for high and low water seasons: (a) Balbina, (b) Cabaliana, and (c) Solimões-Purús (source: Hess et al., 2003).

Balbina (km2 / %) Cabaliana (km2 / %) Purús (km2 / %) Class High Low High Low High Low

Upland Terre-firme 3036 / 72 3036 / 72 2043 / 45 2043 / 45 1071 / 23 1071 / 23

Open water Nonvegetated 140 / 3 16 / 0 523 / 11 257 / 6 412 / 9 351 / 8

Herbaceous 235 / 6 222 / 5 297 / 7 220 / 5 251 / 6 165 / 4

Shrub 3 / 0 0 / 0 72 / 2 2 / 0 94 / 2 0 / 0 Flooded Woodland 479 / 11 479 / 11 177 / 4 177 / 4 104 / 2 104 / 2 24 Wetland

Forest 215 / 5 171 / 4 1077 / 24 137 / 3 2280 / 50 182 / 4

Herbaceous 0 / 0 101 / 3 0 / 0 267 / 6 0 / 0 127 / 3 (or bare soil)

Nonflooded Shrub 0 / 0 39 / 1 0 / 0 146 / 3 0 / 0 114 / 2

Forest 106 / 3 150 / 4 316 / 7 1256 / 28 367 / 8 2465 / 54

Total 4214 / 100 4214 / 100 4505 / 100 4505 / 100 4579 / 100 4579 / 100

2.3 SAR Data and Processing

2.3.1 Interferometric Processing

JERS-1 scenes total to 23 for Balbina, 21 over Cabaliana, and 18 for the

Solimões-Purús thus permitting a variety of interferometric pairs. With the spatially coregistered scenes, all possible combinations of interferometric pairs were generated.

The Balbina site has 253 pairs, Cabaliana has 210, and Solimões-Purús has 153. Figure

2.3 presents the distribution of temporal and perpendicular baselines for all interferometric pairs used in this study. The repeat orbital period of JERS-1 is 44 days thus the minimum temporal baseline in the three study areas is 44 days. Table 2.2 presents the temporal and spatial characteristics of the interferometric pairs considered.

Some time spans are greater than 4 years (data were acquired between 1993 and 1997).

The maximum baselines are all below the JERS-1 theoretical critical baseline of 5.7 km where the correlation between the interferometric signals received by the two radar antennae drops to zero (Eriksson, 2004). The R2 values of the perpendicular baselines with respect to the temporal baselines in all three study regions are less than 0.04, thus a low strength of a linear relationship between the two variables. Overall the interferometric dataset is randomly distributed with respect to both temporal and perpendicular baselines such that there is no preferential sampling of any particular time- span or physical baseline that might skew the resulting coherence data set.

Figure 2.3: Physical and temporal baseline distributions for each study location. Temporal baselines are noted on the x-axis of all three plots whereas physical baselines are on the y-axis.

Temporal Perpendicular Baseline [days] Baseline [m] Location Scenes Pairs Acquisition Period Min. Max. Min. Max.

Balbina 23 253 February 23, 1993 to August 9, 1997 44 1628 128 4967

27 Cabaliana 21 210 February 25, 1993 to August 11, 1997 44 1628 3 4586

Purús 18 153 February 26, 1993 to June 29, 1997 44 1584 156 4231

Table 2.2: The temporal and spatial parameters of acquired SAR images.

Over open water, the transmitted radar pulse specularly reflects away from the off-nadir imaging JERS-1 SAR antenna, yielding low amplitude returns, poor interferometric coherence, and unreliable interferometric phase values. Over inundated vegetation, an L-HH radar pulse follows a path that penetrates the vegetation canopy, reflects specularly from the underlying water surface, backscatters from the vegetation trunks, and returns to the antenna (Richards et al., 1987; Hess et al., 1995). Multi-look amplitude images were generated by averaging 2 looks in range and 6 looks in azimuth to reduce speckle noise. The ground size of a pixel is 28 m in range and 27 m in azimuth.

The classification schemes were coregistrated to the amplitude images with 3rd order polynomials and nearest neighbor interpolation.

2.3.2 Coherence Variations

I produced coherence images for all interferometric pairs. Coherence is a measure of the phase consistency in returned radar energy between two SAR acquisitions and is defined by equation (2.1) (Zebker and Villasenor, 1992)

* s1  s2   (2.1) * * s1  s1 s2  s2

28 where  is coherence, s1 and s2 are complex values in SAR image 1 and SAR image 2, respectively, and s* is the conjugate of s . The braces indicate local spatial averaging around an individual multi-looked pixel: we used a 5 x 5 window, i.e., a 135 m spatial resolution, with a decreasing linear weighting scheme for pixels located away from the center of the window. The larger the window dimension, the higher the estimator accuracy, but the lower the resolution (i.e. low detection probability). In fact, using large estimation windows (i.e., averaging the data over large areas), many stable targets surrounded by noncoherent clutter are lost (Ferretti et al., 2001). In most literature coherence refers to the magnitude of the complex coherence and takes values between 0 and 1.

Total observed coherence is comprised of spatial, temporal, and thermal components, and is described by equation (2.2) (Zebker and Villasenor, 1992)

 sp a tia ltemp o ra lth erma l (2.2)

where  spatial is spatial baseline decorrelation which is a function of the perpendicular baseline, temporal is temporal decorrelation which depends on changes in the scattering centers between the two image acquisitions, and thermal is thermal decorrelation due to radar sensor noise. Also, coherence is affected by vegetation type (Hall-Atkinson and

Smith, 2001) and backscatter amplitude (Luckman et al., 2000). I analyze coherence variations of the hydrologic-vegetation classes on a pixel-by-pixel basis, albeit weighted

29 by the 5x5 window. Coherence is assessed with respect to baseline components, vegetation type, and inundation state. Six classes, terre-firme, open water, flooded herbaceous, flooded woodland, flooded forest and nonflooded forest, exist during both high and low water seasons and were used to estimate coherence variations.

2.4 Results & Discussions

2.4.1 Coherence Variations with Perpendicular Baselines

A perpendicular baseline is the distance, measured perpendicular to the radar look-angle, between two orbits of an interferometric pair. Figure 2.4 shows mean interferometric coherence variations of terre-firme, open water, flooded herbaceous, flooded woodland, flooded forest and nonflooded forest with respect to perpendicular baselines. Mean coherence is the average of all multi-looked pixels within a given hydrologic-vegetation class of a single interferogram. Only those pixels that are in the same class during both high and low water are used in this mean coherence measure

(Figures 2). This selection is necessary to isolate the influence of the physical baseline on coherence from that of vegetation, otherwise there would be a mixing of vegetation influence in Figure 2.4. All statistics reported below and in the tables are calculated from this mean coherence. For each study location, there are over 150 interferometric pairs, thus for display purposes the mean coherence values for each class were further averaged into 50 m bins of incremental perpendicular baseline size and plotted in Figure 2.4.

Short perpendicular baselines yield more topographic relief per phase cycle than long baselines, thus a more reliable estimate of surface change (Zebker and Villasenor,

1992). Assuming the scattering centers do not change substantially between acquisitions, short perpendicular baselines typically should yield better coherence than long baselines because the imaging geometry is more closely parallel (e.g. Kim et al., 2005). In our study area over Amazon floodplain, however, short perpendicular baselines at L-band do not benefit in yielding a quality coherence value (Figure 2.4). Furthermore, coherence values for all three locations are generally low, with Balbina yielding higher values than either Cabaliana or Solimões-Purús. Coherence randomly varies as perpendicular baseline increases: no trends are readily apparent within individual habitat classes. The high coherence peaks, e.g. at Bp = ~4800 m over Balbina and Bp = ~4600 m over

Cabaliana, are likely a result of the short temporal baseline of those particular interferometric pairs. Importantly, the flooded and nonflooded wetland classes have higher coherence values compared to terre-firme or to open water at all three study locations. Terre-firme has lower backscatter amplitudes than wetlands because it does not produce a double-bounce radar return. Over open water the transmitted radar pulse specularly reflects away from a side-looking SAR antenna thus the open water class has the lowest coherence.

Lake Balbina has stronger coherence values for each class and a different distribution of coherence values amongst the classes. The individual mean coherence values of the six habitat classes ranges from 0.28 to 0.14 in Balbina compared to 0.12 to

0.09 in Cabaliana and 0.11 to 0.09 in Solimões-Purús (Table 2.3). In Balbina, flooded

32 woodland and flooded herbaceous habitats have higher coherence values than flooded forest whereas in Cabaliana and Solimões-Purús this distinction is not clear with these habitats having slightly lower coherence values than flooded forest. Use of the nonparametric Mann-Whitney test of population distribution (Davis 1986, Hirsch et al.

1993) and a 99 percent probability level suggests that some habitat classes are identically distributed, i.e. flooded forest and nonflooded forest in Balbina; flooded herbaceous, flooded woodland, and nonflooded forest in Cabaliana; and flooded herbaceous, flooded woodland, flooded forest, and nonflooded forest in Solimões-Purús (Table 2.4). At a 99- percent probability (i.e., =0.01, or 1 percent chance of a type I error), the critical z value for failure in all tests is 2.33. Flooded herbaceous, flooded woodland, flooded forest and nonflooded forest classes have coherence variations that are statistically separate in

Balbina, but they cannot be similarly separated in Cabaliana and Solimões-Purús. It is important to note that open water is statistically distinct from all other classes.

Flooded Flooded Flooded Nonflooded Terre-firme Open water herbaceous woodland forest forest

Balbina 0.14  0.017 0.13 0.018 0.25 0.059 0.28 0.067 0.18 0.033 0.18 0.034

Cabaliana 0.10 0.008 0.09 0.006 0.11 0.008 0.11 0.010 0.12 0.013 0.11 0.010

Purús 0.10 0.007 0.09 0.006 0.11 0.007 0.11 0.010 0.11 0.013 0.11 0.009

Table 2.3: The mean plus and minus the standard deviation of coherences of terre-firme, open water, flooded herbaceous, flooded woodland, flooded forest, and nonflooded forest in Figure 2.4.

Balbina Cabaliana Purús

T O FH FW FF NF T O FH FW FF NF T O FH FW FF NF

T 4.69 14.27 15.08 10.42 11.02 15.88 4.26 5.36 8.18 5.67 13.76 4.92 5.38 4.47 3.99

O 14.57 15.25 11.61 12.14 16.57 16.46 16.85 16.70 14.34 14.35 14.20 14.25

35 FH 2.97 8.34 7.57 1.79 5.22 1.92 1.20 0.57 0.77

FW 10.66 10.06 3.34 0.04 0.44 1.76

FF 1.01 3.35 1.10

I suggest that the stronger coherence values for Lake Balbina, compared to

Cabaliana and Solimões-Purús, results from the difference in vegetation. The dominant vegetation of the inundated areas of Lake Balbina consists mostly of the flooded trunks of dead, leafless trees which are identified as the flooded woodland class (Table 2.1). The radar pulse may more easily penetrate this open canopy and subsequently support strong double-bounce returns. Cabaliana and Solimões-Purús, conversely, are dominated by living, flooded forests with a thicker canopy where leaves help prohibit a strong radar return (compared to Balbina) and thus diminish the strength of the coherence.

Overall, the coherence values at all three study locations are lower than typically used in conventional repeat-pass interferometry, e.g. the study of earthquakes in dry lands

(e.g. Massonnet et al., 1993; Zebker et al., 1994a). Yet, despite the low coherence values,

I suggest that they are greater than the noise floor. Open water represents this noise floor because it is mostly a specular reflector at L-band. Open water consists of constantly changing scattering centers, a result of wave action, which produce non-coherent repeat- pass interferometric phase. While, occasionally, waves within any given pixel may produce a similar scattering center structure for both SAR acquisitions in a repeat-pass interferogram, a spatial average of these is an indication of the noise floor. In Figure 2.4 and Table 2.4 all wetland classes are statistically distinct from open water and thus have coherence values all greater than noise.

2.4.2 Coherence Variations with Temporal Baseline

The temporal baseline indicates the time interval between two image acquisitions that form an interferometric pair. Over land surfaces where changes in soil moisture, vegetation, and freeze/thaw cycling cause random changes in the structure and dielectric properties of the scattering elements, interferometric coherence typically diminishes with increasing time between SAR acquisitions. Figure 2.5 shows mean temporal baseline coherence variations of terre-firme, open water, flooded, and nonflooded habitat classes.

Mean coherence is the average of all multi-looked pixels within a given hydrologic-vegetation class of a single interferogram and only those pixels that are in the same class during both high and low water are used in this mean coherence measure. We further average together coherence values from the three flooded classes to better estimate the influence of flooding on coherence. All statistics reported below and in the tables are calculated from this mean coherence. The repeat cycle of JERS-1 is 44 days, thus for plotting in Figure 2.5 the mean coherence values at each 44-day increment were averaged.

Figure 2.5: Mean coherence values of various vegetation-hydrologic habitat classes (i.e., terre-firme, open water, flooded, and nonflooded) compared to interferometric temporal baselines for each study location. Solid lines are least-squares fit, 2nd order polynomials. Annual periodicity is found in the flooded and nonflooded habitats but absent in the open-water and terre-firme classes.

As temporal baseline increases for the three study areas, mean coherence values of open water generally remain constant whereas all other classes show a decrease. Mean coherence values plotted with temporal baseline are greatest for the flooded classes

(green lines), of intermediate values for the nonflooded wetland classes (red lines), lower for terre-firme (black line), and least for open water (blue line).

The flooded and nonflooded wetland habitat classes show anomalous increases in temporal coherence values which occur annually whereas terre-firme and open water classes do not have similar annual signals (Figures 2.5 and 2.6). Normalized power spectrums, calculated after removal of the trends shown in Figure 2.5, demonstrate a strong annual periodicity for flooded habitats in all three study locations. Nonflooded wetland habitats have a similar or slightly weaker periodicity. In contrast, open water and terre-firme do not show this strong annual amplitude. Note that the annual periodicity in coherence results from temporal baselines stretching across one or multiple years, connecting two low water seasons (or connecting two mid-rising, two high, two mid- recessional, etc. seasons) and that these yield higher coherence values than temporal baselines connecting two differing seasons. This implies that interferometric phase for measuring dh/dt is more reliable with both a one year and same season encompassed in the temporal baseline rather than with a different combination such as six months and different seasons in the temporal baseline.

