Informing grassland ecosystem modeling with in-situ and remote sensing observations

24 Historical grassland aboveground plant productivity (ANPP) was simulated by the DayCent-UV 25 ecosystem model across the midwestern and western conterminous United States. For this study 26 we developed a novel method for informing the DayCent-UV model and validating its plant 27 productivity estimates for grasslands of the midwestern and western conterminous USA by 28 utilizing a wide range of data sources at multiple scales, from field observations to remotely 29 sensed satellite data. The model phenology was informed by the MODIS MCD12Q2 product, 30 which showed good agreement with in-situ observations of growing season commencement and 31 duration across different grassland ecosystems, and with observed historical trends. Model 32

from each simulated grid cell were compared to a remote-sensing ANPP modified version 33 offered by the Analysis Rangeland Platform (RAP). This modified RAP ANPP calculation 34 incorporated total annual precipitation, instead of mean annual temperature, as the control factor 35 for the fraction of carbon allocated to roots. Strong temporal correlations were obtained between 36 RAP and DayCent-UV, especially across the Great Plains. Good agreement was also found when 37 the model results were compared with ANPP observations at the site and county level. The data 38 produced by this study will serve as a valuable resource for validation or calibration of various 39 models that aim to capture accurate productivity dynamics across diverse grassland ecosystems. 40 Plain Language Summary 41 This research used a computer model called DayCent-UV to simulate daily grassland growth 42 across the central and western regions of the contiguous United States. To improve the 43 agreement between the simulations and real-world conditions, we incorporated data from local 44 field measurements and satellite imagery. This data helped determine the start and end dates of 45 the growing season at each location. The simulated annual growth showed good agreement with 46 satellite estimates from the Rangeland Analysis Platform (RAP), another computer application 47 that monitors rangeland vegetation, and with local observations based on harvesting and 48 weighing vegetation, particularly across the Great Plains. These results are valuable for 49 validating and refining other computer models that aim to accurately simulate plant growth in 50 grassland ecosystems; the predictions of these models are crucial for understanding the balance 51 of carbon between plants, soils, and the atmosphere as the climate changes. 52 53 1 Introduction 54 Grasslands, defined as terrestrial ecosystems dominated by herbaceous and shrub vegetation and 55 maintained by fire, grazing, drought and/or freezing temperatures, comprise ~40 percent of the 56 earth’s land surface excluding Greenland and Antarctica (White et al. 2000). In the United 57 States, rangelands, similarly defined as lands on which the native vegetation is predominantly 58 grasses, grass-like plants, forbs, or shrubs suitable for grazing or browsing use, comprise about 59 30% of the entire land cover, totaling about 770 million acres (https://www.nrcs.usda.gov). 60 Grasslands provide forage for domestic livestock and are habitat for wildlife. Additionally, 61 grasslands store approximately 34 percent of the global stock of terrestrial carbon, with ~90% of 62 their carbon stored belowground as root biomass and soil organic carbon (White et al. 2000). 63 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Grasslands are also vulnerable to degradation from climate change impacts including more 64 intense and frequent drought (Bardgett et al. 2021). Because grasslands are so vital to 65 biodiversity, carbon sequestration, and human and animal welfare, it is important to quantify 66 annual and long-term variation grassland plant productivity and its response to changes in 67 climate. 68 Estimation of grassland plant productivity over a large region can be achieved using a 69 combination of in-situ biomass measurements, eddy covariance flux tower data, remotely sensed 70 satellite data, and ecosystem modeling. Field-based measurements of grassland productivity, 71 such as those based on clipping vegetation at peak biomass, provide a direct quantification of 72 above-ground net primary productivity (ANPP), but these measurements are labor intensive and 73 limited in their spatial coverage. Flux tower data provide a reliable estimate daily net ecosystem 74 exchange (NEE) (total C sequestration in plants and soils), but their spatial coverage across is not 75 yet continuous and this data does not directly estimate net primary productivity (NPP) or ANPP. 76 Phenocams are tower-mounted digital cameras that provide on-the-ground information about the 77 phenology of rangeland plants. They can detect interannual variability in growing season 