Is Arctic Greening Consistent with the Ecology of Tundra? Lessons from an Ecologically Informed Mass Balance Model

Environmental Research Letters

LETTER • OPEN ACCESS Is arctic greening consistent with the ecology of tundra? Lessons from an ecologically informed mass balance model

To cite this article: A V Rocha et al 2018 Environ. Res. Lett. 13 125007

View the article online for updates and enhancements.

This content was downloaded from IP address 129.74.231.133 on 02/01/2019 at 15:50 Environ. Res. Lett. 13 (2018) 125007 https://doi.org/10.1088/1748-9326/aaeb50

LETTER Is arctic greening consistent with the ecology of tundra? Lessons OPEN ACCESS from an ecologically informed mass balance model

RECEIVED 9 August 2018 A V Rocha1 , B Blakely1, Y Jiang2, K S Wright1 and S R Curasi1 REVISED 1 University of Notre Dame, Department of Biological Sciences & the Environmental Change Initiative, Notre Dame, IN, 46556, United 23 October 2018 States of America ACCEPTED FOR PUBLICATION 2 Oregon State University, Forest Ecosystems and Society, Corvallis, Oregon, 97331, United States of America 25 October 2018

PUBLISHED E-mail: [email protected] 7 December 2018 Keywords: NDVI, arctic greening/browning, disturbance, arctic ecology and biogeochemistry

Original content from this work may be used under the terms of the Creative Abstract Commons Attribution 3.0 licence. Climate change has been implicated in the widespread ‘greening’ of the arctic in recent decades. Any further distribution of However, differences in arctic greening patterns among satellite platforms and recent reports of this work must maintain attribution to the decreased rate of greening or of browning have made attributing arctic greening trends to a warming author(s) and the title of climate challenging. Here, we compared MODIS greening trends to those predicted by the coupled the work, journal citation and DOI. carbon and nitrogen model (CCaN); a mass balance carbon and nitrogen model that was driven by MODIS surface temperature and climate. CCaN was parameterized using model-data fusion, where model predictions were ecologically constrained with historical ecological ground and satellite based data. We found that, at long temporal and large spatial scales, MODIS greening trends were consistent with ecological and biogeochemical data from arctic tundra. However, at smaller spatial scales, observations and CCaN greening trends differed in the location, extent, and magnitude of greening. CCaN was unable to capture the high rates of MODIS greening in northern wetlands, and the patchy MODIS browning in the southern portion of the North Slope. This model-data disagreement points to disturbance and its legacy impacts on land cover as an important mechanism for understanding greening trends on the North Slope of Alaska.

1. Introduction De Jong et al 2011,Wanget al 2012).Hence,inorderto better understand the observed greening response, it Satellite based observations of the normalized difference must be determined whether the greening signal is vegetation index (NDVI) have illustrated a ‘greening of expected given aprioriknowledge of the ecophysiology the arctic’. Because NDVI is a proxy for leaf area, several and biogeography of arctic vegetation. have hypothesized that these greening trends result from Due to low diversity and low temperatures, pan-

increased vegetation productivity as a result of the rapidly arctic CO2 ﬂuxes are thought to be functionally con- warming temperatures (Walker et al 2003,Goetzet al vergent and can be reasonably modeled with a single 2005, Beck and Goetz 2011,Forkelet al 2016). However, response to light, temperature, and canopy leaf area attributing greening trends to increased vegetation (Street et al 2006; Shaver et al 2007, 2013; Loranty et al productivity is not straightforward (Alcaraz-Segura et al 2011; Street et al 2012; Mbufong et al 2014). This func- 2010). Satellites differ in the magnitude and spatial extent tional convergence is thought to arise from strong of greening, while some regions have exhibited insignif- nitrogen (N) limitation of arctic vegetation, which icant greening trends or declines (i.e. ‘browning’) even as controls both allocation to LAI and leaf photo- the arctic continues to warm (Stow et al 2007,Bhattet al synthetic rates (Williams and Rastetter 1999, Shaver 2013,JuandMasek2016). Moreover, changes in et al 2013). Low mineralization, deposition, and leach- background reﬂectance or satellite performance (e.g. ing rates cause a tight recycling of N among biomass, orbitalorsensordrift) may confound such trends soil organic matter, and available N pools (Shaver et al resulting in a spurious greening signal without an 1992, Stieglitz et al 2006). Consequently, arctic green- increase in leaf area index (LAI)(Rocha and Shaver 2009, ing is dependent on the tight coupling between C