I suggest that this annual periodicity of coherence values is a direct result of the annual Amazon flood. In the central Amazon region, the flood wave peak arrives in late

June to early July whereas the trough can arrive over a slightly broader time range, but generally in October (Richey et al., 1989). Although this timing from peak to peak (or trough to trough, etc.) is not exactly 365 days, it is within the sampling accuracy represented by the 44-day repeat-pass cycle of JERS-1. Seasonal growth cycles in wetland vegetation are also timed with this flood pulse (Middleton, 2002; Junk, 1999).

Thus, the two primary hydro-geomorphic determinants of coherence, i.e., vegetation type and inundation state, are also annual. Conversely, terre-firme does not exhibit an annual inundation state and, likewise, does not exhibit a strong annual signal in temporal coherence (Figures 2.5 and 2.6). Finally, we note that all of the spectra for the flooded and nonflooded wetland classes exhibit more power than the noise floor represented by open-water (Figure 2.6). Thus, despite the overall low values of interferometric coherence for the various wetland classes, the values remain greater than noise for all temporal baselines (Figure 2.5).

Using 253 interferometric SAR coherence images in Balbina, 210 in Cabaliana, and 153 in Solimões-Purús, two-dimensional coherence variation plots were made with krigging interpolation (Figure 2.7). These plots show the spatial and temporal variation in coherence for the various habitat classes. Importantly, open water has the lowest coherence values for all combinations of spatial and temporal baselines whereas the flooded and nonflooded wetland classes show the highest coherence values.

2.4.3 Coherence Variations within High- and Low-Water Seasons

The low and high water seasons represent significant contrasts of inundation state on the Amazon floodplain. Lake Balbina, however, does not have similarly sharp contrasts in habitat with water level fluctuations (e.g. Figure 2.2(a), compare the percentages of flooded and nonflooded habitats between low and high water seasons in

Table 2.1). We use these two seasons to investigate the seasonal similarities and differences of interferometric SAR coherence variations. High water occurs during May,

June, and July whereas low water occurs in October, November, and December. The

GRFM mosaics and their classifications are also based on scene acquisitions from these months. Interferometric pairs constructed from scenes acquired during these months total to 15 (high water) and 15 (low water) for Balbina, 19 (high) and 6 (low) for Cabaliana, and 10 (high) and 10 (low) for Solimões-Purús.

High water coherence values are greater than those of low water for each habitat in Cabaliana and Solimões-Purús, with the exception of open water which shows essentially the same coherence values for both seasons (Figures 2.8 and 2.9). During high water and over a ~1 year time span, both Cabaliana and Solimões-Purús display coherence values for flooded and nonflooded habitats that are distinct from terre-firme and open water classes. For temporal baselines longer than ~1 year at high water, only open water remains distinct from the other Cabaliana and Solimões-Purús classes.

During low water in Cabaliana and Solimões-Purús, coherence values for the flooded,

43 nonflooded, and terre-firme habitats are not well delineated from each other, with only open water remaining clearly separate.

Balbina, however, displays rather similar coherence values for each individual habitat between high and low water seasons. Flooded habitats have higher coherence values and are distinct from nonflooded habitats in both seasons whereas open water has the lowest coherence values for both seasons.

Figure 2.8: Mean coherence values for interferometric pairs where both dates occur during the high-water season. Coherence values of various vegetation-hydrologic habitat classes (i.e. terre-firme, open water, flooded, and nonflooded) are compared to interferometric temporal baselines for each study location. Dots are the habitat mean coherence and circles are the average of these mean coherence values.

Figure 2.9: Mean coherence values for interferometric pairs where both dates occur during the low-water season. Coherence values of various vegetation-hydrologic habitat classes (i.e. terre-firme, open water, flooded, and nonflooded) are compared to interferometric temporal baselines for each study location. Dots are the habitat mean coherence and circles are the average of these mean coherence values.

2.5 Conclusions

Several key conclusions are drawn from our analyses. (1) Interferometric phase coherence does not vary with physical baseline, but does show a decrease with increasing temporal baseline for wetland habitats. It suggests that L-band coherence in wetland shows a function of temporal baseline that are compounding the physical baseline effect since water level and vegetation experience seasonal changes. (2) Interferometric coherence values in wetland habitats show annual periodicity with temporal baseline most likely because of the hydro-geomorphic effects of the annual flood wave. Temporal coherence from baselines spanning annual periods (e.g. 1-year, 2-year, etc.) or from much shorter time spans of less than two months (i.e. 44-days) are greater than those of intervening six month time spans. For example, the mean coherences of flooded habitats in Balbina from 44, 176, and 352 days in temporal baseline are 0.39, 0.27, and 0.32, respectively. Essentially, the interferometric coherence is timed with seasonal variations due to inundation state. (3) Because of the ―double-bounce‖ phenomenon, as radar backscatter amplitudes increase, interferometric coherence also tends to increase.

Flooded habitats are represented by stronger backscatter amplitude and, in general, also by higher coherence values compared to open water, terre-firme, and nonflooded habitats.

Taken together, these analyses suggest that, despite the rather low coherence values, interferometric phase of flooded habitats is a reliable measurement of temporal changes in water levels because these flooded habitats all have coherence values that are greater than open water and have temporal periodicity that most likely results from the annual

47 flood pulse. The low values, however, indicate the need for spatial averaging across areas larger than a single multi-look SAR pixel.

Chapter 3: A Comparison of Congo and Amazon Wetland Hydraulics from Repeat-pass Interferometric Satellite Measurements

3.1 Introduction

Water flow through wetlands controls a number of processes including changes in stored water, biogeochemical cycling, sediment delivery, and nutrient exchange. The floodplains and wetlands of large low-land rivers, such as the Amazon and Congo (Figure

3.1), are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. For example, access to Congo wetlands is difficult resulting in a paucity of published research on the surface water hydraulics. In contrast, in-situ measurements coupled with improvements in remote sensing (e.g., Birkett et al.,

2002) and modeling techniques have significantly added to our measurement and hydrologic mass-balance knowledge in the Amazon Basin (e.g., Alsdorf, 2003).

The Amazon Basin, with an area of ~6.0M km2 and containing the largest tropical rainforest in the world, contributes 15% to 20% of the global river discharge to the oceans (annually averaged discharge of ~200,000 m3/s). The floodplains of the Amazon

River and its tributaries are built from overbank deposits with an overall annually inundated area of about 750,000 km2 (Melack and Forsberg, 2001). The floodplain along just the mainstem Amazon River in Brazil covers at least 92,000 km2 (Sippel et al.,

1992).

The Congo River, with the second-largest discharge and basin area of any river, has not experienced the same degree of new research compared to the Amazon (annually averaged discharge is ~40,600 m3/s and area is ~3.5M km2, Laraque et al., 2001). Most of the primary research on the Congo swamps and wetlands is from the colonial era

(Campbell, 2005) with a limited number of surface water hydrology publications since then. Congo wetlands are known as swamp forests located mostly in broad topographic depressions of the central basin interfluvial regions. The Congo Basin is uniquely located with respect to the Inter-Tropical Convergence Zone (ITCZ) such that precipitation peaks occur twice annually (Kazadi and Kaoru, 1996). Our study locations on the Ubangui and

Mossaka rivers receive their major precipitation peaks in October-November (see flood wave peak for C1, Figure 3.2) whereas rivers draining the southern half of the basin tend to receive their peaks in March-April. Essentially, because the Congo Basin drains from both the Northern and Southern Hemispheres, it does not have the great seasonal fluctuations in water level as compared to the Amazon River (compare C2 to A1, Figure

3.2).

3.2 Study Areas and Interferometric SAR Data

The Global Rain Forest Mapping project (GRFM, e.g., De Grandi et al., 2000) provided regional JERS-1 SAR amplitude mosaics at both high and low water seasons in the Amazon and Congo Basins. The GRFM mosaics over the central Amazon Basin were acquired during the low-water period of September/November, 1995 and during the high- water period of May/June, 1996 whereas the GRFM mosaics over central Africa were acquired during the low-water period of January-March 1996 and during high-water period of October/November 1996. The Amazon study area is at the confluence of

Solimoes and Japura rivers whereas the study area in Congo includes the central part of the Congo basin, which is often called the Cuvette Congolaise (literally, ―saucer,‖ or

―shallow bowl‖, Figure 3.1) and is an immense depression containing Quaternary alluvial deposits that rest on thick sediments of continental origin, consisting principally of sands and sandstones (Sautter, 1966).

Interferometric JERS-1 SAR processing followed the ―two pass‖ method. Raw scenes with the same path were concatenated, and after flat-earth phase removal, the interferometric phase includes the topographic relief as well as any changes in the radar range (i.e. water level change; Alsdorf et al., 2000). The topographic related phase was subtracted using the C-band SRTM elevation data to make differential interferograms having phase values indicative of wetland water level changes. The short perpendicular baselines (Table 3.1) and the C-band SRTM relative height errors of 5.5 m (Farr et al.,

2007) cause less than 0.44 radians of phase change based on the topography and baseline

53 relationship. The resultant error is equivalent to 0.8 cm in the line-of-sight direction or

1.0 cm of water level change (i.e. vertical component).

Amazon Congo

Perp. 195m 97m 621m Baseline Ambiguity 229m/2pi 463m/2pi 72m/2pi Height

JERS-1 SAR April 9, 1997 Oct. 10, 1995 Sept. 28, 1996 Date May 23, 1997 Feb. 19, 1996 Nov. 11, 1996

Time Interval 44 days 132 days 44 days

Topex/Poseidon Water Level +103cm -278cm +126cm Altimetry Change

Landsat Date Oct. 18, 1986 Feb. 18, 2001 Feb. 12, 2002

Table 3.1: Description of Satellite Data.

Backscatter coefficient was used to differentiate between flooded vegetation, nonflooded areas, and open-water in river channels. Flooded vegetation in JERS-1 amplitude images yields brighter returns compared to non-flooded areas because the radar pulse is returned to the antenna when it reflects from water surfaces and scatters from inundated vegetation (i.e., the ―double bounce effect‖). In contrast, open-water river channels show little backscattering return, i.e., a specular reflection at L-band. The radiometric accuracy is considered sufficient for regional scale applications (Rosenqvist and Birkett, 2002). To 54 reduce range dependencies in the backscatter strength, we calculated the average of sigma naught ( 0 ) in each range bin, subtracted the linear trend of the averaged and added the mean of the linear trend. After performing the radiometric correction, the JERS study scenes display a backscatter response which is consistent over the entire range of incidence angles. Siqueira et al. (2003) give a simple classification and interpretation guide of multiseason JERS-1 SAR data for scattering mechanisms in which the ―double bounce‖ of flooded vegetation is greater than -5.5 dB and the noise floor of open water and bare ground is less than -15 dB. Jung and Alsdorf (2010) showed that, in the Amazon basin, flooded and non-flooded areas have a mean interferometric coherence greater than

0.1, including ~4 year temporal baselines, whereas open-water river channels have a mean coherence which is consistently less than 0.1. Thus, in our study we classified the pixels having backscatter brighter than -5.5 dB and interferometric coherence greater than

0.1 as flooded areas.

Measurements of changes in water levels (h/t) were obtained from unwrapping the differential phases of flooded areas with minimum cost flow techniques and a triangular irregular network (Figure 3.3). In the phase unwrapping stage, adaptive radar interferogram filtering was applied to reduce noise and enhance fringe visibility. After removal of the topographic phase (―two pass‖ method), the remaining differential phase depends on water level changes in the flooded areas and on any remaining, uncorrected topographic phase. Because non-flooded areas do not show water level changes, their differential phase is essentially zero. We take this as an indication that any uncorrected topographic phase over the flooded areas is also negligible. Taking into account the

55 wavelength and incidence angle of the JERS-1 SAR images (Alsdorf et al., 2000), interferometrically measured water level changes in the direction of the radar line-of- sight are converted to a vertical displacement where 1.0 radian of interferometric phase is equivalent to 2.4 cm of purely vertical height change.

The interferometric SAR measurements provide relative changes in water levels and thus require a reference datum to convert to absolute values (e.g., Alsdorf et al.,

2007a). In this study, Topex/Poseidon altimetry measurements are used for the reference datum. Topex ellipsoidal heights are converted to orthometric heights with respect to the

WGS84 reference system using the EGM96 geoid model. I spatially average Topex 10-

Hz retracked data using the modified 50% threshold retracker (Lee et al., 2008) over a distance corresponding to the intersection between the satellite ground track (i.e., A1, C1, and C2 yellow lines in Figure 3.1) and water body to construct a time series (Figure 3.2).

While a water mask (Hess et al., 2003) is used to identify the altimeter returns from

Amazon water surfaces, we defined Congo river or wetland waveforms as those which have automatic gain control values larger than 45 dB. The mean RMSEs of A1, C1, and

C2 altimetric measurements are 16, 14, and 12 cm, respectively. Table 3.1 presents altimeter measured water level changes, i.e., datums. In the Amazon and along the

Ubangui River in the Congo, these are reliable indicators of the datum, given the altimetric location in the center of the JERS-1 scenes and with a timing coincident with the acquisitions of the interferometric pair. Along the Mossaka and Sangha Rivers, the datum is less reliable because the altimetric location is 70 km south of the JERS-1 scenes.

3.3 Flow Hydraulics

The patterns of h/t on the Amazon floodplain are spatially complex, with distinct boundaries between pockets of small and large h/t values (R1, Figure 3.4). The

58 floodplain is built from the fluvial delivery of sediments and reworking over a few thousands of years by the migration of both the mainstem and floodplain channels

(Mertes et al., 1996). Discrete, episodic events delivering pulses of sediment are evident in portions of the floodplain several kilometers distal from the mainstem (Aalto et al.,

2003). Sinuous, convoluted, and interconnected floodplain channels are clearly evident in the Landsat and JERS-1 SAR imagery, e.g., low radar amplitudes (Figure 3.4). In the

SRTM DEM, floodplain channels are often flanked by anomalously high elevations most likely indicative of C-band radar returns from the forest canopy rather than the underlying floodplain topography. Sharp changes in the interferometric SAR h/t measurements coincide with floodplain channels (map view R1 and profile P1, Figure

3.4). The entire floodplain, located between terraces near the Japura and Solimoes Rivers

(Figure 3.3), is filled with water having an overall pattern of increasing h/t from upstream to downstream. Within this broad trend, mass continuity suggests that a greater amount of water is flowing to floodplain pockets with a greater h/t, compared to locations of smaller h/t (e.g., black arrows in Figure 3.4; Alsdorf et al., 2007a). There does not appear to be a clear relationship between topography and the h/t pattern: slightly lower (higher) elevations do not appear to have greater (lesser) h/t values.