78 commencement and duration and are important for designing management systems, but do not 79 directly provide an estimate of NPP and are also limited in their spatial extent (Browning et al. 80 2019). Remotely sensed satellite data, such as NDVI, are frequent measurements (about every 81 two weeks) that have existed for several decades at various resolutions and that detect long-term 82 changes in the timing and magnitude of vegetation greenness over large spatial extents. This data 83 has been used to estimate biomass (Morgan et al. 2016) and NPP (Hermance et al. 2015, Chen et 84 al. 2019, Hartman et al. 2020, Jones et al. 2021). In particular, the Rangeland Analysis Platform 85 (RAP) uses NDVI, a plant functional type cover dataset, and linear mixing theory to estimate 86 herbaceous biomass in rangelands (Jones et. al, 2021). In turn, estimates of biomass derived from 87 remotely sensed data are validated against on the ground measurements (Paruelo et al. 1997, 88 Reeves et al. 2021). 89 Ecosystem models represent the co-evolution of both plants and soils and can tie together 90 disparate data sources by using them as model input or for model calibration/validation. The 91 models require regional sources of meterological data and soils data as inputs, phenological data 92 (growing season start and end dates) as input or for model validation, and observations ANPP to 93 compare to model predictions. Particularly, advanced models like the Community Land Model 94 still struggle in defining the start and the end of the growing season, which is crucial to avoid 95 bias in productivity (Li et al. 2022). 96 For this study we developed a novel method for informing the DayCent-UV model and 97 validating its plant productivity estimates for grasslands of the midwestern and western 98 conterminous USA by utilizing a wide range of data sources at multiple scales, from field 99 observations to remotely sensed satellite data. We also demostrate how these myriad data sources 100 are related. We used site-level biomass data at semi-arid and mesic grassland for detailed 101 calibration of model parameters. We used MODIS satellite data with location-specific green up 102 and green down dates to define growing season start and duration for DayCent-UV, and we show 103 how this MODIS phenological data is consistent with PhenoCam and flux tower data. We also 104 used other site-level ANPP data, along with county-level peak biomass data from the Natural 105 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Resource Conservation Service (NRCS), and RAP NPP estimates to validate the model 106 predictions of ANPP at multiple spatial scales. 107 The goal of this research was to accurately simulate the interannual variability of grass plant 108 productivity in the grassland regions of the midwestern and western conterminous USA in order 109 improve our understanding of how these ecosystems respond to changing climatic conditions. 110 Previous studies have shown that grassland productivity is strongly correlated to growing season 111 precipitation or actual evapotranspiration (Sala et al. 1988, Chen et al 2019). Interannual 112 variability in precipitation and production has been linked to longer cycles in sea surface 113 temperatures (Chen et al. 2017). This research serves as a foundation for simulating climate 114 change impacts on grassland ecosystems, as well as to support advances in long-term remote-115 sensing products aiming to assit land managers and to motivate a more exhaustive temporal and 116 spatial validation in regional grasslands simulations. 117 2 Materials and Methods 118 2.1 Spatial domain 119 The spatial distribution of the simulation grid, with 30 km x 30 km spatial resolution, covers 120 most of the grassland ecosystems in the USA, including the Great Plains (Short Grass Steppe, 121 Mixed Grass Prairie and Tall Grass Prairie), cold and warm desert ecosystems (Great Basin, 122 Colorado Plateau and Chihuahuan Desert), inter-mountain grasslands (Montana valley, 123 Wyoming Steppe and Palouse), and the annual grasslands in California (Figure 1). Each of the 124 2712 grassland grid cells were defined based on the USGS's 24-category land cover 125 Figure 1. The grassland simulation grid (hollow black squares) overlaying different ecosystems in the western and midwest United States.The Great Plains region includes the Short Grass Steppe, Mixed Grass Priarie, and Tall Grass Prairie in the central part of the map from Canada (in the north) and Texas (in the south). The black stars indicate the locations where we validated e satellite- derived phenology with PhenoCam and other data. Map modified from Olson et al. (2001) .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal classification map, adapted to the Climate-Weather Research and Forescated model (He et al. 126 2022). 