© 2018 The Author(s). Published by IOP Publishing Ltd Environ. Res. Lett. 13 (2018) 125007 and N cycles. This conceptual framework was imple- region comprises nine out of fifteen arctic vegetation mented into a parsimonious mass balance model of community types as defined in Raynolds et al (2006), coupled C and N cycles (e.g. the coupled carbon and and included erect-shrub tundra in the south, acidic nitrogen (CCaN) model)(Wright and Rocha 2018). and non-acidic graminoid tundra in the foothills of CCaN was parameterized with model-data fusion, the North Slope, and wetlands in the north. The North where prior observed ecological information (i.e. eddy Slope also is a region where satellite based NDVI covariance, NDVI, and vegetation and soil data from trends have diverged across satellite platforms, indi- the Arctic Long Term Ecological Research (LTER) cating a need for further validation of arctic greening program) provided realistic biogeochemical con- trends (Stow et al 2007, Ju and Masek 2016). straints on model parameters and predictions. CCaN performed well at capturing the seasonal and spatial 2.2. CCaN model and parameterization variability in NDVI across arctic systems on the North Details on CCaN formulation and parameterization ( ) Slope of Alaska Wright and Rocha 2018 . Conse- are described in Wright and Rocha (2018). Briefly, quently, it was hypothesized that CCaN along with a CCaN couples C and N cycles through the linear model-data fusion approach would provide insight relationship between LAI and foliar N (Williams and into arctic greening trends. Rastetter, 1999), a shared litter-fall rate for C and N, The modern landscape temperature sensitivity of and plant N uptake limited by root biomass. The arctic productivity may place a strong constraint on model is driven with daily light and temperature, and arctic greening trends. Temperature and NDVI on the predicted daily net ecosystem exchange of CO2 (NEE), North Slope of Alaska are tightly linked and follow a LAI, and NDVI using the functional convergence general north to south gradient of increasing temper- model of Shaver et al (2007, 2013) and Loranty et al ( ature and NDVI Raynolds et al 2008; Wright and (2011). NDVI was obtained by inverting the exponen- ) Rocha 2018 . Such sensitivity is likely related to tial relationship between ground based LAI and NDVI the long-term equilibrium response of arctic bio- observations (Shaver et al 2007, Rocha and Shaver geochemistry to temperature; providing an upper 2009). CCaN was constrained with field measured C limit to the expected shifts in ecosystem productivity and N cycling processes and their uncertainties, and with warmer temperatures. Here model-data fusion was parameterized with model-data fusion (Richard- was used with CCaN and latitudinal MODIS NDVI son et al 2010, Keenan et al 2012). observations on the North Slope to constrain the The functional convergence model of Shaver et al temperature sensitivity of LAI under tightly coupled C (2007, 2013) and Loranty et al (2011) hypothesizes that and N cycles. Model-data fusion is an iterative techni- NEE in all arctic tundra communities responds simi- que used to find the best set of parameters that fita larly to temperature, light, LAI, and NDVI throughout proposed model to data with quantified uncertainties time. In accordance with this hypothesis, CCaN was (Keenan et al 2012). Parameters were constrained originally parameterized with NEE and NDVI obser- within a given range, given a priori information vations from a single site (Wright and Rocha 2018). from historical LTER measurements (Wright and With this parameterization, CCaN captured a major- Rocha 2018). Model-data fusion provided a posteriori ity of NDVI observations over 8 years at a single site statistical likelihood parameter distribution based on and NDVI observations across a latitudinal gradient. model structure, data uncertainty, and parameter con- The gradient was comprised of 32 tundra sites along a straints. We hypothesize that such constraints will North–South transect that included all major vegeta- allow for attribution of the observed greening trends tion communities found on the North Slope of Alaska from 2000–2015 to increased temperatures in the (Wright and Rocha 2018). However, this original region. parameterization underestimated NDVI at sites with high LAI (Wright and Rocha 2018). Consequently, 2. Methods for this study we reparameterized CCaN using the MODIS NDVI from the 32 tundra sites previously 2.1. Region of interest mentioned, with a focus on constraining the temper- Analyses focused on the foothills and coastal plains of ature sensitivity of the proportion of N in leaves, which the North Slope of Alaska, but excluded the Brooks ultimately determined LAI and NDVI. Range mountains (figure 1). The region of interest CCaN was reparameterized using a weighted least- encompassed 151 427 km2 and included a diverse squares model-data fusion technique based on the array of topography, landscape age, soil types, and approaches of Keenan et al (2012) and Richardson et al ecosystem communities. Elevation generally decreases (2010). To determine starting values for CCaN pools, a with latitude from ∼1000 m above sea level (asl) in the model-spin up was run for thirty years using decadal South to <5 m asl in the North. Soils in the region vary averaged daily forcing data. This was done to ensure widely in age from 14 kya to >900 kya and pH from that the NDVI trends were not a result of the initial <5.5 to >6.5 as a result of past glaciation events and model conditions being out of equilibrium with the climate (Walker et al 1998, Raynolds et al 2006). The current climate. The pool starting values for the model