Thus, while the main flood wave moves downstream, floodplain channels deliver water with greater amounts to distinct, small pockets of the floodplain, yet predicting the spatial details of this pattern based on local SRTM topography is not evident.

Flooded areas in the Congo study region are broadly distributed, lacking well- defined boundaries and do not appear to have abundant floodplain channels (Figure 3.3).

Compare, for example, Congo regions 2, 3, and 4 with the same sized region in the

Amazon where channels are many (Figure 3.4). Connectivity of Congo rivers to the adjoining, interfluvial wetland areas is thus limited, compared to the Amazon.

Furthermore, unlike the Amazon floodplain spatially limited by its terre-firme, the overall flooding pattern of the Congo does not have similarly distinct boundaries (Figures 3.1 and 3.3). Locally, however, the following examples highlight Congo flow hydraulics and relationships to topography.

Inundation of region 2 is complex and does not appear to simply follow SRTM topography. Both the Landsat and JERS-1 amplitude images show an overall, broad flooding pattern that covers the entire northern half of region 2 (e.g., regions of stronger radar amplitudes). Yet, SRTM topography over the same area contains both low and high elevations. In detail, h/t values are greater within small, low lying areas (red circles and profile P2, in region R2, Figure 3.4). The degree to which SRTM elevations are indicative of canopy or underlying topography is not clear in region 2. One channel is identified in the imagery and may represent a route for infilling of the interfluvial region.

Because radar returns are stronger in flooded vegetation, raphia palm areas (RP) in region 3 appear slightly more inundated compared to swamp forests (SF). SRTM elevations can be associated with canopies and thus elevation variations in region 3 may be indicative of changes in vegetation heights. In region 3, grass savanna (GS) is essentially the ground height (305m a.s.l.), raphia palms are 10m to 15m in height

61 whereas swamp forest trees are taller, between 15m and 20m. In profile P3, the central portion is marked by swamp forests about 5m higher than the raphia palms. Nearby grass savannas are about 15m lower than the adjacent raphia palms. While the topography underlying the canopy is likely more subtle than indicated by SRTM, there does appear to be a small, ~5m high ridge separating the Likouala-aux-Herbes River from the interfluvial areas to the northeast. This ridge is bisected by just one channel (yellow circle, Figure 3.4). Water level changes in region 3 are localized and decrease ~10 cm more in the middle compared to the edges of the ―bulls-eye‖ pattern in Figure 3.4. The

h/t values extend into the higher swamp forest, i.e., the ―island‖ along profile P3. Thus the degree to which SRTM topography is a control on water flow is not apparent (i.e., this ―island‖ of higher topography should not have an associated h/t low). Within the interfluvial wetland area, the h/t pattern is proximal to the isolated channel suggesting an evacuation route.

Similar to region 3, interfluvial topography in region 4 does not appear to be a control on h/t patterns. Water level changes along profile P4 are not bounded by

SRTM topography: h/t values show the least change at the ends of the profile but

SRTM topography shows essentially no bounding elevations. Like region 3, the h/t spatial pattern is localized and adjacent to one channel (yellow circle) which connects with the mainstem Congo River.

3.4 Conclusions

Our results suggest that a straightforward relationship is not apparent between

SRTM topography and Amazon or Congo flow hydraulics. Certainly, water is confined between prominent high relief areas, e.g., Amazon terre-firme. However, the SRTM topography across the Amazon floodplain does not clearly govern water flow, i.e., various pockets all having the same elevation do not have the same h/t. SRTM topography of Congo wetlands is confounded by vegetation but it is possible that much of the wetland area has topography more subtle than indicated by SRTM. In agreement with this suggestion of subtle topography is the lack of Congo wetland channels and corresponding broad, diffuse patterns of h/t. The many Amazon floodplain channels with identifiable connections to the mainstem contrasts sharply with the very few connections between Congo rivers and interfluvial areas. Essentially, the scale and magnitude of floodplain building processes in the Amazon are not similarly found in the

Congo.

Chapter 4: Flood Inundation Mapping in the Logone Floodplain from Multi-temporal Landsat ETM+ Imagery

4.1 Introduction

The Logone floodplain in the Lake Chad Basin is one of the excellent examples of coupled human and natural systems in most African floodplains (CHANS; Liu et al.,

2007). There are strong couplings between the hydrological, ecological and social systems as the extent, depth, and duration of seasonal flooding have an impact on vegetation quality and quantity, fish and other animal populations as well as human livelihoods. Local communities make use of the floodplain for agriculture, fishing and dry season grazing for more than 20 million people (Scholte, 2005). The productivity and carrying capacity of the Logone floodplain is highly correlated with the extent of the flooding (Loth, 2004). Wetland loss in the Logone floodplain has been accelerated due primarily to anthropogenic and natural processes, which impact the magnitude of flooding in the basins, and threatens the ecosystems. From 1988 to 2003, the Waza

Logone Project of the International Union for Conservation of Nature (IUCN) has been instrumental in rehabilitating the degraded Logone floodplain. The general objective of

64 the Waza Logone project was to achieve long-term enhancement of the biodiversity of the Logone area and to provide a sustainable improvement to the welfare of its rural population (Loth, 2004; Scholte, 2005). Climate variability and increased human water consumption have caused large changes to the water balance of the Lake Chad Basin. For example, the discharge of the Logone/Chari River system at N’Djamena has decreased by almost 75% over the last 40 years, from about 40 km3/yr in the early 1960s to 10-15 km3/yr in the 1980s and 1990s (Olivry et al., 1996; Coe and Foley, 2001).

Monitoring flood inundation areas of basins as well as water level changes of rivers and floodplains is necessary for understanding flood hazards, methane production, sediment transport, and nutrient exchange. The Logone floodplain includes flood-prone nations situated within this International River Basin (IRB), forming part of the international border between Chad and Cameroon. The challenge of issuing effective flood forecasts can be particularly difficult to overcome when there is no political agreement between riparian nations to share hydrologic information in real time for proactive flood management (Hossain and Katiyar, 2006). The member countries of the

Lake Chad Basin Commission (LCBC) have signed a data exchange protocol that covers this issue (LCBC, 2010). However, the wetlands of lowland rivers and lakes are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. Most of the major river systems in the Sahel region of West Africa contain extensive floodplains. In an average year the total inundated area of the major floodplains in the Sahel is ~67K km2 (Loth, 2004). The Logone floodplain in northern

Cameroon contains 10% of the total surface areas summed from major inland wetlands in

65 the West African Sahel (Wesseling et al., 1994). Another difficulty when studying the floodplain is the complexity of flood dynamics. Floods are two-dimensional. Water flow across wetlands is more complex than implied by 1D, point-based measurements. Flow paths and water sources are not fixed in space and time, but rather vary with floodwater elevations. Taken together with the size and complexity of the Logone floodplain dynamics, satellite remote sensing can play an important role of flood monitoring in the

Logone floodplain.

The growing availability of satellite data has increased the opportunities for flood inundation mapping from space (Smith, 1997; Alsdorf et al., 2007b). Previously, remote sensing techniques and hydrological modeling were applied to inform hydrology and floodplain dynamics in the study area. For measuring water surface elevations of rivers and wetlands in the Lake Chad Basin, TOPEX/POSEIDON radar altimetry was processed

(e.g. Birkett, 2000; Coe and Birkett, 2004; Crétaux and Birkett, 2006). Another is to identify flooded areas and measure annual flooding extents using Landsat 1 (e.g. Benech et al., 1982), ENVISAT Advanced Synthetic Aperture Radar (ASAR) and SPOT optical image (e.g. Westra et al., 2005), and Moderate Resolution Image Spectrometer (MODIS)

(e.g. Westra and De Wulf, 2009). Other approaches include combining hydrological modeling with Shuttle Radar Topography Mission (SRTM) data to simulate the water balance of the basin (e.g. Coz et al., 2009) and developing a sophisticated hydrodynamic model with in-situ measurements to assess the restoration potential of the Logone floodplain (e.g. Evans et al., 2003).

Here, we present the first study of the relationship between flooding extents and water height variations in the Logone floodplain using space borne data. Flooding extents are calculated from the most recent data from the Landsat sensor Enhanced Thematic

Mapper Plus (ETM+) whereas water height variations are provided from ENVISAT altimetry in the floodplain and gauge stations in the river. The results will significantly add to our understanding of the Logone floodplain dynamics and provide an opportunity to investigate the impacts of flood hazards in the highly interconnected ecological and social systems in the Logone floodplains.

4.2 Study Area

The Logone floodplain, known locally as the Yaéré, is located in the Lake Chad

Basin, Africa (Loth, 2004). The Lake Chad Basin is immense, covering an area of 2.5M km2 (Crétaux and Birkett, 2006; Coz et al., 2009) as compared with the Amazon Basin area of 6.0M km2 and the Congo Basin area of 3.5M km2. The Logone River is a major tributary of the Chari River. The Logone/Chari River waters flow into the Lake Chad

Basin from the south and gradually move northwards, supplying permanent open water and seasonally inundating the marsh regions. Approximately 90% of the Lake Chad’s water stems from the Logone/Chari River system. These rivers have their origins in

Cameroon and the Central African Republic, respectively. The remaining 10% stems from other tributaries and local precipitation (FEWS, 1997).

The Logone floodplain lies at the south of Lake Chad and the northeast of

Mandara Mountains (Figure 4.1). The climate is semi-arid and the average annual rainfall varies from 750 mm/yr in the south to 550 mm/yr in the north (Westra and De Wulf,

2009). A Landsat ETM+ frame, with a size of 176×171 km, includes the floodplain

(Table 4.1). The superimposed SRTM elevation map in Figure 4.1(a) shows that the study area is so flat as the topography slope is ~0.6 m/km. The basin topography apart from some local mountains is quite flat as indicated by the overall median slope value of

~1.3% south-north gradient (Coz et al., 2009). This flatness leads to the existence of extensive floodplains, which play a significant role in the regional water balance by redistributing water through evaporation (Gac, 1980; Olivry et al., 1996). The flooded area appears very dark green in Figure 4.1(b). The inundation mechanism depends to a great degree on rainfall patterns. The first rains, which normally occur in May, saturate the soils and begin to fill the deepest depressions. Overbank flow is, however, by far the biggest contributor to the inundation of the floodplain, which takes place from September to October (Mott MacDonald, 1993, 1999). The annual overbank flooding reduces both the peak flows and total volume of water in the Logone River. Water stored on the floodplain is returned during the recession period. The floodplain regulates the river by distributing flows throughout the year (Loth, 2004).

4.3 Materials and Methods

4.3.1 Landsat ETM+ Data

Landsat ETM+ (i.e. Landsat 7) is the most recent in a series of Landsat sensors that have a 30×30 m spatial resolution and with a 16-day revisit capability to provide a balance between requirements for localized high spatial resolution studies and large area monitoring (Arvidson et al., 2001; Goward et al., 2001; Williams et al., 2006). Table 4.1 summarizes the Landsat ETM+ dataset from 2006 to 2008 used in this study. The 33 69 scenes are available in the USGS Earth Resources Observation and Science Center

(EROS) archive at no charge. All the acquisitions are processed to Standard Terrain

Correction (Level 1T) for the research.

Cloud/cloud- Site location *Flooded Nonflooded Acquisition date shadow (km2 / (path / row) (km2 / %) (km2 / %) %) 4 January 2006 678 / 2 29175 / 98 0 / 0 5 February 2006 211 / 1 29000 / 97 642 / 2 25 March 2006 34 / 0 29819 /100 0 / 0 10 April 2006 15 / 0 28360 / 95 1478 / 5 26 April 2006 65 / 0 24984 / 84 4805 / 16 28 May 2006 77 / 0 29777 /100 0 / 0 15 July 2006 129 / 0 28238 / 95 1487 / 5 1 September 2006 526 / 2 28997 / 97 330 / 1 17 September 2006 1199 / 4 28165 / 94 489 / 2 3 October 2006 2702 / 9 23557 / 79 3594 / 12 20 November 2006 3958 / 13 25896 / 87 0 / 0 6 December 2006 2847 / 10 27006 / 90 0 / 0 22 December 2006 1385 / 5 28469 / 95 0 / 0 7 January 2007 453 / 2 29401 / 98 0 / 0 8 February 2007 100 / 0 29753 / 100 0 / 0 Logone 24 February 2007 1 / 0 29745 / 100 108 / 0 floodplain 12 March 2007 17 / 0 29837 / 100 0 / 0 (P184 / 28 March 2007 15 / 0 29152 / 98 687 / 2 R052) 13 April 2007 18 / 0 22119 / 74 7716 / 26 31 May 2007 49 / 0 29379 / 98 425 / 1 23 November 2007 2731 / 9 27123 / 91 0 / 0 9 December 2007 1663 / 6 28191 / 94 0 / 0 25 December 2007 843 / 3 29010 / 97 0 / 0 26 January 2008 253 / 1 29601 / 99 0 / 0 11 February 2008 158 / 1 29695 / 99 0 / 0 27 February 2008 89 / 0 29765 / 100 0 / 0 14 March 2008 17 / 0 29178 / 98 658 / 2 30 March 2008 60 / 0 29793 /100 0 / 0 15 April 2008 32 / 0 29821 / 100 0 / 0 18 June 2008 43 / 0 23228 / 78 6583 / 22 24 October 2008 5764 / 19 24089 / 81 0 / 0 9 November 2008 4894 / 16 23538 / 79 1422 / 5 25 November 2008 5014 / 17 24840 / 83 0 / 0

Table 4.1: Summary of Landsat ETM+ dataset and classification results. *Flooded class excludes open water of Logone/Chari Rivers and Lake Maga with an area of ~ 279 km2 in the Landsat ETM+ swath.

Two primary limitations to the utility of Landsat ETM+ data are (1) the availability of cloud-free surface observations and (2) the failure of the Scan Line

Corrector (SLC) that compensates for the forward motion of the satellite. Clouds are common features of visible and infrared remotely-sensed images collected from many tropical, humid, mountainous, and coastal regions of the world (Martinuzzi et al., 2007).