127 2.2 The DayCent-UV model 128 The DayCent ecosystem model is a process-based model that simulates daily fluxes of water, 129 carbon and nutrients between the atmosphere, soil and vegetation (Parton et al. 1998). DayCent 130 can simulate the plant production, soil carbon decomposition and formation, nutrient cycling, 131 water fluxes, trace gas fluxes, and soil temperature dynamics across the vertical soil profile (Del 132 Grosso et al. 2011), and is able to incorporate alterations introduced by the environment and 133 management practices including fire, grazing, fertilizer application, tillage, and harvest. 134 In DayCent, the plant submodel computes the above and below ground biomass based on a 135 plant’s maximum potential production, which depends on the solar incoming radiation, the light 136 use efficiency of each plant type, water availability, and temperature. For grasses and other plant 137 types, the total NPP corresponds to the maximum potential production reduced by nutrient 138 limitation. The NPP is allocated to above-ground shoots and below-ground juvenilte and mature 139 fine roots according to the shoot/root ratio, which is a function of soil moisture and nutrient 140 availability (Parton et al. 1987, Gherardi & Sala 2020). In this work we used the DayCent-UV 141 version, which also simulates the C losses associated with the photodegradation of surface and 142 standing dead plant litter due to incident ultra-violet radiation (Chen et al. 2016). We used a 143 point version of the model for calibrating parameters at the site-level and a gridded version for 144 the regional simulations; these two versions differ only in the input data supplied to the models. 145 2.3 DayCent-UV optimization 146 To prepare DayCent-UV to simulate all grasslands in the region (Figure 1) we first optimized the 147 point version of the model for two distinct grassland ecosystems: Central Plains Experimental 148 Range (CPER) in the Shortgrass Steppe of Colorado and Konza Tallgrass Prairie in Kansas 149 (Table 1) where long-term observations of aboveground plant productivity (ANPP) were 150 available (Blair and Nippert 2024, Dorich et al. 2021). To minimize temporal bias and maximize 151 the coefficient of determination (R²) between observations and model predictions of ANPP, we 152 employed a Bayesian optimization approach to identify the optimal parameters for modeling 153 annual grassland ANPP. For this optimization we used site-specific meterological inputs from 154 the Applied Climate Information System (ACIS) Web Services (http://data.rcc-acis.org/), and 155 specific soil properties data from STATSGO (Schawrz & Alexander, 1995). The observed 156 annual average ANPP for three grazing treatments (light, moderate and heavy) were used for the 157 CPER site while ANPP data from the annually burned ridge site were used for the Konza site. 158 Detailed information on the necessary model adjustments to accurately simulate ANPP time 159 series for both sites is provided in Section SP1 of the supplementary material. Live root biomass 160 at both sites was assumed to be around 750 g biomass m-2, while the ratio of live root production 161 to live shoot production was assumed to be 1.2 and 0.9 for CPER and Konza, respecitively 162 (Milchunas & Lauenroth 2001, McCulley et al. 2009). In addition to calibrating plant growth 163 parameters through the Bayesian optimaization, we adjusted live shoot death rates in the model 164 so the simulated live leaf biomass for CPER and Konza followed the patterns of observed 165 MODIS NDVI data at the two sites. Grassland plant attributes for each of the 2712 simulated 166 grid cells, with the exception of annual grasslands in California, were assigned based on 167 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal historical annual mean precipitation. Grid cells with annual precipitation below 500 mm were 168 assigned plant attributes from the calibrated CPER site, while those exceeding 500 mm 169 precipitation were assigned plant attributes from the Konza site. For the California annual 170 grasslands we used plant parameters determined from previous work (Ryals et al. 2015). 171 2.4 DayCent-UV gridded inputs 172 Daily meteorological inputs from 1979-2015 for each grid cell in this study were obtained from 173 the National Centers for Environmental Prediction North American Regional Reanalysis (NCEP, 174 2005). The daily metereological inputs necessary to run the gridded DayCent-UV model 175 included precipitation, minimum air temperature, maximum air temperature, relative humidity, 176 wind speed, and incoming solar radiation. 177 The vertical soil structure for each grid was based on the ten-layer configuration of the Common 178 Land Model (Dai et al., 2003). Its hydraulic and textural properties were derived from a 179 combination of two datasets: the Continental United States Multi-Layer Soil Characteristics 180 Dataset (CONUS-SOIL) and the Food and Agriculture Organization of the United Nations 181 Educational, Scientific, and Cultural Organization (FAO-UNESCO) Soil Map of the World 182 (Liang et al., 2005). 