2 Environ. Res. Lett. 13 (2018) 125007

Figure 1. Map of the North Slope of Alaska overlaid with a vegetation community map and the regional quadrats (i.e. Northwest, Northeast, Southwest, Southeast) that will be referenced in the Results. Vegetation community map modified from Raynolds et al 2006, and description of modification listed in Methods section 2.4. spin-up were determined from previous publications 2010). In step two, uncertainty-weighted data-model (Jiang et al 2016). The model-data fusion routine was disagreements were normalized based on the variance an adaptive Markov-chain Monte Carlo that used of the j’s obtained in step one, and parameter sets were Metropolis–Hastings sampling (Metropolis et al 1953) only accepted if the cost function for each data stream and incorporated measurement uncertainty into the (ji) passed a chi-square test at 90% confidence. (Franks parameterization, allowing propagation of parameter et al 1999, Richardson et al 2010, Keenan et al 2012). uncertainty into model predictions. In the first step of Step two was run until 1000 parameter sets were the optimization, an individual cost function (ji) was accepted, providing distributions and uncertainty calculated as the total uncertainty-weighted squared estimates for each parameter. data-model disagreement for each data type (Richard- son et al 2010) [equation (1)]. 2.3. CCaN model drivers and MODIS NDVI data ⎛ 2⎞ We obtained daily incoming shortwave radiation for Ni ⎛ yt()- pt ()⎞ j = ⎜å⎜ ii⎟ ⎟,1() the period of 2000–2015 from the North American i ⎝ ⎝ d ()t ⎠ ⎠ t-1 i Regional Reanalysis (NARR) for Arctic Research ( // / / / /) where Ni was the number of observations for data type http: www.esrl.noaa.gov psd data narr . The spa- ° ° i, yi(t) was the data at time t, pi(t) was the value tial resolution of NARR grid data was 0.3 by 0.3 .We × predicted by the model, and δi(t) was the measurement downscaled the data to 1 1 km spatial resolution uncertainty. NDVI uncertainty was quantified as a using bilinear interpolation and extracted the spatial constant 0.07 based on analyses of ground and MODIS domain of the North Slope as in Jiang et al (2016). NDVI observations (Wright and Rocha, 2018). In the Bilinear interpolation is a resampling method that uses initial parameterization, data on soils and vegetation the distance weighted average of the surrounding coarse from the Toolik LTER were used to assign biologically scale pixels to estimate the value of the newly finer valid maximum and minimum parameter values downscaled resolution (Zhang et al 2017). Shortwave ( ) Wright and Rocha 2018 . For the reparameterization radiation was multiplied by 0.1686 to translate the of the temperature sensitivity of the proportion of N in − − − unit from W m 2 to mol PAR m 2 d 1 (Papaioannou leaves, parameter values were randomly drawn from a et al 1993). normal distribution with a mean equal to the initial We used 15 years (2000–2015) of 1 km, 8 d parameter (Wright and Rocha 2018) and standard deviation equal to a fraction of the initial parameter MODIS Land Surface Temperature and Emissivity ( ) range that was adjusted to achieve an acceptance rate MOD11A2 to characterize land surface temperature ( ) of 25%–30%, preventing the routine from getting Wan et al 2004 . We used surface temperature instead stuck at local minima (Richardson et al 2010, Keenan of air temperature for several reasons. First, perma- et al 2012). The best parameter value was determined frost creates a strong heat sink, and as a result, surface by minimizing the overall cost, which was the product and atmospheric temperature rarely deviate from each of the individual cost functions (Richardson et al other during the growing season in tundra ecosystems

3 Environ. Res. Lett. 13 (2018) 125007

(Eugster et al 2000, Rocha and Shaver 2011). Second, category. A positive difference indicated that MODIS surface temperature better reflects the thermal produced larger greening trends, while a negative environment experienced by vegetation (Eugster et al difference indicated the CCaN produced larger green- 2000). Third, surface temperature can be reliably mea- ing trends. sured at large spatial resolution with the MODIS satel- To eliminate low NDVI values due to open water lite, reducing potential errors from extrapolation of and exposed floodplain sediment, we removed MODIS sparse meteorological measurements. We removed and CCaN pixels that were located on or near water pixels with poor data quality and gap-filled with a bodies. Analyses were carried out both with and with- modified Savitsky–Golay filter (Chen et al 2004). The out the water mask in place. Waterbody vector data for Savitsky–Golay filter also was used to transform the 8 d the North Slope was obtained from the National surface temperature product into daily surface temp- Hydrology Dataset (http://catalog.northslope.org/ or erature product for the CCaN model. MODIS surface https://nhd.usgs.gov/). These data were used to locate temperature was retrieved from the Land Processes large rivers (>100 miles in length) and lakes (>1km2 in Distributed Active Archive Center (LP-DAAC) using area) that were masked out of subsequent NDVI and the R package ‘ModisDownload’ (r-gis.net). NDVI trend analyses with a 250–500 m buffer on We used 1 km, 8 d MODIS NDVI (MOD13A2) to each side. characterize vegetation greenness across the North – Slope of Alaska from 2000 2015. MODIS Ts and 2.5. Statistical analyses NDVI products were downloaded from LP-DAAC NDVI trends (i.e. greening or browning trends) were ‘ ’ ( ) using the R package ModisDownload r-gis.net determined with a modified Mann Kendall and a Theil fi for the North Slope region. MODIS NDVI ll-value Sen robust linear regression at the 95% confidence ( ) poor quality pixels from cloud or snow cover were level. The modified Mann Kendall is a non-parametric fi fi – removed and gap- lled with a modi ed Savitsky test for a monotonic trend that also takes account of fi ( ) Golay lter Chen et al 2004 . For CCaN parameteriza- autocorrelation in the data (Hamed and Ramachandra tion, we extracted a 15 year NDVI time series from 32 Rao 1998). The test only determines the presence of a ( × ) – sites 2 2 km in area along a North South transect trend in the data without calculating a regression – ° ’ extending from 65 70 N that ran along Alaska s Dal- slope. The regression slope (i.e. the greening trend) ( ) ton Highway Wright and Rocha 2018 . This transect was determined with a Theil Sen robust linear regres- included all vegetation communities and soil types sion, which is also a non-parametric test that calculates found on the North Slope of Alaska and ranged from the median of the slopes of all lines through pairs of erect-shrub tundra in the south, acidic and non-acidic points (Gilbert 1987). Non-parametric tests were graminoid tundra in the foothills of the North Slope, chosen over traditional least squares regression meth- and wetlands in the far north. The area encompassed ods to account for skewed or heteroscedastic data by the 32 sites represented less than 0.1% of the North and outliers that may influence the estimation of Slope area. These data were used for CCaN para- trends (Christensen 2001). Greening trends that were meterization and determination of MODIS NDVI below the 95% confidence level were not included in ( ) uncertainty Wright and Rocha 2018 . Lastly, we subsequent analyses. ( – aggregated MODIS NDVI into growing season June Differences in MODIS and CCaN greening trends ) August averages and then further aggregated to a 15 for each vegetation type were determined with a stu- year average growing season NDVI map for the North dent’s t-test. Both MODIS and CCaN data demon- Slope. This 15 year average growing season MODIS strated a high degree of spatial autocorrelation, which NDVI map was used to validate the CCaN NDVI map violates the assumption of the t-test that values should for the same period using the same growing season be independent of one another. This spatial auto- average. correlation inflates the degrees of freedom of the statistical test and increases the likelihood of type II 2.4. Arctic vegetation map and water mask statistical errors (Christensen 2001, Zhang et al 2017). Tundra vegetation was delineated using the Circum- Consequently, the degrees of freedom was determined polar Arctic Vegetation Map (CAVM). The CAVM by the number of independent greening clusters in the includes the spatial distribution of 85 plant commu- MODIS and CCaN datasets using a density-based nity descriptions across Alaska at a spatial resolution of clustering analysis (self adjusting HDBSCAN method) ∼14 km (Raynolds et al 2006). Vegetation types on the in ArcGIS pro. The density-based clustering analysis North Slope were grouped according to the following identified spatially correlated groups of greening main categories: Barrens (Class: B3, B4), Graminoid trends and delineated a number of unique clusters tundra (Class: G3, G4), Shrub tundra (Class: S1, S2), across the spatial domain. This removed the spatial and Wetlands (Class: W1, W2, W3)(figure 1). Green- autocorrelation in the data and allowed for indepen- ing trends observed in MODIS and CCaN NDVI were dence among the samples. We delineated greening compared using the difference between the average trend clusters for each vegetation type and used the MODIS and CCaN greening trend for each vegetation number of clusters in each sample as the degrees of