Cloud cover reduces the number of Landsat surface observations. To construct a time- series dataset at a finer temporal resolution, despite masking out cloud areas, Landsat

ETM+ data were collected with a cloud cover of less than 40% from USGS Global

Visualization Viewer for the 3-year study period. Cloud and cloud-shadow were determined from the blue reflectance value. The cloud cover areas were derived from a majority analysis of 3 by 3 window size for pixels in which blue reflectance (i.e. Landsat

ETM+ band 1) is greater than 0.2. The cloud-shadow areas were then masked by the cloud areas including 10 pixel buffer zones (Sakamoto et al., 2007). A SLC instrument malfunction occurred onboard Landsat 7 on 31 May 2003 (NASA, 2009). The Landsat 7

ETM+ is still capable of acquiring useful image data with the SLC turned off, reducing the usable data in each SLC-off scene by about 22% (Maxwell et al., 2007). The NASA gap filling software developed in IDL programming (Storey et al., 2005; NASA, 2009) goes through two steps to fill gaps in the Landsat ETM+ dataset used in the study. The first step, re-framing, processes all of the input imagery to create images that have the same dimensions in line length and number of lines. The second step replaces the no-data pixels in the image by linear least squares regression analysis using their counterparts in temporally close and in anniversary scenes.

The short-wave infrared (SWIR) data is used to identify flooded area in the study area. SWIR data are highly sensitive to moisture content in the soil and the vegetation canopy (Sakamoto et al., 2007). Westra and De Wulf (2009) have recently demonstrated that the Moderate-Resolution Imaging Spectroradiometer (MODIS) band 7 (wavelength:

2.105-2.155 m) provides better results for delineating the flooding extent in the Logone floodplain rather than the MODIS Normalized Difference Vegetation Index (NDVI) and the MODIS Normalized Difference Water Index (NDWI). A threshold of 0.08 in reflectance unit was used to calculate the flooding extent for the 2000-2005 periods. To apply the approach to all of the acquired Landsat ETM+ data used in the study, the digital numbers (DN) of the original Landsat ETM+ band 7 (wavelength: 2.09-2.35 m) are converted through spectral radiance at the sensor’s aperture into top of atmosphere

(TOA) reflectance. The detail of the conversion follows equations described in the

Landsat 7 science data user’s handbook (NASA, 2009). The converted planetary TOA reflectance incorporates the solar zenith angle, earth-sun distance, and calibrated radiance related to outer space radiance, thus considering the conversion from DN to reflectance unit as requiring correction for quantitative remote sensing applications (Liang et al.,

2002; Ouaidrari and Vermote, 1999). Open water of the Logone/Chari Rivers and of

Lake Maga is classified from a Landsat 7 SLC-on image dated 21 October 2001. The water bodies mask out the classified flooded/nonflooded areas to focus on floodplain inundation dynamics for the 3-year study period. The Iterative Self-organizing Data

Analysis (ISODATA) clustering method is implemented to classify the open water.

ISODATA is an unsupervised classification scheme that uses an iterative approach

73 incorporating a number of heuristic procedures to compute classes (Tou and Gonzalez,

1974; Melesse and Jordan, 2002). The ISODATA utility in the study repeats the clustering of the image into 20 classes until the minimum percentage of the cluster is set to 0.05%.

4.3.2 ENVISAT Altimetry Data

The ENVISAT altimeter data are processed and selected from January 2006 to

December 2008. The ENVISAT orbits on a 35-day repeat cycle with 98.5° inclination.

The ENVISAT Geophysical Data Record (GDR) contains 18-Hz retracked measurements, corresponding to an along-track ground spacing of approximately 350m.

In this study, ICE-1, which has been proved to perform well over inland water bodies

(Frappart et al., 2006; Lee et al., 2010), retracked measurements are used. The instrument corrections, media corrections (i.e. dry troposphere correction, wet troposphere correction calculated by the French Meteorological Office (FMO) from the European Centre for

Medium-Range Weather Forecasts (ECMWF) model, and the ionosphere correction based on Global Ionosphere Maps (GIM)), and geophysical corrections (i.e. solid Earth tide and pole tide) have been applied. To identify the radar returns from the water surface, the 18-Hz ICE-1 backscattering coefficients are examined over the 18-Hz locations selected (Lee et al., 2009). The backscattered energy is generally higher over the floodplain than the surrounding dry land with moderate vegetation cover. We select the 18-Hz radar returns that have the backscattering coefficient higher than 20 dB, and

74 spatially average them (i.e. red dots in Figure 4.1(b)) to construct a time series in the

Logone floodplain on ascending pass 973 and descending pass 272 (Figure 4.3). The mean RMSEs of ENVISAT 973 and ENVISAT 272 altimetric measurements are 7 and 3 cm, respectively (see error bars in Figure 4.3).

4.3.3 Ground-based Data

In situ measurements of daily gauge height at Katoa, Logone Gana, and

N’Djamena in the Logone River (see Figure 4.1(b)) are collected to compare with the

Landsat-derived flooding extents and the ENVISAT radar altimetric water height variations. Missing data at Katoa and N’Djamena are filled with the average daily values for the previous years from 2000 to 2005. N’Djamena, Chad's capital city, is at the location where the Logone River empties into the Chari River. Logone Gana is located at the middle of the study area and Katoa lies at the east of Lake Maga to provide a time series of water heights in the Logone River.

4.4 Results and Discussion

4.4.1 Flooding Extent

Flood inundation maps are generated from 33 Landsat ETM+ scenes summarized in Table 4.1 (Figure 4.2). The time interval between the successive flood maps varies

75 from 16 days in the satellite repeat cycle to 176 days from June to October 2007 during low water. Flooded area is classified and colored red in Figure 4.2. The flooding inundates a large area of the Logone floodplain between Katoa and N’Djamena.

Variability in flooding extents during high water are comparatively greater than during low water, thus illustrating yearly flooding in the study area. The flooding extents are less changeable during low water. The duration of the floodwater is most likely 5 months from September to January. The finding supports that the saturated soils during rainy season (i.e. June to August) have no more capability for storing water and allow inundation of the floodplain for three to five months after the overbank flooding from the

Logone River (Westra and De Wulf, 2009). The largest flooding extent for the study period reaches ~5.8K km2 and occupies ~19% in the image on 24 October 2008 (Table

4.1). Despite excluding ~1.3K km2 potential floodplain at the south of Lake Maga out of the used Landsat ETM+ swath, the result is consistent with 6.71.8K km2 in the mean

MODIS-observed maximum flooding extent from 2000 to 2005 (Westra and De Wulf,

2009). The maximum annual flooding extents of ~4.0K km2 on 20 November 2006 and

~2.7K km2 on 23 November 2007 are much less than that of 2008 because Landsat

ETM+ dataset in 2006 and 2007 do not include the maximum flooding period from late

October to early November. Besides the sampling issue, the maximum extent of the flooding varies highly from one year to another depending on the yearly amount of runoff and the soil moisture of the floodplain prior to flooding (Loth, 2004).

Figure 4.2: Flood inundation maps derived from multi temporal Landsat ETM+ imagery. See text for processing details.

Cloud/cloud-shadow occupies up to 26% of the image on 13 April 2007. All of the images acquired during high water have less than 5% cloud/cloud-shadow covered except an image on 3 October 2006. A time series of flooding extents in the uppermost of

Figure 4.3 shows a smooth trend of the flooding extent variation despite the cloud contamination. It suggests that most of the cloud/cloud-shadow areas are most likely nonflooded. But, flooding extent seems to be underestimated by about 5% (i.e. ~1.4K km2 cloud/cloud-shadow coverage on 9 November 2008), thus having a lower size of flooding extent rather than both prior and posterior flood maps on 24 October and 25

November 2008 (see the last three bar graphs in Figure 4.3).

4.4.2 Water Height Variation

Time series of water height variations are generated using ENVISAT altimetry in the floodplain and in-situ measurements in the river (Figure 4.3). The altimeter-observed height is relative to the reference ellipsoid whereas the river gauge height is with respect to a local datum. To compare relative water height variations from both the altimetry and the river gauges, they are converted into water heights above observed minimum in

Figure 4.3. ENVISAT altimetry provides 35-day repeated water height measurements in the floodplains (see ENVISAT pass 973 and ENVISAT pass 272 shown in Figure

4.1(b)). The ENVISAT altimetric measurements are linearly interpolated to estimate daily water height variations between two successive altimetric measurements marked with circles in Figure 4.3. The amplitude of water heights in ENVISAT pass 272 is ~2.5 m, which is greater than the ~1 m amplitudes observed in ENVISAT pass 973. Since

ENVISAT pass 272 is in the proximity of the Logone River, it is more influenced by the overbank flooding from the river. The observed minimum altimetric ellipsoidal heights are 316.41 m in ENVISAT pass 973 and 307.79 in ENVISAT pass 272. The height difference can be indicative of the local slope in the floodplain, thus suggesting that floodwaters in ENVISAT pass 973 flow downhill to the north. Three local river gauge stations provide daily water heights at Katoa, Logone Gana, and N’Djamena. The river water amplitudes become greater downstream, compared to upstream, and they are all greater than those of altimetric measurements in the floodplain. The river water amplitudes are ~3.5 m at Katoa, ~5.5 m at Logone Gana, and ~5.8 m at N’Djamena. The

80 results are consistent with seasonal water height amplitudes of 5-6 m for the most southerly zones of the Logone and Chari Rivers and of 1-2 m for the wetland areas using

TOPEX/POSEIDON altimetry from 1993 to 1998 (Birkett, 2000). Although Logone

Gana and N’Djamena show similar amplitudes of the water height variations, the

N’Djamena hydrograph becomes sharply increased and decreased more than Logone

Gana. N’Djamena is located immediately after the confluence between Logone and Chari

Rivers, thus the N’Djamena heights are a summation of the flows in these two rivers. The peak-level at Katoa is observed in late September, which is earlier than the late October peak-level periods at Logone Gana and N’Djamena. Birkett (2000) showed the phase lag of one to two months in water heights between the upstream Chari River and Lake Chad during the 1990s.

4.4.3 Correlation between Flooding Extent and Water Height Variation

A positive relationship with a time lag is found between flooding extents and water height variations for all five sites (Figure 4.4). The correlations are calculated based on a second order polynomial regression model and time shifting. On the graphs, the y-axis represents flooding extent and the x-axis represents the water height proportion of maximum height above observed minimum at each site. Coefficients of determination

(i.e. R2) range from 0.50 to 0.89 before time shifting. Altimetric water heights are more highly correlated with flooding extents than river gauge heights. The altimetric measurements are collected in the floodplain and they do not anticipate a large time lag

81 between their water height variations and flooding extents as much as river gauge heights. After performing the optimized time shifting, the correlations at river gauges increased to greater than 0.95. The delayed time shifting (i.e. phase lag) increases downstream compared to upstream. River gauge data at Katoa, Logone Gana, and

N’Djamena have the highest correlations as they are shifted +36, +22, and +13 days, respectively. The correlations of the altimetric measurements also increase as they are shifted +4 and -15 days within the ENVISAT repeat cycle of 35 days. This is caused by linear interpolation problem and ENVISAT altimetric measurement error as well as a small time lag.

The regression model calculates flooding extents only when water heights are above a threshold level of flooding. The model does not consider flooding when water heights are less than an x-intercept defined by the second order polynomial in Table 4.2.

The water heights in x-intercepts at river stations are converted into flow rates based on local discharge rating curves from their absolute river gauge heights. The flow rate could be a good index of flooding intensity as floodwaters approach overbank flooding levels.

The regression model during high water on the x-axis becomes more sensitive to flooding extents on the y-axis due to lower degree of the freedom (i.e. more severe slope of the second polynomial fitting line) as compared to during low water. Table 4.3 summaries the corresponding flow rates to flooding extents during high water in the regression models with time shifting. The result suggests that flooding to 4000 km2 in the study area is most likely to occur 36, 22, and 13 days after the flow rates exceed 949 m3/s at Katoa,

793 m3/s at Logone Gana, or 2010 m3/s at N’Djamena, respectively.

Regression model (y = a∙x2 + b ∙x + c; X-intercept Site (time shifting) y: flooding extent R2 x: water height, *H1 **H2 ***Q a,b,c: constants) (prop.) (cm) (m3/s) ENVISAT 973 8867∙x2-4493 ∙x+507 0.89 0.34 31670 ENVISAT 973 (+4 days) 10169∙x2-5800 ∙x+650 0.91 0.42 31677 ENVISAT 272 13110∙x2-9436 ∙x+1598 0.89 0.45 30891 ENVISAT 272 (-15 days) 12651∙x2-7972 ∙x+1112 0.94 0.42 30884 Katoa -10710∙x2+13650 ∙x-666 0.50 0.05 78 30 Katoa (+36 days) 5976∙x2-998 ∙x+89 0.98 0.08 88 43 Logone Gana 1576∙x2+2753 ∙x-80 0.71 0.03 21 13 Logone Gana (+22 days) 2 85 6832∙x -2372 ∙x+222 0.97 0.17 100 116

N’Djamena 4767∙x2+1089 ∙x-81 0.86 0.06 89 81 N’Djamena (+13 days) 6217∙x2-930 ∙x+65 0.96 0.07 98 98

Table 4.2: Summary statistics for regression models in Figure 4.4.

*H1 corresponds x-intercept in Figure 4.4. **H2 is the corresponding absolute water height to H1 with respect to the reference ellipsoid for ENVISAT altimetry and local datum for river gauge. ***Q is the corresponding flow rate to H2 based on local discharge rating curve from river gauge height.

Flow rate (m3/s) Flooding extent (km2) Katoa Logone Gana N’Djamena

1000 374 386 835

2000 569 532 1280

3000 765 664 1660

4000 949 793 2010

Table 4.3: The estimation of flow rates for flooding extents in the regression models with time shifting.

4.4.4 Logone Floodplain Dynamics

The Logone floodplain dynamics are analyzed with monthly average flood probability maps (Figure 4.5). The flood probability is calculated on a pixel-by-pixel basis from flood maps in the same month for the 3-year study period. The probability varies from 100% in flooded (i.e. red) into 0% in nonflooded (i.e. white). Blue represents open water such as Logone/Chari Rivers and Lake Maga. For instance, red areas on

November map in Figure 4.5 represent flooding in all November flood maps of Figure

4.2 (i.e. 20 November 2006, 23 November 2007, 9 November 2008, and 25 November

2008) whereas orange areas represent flooding in only two out of the four flood maps.

The flood probabilities multiplied by a single pixel area of 900 m2 are summed to calculate monthly average flooding extents. The monthly average flooding extent takes up 863 km2 in September, reaches up to 4233 km2 in October and 4149 km2 in November during maximum flooding period, and reduces into 1684 km2 in December and 863 km2 in January during the drainage period.