183 2.4 Phenological Parameterization 184 We used the MODIS MCD12-Q2 phenology product to define the grid cell-specific growing 185 season parameters used by the DayCent-UV model (Friedl et al. 2022). The 'GreenUp' band, 186 representing the first date when the EVI timeseries surpassed its 15% amplitude, was used to 187 identify the growing season start. Similarly, the 'MidGreenDown' band, signifying the last date 188 when the EVI timeseries dipped below its 50% amplitude, determined the growing season's end. 189 MODIS MCD12-Q2 phenology product also provides the dates of peak biomass, Senescence, 190 and Dormancy which we did not use. For each simulated grid, we extracted the mean and 191 standard deviation for the Greenup and MidGreenDown dates over a 20-year period (2001-2020) 192 from all MODIS pixels within the grid. The mean Greenup date minus one standard deviation 193 determined the potential start of the growing season and the mean MidGreenDown date 194 determined the onset of plant senescence. The actual start of the growing season was simulated 195 on or after the potential start date and varied each year once air temperatures exceeded the 196 threshold for growth. 197 To determine which of the MODIS MCD12-Q2 derived phenology values were most appropriate 198 for our study, we selected nine sites across diverse grassland ecosystems (black stars in Figure 199 1). Daily green chromatic coordinate (GCC) data from the PhenoCam Dataset version 2.0 200 (Seyednasrollah et al. 2019) and daily Gross Primary Productivity (GPP) or Net Ecosystem 201 Exchange (NEE) data from AmeriFlux eddy covariance towers were used for comparison. GPP 202 reflects light use efficiency for carbon fixation, making it suitable for phenology assessment. The 203 start of the growing season was defined as the date when the GPP timeseries began to rise from 204 zero, while the end of the growing season was defined as its decline to zero. For NEE, the start of 205 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal the growing season was defined when the NEE flux drops from positive to negative while the 206 end of the growing season was defined when NEE becomes positive again. 207 To minimize the influence of non-grassland areas, the MODIS MCD12-Q2 products were 208 masked using the annual grassland classification from the MODIS MCD12Q1 Version 6.1 209 product by the University of Maryland (UMD) (Friedl et al., 2022b). This process, performed 210 through the Google Earth Engine platform, ensured the final processed time series represented 211 only the relevant grassland areas. 212 2.5 ANPP estimated by the Rangeland Analysis Platform (RAP) 213 The Rangleland Analysis Platform (RAP) was developed to assist and monitor rangeland across 214 USA (Jones et al. 2021). It estimates annual NPP by composing daily NDVI time series from 16-215 days Landsat imagery. RAP products have a 60m pixel resolution and a temporal window from 216 1986 to the current time. To estimate anuual ANPP, RAP multiplies the annual NPP by the 217 fraction of carbon allocated in shoots 𝑓𝐴𝑁𝑃𝑃 , based on the Hui and Jackson empirical 218 relationship (Hui & Jackson, 2006) 219 𝑓𝐵𝑁𝑃𝑃 = 0.8290 − 0.0129 ⋅ 𝑀𝐴𝑇 𝑤ℎ𝑒𝑟𝑒 𝑓𝐴𝑁𝑃𝑃 = 1 − 𝑓𝐵𝑁𝑃𝑃 Eqn. (1) 220 Here, 𝑓𝐵𝑁𝑃𝑃 is the fraction carbon allocated in roots and MAT is the mean annual temperature 221 (°C). We propose an alternative approach incorporating the empirical relationship established by 222 Gerardhi & Sala (2020). Their research suggests a stronger correlation between ANPP and 223 annual precipitation (APPT, mm) compared to the formula in Eqn.(1), given by 224 𝑓𝐵𝑁𝑃𝑃 = 0.8845 − 0.0005 ⋅ APPT Eqn. (2) 225 This relationship was also supported by an independent study in Sun et al. (2021). 226 We regridded RAP NPP estimates obtained from the RAP Google Earth Engine version 227 (http://rangeland.ntsg.umt.edu/data/rap/rap-vegetation-cover/) to our 30 km x 30 km grassland 228 grid and calculated two estimates of ANPP using Eqn. (1) and Eqn. (2). For Eqn. (2) we accessed 229 the GRIDMET database to get the annual precipitation (Abatzoglou, 2013). We evaluated our 230 proposed alternative RAP ANPP estimate and the original RAP ANPP calculation against 231 DayCent-UV results and against site-level and county-level ANPP data. 232 2.6 Site and county level ANPP observations 233 We assessed the DayCent-UV model’s ability to simulate site-level temporal variability by 234 comparing in-situ ANPP observations from long-term experiments (Table 1) with the 235 corresponding modeled ANPP values from the nearest simulated grid cell. Note that although the 236 point version of DayCent-UV was used to calibrate the plant model for CPER and Konza using 237 local meterological and soils data, the grid cell-level results that we compared to in-situ 238 observations for CPER, Konza, and the other four locations used grid cell averaged 239 meteorological inputs and soil properties. 