4 Environ. Res. Lett. 13 (2018) 125007

Figure 2. MODIS [Panel A] and CCaN [Panel B] average growing season NDVI from 2000–2015, MODIS-CCaN NDVI differences [Panel C], and the 1:1 relationship between CCaN and MODIS NDVI [Panel D]. Grey dots indicate MODIS and CCaN NDVI across the North Slope of Alaska, while open circles indicate latitudinal average NDVI. The blue line is the linear regression of the latitudinal average NDVI, and error bars are 95% confidence intervals. freedom. Statistical significance was determined at the (figures 3(A), (B)). A majority of the NDVI trends were 95% confidence level. positive, indicating greening, but 0.01% of MODIS and 0.001% of CCaN NDVI trends were negative, 3. Results indicating browning. MODIS exhibited larger variation in greening trends than CCaN, ranging from − 0.003 to −0.001 ΔNDVI y 1 for MODIS and 0.001 to CCaN captured the broad scale spatial patterns of − −0.001 ΔNDVI y 1 for CCaN. average growing season NDVI from 2000–2015 There were noticeable differences in the spatial (figure 2). Both MODIS and CCaN NDVI generally declined with latitude with distinct latitudinal bands extent of greening and browning trends for MODIS (fi ) fi ( across the North Slope (figures 2(A), (B)). MODIS and CCaN gure 3 . At very ne scales 1 km resolu- ) fi NDVI latitudinal differences were more noticeable tion , spatial greening patterns were dif cult to cap- ∼ than those predicted by CCaN, especially in the ture by the model with 30% of trends overlapping western region of the North Slope with a spatial for both model and observation. CCaN browning coefficient of variation for NDVI that was double that NDVI trends were localized and formed several dis- of CCaN (figure 2(C)). CCaN generally under pre- tinct clumps on the northern region of the North dicted NDVI in the south western central region, and Slope. MODIS browning NDVI trends were scattered overestimated NDVI in the alluvial dominated North- with a majority of patches occurring in the south cen- east region of the North Slope (figure 2(C)). MODIS tral region of the North Slope. Strong MODIS green- NDVI had larger NDVI range than CCaN across the ing trends occurred along the Northern coastal plains, North Slope, ranging from 0.24 to 0.71 for MODIS whereas strong CCaN greening trends occurred along and from 0.34 to 0.63 for CCaN (figure 2(D)). The the southwestern region of the North Slope. This average latitudinal gradient of MODIS NDVI was well resulted in high positive model disagreement in the correlated with CCaN with an R2 of 0.82 (figure 2(D)). North, indicating larger MODIS NDVI trends, and This compares an R2 of 0.65 for the MODIS and CCaN high negative model disagreement in the Southwest, NDVI data from the subset of 32 sites used to inform indicating larger CCaN NDVI trends (figure 3(C)). CCaN parameterization. NDVI greening trends were statistically similar for Both CCaN and MODIS indicated widespread both CCaN and MODIS across the entire region (p- NDVI changes across the North Slope from value: 0.11)(figure 3(D)). Although not statistically 2000–2015 (figure 3). CCaN indicated more sig- significant, MODIS produced a slightly larger average − nificant NDVI trends than MODIS. MODIS indicated greening trend (0.004 ΔNDVI y 1) than CCaN (0.003 − − a statistically significant NDVI trend for 35% of the ΔNDVI y ΔNDVI y 1)(figure 3(D)). North Slope, whereas CCaN indicated a statistically There was significant variation in CCaN and significant NDVI trend for 71% of the North Slope MODIS greening trends with latitude and vegetation

5 Environ. Res. Lett. 13 (2018) 125007

Figure 3. MODIS [Panel A] and CCaN [Panel B] NDVI changes from 2000–2015, spatial differences between CCaN and MODIS NDVI trends for the North Slope of Alaska [Panel C], and histogram of the MODIS (red line) and CCaN (blue line) NDVI greening trends [Panel D]. Frequency in panel D corresponds to fractional area of total North Slope region of interest.