The Logone floodplain clearly shows the spatial variation of flooding as well as the size of flooding extent from September to January. Figure 4.5 marks the most frequently flooded location (i.e. the mode in statistics) in latitude and longitude with ―X‖ at each map. The monthly mode locations suggest that floodwater drains to the northwest and eventually into Lake Chad. In September, most flooded areas are located at the east side of the Logone River. In October and November, the flooded areas are spread out to both sides of the river, but the mode locations move from the east to the west. In

December and January, the mode locations move rapidly toward the north with decreasing size of the flooding extent.

The north of Lake Maga is mostly nonflooded even during the maximum flooding months of October and November in Figure 4.5. This is consistent with the finding that

Maga dam eliminates flooding and results in a decrease of depth and extent of the 88 flooding because the dam intercepts discharge from the Logone River and the runoff from the Mandara Mountains (Loth, 2004). To date, the dam and embankment have altered the tributary network and discharge in the Logone floodplain since the construction in 1979.

4.5 Conclusions

This study is the first demonstration of a strong correlation between flooding extents and water height variations in the Logone floodplain using space borne data. The regression model can facilitate a flood monitoring system and can support a flood prediction system with a few weeks prior to the overbank flooding from the Logone

River and in combination with river gauge data.

The multi-temporal Landsat ETM+ imagery renders the first time series of flooding extents in the Logone floodplain for the recent three years. The high-resolution flood inundation maps provide a better understanding of the complex floodplain dynamics despite cloud/cloud-shadow contamination. The overbank flooding starts at the east of the Logone River in September. The monthly average flooding extent increases to

~4.2K km2 in October and November. As inflow reduces, the floodwater drains to the northwest in December and January. Loth (2004) pointed out that part of it returns back to the Logone River, part of it contributes to the groundwater through infiltration, and part of it is lost through evapotranspiration during recession period. The results are the first real confirmation to the claims of the local people that Maga dam and embankment

89 influence the tributary network and discharge in the Logone floodplain to date since the construction in 1979.

Rainfall, runoff, soil moisture, and evapotranspiration by plants are not explored in parallel with water height variation to study the relationship with the degree of inundation in this study. These are less likely to influence the flooding system in the

Logone floodplain, but the data deficiencies could explain some errors in the regression analysis. Another weakness of the study is that a time series of flooding extents has no flood map in August for the 3-year study period due to the unavailability of Landsat images.

Volumetric storage change can be estimated by multiplying flooding extent and water height change. Given the largest flooding extent of 5764 km2 on 24 October 2008 and the highest water height variation of 1 m/yr on ENVISAT pass 973 in the middle of the floodplain, the first order of the annual floodplain storage change is ~5.8 km3/yr. But, the depth of floodwaters varies too much in space and time in the floodplain. For a promising future research, having more dense water height measurements in the floodplain makes it feasible to estimate the storage change in the catchment.

Chapter 5: Hydrodynamic Modeling of the Congo Wetlands Using LISFLOD-FP and Spaceborne Data

5.1 Introduction

The Congo Wetland is the largest wetland in Africa and plays an important role in global climatic and hydrological system. The Congo Wetland provides significant scales, a dynamic hydrologic cycle, and an opportunity to investigate the impacts of drought and deforestation on tropical surface water stores and fluxes. Water flow through wetlands controls a number of processes including changes in stored water, biogeochemical cycling, sediment delivery, and nutrient exchange. The floodplains and wetlands of large low-land rivers, such as the Amazon and Congo are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. However, the Congo Wetland, with the second-largest discharge and basin area of any river, has not experienced the same degree of new hydraulic research compared to the Amazon

(annually averaged discharge is ~40,600 m3/s and area is ~3.5M km2; Laraque et al.,

2001).

Hydrodynamic and hydrological modeling efforts have begun to rely on remotely sensed observations that either directly record water surface elevations using satellite

91 radar altimetry (e.g. Koblinsky et al., 1993) or infer stage and discharge from relationships between main channel gauge data and remotely sensed inundated area (e.g.

Smith et al., 1995). My previous work has recently been demonstrated to measure water level changes with time (Jung et al., 2010). Here, I quantify the spatial and temporal distribution of water level and storage changes in the central Congo wetland using

LISFLOOD-FP hydrodynamic modeling and spaceborne data. The hydrodynamic modeling in the Congo wetland will provide water flow dynamics in the rivers and wetlands and thus help to suggest an opportunity to investigate the impacts of drought and deforestation on tropical surface water stores and fluxes.

5.2 Methods

5.2.1 LISFLOOD-FP

The hydraulic routing numerical technique is modeled after Garbrecht and Brunner

(1991) and uses the Muskingum-Cunge routing model. The hydraulic computations are split between the channel and floodplains using an energy balance and bankfull discharge approach described by Garbrecht and Brunner (1991). This research uses the LISFLOOD-

FP hydraulic model which was originally developed by Bates and De Roo (2000) and yielded good predictions of maximum inundation extent for fluvial flooding problems (e.g.

Bates and De Roo, 2000). I compiled LISFLOOD-FP on the Ohio Super Computer (OSC) and made it run within the computational time of 90 hours.

Within LISFLOOD-FP, channel flow is handled using a one-dimensional approach that is capable of capturing the downstream propagation of a floodwave and the response of flow to free surface slope, which can be described in terms of continuity and momentum equations as:

Q A   q (5.1) x t

4 n2 P 3Q2 h S    0 0 10   (5.2) A 3 x

where Q is the volumetric flow rate in the channel, A the cross sectional area of the flow, q the flow into the channel from other sources (i.e. from the floodplain or possibly tributary

channels), S0 the down-slope of the bed, n the Manning’s coefficient of friction, P the wetted perimeter of the flow, and h the flow depth.

Floodplain flows are similarly described in terms of continuity and momentum equations. The flow between two cells is assumed to be simply a function of the free surface height difference between those cells:

hi, j Qi1, j Qi, j Qi, j1 Qi, j  x x y y (5.3) t xy

5 / 3 i1, j i, j h flo w h  h Q i, j  ( )1/ 2 y (5.4) x n x

93 where hi, j is the water free surface height at the node (i,j), Δx and Δy are the cell dimensions, n is the effective grid scale Manning’s friction coefficient for the floodplain,

Q Q and x and y describe the volumetric flow rates between floodplain cells. The flow depth, h flow , represents the depth through which water can flow between two cells, and is defined as the difference between the highest water free surface in the two cells and the highest bed elevation (this definition has been found to give sensible results for both wetting cells and for flows linking floodplain and channel cells).

The key parameters in this model that will be estimated from satellite based measurements include river parameters (e.g. river channel width and depth, river bank and bed elevation), boundary conditions (e.g. water surface elevation, river discharge), and floodplain bare ground elevation.

5.2.2 River Parameters

The study area is located at the central part of the Congo Basin, which is often called the Cuvette Congolaise (literally, ―saucer,‖ or ―shallow bowl‖, Figure 5.1). This is the confluence of the major Congo, Ubangi, and Sangha Rivers and has been studied as part of the largest wetland in Africa. Global Rain Forest Mapping (GRFM) mosaics,

Shuttle Radar Topography Mission (SRTM) data, and Hydrological data and maps based on Shuttle Elevation Derivatives at multiple Scales (HydroSHEDS) are used to input river parameters into the LISFLOOD-FP model. All of the available spaceborne images

94 were map-projected into UTM (Universal Transverse Mercator) South 33 and WGS

(World Geodetic System) 84 coordinate before the river parameters were derived.

River channel width was calculated using GRFM mosaics in the central Africa

(Figure 5.1). Open water appears very dark in radar amplitude images, in contrast with non-water, thus river channels can be simply extracted in the radar mosaics with a density slicing method (Hess et al., 2003). The GRFM mosaics were originally mapped in side- looking radar coordinates so orthorectification were carried out to make corrections for geometric distortion. Co-registration of the mosaic with the SRTM follows methods demonstrated by Sheng and Alsdorf (2005). Multiple channels in the Congo and Ubangui

Rivers are refined as single channel rivers for simplicity and efficiency of the model.

River network vector lines were obtained from the HydroSHEDS in 15 arc-seconds.

River bank elevation and river bed elevation were generated from the SRTM data. The river bank elevations were low-pass filtered with a median filter to reduce spiky noise elevations due to canopy heights. The first order river bed elevation was calculated by subtracting water depth from river bank elevation in the SRTM data. The water depth was calculated based on power law relationship between river discharge and river water depth (Moody and Troutman, 2002). The regression relations obtained are:

Dˆ  0.27Q0.39 (5.5)

where Dˆ is the regression estimated water depth and Q is river discharge.

Figure 5.1: LISFLOOD-FP Hydraulic model area in SRTM DEM (left-upper), water mask (left-bottom), and river networks in GRFM mosaic (right). This area is composed of several river networks: R0-Congo, R1-Giri, R2-Ubangui, R3-Likoula-aux-Herbes, R4- Sangha, R5-Mambili, R6-Likoula (Mossaka), R7-Kouyou, R8-Alima, and R9-Nkeni Rivers. Red dots indicate 10 input discharge locations in the rivers for upstream boundary conditions and green dot indicates an input river height location for downstream boundary condition. Yellow box locates ALOS PALSAR interferometric swath and orange line locates ENVISAT altimetric measurements.

5.2.3 Boundary Conditions

River discharge data for the 10 upstream boundary conditions were obtained from

Hillslopes River Routing (HRR) hydrological model developed by Beighley et al. (2009).

The HRR model framework integrates two models: (1) a water balance model (WBM) for the vertical fluxes and stores of water in and through the canopy and soil layers based on the conservation of mass and energy, and (2) a routing model for the horizontal routing of surface and subsurface runoff and channel and floodplain waters based on kinematic and diffusion wave methodologies. The WBM is driven by satellite-derived precipitation (TRMM_3B42; Huffman et al., 2007) and air temperature (MOD08_M3;

Seemann et al., 2003). A key feature of the approach is the topographic method used to subdivide the landscape and define the model unit boundaries (i.e. irregular computational grid). The discretization framework used for this study is based on concepts first articulated by Pfafstetter (1989) and implemented for the globe by Verdin and Verdin (1999). This framework is a natural system, based on topographic subdivision of the land surface and the resulting topology of the hydrographic network. Pfafstetter units can be small to handle high drainage density.

Water surface elevations for the downstream boundary condition on the Congo mainstem river were measured from ENVISAT altimetry. The altimetric measurements are ellipsoidal heights in Figure 5.2.

5.2.4 Floodplain Bare Ground Elevation

Bare ground elevations are necessary to measure floodplain water flows in the model since overbank flow from the river and runoff from the upland are calculated based on the ground topographic slopes. But the SRTM elevations include tree canopy heights in vegetated areas. Where tree cover is present, the SRTM elevation is biased upward above the ground into the canopy and corresponds closely to the centroid signal of the Ice, Cloud, and Land Elevation Satellite (ICESat). The depth of SRTM penetration into the canopy varies along the length of the profile (Carabajal and Harding, 2005). The difference between ICESat and SRTM elevation data as a function of vegetation cover and relief were evaluated using MODIS Vegetation Continuous Fields (VCF) and SRTM data (Carabajal and Harding, 2006). Vegetation cover is obtained from MODIS VCF and vegetation relief is measured from 3 by 3 window of standard deviation in the SRTM data. With higher vegetation cover and relief, the difference (i.e. canopy height) becomes greater.

In this study, ICESat (Ice, Cloud, and land Elevation Satellite) data (GLA14;

Zwally et al,, 2002) are implemented in the mission period of L1A from February to

March, 2003 that are most closely tied with the same season as the SRTM data acquisition in February, 2000. The difference between SRTM and ICESat elevations from the signal start, centroid, and end are shown in Figure 5.3. SRTM elevations are

99 similar (or little higher) to ICESAT centroid elevations. Canopy heights are closely tied to the difference between SRTM and ICESat lowest elevation from the signal end.

ICESat elevation is a point measurement. To generate two dimensional canopy heights in the model area, I calculated the relationship of ICESAT minus SRTM elevation differences for SRTM roughness and MODIS VCF in the model area using a least square approximation as follows:

Y  A X, AYX'(XX')1 (5.6) where

- Y (1 by N ): the difference between SRTM and ICESAT lowest elevations,

- X (2 by N): [SRTM roughness (1 by N); MODIS VCF (1 by N)],

- A (1 by 2): coefficients.

Based on the calculated coefficients, I calculated canopy heights in the model area using SRTM and MODIS VCF images, subtracted the heights from SRTM, and generated bare ground elevations in the floodplain.

100

101

5.3 Results

5.3.1 Hydrodynamics Models

I applied the LISFLOOD-FP hydrodynamic model to the central Congo wetlands with a spatial resolution of 500 m from January 2007 to February 2008. The first 60 days are used as a spin-up usage period. The model results are obtained from diffusive solutions in the channel and floodplain flows. The model shows the spatial and temporal variations of flood extent and water depth in the floodplain (Figure 5.4). Open water and the dry uplands are masked out in white to focus on the inundation in the study wetlands.

Figure 5.4 shows the complexity of floodplain flood dynamics during high water from

September to February (See the hydrographs in Figure 5.2). Comparisons to the 35-day repeated ENVISAT altimetric measurements are shown in Figure 5.6. The riverine areas in proximity to the Sangha and Likoula-aux-Herbes River are most flooded. Also, the interfluvial areas between the Congo, Sangha, and Likoula-aux-Herbes Rivers are shown as most flooded. The upstream areas of the Congo and Ubangi Rivers are rarely flooded and the southeastern area of the study area does not show any inundation even during high water.

Water elevations in the Congo, Ubangi, and Sangha Rivers are shown during high water from September to February in Figure 5.5. The large rivers have discharges greater than 8000 m3/s during high water whereas the other rivers have discharges less than 2000 m3/s (Figure 5.2). The large rivers show water level changes of ~4 m in the profiles of

102 water elevations. This is consistent with ENVISAT altimetry measurement at the downstream boundary condition on the Congo River.

Figure 5.4: The spatial and temporal variations of flood extent and water depth (m) in the study wetland.

103

Figure 5.5: The profiles of water elevations in the Congo, Ubangui and Sangha Rivers as compared to river bed elevations.

104

5.3.2 Comparisons of the Models with Altimetry and SAR Interferometry

The model results are compared to and evaluated by ENVISAT altimetry and

ALOS PALSAR interferometry. ENVISAT altimetry provides 35 day repeat water surface elevations. ALOS PALSAR interferometry provides water level changes over 92 days.