240 Annual ANPP in the period 2004 – 2014 were collected every few years by the Natural Resource 241 Conservation Service (NRCS) across multiple sampling locations in each county in the Great 242 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Plains and were averaged by county to compute mean county-level ANPP (Chen et al. 2019). We 243 averaged each simulated grid cell falling within each county to validate the DayCent-UV and 244 RAP ANPP estimates against the NRCS data. 245 Table 1. Sitelevel ANPP observations 246 Name State Ecosystem Methods and References CPER Colorado Short Grass Steppe During the end of the season, fifteen biomass samples in 0.25 and 0.1 m2 were hand-clipped. Levels of grazing were reached introducing 9.3 animal unit days (DAU) per hectarea for light grazing (20% utilization of peak biomass), 12.5 AUD/ha for moderate grazing (40% utilization) and 18.6 AUD/ha for intensive grazing (60% utilization). Biomass was dried at 60°C to constant mass. (Irrisari, et al. 2016) Konza Kansas Tall Grass Prairie During the end of the season, over the annually burned 1D watershed in the Florence soil type, twenty biomass samples in 0.1 m2 were clipped. Samples were dried at 60 C for 24-hours prior to weighing, to obtain the standing crop biomass (grams per square meter) of live graminoids, forbs, woody plants. (Blair and Nippert, 2024) Miles City Montana Short Grass Steppe Clipped at the peak biomass productionof the year (Dorich et al. 2021) Cotonwood South Dakota Mixed Grass Prairie Clipped at the peak biomass productionof the year, averaging on plots with heavy, moderate and light grazing intensities (Dorich et al. 2021) Hays Kansas Mixed Grass Prairie Clipped at the peak biomass productionof the year, averaging on plots with heavy, moderate and light grazing intensities (Dorich et al. 2021) Jornada New Mexico Chihuahuan Desert Clipped at the peak biomass productionof the year. Low productivity sites were selected: CALI, GRAV, EAST, TAYL, and WEST. Data is available from Dorich et al. (2021). 4 Results 247 4.1 Bayesian optimization 248 Given that seasonal precipitation is a major driver of annual ANPP variability (Lauenroth and 249 Sala, 1992), we calibrated the DayCent model at long-term experimental sites representing 250 contrasting precipitation regimes, CPER and Konza (Table 1). The best fit parameters for the 251 CPER and Konza sites were obtained after performing a Bayesian optimization, constrained to a 252 minimum bias and high temporal correlation (𝑅2) between site-level annual observed and 253 simulated ANPP. The best temporal performance in both sites had 𝑅2 values close to 0.5 254 (Figure S1). Though simulated above ground shoot biomass was not directy used in the 255 optimization process, continuous simulated above ground shoots coincided well with MODIS-256 NDVI values, particularly at CPER (Figure S1). 257 4.2 Evaluation of phenology data 258 Using an dependable source of data to define model phenology is crucial in regions where the 259 start and the end of growing season varies spatially. We utilized the 'GreenUp' and 260 'MidGreenDown' bands from the MODIS MCD12-Q2 product to define the start and the end of 261 the growing season, respectively. 262 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal 263 Figure 2 Validating MODIS MCD12-Q2 phenology across different ecosystem sites in Figure 1: Temporal variation for GPP in 264 𝜇𝑚𝑜𝑙/𝑚2𝑠 (green filled), NEE in 𝜇𝑚𝑜𝑙/𝑚2𝑠 (gray filled) and GCC (black solid line) and the corresponding MODIS dates 265 indicating the Green-up (vertical blue dashed line), Senescence (vertical brown dashed line), Mid Green-Down (vertical red 266 dashed line) and Dormancy (vertical orange dashed line). 267 We confirmed that the MODIS GreenUp and MidGreenDown dates were consistent with other 268 available ecosystem data. To validate these key dates we showed that there was good agreement 269 in the start/end of the growing season between MODIS phenology and the rise/fall in 270 photosynthesis activity informed by GPP (green filled) or NEE (gray filled) (Figure 2). We 271 observed a strong agreement between the temporal patterns in sites with GPP and GCC 272 information. This agreement allows us to validate MODIS phenology dates at sites with only 273 PhenoCam information. Overall, for the nine sites representing diverse grassland ecosystems, 274 MODIS GreenUp and MidGreenDown dates successfully captured the start and end of the 275 growing season, with the exception of sites across the central Great Plains (Oklahoma and 276 Colorado). These central Great Plains sites exhibited greater inter-annual variability, 277 particularly in the end-of-season detection. 