Figure 4. MODIS and CCaN trend differences across the North slope with solid black line denoting average of black dots [Panel A]. Average MODIS and CCaN trend difference for each vegetation community from figure 1 map [Panel B]. Error bars are 95% confidence intervals and stars indicate significant difference from zero at the 95% confidence level. communities (figure 4). At low latitudes and along the (Barrens [p-value: 0.92]; Gramminoid [p-value: 0.13]; foothills of the North Slope (<69 °N), averaged CCaN Shrubs [p-value: 0.7]). and MODIS greening trends did not significantly differ (p-value: 0.7). Differences between CCaN and MODIS greening trends were more pronounced at 4. Discussion higher latitudes (figure 4(A)). In general, greening MODIS and CCaN trend differences increased with Our analyses revealed that greening on the North increasing latitude with substantially higher greening Slope of Alaska from 2000–2015 was consistent with trends reported for MODIS than CCaN. On average, the landscape temperature sensitivity of arctic NDVI/ greening trends were twice as large for MODIS than LAI. In general, increased temperatures increased CCaN above 70 °N. This difference was attributed to NDVI/LAI, which is consistent with experimental Wetland communities where MODIS trends were warming manipulations with greenhouses and open- 44% higher than predicted by CCaN (figure 4(B)). top chambers (Shaver et al 1992, Rusted et al 2001, Differences in greening trends in wetland commu- Elmendorf et al 2011, Shaver et al 2014, Buruah et al nities were significantly different (p-value <0.01), 2017). Our approach of substituting the spatial temp- whereas differences in greening trends among erature sensitivity of NDVI with the temporal sensitiv- other vegetation communities were non-significant ity of NDVI was similar to Raynolds et al (2008).

6 Environ. Res. Lett. 13 (2018) 125007

Raynolds et al (2008) used a qualitative statistical of congruency indicated areas of warming, but no model to determine the temperature response of expected positive change in NDVI. This lack of tundra NDVI, which precluded a mechanistic understanding NDVI response to temperature has been shown else- of NDVI trends. However, our approach was mechan- where and supports a more nuanced explanation of istic allowing for the prediction and attribution of attributing greening to temperature alone (Raynolds NDVI trends to temperature by using a mass balance et al 2013, Lara et al 2018). It is likely that the mismatch model that was constrained with uncertainty assessed among greening, no significant change, and browning field observations. Consequently, the model-data patches across the North Slope highlights the impor- fusion approach tested our ecological knowledge of tance of process other than temperature in determin- tundra systems and identified significant model-data ing NDVI trends. These processes are obviously mismatches (Rastetter 2017, Wright and Rocha 2018) occurring at much smaller scales than the large spatial where NDVI is influenced by factors other than surface temperature increases on the North Slope temperature. imply, due to the patchier distribution of MODIS It is clear that, at large spatial and long temporal NDVI trends as compared with CCaN. scales, Arctic NDVI/LAI responses to temperature are MODIS tundra browning was much patchier with tightly constrained. This is supported by CCaN’s abil- a higher spatial standard deviation than predicted by ity to capture the average latitudinal NDVI gradient CCaN. A majority of these small scale browning pat- and the range of greening trends with a single para- ches occurred in southern graminoid communities, meterization with information from <1% of the total and could be the result of several factors including region. These observations support a functionally con- increased litter production through either increased vergent response of coupled carbon and nitrogen plant productivity, changes in vegetation composition cycles to temperature and radiation across the arctic (i.e. shrubification), or increased plant mortality. Lit- (Loranty et al 2011, Wright and Rocha 2018). It was ter decreases NDVI by brightening the surface and also surprising that CCaN was able to capture broad decreasing the amount of green leaf exposure through scale NDVI patterns without the inclusion of impor- shading (Van Leeuwen, Huete 1996, Rocha et al 2008). tant arctic processes that control C and N cycling such Litter production in graminoid communities is high as thaw depth and soil moisture. However, surface and can be the result of both proposed mechanisms. temperature, thaw depth and soil moisture covary Tundra greening trends have decreased or browned in through the energy budget, and these processes are recent years even as the climate continues to warm likely represented in CCaN at longer decadal time- (Bhatt et al 2013, Pheonix and Bjerke 2016, Vickers scales and larger spatial scales through this covariation et al 2016). Early snow melt and frosting events also (Williams and Rastetter 1999, Shaver et al 2013). have been identified as an important process in tundra Although CCaN captured the broad scale arctic browning (Pheonix and Bjerke 2016). The ecological NDVI patterns, variability at finer spatial scales was mechanisms regulating tundra browning remain elu- more challenging to replicate. Some of the larger dif- sive. However, our analyses highlights that these pro- ferences between MODIS and CCaN NDVI also coin- cesses occur independently of large scale temperature cided with spatial variability in pH and underlying increases. bedrock. The large difference in NDVI in the north- CCaN assumed functional convergence in ecosys- east coincides with a patch of younger tertiary bedrock tem cycles of C and N, which indicated that the para- (Wilson et al 2015). A majority of the North Slope is meters driving flux rates in all arctic community types comprised of older Mesozoic bedrock, and landscape are constant once light, temperature, and LAI are age is a strong determinant of soil pH in this region taken into account. In this model, all arctic vegetation (Tieszen 1978; Raynolds et al 2006). Soil pH is not communities function the same, and hence, individual included in CCaN, but is important because it influ- parameterizations based on vegetation community ences arctic species diversity, nutrient availability, type are unnecessary allowing for pan-arctic predic- vegetation productivity, and land surface energy tion of ecosystem function. CCaN also used the mod- exchange (Walker et al 1998, Gough et al 2000, Walker ern latitudinal temperature sensitivity of LAI and et al 2003, Walker et al 2012). These large mismatches NDVI to provide an upper bound to their temporal between model and observations highlight the impor- temperature sensitivity. Such assumptions simplified tance of understanding soil impacts on vegetation and the analyses and highlighted differences in the func- NDVI across the arctic. tioning among arctic vegetation community types. On On average, MODIS and CCaN NDVI trends average, most vegetation communities followed func- agreed until 70 °N and were similar for Barrens, tional convergence in C and N fluxes provided the Graminoid, and Shrub communities. However, there modern day latitudinal temperature sensitivity of were areas where CCaN indicated either a greening or NDVI and LAI, with the exception of northern wet- browning without congruent trends in MODIS. Since land communities. These communities had higher CCaN was largely driven by temperature, this lack temperature sensitivities than other community types