The profile of water elevations in the interfluvial areas between the Congo,

Sangha, and Likoula-aux-Herbes Rivers are calculated from ENVISAT altimetry (Figure

5.6). The altimetric elevations are shown in solid lines whereas the model elevations are plotted in dots. Both profiles have water increasing in September and October, the peak in either November or December, and water decreasing in January and February. In the altimetry profile, river areas of (4) and (5) show greater water level fluctuations of 4-5 m during high water as compared to less than 1 m in wetland areas of (1), (2), and (3). But, the water fluctuations in the model profiles become gradually decreasing from 4 to 1 m as the areas move away from the main Congo River. Essentially, the model is showing over

3 m of water level variations in wetland locations where the altimetric measurements show less than a meter of variation. This is most likely caused by the model’s overestimation of over bank flooding from the Congo River given its proximity to wetland areas (1), (2), and (3). The altimetric profiles include the complexity of water level fluctuations whereas the model’s profiles have simple linear patterns decreasing as moving away from the mainstem Congo River. Water levels in November and December

105 show a good correlation between altimetry and model, but with an offset of ~6 m (see the upper right in Figure 5.6).

106

ALOS PALSAR interferometry calculates water level changes for 92 days from

September 17 to December 18, 2007. The interferometric measurements are compared with model elevation differences during this period (Figure 5.7). Three frames of 7160,

7170, and 7180 in path 648 are concatenated into one image to determine the interfluvial wetland’s water level change. The topographic phases are removed using the SRTM

DEM subsampled into 1 arc second to make the differential interferogram in a pixel size of ~30 m. The perpendicular baseline is ~26 m, thus most phases are representative of water level changes with the small portion of atmospheric artifacts and system decorelation noises. I unwrap the differential interferometric phases across the wetlands from the Likoula-aux-Herbes River to Congo River. The unwrapped phases are converted from radian to centimeter scales in the vertical projection to compare with model-derived water level change. Both profiles have water level changes decreasing from north to south. Despite the coarse spatial resolution, the model profile shows a good agreement with the PALSAR interferometry.

107

Figure 5.7: Water level changes (cm) for 92 days from December 18 to September 17, 2007 are calculated from ALOS PALSAR differential interferograms (upper left) and model (upper right). The differential interferogram is unwrapped to compare with model- derived water level change along a black profile at Easting 282 (bottom).

108

5.4 Discussions and Conclusions

The LISFLOOD-FP hydrodynamic model shows the complexity of the central wetlands in the Congo Basin. It would be worthwhile to continue studying, with more detail of the hydrodynamics in the interfluvial wetlands and with spaceborne measurements since many studies suffer from a lack in-situ measurements as well as any local expertise.

The time step of 20 seconds is fixed for this model result. However, an adaptive time step solution should be carried out to avoid chequerboard oscillation errors between two adjacent cells in the raster based model (Hunter et al., 2005). The optimal time step should be large enough for computational efficiency and small enough for stability. The stability depends on water depth, free surface gradients, Manning’s coefficient, and grid cell size.

River bed elevation determines the slope in the Manning equation to calculate channel flows in rivers. I compute the river bed elevation by subtracting water depth derived from an empirical approach (Moody and Troutman, 2002) using SRTM water surface elevations. However, there are still some rather abrupt changes in slope of the tributaries. I should attempt to smooth out the sharp change in slope by fitting a 2nd order polynomial to the river bed elevation or estimating bathymetric depth from data assimilation of available altimetric information.

109

Chapter 6: Summary and Conclusions

The wetlands of many tropical low-land rivers and lakes are massive in size and in volumetric fluxes, which greatly limits a thorough understanding of their flow dynamics. The complexity of floodwater flows has not been well captured because flood waters move laterally across wetlands and this movement is not bounded like that of typical channel flow. In particular, wetland loss in the Lake Chad Basin has been accelerated due primarily to natural and anthropogenic processes, making an impact on the magnitude of flooding in the basin, and also threatens the ecosystem. In comparison, the three wetlands of the Amazon, Congo, and Logone regions are different in size and location, but all are associated with rivers. These are representative of riparian tropical, swamp tropical and inland Saharan wetlands, respectively.

I analyze repeat-pass interferometric coherence variations in JERS-1 (Japanese

Earth Resources Satellite) L-band SAR (Synthetic Aperture Radar) data at three central

Amazon sites. Because radar pulse interactions with inundated vegetation typically follow a double-bounce travel path, which returns energy to the antenna, coherence will vary with vegetation type as well as with physical and temporal baselines. Lake Balbina consists mostly of upland forests and inundated trunks of dead, leafless trees as opposed to the Cabaliana and Solimões-Purús regions which are dominated by flooded forests. 110

Balbina has higher coherence values than either Cabaliana or Solimões-Purús likely because the dead, leafless trees support strong double-bounce returns. The coherences of flooded habitats in one year are 0.32 in Balbina, 0.13 in Cabaliana, and 0.12 in Solimões-

Purús. With increasing temporal baselines, flooded and non-flooded wetland habitats show a steadily decreasing trend in coherence values whereas terre-firme and especially open water habitats have little variation and remain lower in value. Flooded and nonflooded wetland coherence varies with the season whereas terre-firme and open water do not have similarly evident seasonal variations. For example, flooded habitats in all three study regions show annual peaks in coherence values that are typically half of the peak power greater than coherence values from temporal baselines 180 days later in the power spectrum analysis, yet open water shows no variation with time. Our findings suggest that, despite overall low coherence values, repeat-pass interferometric coherence of flooded habitats is capable of showing the annual periodicity of the Amazon flood wave.

Despite being large, low-relief, tropical river systems, the floodplains and wetlands of the Amazon and Congo Basins show markedly different surface water flow hydraulics. Interferometric processing of JERS-1 SAR data from the central portions of both basins provides centimeter-scale measurements of water level change (h/t). The

Amazon is marked by a myriad of floodplain channels, but the Congo has comparatively few. Amazon floodplain channels, lakes and pans are well interconnected, whereas the

Congo wetlands are expanses with few boundaries or flow routes. Amazon patterns of

h/t are well defined with clear boundaries whereas the Congo patterns are not well

111 defined and have diffuse boundaries. Amazon h/t during inundation often shows sharp changes across floodplain channels. The Congo, however, does not show a similar spatial distinction of sharp changes, whether during infilling or evacuation. Overall, the hydraulic processes that build the Amazon floodplain are not similarly apparent in the

Congo.

Yearly flooding in the Logone floodplain has a direct impact on agricultural, pastoral, and fishery systems in the Lake Chad Basin. Since the flooding extent, depth, and duration are highly variable, flood inundation mapping helps us make better use of water resources and have more knowledge of the coupled human-natural system in the

Logone floodplain. The flood maps are generated from 33 multi-temporal Landsat

Enhanced Thematic Mapper Plus (ETM+) images acquired during three years from 2006 to 2008. Flooded area is classified using a short-wave infrared band whereas open water is classified by Iterative Self-organizing Data Analysis (ISODATA) clustering. The maximum flooding extent in the study area increases up to ~5.8K km2 in late October

2008. The study also provides strong correlation between the flooding extents and water height variations in both the floodplain and the river based on a second-order polynomial regression model. The water heights are from ENIVSAT altimetry in the floodplain and gauge measurements in the river. Coefficients of determination between flooding extents and water height variations are greater than 0.91 with 4 to 36 days in phase lag.

Floodwater drains back to the river and to the northwest during the recession period in

December and January. The study contributes to a better understanding of the Logone

112 floodplain dynamics with details of the spatial pattern and size of the flooding extent as well as assisting the flood monitoring and forecasting systems in the catchment.

The Congo River, with the second-largest discharge and basin area of any river, has not had the same degree of research compared to the Amazon. Access to Congo wetlands is difficult resulting in a paucity of published research on the surface water hydraulics. Most of the primary research on the Congo swamps and wetlands is from the colonial era with a limited number of surface water hydrology publications since then.

Hydrodynamic or hydrological modeling efforts instead rely on remotely sensed observations that either directly record water surface elevations using satellite radar altimetry or infer stage and discharge from relationships between main channel gauge data and remotely sensed data. Here I quantify the spatial and temporal distribution of water level and storage changes in the central Congo wetland using spaceborne data and the LISFLOOD-FP hydrodynamic model. This model provides 1-D kinematic and diffusive channel flow and 2-D dynamic floodplain flow. I derive model parameters such as local topography, channel width, and water depth from the Shuttle Radar Topography

Mission (SRTM), from the Hydrological data and maps based on Shuttle Elevation

Derivatives at multiple Scales (HydroSHEDS), and from L-band JERS-1 SAR data from the Global Rain Forest Mapping project (GRFM). I use water elevations from altimetric measurements as a downstream boundary condition and water flow derived from empirical equations or other hydrological models as initial conditions. The model results show centimeter scale water level changes on the main stem Congo River and in its tributaries (e.g. Ubangi, Sangha, Likouala-aux-Herbes, and Likouala Rivers) at 500-

113 meters/pixel spatial resolution. ALOS PALSAR interferometry shows a good agreement with the model’s water level changes (h/t).

These results will significantly add to our understanding of global wetland hydrodynamics and suggest an opportunity to utilize satellite measurements in combination with hydrodynamic models for remote and inaccessible wetland areas.

Future studies for expanding the results of my dissertation research include (1) wetland hydrodynamics in the Atchafalaya and Tonle Sap Basins, (2) statistical error assessments on the satellite-derived hydraulic measurements, (3) data assimilation into the Congo hydrodynamic model for river discharge and water depth estimation, and (4) studies on a time-varying limnological network controlling water, sediment, carbon, and nutrient exchange between floodplain wetlands and the main channel.

114

REFERENCES

Aalto, R., Maurice-Bourgoin, L., Dunne, T., Montgomery, D. R., Nittrouer, C. A., and Guyot, J., 2003, Episodic sediment accumulation on Amazonian flood plains influenced by El Nino/Southern Oscillation, Nature, 425, 493-497.

Alsdorf, D. E., Melack, J. M., Dunne, T., Mertes, L. A. K., Hess, L. L., and Smith, L. C., 2000, Interferometric radar measurements of water level changes on the Amazon floodplain. Nature, 404, 174-177.

Alsdorf, D., Birkett, C., Dunne, T., Melack, J., and Hess, L., 2001a, Water level changes in a large amazon lake measured with spaceborne radar interferometry and altimetry. Geophysical Research Letters, 28, 2671-2674.

Alsdorf, D. E., Smith, L. C., and Melack, J. M., 2001b, Amazon floodplain water level changes measured with interferometric SIR-C radar, IEEE Transactions on Geoscience and Remote Sensing, 39, 423-431.

Alsdorf, D. E., 2003, Water storage of the central Amazon floodplain measured with GIS and remote sensing imagery. Annals of the Association of American Geographers, 93 (1), 55-66.

Alsdorf, D. E., Dunne, T., Melack, J. M., Smith, L. C., and Hess, L. L., 2005, Diffusion modeling of recessional flow on central Amazonian floodplains. Geophysical Research Letters, 32, L21405, doi:10.1029/2005GL024412.

Alsdorf, D., Bates, P., Melack, J., Wilson, M., and Dunne, T., 2007a, Spatial and temporal complexity of the Amazon flood measured from space. Geophysical Research Letter, 34, L08402.

Alsdorf, D. E., Rodriguez, E., and Lettenmaier, D. P., 2007b, Measuring surface water from space. Reviews of Geophysics, 45, 1–24.

Aronica, G., Bates, P. D. and Horritt, M. S., 2002, Assessing the uncertainty in distributed model predictions using observed binary pattern information within GLUE. Hydrological Processes, 16, 2001-2016. 115

Arvidson, T., Gasch, J., and Goward, S. N., 2001, Landsat 7′s long-term acquisition plan — an innovative approach to building a global imagery archive. Remote Sensing of Environment, 78, 13−26.

Askne, J. I., Dammert, P. B., Ulander, L. M., and Smith, G., 1997, C-band repeat-pass interferometric SAR observations of the forest. IEEE Transactions on Geoscience and Remote Sensing, 35 (1), 25-35.

Bates, P. D., Anderson, M. G., Baird, L., Walling, D. E. and Simm, D., 1992, Modelling floodplain flow with a two-dimensional finite element scheme. Earth Surface Processes and Landforms, 17, 575-588.

Bates P. D., Anderson M. G and Hervouet J. M., 1995, Initial comparison of two two- dimensional finite element codes for river flood simulation. Proceedings of the Institution of Civil Engineers, Water Maritime and Energy, 112, 238-248.

Bates, P. D. and De Roo, A. P. J., 2000. A simple raster-based model for floodplain inundation. Journal of Hydrology, 236, 54-77.

Beighley, R. E., Eggert, K. G., Dunne, T., He, Y., Gummadi, V., and Verdin, K. L., 2009, Simulationg hydrologic and hydraulic processes throughout the Amazon River Basin. Hydrological Processes, 23, 1221-1235.

Benech, V., Quensiere, J., and Vidy, G., 1982, Hydrologie et physico-chimie des eaux de la plaine d'inondation du Nord Cameroun. Cahiers ORSTOM series hydrologique XIX(1), 15-35.

Bennett, M. W. A., 1987, Rapid monitoring of wetland water status using density slicing, Proceeding of the 4th Australian Remote Sensing Conference, 14-18 September, Adelaide, 682-691.

Birkett, C. M., 1998, Contribution of the TOPEX NASA radar altimeter to the global monitoring of large rivers and wetlands. Water Resources Research, 34, 1223-1239.

Birkett, C. M., 2000, Synergistic remote sensing of Lake Chad: variability of basin inundation, Remote Sensing of Environment, 72, 218-236.

Birkett, C. M., Mertes, A. A. K., Dunne, T., Costa, M. H., and Jasinski, M. J., 2002, Surface water dynamics in the Amazon Basin: application of satellite radar altimetry, Journal of Geophysical Research, 107, doi:10.1029/2001JD000609.

Booker, D. J., Sear, D. A., and Payne, A. J., 2001, Modelling three-dimensional flow structres and patterns of boundary shear stress in a natural pool-riffle sequence. Earth Surface Processes and Landforms, 26, 553-576.

116

Campbell, D. G., Stone, J. L., and Rosas, A., 1992, A comparison of the phytosociology and dynamics of three floodplain (Varzea) forests of known ages, Rio Jurua, western Brazilian Amazon, Journal of the Linnean Society of London, 108, 213-237.