278 279 We further investigated the utility of the Senescence and Dormancy bands from the MCD12-Q2 280 product. However, both bands exhibited limitations for our purposes, particularly in the Great 281 Plains region. The Senescence band, defined when the EVI time series falls below 90% of its 282 amplitude, typically occurred around one month earlier than the actual end of the growing 283 season in the Great Plains (which usually falls around late August). This earlier timing leads to 284 underestimation in plant productivity (results not shown). Conversely, the Dormancy band, 285 defined when the -EVI time series drops below 15% of its amplitude, lags the actual end of the 286 growing season by about a month, causing productivity overestimations (results not shown). 287 288 To explore the potential for shifts in the start and end of the growing season between 2001 and 289 2020, we performed a temporal trend analysis for the GreenUp and MidGreenDown dates over 290 this 20-year period (Figure S2). The start of the growing season shows a negative trend in the 291 northern Great Plains, indicating an earlier onset of the growing season by about one day per 292 year. In addition, a positive trend, indicating a later onset of the growing season, was observed 293 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal in the Mixed Grass Prairie in North Kansas. Regarding the end of the growing season, a delayed 294 senescence was observed in South Dakota and North Dakota. 295 296 4.3 Comparing DayCent-UV simulated ANPP against RAP estimates 297 After optimizing plant growth parameters, and specifying the average start and end of the 298 growing season for each grid cell, each grassland grid cell in the midwestern and western USA 299 was simulated at a daily time scale from 1979 until 2015 using grid-cell specific weather and 300 soil information. The simulated average annual ANPP and RAP remote-sensing estimations 301 using Eqn. (2) to estimate ANPP from total NPP (1986-2015) showed similar pattern across all 302 simulated ecosystems (Figure 3). Starting from the west, ANPP for California grasslands 303 averaged from 150 to 300 g/m2 for both RAP and DayCent-UV. Simulated productivity for 304 grasslands between California and the Great Plains was generally below 100 g/m2 and slightly 305 higher than RAP estimations. Going from west to east across the Great Plains, the both RAP 306 and DayCent-UV ANPP gradually increased from 100 g/m2, in the Short Grass Steppe, to 307 values above 400 g/m2 in the Tall Grass Prairie. 308 309 Figure 3 Average ANPP estimated by RAP (left) and by DayCent-UV model (right) 310 311 Figure 4 Temporal R2 between DayCent-UV ANPP and RAP ANPP (left) when Eqn 2 was used to estimate RAP ANPP. The 312 corresponding relative bias between DayCent-UV ANPP and RAP ANPP (right) 313 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal To evaluate the temporal performance of DayCent-UV model against RAP estimations, we 314 calculated the coefficient of determination (R2) and relative bias (RB) for each simulated grid 315 (Figure 4). The simulated ANPP for California grasslands showed a low RB, but a low temporal 316 correlation (R2 < 0.3). For most of the grasslands in between California and the Great Plains, 317 DayCent-UV’s ANPP was more than 100% of RAP’s, but we also found a high heterogeneity 318 in the temporal correlation. We observed the best model agreement between DayCent-UV and 319 RAP in the Great Plains, where DayCent-UV’s ANPP estimates were only around 30% less 320 than RAP’s for most of the Short Grass Steppe (western Great Plains) and Tall Grass Prairie 321 (eastern Great Plains), while DayCent’s APP estimates were higher than RAPS’ in most of 322 Mixed Grass Prairie. 323 The strength of the correlation between DayCent-UV and RAP was generally lower when RAP 324 ANPP was estimated using the empricial relationship from Hui and Jackson (Eqn. 1) (Figure 325 S3) than when using the one from Gerardhi and Sala (Eqn. 2) (Figure 4). In general, the 326 grassland sites west of the Great Plains showed more regions with lower R2 in Figure S3 327 compared to Figure 4. For the Short Grass Steppe, Figure S3 results are similar to those of 328 Figure 4 but with lower R2 at southern regions. For the Mixed and Tall Grass Prairie R2 values 329 were lower and relative bias was increased in Figure S3 relative to Figure 4. 330 331 Figure 5 Average county-level ANPP for the period 2004 to 2014: NRCS data (left panel), DayCent-UV (middle panel), RAP 332 ANPP (right panel). 333 4.4 Validating county-level simulated ANPP against NRCS and RAP estimates 334 335 Maps of the county-level DayCent-UV simulated mean ANPP, mean RAP ANPP, and observed 336 mean NRCS ANPP (2004 – 2014) for all of the Great Plains (Figure 5) showed a similar pattern 337 with mean ANPP increasing from west to east across the Great Plains. The regression of the 338 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal mean county-level ANPP between NRCS data and DayCent-UV predictions showed a good 339 spatial correlation (R2 = 0.53). Mean DayCent-UV ANPP also had a high spatial correlation with 340 mean RAP ANPP for these counties (R2 = 0.64). DayCent-UV ANPP estimates are most similar 341 to NRCS ANPP for the Mixed Grass Prairie and Tall Grass Prairie in the eastern Great Plains. 