7 Environ. Res. Lett. 13 (2018) 125007 on the North Slope because of the larger than expected vegetation changes measured with remote sensing (i.e. by the latitudinal NDVI temperature response in with ground based observations. Here we demonstrate CCaN) northern wetland MODIS greening trends the utility of combining a parsimonious mass balance (figures 3 and 4). This suggests that either northern model with historical ground based measurements of wetlands do not follow the functional convergence arctic tundra function to develop a mechanistic under- hypothesis and the intrinsic latitudinal temperature standing of arctic greening. We demonstrate that sensitivity of NDVI, which would be inconsistent with northern wetlands are greening at a much higher rate Shaver et al (2007, 2013) and Raynolds et al (2008),or than expected, and that many areas on the North Slope points to a mechanism that allows these systems to are warming but not greening. We hypothesize that more rapidly respond to temperature increases than these patterns indicate that disturbance and succes- anticipated from the latitudinal NDVI-temperature sional processes are important for interpreting green- gradient. ing trends. Further testing of this hypothesis using Northern wetland areas have been subjected to such an approach would be useful for constraining the large changes in hydrology as a result of permafrost response of vegetation to warming and improving thaw, resulting in decreased lake cover and surface predictions of the arctic C cycle in a future warmer moisture (Liljedahl et al 2016). These changes create world. disturbance and initiate succession of newly colonized vegetation that stimulates vegetation productivity Acknowledgments (Odum 1969). Such dynamics accelerate temporal changes in NDVI through the initiation of succession This work was supported by a NASA Earth and Space and the stimulation of vegetation productivity Science Fellowship to Kelseyann Wright, NSF grants during mid-successional stages (McMillan and #1065587 and #1026843 to the Marine Biological Goulden 2008, Rocha et al 2012). The patchiness of Laboratory, and NSF grant #1556772 to the Uni- greening trends also support a disturbance induced versity of Notre Dame. Any use of trade, product, or greening as disturbances tend to be smaller in scale firm names is for descriptive purposes only and does than regional temperature increases. Successional not imply endorsement by the US Government. dynamics also may explain why arctic greening has recently slowed even as temperatures continue to increase. Older disturbed patches will enter a period of ORCID iDs senescence due to increased nutrient limitation ( ) Rocha et al 2012 . Permafrost thaw, subsidence, and A V Rocha https://orcid.org/0000-0002- thermokarst events are increasing in the arctic and 4618-2407 impart successional land cover legacies that can S R Curasi https://orcid.org/0000-0002-4534-3344 decouple vegetation from year to year temperature change (Raynolds et al 2008, Liljedahl et al 2016, References Pastick et al 2018). Other disturbances could include increased productivity and shrub cover in water Abbott B W et al 2016 Biomass offsets little or none of permafrost tracks, which are likely more common in southern carbon release from soils, streams, and wildfire: an expert parts of the North Slope (Curasi et al 2016). The role of assessment Environ. Res. Lett. 11 034014 Alcaraz-Segura D, Chuviecos E, Epstein H E, Kasischke E S and disturbance in explaining greening trends has gained Trishchenko A 2010 Debating the greening versus browning recent attention (Lara et al 2018), and our results cor- of North American boreal forest: differences between satellite roborate the importance of land cover legacies in datasets Glob. Change Biol. 16 760–70 understanding arctic greening trends. Baruah G, Molau U, Bai Y and Alatalo J M 2017 Community and species-specific responses of plant traits to 23 years of experimental warming across subarctic tundra plant communities Sci. Rep. 7 2517 5. Conclusions Beck P S A and Goetz S J 2011 Satellite observations of high northern latitude vegetation productivity changes between 1982-2008: Understanding links between climate change, arctic ecological variability and regional differences Environ. Res. greening, and C cycling are critical in constraining the Lett. 6 045501 Bhatt U S et al 2013 Recent declines in warming and vegetation temperature sensitivity of arctic vegetation productiv- greening trends over pan-Arctic tundra Remote Sens. 5 ity and predicting future arctic C cycling (Schuur et al 4229–59 2015, Abbott et al 2016, Fisher et al 2018). Because of Chen J, Jönsson P, Tamura M, Gu Z, Matsushita B and Eklundh L its large spatial extent, remote sensing provides a 2004 A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky–Golay filter valuable tool in understanding these links across the Remote Sens. Environ. 91 332–44 arctic, but remains limited because it includes satellite Christensen R 2001 Advanced Linear Modeling 2nd edn (Berlin: measurement artifacts and provides no mechanistic Springer) pp 398 understanding due to the indirect measure of vegeta- Curasi S R, Loranty M M and Natali S M 2016 Water track distribution and effects on carbon dioxide flux in an eastern tion productivity through a spectral reflectance proxy. Siberian upland tundra landscape Environ. Res. Lett. 11 Consequently, there is a need to validate and verify 045002