Campbell, D., 2005, The Congo River basin, The world’s largest wetlands: ecology and conservation, Cambridge Univ. Press, 149-165.

Carabajal, C. C., and Harding, D. J., 2005, ICESat validation of SRTM C-band digital elevation models, Geophysical Research Letters, 32, L22S01, doi:10.1029/2005GL023957.

Carabajal, C. C., and Harding, D. J., 2006, SRTM C-band and ICESat laser altimetry elevation comparisons as a function of tree cover and relief, Photogrammetric Engineering and Remote Sensing, 72, 287-298.

Coe, M. T., 1998, A linked global model of terrestrial hydrologic processes: simulation of modern rivers, lakes, and wetlands. Journal of Geophysical Research, 103, 8885- 8899.

Coe, M. T., and Foley, J. A., 2001, Human and natural impacts on the water resources of the Lake Chad basin. Journal of Geophysical Research, 106, 3349-3356.

Coe, M. T., and Birkett, C. M., 2004, Calculation of river discharge and prediction of lake height from satellite radar altimetry: example for the Lake Chad Basin. Water Resources Research, 40, doi: 10.1029/2003WR002543.

Coz, M. L., Delclaux, F.,Genthon,,P., and Favreau G., 2009, Assessment of Digital Elevation Model (DEM) aggregation methods for hydrological modeling: Lake Chad Basin, Africa. Computers & Geosciences, 35,1661–1670.

Crétaux, J. F., and Birkett, C., 2006, Lake studies from satellite radar altimetry, Comptes Rendus Geosciences, 338, 1098-1112.

Davis, J. C., 1986, Statistics and data analysis in geology. New York: John Wiley and Sons.

De Grandi, G. D., Mayaux, P., Rauste, Y., Rosenqvist, A., Simard, M., and Saatchi, S. S., 2000, The global rain forest mapping project JERS-1 radar mosaic of tropical Africa: development and product characterization aspects, IEEE Transactions on Geoscience and Remote Sensing, 38, 2218-2233.

DHI Water and Environment, 2001, MIKE11 hydrodynamic reference manual, DHI, Horshølm, Denmark

Engdahl, M. E. and Hyyppa, J. M., 1997, Forest inventory using interferometric SAR techniques. In: Proceddings of the 3rd ERS symposium on space at the service of our 117

environment, Florence, Italy (pp. 345-350), The Netherlands: Estec, Noodrwijk (ESA SP-414).

Eriksson, L. E. B., Santoro, M., and Wiesmann, A., 2003, Multitemporal JERS repeat- pass coherence for growing-stock volume estimation of siberian forest. IEEE Transactions on Geoscience and Remote Sensing, 41, 1561-1570.

Eriksson, L. E. B., 2004, Satellite-borne L-band interferometric coherence for rorestry applications in the boreal zone. Ph.D. thesis, Friedrich Schiller University, Jena, Germany.

Ervine D. A. and MacCleod, A. B., 1999, Modelling a river channel with distant floodbanks. Proceedings of the Institution of Civil Engineers, Water Maritime and Energy, 136, 21-33.

Evans, S. Y., Bradbrook, K., Braund, R., and Bergkam, G., 2003, Assessment of the restoration potential of the Logone floodplain (Cameroon). Water and Environment Journal, 17, 123-128.

Farr, T., and 17 others, 2007, The Shuttle Radar Topography Mission, Reviews of Geophysics, 45, RG2004, doi:10.1029/2005RG000183.

Fearnside, P. M., 1989, Brazil’s Balbina Dam: environment versus legacy of the pharaohs in Amazonia, Environmental Management, 13, 401-423.

Federal Geographic Data Committee, 1997, National vegetation classification standard (FGDC-STD-005).

Feldhaus, R., Höttges, R., Brockhaus, T. and Rouvé, G., 1992, Finite element simulation of flow and pollution transport applied to part of the River Rhine. In: Hydraulic and environmental modelling: estuarine and river waters, R.A. Falconer, K. Shiono and R.G.S. Matthews (eds.), Ashgate Publishing, Aldershot, 323-334.

Ferretti, A., Prati, Claudio, and Rocca, F., 2001, Permanent scatters in SAR interferometry. IEEE Transaction on Geoscience and Remote Sensing, 39 (1), 8-20.

FEWS, 1997, Lake Chad untapped potential, Famine Early Warning System, Special Report 97-4.

Frappart, F., Calmant, S., Cauhopé, M., Seyler, F., and Cazanave, A., 2006, Preliminary results of ENVISAT RA-2-derived water levels validation over the Amazon basin, Remote Sensing of Environment, 100, 252-264.

Frazier, P., and Page, K., 2000, Water body detection and delineation with Landsat TM data, Photogrammetric Engineering and Remote Sensing, 66, 1461-1467.

118

Fread, D. L., 1984, Flood routing. In: M.G. Anderson and T.P. Burt (eds), Hydrological Forecasting, John Wiley and Sons, Chichester, Chapter 14.

Fread, D. L., 1993, Flood routing. In D.R. Maidment (ed), Handbook of Applied Hydrology, Mc-Graw Hill, New York, Chapter 10.

Garbrecht, J., and Brunner, G., 1991, Hydrologic channel-flow routing for compound sections, Journal of Hydraulic Engineering, 177, 629-642.

Gac, J. Y., 1980, Géochimie du bassin du Lac Tchad (Geochemistry of Lake Chad basin). Thèse de Doctorat d’Etat, Travaux et Documents de l’ORSTOM, ORSTOM (Office de la Recherche Scientifique et Technique d’Outre-Mer), Paris, pp. 252.

Goldstein, R. M., Engelhardt, H., Kamb, B., and Frolich, R. M., 1993, Satellite radar interferometry for monitoring ice sheet motion : Application to an Antarctic ice stream, Science, 262, 1525-1530.

Goward, S. N., Masek, J. G., Williams, D. L., Irons, J. R., and Thompson, R. J., 2001, The Landsat 7 mission, terrestrial research and applications for the 21st century. Remote Sensing of Environment, 78, 3-12.

Hall-Atkinson, C., and Smith, L. C., 2001, Delineation of delta ecozones using interferometric SAR phase coherence Mackenzie River Delta, N.W.T., Canada. Remote Sensing of Environment, 78, 229-238.

Hanssen, R. F., Weckwerth, T. M., Zebker, H. A., and Klees, R., 1999, High-resolution water vapor mapping from interferometric radar measurements, Science, 283, 1297- 1299.

Harris, J., and Digby-Arbus, S., 1986, The detection of wetlands on radar imagery. In: Proceedings of the Tenth Canadian remote sensing symposium, Edmonton, Alberta (pp. 529-543). Ottawa: Canadian Aeronautics and Space Institute.

Hedges, J. I., Clark, W. A., and Quay, P. D., 1986, Composition and fluxes of particulate organic material in the Amazon river. Limnol Oceanogr, 31, 717-738.

Hess, L. L, Melack, J. M., and Simonett, D. S., 1990, Radar detection of flooding beneath the forest canopy: a review. International Journal of Remote Sensing, 11 (7), 1313- 1325, 1990.

Hess, L. L., and Melack, J. M., 1994, Mapping wetland hydrology and vegetation with synthetic aperture radar. International Journal of Ecology and Environmental Sciences, 20, 197-205.

119

Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Delineation of inundated area and vegetation along the Amazon floodplain with the SIR-C synthetic aperture radar. IEEE Transactions on Geoscience and Remote Sensing, 33, 896-904.

Hess, L. L., Melack, J. M., Novo, E. M. L. M., Barbosa, C. C. F., and Gastil, M., 2003, Dual-season mapping of wetland inundation and vegetation for the central Amazon basin. Remote Sensing of Environment, 87, 404-428.

Hirsch, R. M., Helsel, D. R., Cohn, T. A., and Gilroy, E. J., 1993, Statistical analysis of hydrologic data. In Handbook of hydrology, New York: McGraw-Hill.

Hoffmann, J., Zebker, H. A., Galloway, D. L., and Amelung, F., 2001, Seasonal subsidence and rebound in Las Vegas Valley, Nevada, observed by synthetic aperture radar interferometry. Water Resources Research, 37, 1551-1566.

Horritt, M. S. and Bates, P. D., 2001, Predicting floodplain inundation: raster-based modelling versus the finite element approach. Hydrological Processes, 15, 825-842.

Hossain, F., and Katiyar, N., 2006, Improving flood forecasting in international river basins. EOS Transactions American Geophysical Union, 87, 49-60.

HR Wallingford, 1997, ISIS user reference manual, HR Wallingford, Wallingford.

Huffman, G. J., Adler, R. F., Bolvin, D. T., Gu, G., Nelkin, E. J., Bowman, K.P., Hong, Y., Stocker, E.F., and Wolff, D. B., 2007, The TRMM multi-satellite precipitation analysis: quasi-global, multi-year, combined-sensor precipitation estimates at fine scale. Journal Hydrometeorology, 8, 38–55.

Hunter, N. M., Horritt, M. S., Bates, P. D., Wilson, M. D., and Werner, M. G. F., 2005, An adaptive time step solution for raster-based storage cell modelling of floodplain inundation. Advances in Water Resources, 28, 975-991.

Hunter, N. M., Bates, P. D., Horritt, M. S., and Wilson, M. D., 2007, Simple spatially- distributed models for predicting flood inundation: A review. Geomorphology, 90, 208-225.

Islam, A. S., Bala, S. K., and Haque, M. A., 2010, Flood inundation map of Bangladesh using MODIS time-series images, Journal of Flood Risk Management, 3, 210-222.

Jain, S. K., Singh, R. D., Jain, M. K., and Lohani, A. K., 2005, Delineation of flood- prone areas using remote sensing techniques, Water Resources Management, 19, 333- 347.

Jung, H. C., and Alsdorf, D., 2010, Repeat-pass multi-temporal interferometric SAR coherence variations with Amazon floodplain and lake habitats, International Journal of Remote Sensing, 31, 881-901. 120

Jung, H. C., J. Hamski, M. Durand, D. Alsdorf, F. Hossain, H. Lee, A. K. M. A. Hossain, K. Hasan, A. S. Khan, and A. K. M. Z. Hoque, 2010, Characterization of complex fluvial systems via remote sensing of spatial and temporal water level variations, Earth Surface Processes and Landform, 35, 294-304.

Junk, W. J., 1993, Wetlands of tropical South America. In D. F. Whigham, D. Dykyjova, and S. Hejny, eds. Wetlands of the World, I: Inventory, Ecology, and Management. Kluwer academic Publishers, Dordrecht, 679-739.

Junk, W. J., 1999, The flood pulse concept of large rivers: Learning from the tropics. Archiv für Hydrobiologie, 115, 261-280.

Kim, S., Hong, S., and Won, J., 2005, An application of L-band synthetic aperture radar to tide height measurement. IEEE Transactions on Geoscience and Remote Sensing, 43, 1472-1478.

Kasischke, E. S., and Bourgeau-Chavez, L. L., 1997, Monitoring south Florida wetlands using ERS-1 SAR imagery. Photogrammetric Engineering and Remote Sensing, 63, 281-291.

Kazadi, S-N, and Kaoru, F., 1996, Interannual and long-term climate variability over the Zaire River Basin during the last 30 years. Journal of Geophysical Research, 101, 21351-21360.

Khan, S. I., Hong, Y., Wang, J., Yilmaz, K., Gourley, J. J., Adler, R. F., Brakenridge, G. R., Policelli, F., Habib, S., and Irwin, D., 2010, Satellite remote sensing and hydrologic modeling for flood inundation mapping in Lake Victoria Basin: Implications for hydrologic prediction in ungagged basins. IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2010.2057513.

Klinge, H., Junk, W. J., and Revilla, C. J., 1990, Status and distribution of forested wetlands in tropical South America. Forest Ecology and Management, 33/34, 31-101.

Koblinsky, C. J., Clarke, R. T., Brenner, A. C., and Frey, H., 1993, Measurement of river level variations with satellite altimetry. Water Resources Research, 6, 1839-1848.

Kushwaha, S. P. S., Dwivedi, R. S., and Rao, B. R. M., 2000, Evaluation of various digital image processing techniques for detection of coastal wetlands using ERS-1 SAR data. International Journal of Remote Sensing, 21, 565-579.

Lane, S. N., Bradbrook, K. F., Richards, K. S., Biron, P. A., and Roy, A. G., 1999, The application of computational fluid dynamics to natural river channels: three- dimensional versus two-dimensional approaches. Geomorphology, 29, 1-20.

121

Laporte, N., Goetz, S. J., Justice, C. O., and Heinicke, M., 1998, A new land cover map of central Africa derived from multi-resolution, multi-temporal AVHRR Data, International Journal of Remote Sensing, 19, 3537-3550.

Laraque, A., Mahe, G., Orange, D., and Marieu, B., 2001, Spatiotemporal variations in hydrological regimes within Central Africa during the XXth century, Journal of Hydrology, 204, 104-117.

LCBC, 2010, Lake Chad Basin project. (accessed October 2010).

Lee, H., Shum, C. K., Yi, Y., Braun, A., and Kuo, C. Y., 2008, Laurentia crustal motion observed using TOPEX/POSEIDON radar altimetry over land, Journal of Geodynamics, doi: 10.1016/j.jog.2008.05.001.

Lee, H., Shum, C., Yi, Y., Ibaraki, M., Kim, J.-W., Braun, A., Kuo, C.-Y., and Lu, Z., 2009, Louisiana wetland water level monitoring using retracked TOPEX/POSEIDON altimetry, Marine Geodesy, 32, 284-302.

Leon, J. G., Calmant, S., Seyler, F., Bonnet, M. P., Cauhopé, M., Frappart, F., Filizola, N. and Fraizy, P., 2007, Rating curves and estimation of average water depth at the upper Negro River based on satellite altimeter data and modeled discharges. Journal of Hydrology, 328, 481-496.

Liang, S., Fang, H., Morisette, J. T., Chen, M., Shuey, C. J., and Walthall, C. L., 2002, Atmospheric correction of Landsat ETM+ land surface imagery: II. Validation and applications. IEEE Transactions on Geoscience and Remote Sensing, 40, 2736−2746.

Liu, J., Dietz, T., Carpenter, S. R., Alberti, M., Folke, C., Moran, E., Pell, A. N., Deadman, P., Kratz, T., Lubchenco, J., Ostrom, E., Ouyang, Z., Provencher, W., Redman, C. L., Schneider, S. H., and Taylor, W. W., 2007, Complexity of coupled human and natural systems, Sciences, 317, 1513-1516.