342 However, in western counties, the NRCS ANPP shows higher productivity estimates compared 343 to the DayCent-UV model. This bias is confirmed by the correlation analysis in counties with 344 ANPP lower than 200 g/m² (Figure S4), where the slope of the trend line indicates an almost 345 60% greater ANPP by NRCS compared to DayCent-UV model. This may be due in part to the 346 fact that NRCS data was not collected continuously each year (2004-2014), but DayCent-UV and 347 RAP ANPP means include all years during this time period. In fact, DayCent-UV ANPP was 348 more similar to RAP ANPP than it was to NRCS ANPP for this western region of the Great 349 Plains. 350 4.5 Evaluating grid cell ANPP results against site-level data 351 We compared annual ANPP estimates from gridded DayCent-UV and gridded RAP against in 352 situ observations at six sites in the Great Plains (Table 1) (Figure 6). Although DayCent-UV was 353 calibrated for CPER and Konza using local meteorological drivers and soil texture, when 354 DayCent-UV was run for the grid, it used grid cell-level weather drivers and soils data that were 355 not completely consistent with site-level data. Nevertheless, DayCent’s grid-cell level estimates 356 of ANPP for CPER had little bias compared to observations and were strongly temporally 357 correlated to ANPP observations from 1991-2015. For Konza, DayCent-UV results were much 358 higher than observed ANPP but were close to RAP estimates; however, grid cell-level sand 359 content for the grid cell containing Konza was 10% while site-level sand content was 25% and 360 higher sand content generally results in lower observed and simulated plant productivity. For 361 Miles City, MT, DayCent-UV overestimated ANPP and showed the lowest R2 of the six sites, 362 while RAP ANPP followed observations closely. The second lowest agreement between 363 DayCent-UV and observations was for Jornada where both DayCent-UV and RAP 364 underestimated observed ANPP; however the nearest grid cell to Jornada was 150 km away. For 365 Cottonwood, SD, DayCent-UV underestimated observed ANPP. For Hays, KS, both DayCent-366 UV and RAP underestimated observed ANPP. Overall, DayCent-UV’s grid cell-level ANPP 367 estimates seemed to correlate temporally with observed -site-level ANPP, showing similarities in 368 interannual variability despite differences in scale, precipitation inputs, and soil texture, and with 369 no consistent direction in the bias when it existed. DayCent-UV’s gridded ANPP estimates were 370 often closer to RAPs grid cell level estimates than to site-level observations. 371 372 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal 373 Figure 6 Site level comparison between ANPP field observations (blue lines) and the simulated ANPP (solid black lines) and 374 RAP (dashed black lines) at nearest simulated grid. Temporal R2 are between site-level observations and DayCent-UV 375 376 5 Discussion 377 The DayCent-UV model was calibrated to simulate biomass productivity across various 378 grassland ecosystems in the western and midwestern regions of the contiguous United States. We 379 implemented a novel approach to define the growing season period within the model. This 380 approach leverages phenological information derived from the MODIS MCD12-Q2 remote 381 sensing product, specifically the Green-Up and Mid-Green-Down bands. The validity of this 382 approach was supported by the observed -visual agreement between plant photosynthetic activity 383 site records (GPP and GCC) and the growing season dates determined using the Green-Up and 384 Mid-Green-Down bands (Figure 2). These findings align with those reported by Cui et al. 385 (2019), who demonstrated good temporal correlations between GCC and daily MODIS time 386 series data at four grassland sites. 387 Within the model, we defined the grid cell specific end of the growing season (beginning of the 388 month-long senescence) as a fixed day. This day corresponds to the 20-year average of the Mid-389 Green Down date derived from the MODIS MCD12-Q2 product. After the simulation reached 390 this predetermined day, aboveground live shoot biomass stoppped growing, gradually died, and 391 was reduced to zero after one month. We found that the MidGreen down date was more 392 appropriate for defining the onset of senescence for the DayCent-UV model than either the 393 MODIS “Senescence” date or the MODIS “Dormancy” date. Not only did the MidGreenDown 394 dates correspond to the end of the growing season as determined by the other ecological data 395 (GPP, NEE, and phenocam GCC) (Figure 2), but DayCent’s estimates of plant productivity for 396 the Great Plains (results not shown) were underestimated when the model used the “Senescence” 397 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal date and were overestimated when the model used the “Dormancy” date to define the end of the 398 growing season. 