8 Environ. Res. Lett. 13 (2018) 125007

De Jong R, de Bruin S, de Wit A, Schaepman M E and Dent D L 2011 ecosystem change and causation across Alaska Glob. Change Analysis of monotonic greening and browning trends from Biol. (https://doi.org/10.1111/gcb.14279) global NDVI time-series Remote Sens. Environ. 115 692–702 Pheonix G K and Bjerke J W 2016 Arctic browning: extreme events Elmendorf S C et al 2011 Global assessment of experimental climate and trends reversing arctic greening Glob. Change Biol. 22 warming on tundra vegetation: heterogeneity over space and 2960–2 time Ecology Letters 15 164–75 Rastetter E B 2017 Modeling for understanding v. modeling for Eugster W et al 2000 Land-atmosphere energy exchange in arctic numbers Ecosystems 20 225–221 tundra and boreal forest: available data and feedbacks to Raynolds M A, Walker D A and Maier H A 2006 Alaska arctic tundra climate Glob. Change Biol. 6 84–115 vegetation map. Scale: 1:4000000. Conservation of Arctic Fisher J B et al 2018 Missing pieces to modeling the Arctic-Boreal Flora and Fauna (CAFF) Map No. 2, U.S. Fish and Wildlife puzzle Environ. Res. Lett. 13 020202 Service, Anchorage, Alaska Forkel M, Carvalhais N, Rodenbeck C, Keeling R, Heimann M, Raynolds M K, Comiso J C, Walker D A and Verbyla D 2008 Thonicke K, Zaehle S and Reichstein M 2016 Enhanced Relationship between satellite-derived land surface seasonal CO2 exchange caused by amplified plant temperatures, arctic vegetation types, and NDVI Remote Sens. productivity in northern ecosystems Science 351 696–9 Environ. 112 1884–94 Franks S W, Beven K J and Gash J H C 1999 Multi-objective Raynolds M K, Walker D A, Verbyla D and Munger C A 2013 conditioning of a simple SVAT model Hydrol. Earth Syst. Sci. Patterns of changes within a tundra landscape 22 year Landsat Discuss. 3 477–88 trends in an area of the northern foothills of the Brooks range, Gilbert R O 1987 Sen’s nonparametric estimator of slope Statistical Alaska Arct., Antarct., Alpine Res. 45 249–60 Methods for Environmental Pollution Monitoring (New York: Richardson A D et al 2010 Estimating parameters of a forest Wiley) pp 217–9 ecosystem C model with measurements of stocks and fluxes Goetz S J, Bunn A G, Fiske G J and Houghton R A 2005 Satellite- as joint constraints Oecologia 164 25–40 observed photosynthetic trends across boreal North America Rocha A V, Loranty M M, Higuera P E, Mack M C, Hu F S, associated with climate and fire disturbance Proc. Natl Acad. Jones B M, Breen A L, Rastetter E B, Goetz S J and Shaver G R Sci. 102 13521–5 2012 The footprint of Alaskan tundra fires during the past Gough L, Shaver G R, Carroll J, Royer D L and Laundre J A 2000 half-century: implications for surface properties and radiative Vascular plant species richness in Alaskan arctic tundra: the forcing Environ. Res. Lett. 7 044039 importance of soil pH J. Ecol. 88 54–66 Rocha A V, Potts D L and Goulden M L 2008 Standing litter as a