Loth, P., 2004, The return of the water: restoring the Waza Logone floodplain in Cameroon, IUCN, Gland, Switzerland and Cambridge, UK.

Lu, Z., Crane, M., Kwoun, O., Wells, C., Swarzenski, C., and Rykhus, R., 2005, C-band radar observes water level change in swamp forests. EOS, 86 (14), 141-144.

Lu, Z., and Kwoun, O., 2008, Radarsat-1 and ERS InSAR analysis over southeastern coastal Louisiana: Implications for mapping water-level changes beneath swamp forests. IEEE Transactions on Geoscience and Remote Sensing, 46, 2167−2184.

122

Luckman, A., Baker, J., and Wegmuller, U., 2000, Repeat-pass interferometric coherence measurements of disturbed tropical forest from JERS and ERS satellites. Remote Sensing of Environment, 73, 350-360.

Maheu, C., Cazenave, A., and Mechoso, C. R., 2003, Water level fluctuations in the Plata Basin (South America) from Topex/Poseidon Satellite Altimetry. Geophysical Research Letter, 30, 10.1029/2002GL016033.

Martinuzzi, S., Gould, W. A., and González, O. M. R., 2007, Creating cloud-free Landsat ETM+ data sets in tropical landscapes: cloud and cloud-shadow removal. USDA General Technical Report IITF-GTR-32.

Massonnet, D., M. Rossi, C. Carmona, F. Adragna, G. Peltzer, K. Feigl, and T. Rabaute, 1993, The displacement field of the Landers earthquake mapped by radar interferometry, Nature, 364, 138–142.

Matthews, E. and Fung, I., 1987, Methane emission from natural wetlands: Global distribution, area, and environmental characteristics of sources. Global Biogeochemical Cycles, 1, 61-86.

Matthews, E., 1993, Wetlands (In: atmospheric methane: sources, sinks, and role in global change, pp.314-361). Springer-Verlag, Berlin.

Maxwell, S. K., Schmidt, G. L., and Storey, J. C., 2007, A multi-scale segmentation approach to filling gaps in Landsat ETM+ SLC-off images. International Journal of Remote Sensing, 28, 5339−5356.

Melack, J. M. and Yang, Y., 1998, Delineation of flooded area and flooded vegetation in Balbina Reservoir (Amazonas, Brazil) with synthetic aperture radar. Verh. Internat, Verein. Limnol., 26, 2374-2377.

Melack, J. M. and Forsberg, B. R., 2001, Biogeochemistry of Amazon floodplain lakes and associated wetlands (In: the biogeochemistry of the Amazon Basin and its role in a changing world, pp. 235–276). Oxford Univ. Press, New York.

Melesse, A., and Jordan, J. D., 2002, A comparison of fuzzy vs. augmented-ISODATA classification algorithms for cloud-shadow discrimination from Landsat images. Photogrammetric Engineering and Remote Sensing, 68, 905-911.

Mertes, L. A. K., Dunne, T., and Martinelli, L. A., 1996, Channel-floodplain geomorphology along the Solimoes-Amazon River, Brazil, GSA Bulletin 108, 1089- 1107.

Middleton, B. A., 2002, Flood pulsing in wetlands: restoring the natural hydrological balance. John Wiley & Sons, New York.

123

Mitchell, G. J., 1990, Foreward to wetland creation and restoration (In: Wetland Creation and Restoration: The Status of the Science, pp. ix-x). Island Press, Washington D.C.

Moody, J. A., and Troutman, B. M., 2002, Characterization of the spatial variability and channel morphology. Earth Surface Processes and Landforms, 27, 1251-1266.

Morvan, H., Pender, G., Wright, N. G., and Ervine, D. A., 2002, Three-dimensional hydrodynamics of meandering compound channels. Journal of Hydraulic Engineering, 128, 674-682.

Mott MacDonald, 1999, Logone floodplain model study report. Mott MacDonald, Cambridge. UK.

Mott MacDonald, 1993, Mathematic model of the hydrological behaviour of Lake Chad and its feeder rivers. Mott MacDonald, Cambridge. UK.

NASA, 2009, Landsat 7 science data users handbook, Chapter 11-Data products. (last updated June 2009).

Olivry, J. C., Chouret, A., Vuillaume, G., Lemoalle, J., and Bricquet, J. P., 1996, Hydrologie du Lac Tchad (Lake Chad hydrology). Monographie hydrologique 12, ORSTOM (Office de la Recherche Scientifique et Technique d’Outre-Mer) éditions, pp. 266.

Ouaidrari, H., and Vermote, E. F., 1999, Operational atmospheric correction of Landsat TM data. Remote Sensing of Environment, 70, 4−15.

Ozesmi, S. L., Bauer, M E., 2002, Satellite remote sensing of wetlands. Wetlands Ecology and Management, 10, 381-402.

Pfafstetter, O., 1989, Classification of Hydrographic Basins; Coding Methodology, Departamento Nacional do Obras de Sanearnento, Rio de Janeiro, Brazil. [Translated by Verdin JP. U.S. Bureau of Reclamation: Denver, CO; 13].

Richards, J. A., Woodgate, P. W., and Skidmore, A. K., 1987, An explanation of enhanced radar backscattering from flooded forests. International Journal of Remote Sensing, 8 (7), 1093-1100.

Richey. J. E., Mertes, L. A. K., Dunne, T., Victoria, R. L., Forsberg, B. R., Tancredi, A. C. N. S., and Oliveira, E., 1989, Sources and routing of the Amazon River flood wave. Global Biogeochemical Cycles, 3, 191-204.

Rodriguez, E., and Martin, J. M., 1992, Theory and design of interferometric synthetic aperture radars. IEE Proceedings, Part F: Radar and Signal Processing, 139, 147- 159. 124

Rosen, P. A., Hensley, S., Zebker, H. A., Webb, F. H. and Fielding, E. J., 1996, Surface deformation and coherence measurements of Kilauea Volcano, Hawaii, from SIR-C radar interferometry. Journal of Geophysical Research, 101, 23109-23125.

Rosenqvist, A., Shimada, M., Chapman, B., and Freeman, A., 2000, The global rain forest mapping project-a review. International Journal of Remote Sensing, 21, 1201- 1234.

Sakamoto, T., Nguyen, N. V., Kotera, A., Ohno, H., Ishitsuka, N., and Yokozawa, M., 2007, Detecting temporal changes in the extent of annual flooding within the Cambodia and the Vietnamese Mekong Delta from MODIS time-series imagery. Remote Sensing of Environment, 109, 295-313.

Samuels, P. G., 1990, Cross section location in one-dimensional models. In W.R. White (ed), International Conference on River Flood Hydraulics. John Wiley and Sons, Chichester, 339-350.

Samuels, P G., and Gray, M. P., 1982, The FLUCOMP river model engineers guide, report ex 999, HR Wallingford.

Sautter, G., 1966, De l’Atlantique au fleuve Congo: Une géographie du sous-peuplement (Paris/La Haye: Mouton), 2.

Scholte, P., 2005, Floodplain rehabilitation and the future of conservation and development: adaptive management of success in Waza-Logone, Cameroon, Leiden University.

Seemann, S. W., Li, J., Menzel, W. P., Gumley, L. E., 2003, Operational retrieval of atmospheric temperature, moisture, and ozone from MODIS infrared radiances. Journal of Applied Meteorology, 42, 1072–1091.

Sheng, Y., and Alsdorf, D., 2005. Automated geo-referencing and ortho-rectification of Amazon Basin-wide SAR mosaics using SRTM DEM data, IEEE Transactions on Geoscience and Remote Sensing, 43, 1929–1940.

Sioli, H., 1968, Hydrochemistry and geology in the Brazilian Amazon region. Amazoniana 1, 267-277.

Sippel, S. J., Hamilton, S. K., and Melack, J. M., 1992, Inundation area and morphometry of lakes on the Amazon River floodplain, Brazil. Archives of Hydrobiology, 123, 385- 400.

Sippel, S. J., Hamilton, S. K., Melack, J. M., and Novo, E. M. M., 1998, Passive microwave observations of inundation area and the area/stage relation in the Amazon River floodplain. International Journal of Remote Sensing, 19, 3055-3074.

125

Siqueira, P., Chapman, B., and McGarragh, G., 2003, The coregistration, calibration, and interpretation of multiseason JERS-1 SAR data over South America, Remote Sensing of Environment, 87, 389-403.

Smith, L. C., Isacks, B. L., Forster, R. R., Bloom, A. L., and Preuss, I., 1995, Estimation of discharge from braided glacial rivers using ERS-1 SAR: First results. Water Resources Research, 31 (5), 1325-1329.

Smith, L. C., Isacks, B. L., Bloom, A. L., and Murray, A. B., 1996, Estimation of discharge from three braided rivers using synthetic aperture radar (SAR) satellite imagery: Potential application to ungaged basins. Water Resources Research, 32 (7), 2021-2034.

Smith, L. C., 1997, Satellite remote sensing of river inundation area, stage, and discharge: A review, Hydrological Processes, 11, 1427-1439.

Storey, J., Scaramuzza, P., Schmidt, G., and Barsi, J., 2005, Landsat 7 scan line corrector-off gap filled product development, In: Pecora 16 "Global Priorities in Land Remote Sensing" Proceedings. American Society for Photogrammetry and Remote Sensing, Sioux Falls, South Dakota.

Sun, G., 1990, Radar backscattering modeling of coniferous forest canopies. Ph.D. dissertation, Univ. California, Santa Barbara, California.

Tou, J. T., and Gonzales, R. C., 1974, Pattern recognition principles. Isodata algorithm, pattern classification by distance functions, Addison-Wesley, Reading, Massachusetts, pp. 97-104.

Townsend, P. A., and Walsh, S. J., 1998, Modeling floodplain inundation using an integrated GIS with radar and optical remote sensing. Geomorphology, 21, 295-312.

USACE, 2001, HEC-RAS river analysis system, hydraulic reference manual version 3.0, US Army Corps of Engineers, Davis, CA.

USGS, 2010, The USGS Flood inundation mapping prospectus (accessed November 2010).

Verdin, K. L., and Verdin, J. P., 1999, A topological system for delineation and codification of the earth’s river basins. Journal of Hydrology, 218, 1–12.

Vorosmarty, C. J., Moore III, B., Grace, A. L., and Gildea, M. P., 1989, Continental scale models of water balance and fluvial transport: An application to South America. Global Biogeochemical Cycles, 3, 241-265.

Vorosmarty, C. J., Willmott, C. J., Choudhury, B. J., Schloss, A. L., Stearns, T. K., Robeson, S. M., and Dorman, T. J., 1996, Analyzing the discharge regime of a large 126

tropical river through remote sensing, ground-based climatic data, and modeling, Water Resources Research, 32, 3137-3150.

Wang, Y., Hess, L. L., Filoso, S., and Melack, J. M., 1995, Understanding the radar backscattering from flooded and nonflooded Amazonian forests: Results from Canopy Backscatter Modeling. Remote Sensing of the Environment, 54, 324-332.

Wang, Y., Colby, J. D., and Mulcahy, K. A., 2002, An efficient method for mapping flood extent in a coastal flood plain using Landsat TM and DEM data. International Journal of Remote Sensing, 23, 3681-3696.

Wegmuller, U. and Werner, C. L., 1995, SAR interferometric signature of forest. IEEE Transactions on Geoscience and Remote Sensing, 33 (5), 1153-1161.

Wegmuller, U., Strozzi, T., Farr, T., and Werner, C. L., 2000, Arid land surface characterization with repeat-pass SAR interferometry. IEEE Transaction on Geoscience and Remote Sensing, 38 (2), 776-781.

Wesseling, J. W., Drijver, C. A., Nah E., Namba, A., and Zuiderwijk, A., 1994, Waza Logone flood restoration study: identification of options for re-flooding. Leiden and Delft, the Netherlands: Center of Environmental Sciences, Leiden University and Delft Hydraulics.

Westra, T., Mertens, K.C., and De Wulf, R.R., 2005, ENVISAT ASAR wide swath and SPOT-VEGETATION image fusion for wetland mapping: evaluation of different wavelet-based methods. Geocarto International, 20, 21-31.

Westra, T., and De Wulf, R. R., 2009, Modelling yearly flooding extent of the Waza- Logone floodplain in northern Cameroon based on MODIS and rainfall data. International Journal of Remote Sensing, 30, 5527-5548.

Wicks, C., Thatcher, W., and Dzurisin, D., 1998, Migration of fluids beneath Yellowstone caldera inferred from satellite radar interferometry. Science, 282, 458- 462.

Williams, D. L., Goward, S., and Arvidson, T., 2006, Landsat: Yesterday, today, and tomorrow. Photogrammetric Engineering and Remote Sensing, 72, 1171−1178.

Wilson, C. A. M. E., Boxall, J. B., Guymer, I., Olsen, N. R. B., 2003, Validation of a three-dimensional numerical code in the simulation of pseudo-natural meandering flows. Journal of Hydraulic Engineering, 129, 758-768.

Zcbker, H. A., and Goldstein, R. M., 1986, Topographic mapping from interferometric synthetic aperture radar observations. Journal of Geophysical Research, 91, 4993- 4999.

127

Zebker, H. A. and Villasenor, J., 1992, Decorrelation in interferometric radar echoes. IEEE Transaction on Geoscience and Remote Sensing, 30 (5), 950-959.

Zebker, H.A., P.A. Rosen, R.M. Goldstein, A. Gabriel, and C.L. Werner, 1994a, On the derivation of coseismic displacement fields using differential radar interferometry: The Landers earthquake. Journal of Geophysical Research, 99, 19617-19634.

Zebker, H. A., Werner, C. L., Rosen, P. A., and Hensley, S., 1994b, Accuracy of topographic maps derived from ERS-1 interferometric radar. IEEE Transaction on Geoscience and Remote Sensing, 32 (4), 823-836.

Zebker, H. A., Rosen, P. A., and Hensley, S., 1997, Atmospheric effects in interferometric synthetic aperture radar surface deformation and topographic maps. Journal of Geophysical Research, 102, 7547-7563.

Zwally, H. J., Schutz, B., Abdalati, W., Abshire, J., Bentley, C., Brenner, A., Bufton, J., Dezio, J., Hancock, D., Harding, D., Herring, T., Minster, B., Quinn, K., Palm, S., Spinhirne, J., and Thomas, R., 2002, Icesat’s laser measurements of polar ice, atmosphere, ocean, and land. Journal of Geodynamics, 34, 405–445.

128