Hamed K H and Ramachandra Rao A 1998 A modified Mann- driver of interannual CO2 exchange variability in a freshwater Kendall trend test for autocorrelated data J. Hydrol. 204 Marsh J. Geophys. Res.: Biogeosciences 113 G04020 182–96 Rocha A V and Shaver G R 2009 Advantages of a two band EVI Jiang Y, Rocha A V, Rastetter E B, Shaver G R, Mishra U, calculated from solar and photosynthetically active radiation Zhuang Q and Kwiatkowski B L 2016 C-N-P interactions fluxes Agric. Forest Meteorol. 149 1560–3 control climate driven changes in regional patterns of C Rocha A V and Shaver G R 2011 Postfire energy exchange in arctic storage on the North Slope of Alaska Landscape Ecol. 1 tundra: the importance and climatic implications of burn 195–213 severity Glob. Change Biol. 17 2831–41 Ju J and Masek J G 2016 The vegetation greenness trend in Canada Rustad L E, Campbell J L, Marion G M, Norby R J, Mitchell M J, and US Alaska from 1984-2012 Landsat data Remote Sens. Hartley A E, Cornelissen J H C, Gurevitch J and GCTE- Environ. 176 1–16 NEWS 2001 A meta-analysis of the response of soil Keenan T F, Davidson E, Moffat A M, Munger W and respiration, net nitrogen mineralization, and aboveground Richardson A D 2012 Using model-data fusion to interpret plant growth to experimental ecosystem warming Oecologia past trends, and quantify uncertainties in future projections, 126 543–62 of terrestrial ecosystem carbon cycling Glob. Change Biol. 18 Schuur E A G et al 2015 Climate change and the permafrost carbon 2555–69 feedback Nature 520 171–9 Lara M J, Nitze I, Grosse G, Martin P and McGuire A D 2018 Shaver G R, Billings W D, Chapin F S, Giblin A E, Nadelhoffer K J, Reduced arctic tundra productivity linked with landform and Oechel W C and Rastetter E B 1992 Global change and the climate change interactions Sci. Rep. 8 2345 carbon balance of Arctic ecosystems BioScience 42 433–41 Liljedahl A K et al 2016 Pan-Arctic modern sub-decadal permafrost Shaver G R et al 2014 Terrestrial Ecosystems at Toolik Lake, Alaska. ice-wedge degradation and hydrologic implications Nat. Alaska’s Changing Arctic: Ecological Consequences for Tundra, Geosci. 9 312–8 Streams, and Lakes ed J E Hobbie and G W Kling (New York: Loranty M M, Goetz S J, Rastetter E B, Rocha A V, Shaver G R, Oxford Universtiy Press) pp 90–142 Humphreys E R and Lafleur P M 2011 Scaling an Shaver G R, Rastetter E B, Salmon V, Street L E, van de Weg M J, instantaneous model of tundra NEE to the arctic landscape Rocha A, van Wijk M T and Williams M 2013 Pan-Arctic Ecosystems 14 76–93 modelling of net ecosystem exchange of CO2 Phil. Trans. R. Mbufong HN M et al 2014 Assessing the spatial variability in peak Soc. B 368 20120485–20120485 season CO2 exchange characteristics across the Arctic tundra Shaver G R, Street L E, Rastetter E B, Van Wijk M T and Williams M using a light response curve parameterization Biogeosciences 2007 Functional convergence in regulation of net CO2 flux in 11 4897–912 heterogeneous tundra landscapes in Alaska and Sweden McMillan A and Goulden M 2008 Age‐dependent variation in the J. Ecol. 95 802–17 biophysical properties of boreal forests Glob. Biogeochem. Stieglitz M, McKane R B and Klausmeier C A 2006 A simple model Cycles 22 GB2019 for analyzing climatic effects on terrestrial carbon and Metropolis N, Rosenbluth A W, Rosenbluth M N, Teller A H and nitrogen dynamics: an arctic case study Glob. Biogeochem. Teller E 1953 Equation of state calculations by fast computing Cycles 20 GB3016 machines J. Chem. Phys. 21 1087 Stow D A, Petersen A, Hope A, Engstrom D R and Coulter L 2007 Odum E P 1969 Strategy of ecosystem development Science 164 Greenness trends of arctic tundra vegetation in the 1990s 262–70 comparison of two NDVI datasets from NOAA AVHRR Papaioannou G, Papanikolaou N and Retalis D 1993 Relationships systems Int. J. Remote Sens. 28 4807–12 of photosynthetically active radiation and shortwave Street L E, Shaver G R, Rastetter E B, van Wijk M T, Kaye B A and irradiance Theor. Appl. Climatol. 1 23–7 Williams M 2012 Incident radiation and the allocation of Pastick N J, Jorgenson M T, Goetz S J, Jones B M, Wylie B K, nitrogen within Arctic plant canopies: implications for Minsley B J, Genet H, Knight J F, Swanson D K and predicting gross primary productivity Glob. Change Biol. 18 Jorgenson J C 2018 Spatiotemporal remote sensing of 2838–52

9 Environ. Res. Lett. 13 (2018) 125007

Street L E, Shaver G R, Williams M and van Wijk M T 2006 What is Walker D A, Epstein H E E P, Raynods M K, Kuss P, Kopecky M A, the relationship between changes in canopy leaf area and Frost G V, Daniels F J A, Leibman M O, Moskalenko N G and changes in photosynthetic CO2 flux in arctic ecosystems? Matyshak G V 2012 Environment, vegetation and greenness J. Ecol. 95 139–50 (NDVI) along the north American and Eurasia Arctic Sulla-Menashe D, Woodcock C E and Friedl M A 2018 Canadian transects Environ. Res. Lett. 7 015504 boreal forest greening and browning trends: an analysis of Wan Z, Zhang Y, Zhang Q and Li Z L 2004 Quality assessment and biogeographic patterns and the relative roles of disturbance validation of the MODIS global land surface temperature Int. versus climate drivers Environ. Res. Lett. 13 014007 J. Remote Sens. 25 261–2674 Tieszen L L 1978 Vegetation Production Ecology of an Alaskan Arctic Wang D, Morton D, Masek J, Wu A, Nagol J, Xiong X, Levy R, Tundra. Ecological Studies vol 29 (Berlin: Springer) pp 686 Vermote E and Wolfe R 2012 Impact of sensor degradation on Van Leeuwen W J D and Huete A R 1996 Effects of standing litter on the MODIS NDVI time series Remote Sens. Environ. 119 55–61 the biophysical interpretation of plant canopies with spectral Williams M and Rastetter E B 1999 Vegetation characteristics and indices Remote Sens. Environ. 55 123–8 primary productivity along an arctic transect: implications Vickers H, Hogda K A, Solbo S, Karlsen S R, Tommervik H, for scaling-up J. Ecol. 87 885–98 Aanes R and Hansen B B 2016 Changes in greening in the Wilson F H, Hults C P, Mull C G and Karl S M 2015 Geologic map of high Arctic: insights from a 20 year AVHRR max NDVI Alaska: US Geological Survey Scientific Investigations Map dataset for svalbard Environ. Res. Lett. 1 105004 3340, pamphlet 196 p., 2 sheets, scale 1:1584000 Walker D A et al 2003 Phytomass, LAI, and NDVI in northern Wright K S and Rocha A V 2018 A test of functional convergence in Alaska relationships to summer warmth, soil pH, plant carbon fluxes from coupled C and N cycles in Arctic tundra functional types, and extrapolation to the circumpolar Arctic Ecol. Modelling 383 31–40 J. Geophys. Res. - Atmos. 108 8169 Zhang J, Atkinson P and Goodchild M F 2017 Scale in Spatial Walker D A et al 1998 Energy and trace-gas fluxes across a soil Information and Analysis 1st edn (Boca Raton, FL: CRC Press) pH boundary in the Arctic Nature 394 69–472 pp 367