399 Recent papers suggest that potential impact of climate change in the Great Plains include 400 increases in precipitation variability, earlier start to the growing season due to higher spring air 401 temperatures, and increases in growing season air temperatures (Petrie et al. 2016, Post et al. 402 2022, Hayek & Knapp 2022, Burke et al. 2024). Chen et al. (2017) showed that variability of 403 grassland plant production and growing season precipitation have increased from 1950 to 2020 404 for grasslands in eastern Colorado. In this paper, we utilized the observed remote sensing 405 changes in plant phenology patterns (2001 to 2020) to inform the DayCent-UV model of the 406 location-specific mean growing season onset and cessation. Our method defined the potential 407 start of the growing season as the GreenUp date minus one standard deviation, allowing air 408 temperature to trigger the actual start. However, the start of senescence was the same from year 409 to year. This method may have been adequate for the timeframe we modeled (1979-2015) but 410 trends in the start and end of the growing season will be increasingly important to include when 411 simulating long-term changes in ANPP with climate change. 412 A trend analysis of the growing season onset (2001-2020) aligns with previous findings (Liu & 413 Zhang, 2020) of earlier green-up in the northwestern Great Plains and a later spring onset north 414 of Kansas (Figure S2). This earlier green-up in the north is likely attributed to a broad spring 415 temperature shift across the region (Liu & Zhang, 2020). However, only a small portion of these 416 sites exhibited a statistically significant trend (p < 0.05) towards an earlier start to the growing 417 season, suggesting the site-specific dependence in phenology shifts (Post et al., 2022). The 418 observed later spring onset north of Kansas could be associated with shifts in spring precipitation 419 patterns (Ren et al., 2018). Furthermore, we observed a significant trend towards a later end of 420 the growing season in the northern Great Plains (Figure S2). This is likely attributable to 421 increased autumn temperature and precipitation (Ren et al., 2018). 422 During the model validation process, we evaluated the utility of the remote-sensing ANPP 423 product offered by RAP. Our analysis revealed that incorporating an annual precipitation 424 dependence into the fraction of carbon allocated to roots (Eqn. 2) improved RAP ANPP 425 estimates compared to using a temperature dependence (Eqn. 1). With this improvement, 426 DayCent-UV modeled productivity exhibited strong temporal correlation and low bias with RAP 427 ANPP data in Great Plains grassland ecosystems, which aligns with the ecosystems used for 428 model calibration (Figure 4). Model results for other grasslands, such as those in the Great Basin, 429 Wyoming Steppe, Palouse Steppe, and Montana Valley, displayed a mixed pattern. In some 430 locations DayCent-UV ANPP good temporal correlation and low bias when compared to RAP 431 ANPP, while others exhibited higher bias. This finding presents a valuable opportunity to focus 432 on long-term estimations in these understudied but important ecosystems. On the other hand, 433 DayCent-UV ANPP for annual grasslands in California exhibited low bias but low temporal 434 correlation with RAP ANPP. This discrepancy could be attributed to the way we accumulated 435 productivity within each calendar year, since California grasslands start their growing season at 436 the end of previous year. The high bias in DayCent-ANPP for cold and warm desert grasslands 437 may be due in part to the limitations of remote sensing in arid areas, including the increasing 438 concentration of invasive species in the Great Basin since 1990 (Smith et al., 2022), high co-439 dominance of shrubs and grasses in the Chihuahuan Desert (Smith et al., 2019), and the strong 440 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal response of green-up and senescence to precipitation patterns in these ecosystems (Currier & 441 Sala, 2022). 442 The gridded model productivity was also compared site-level ANPP observations (Figure 6) and 443 average county ANPP observations across the Great Plains (Figure 5). While most of the site-444 level exhibited a good termporal correlation (R2 > 0.4), a high bias was observed. This bias is 445 likely attributable to a combination of factors, including differences in precipitation between 446 modeling grids and actual measurement sites, variation in soil layer properties across the study 447 area, and the high dependence of plant productivity on topographic gradients (Nippert et al. 448 2011, Hoover et al. 2021). Regarding site-level measurements in desert grasslands, the 449 underestimation of ANPP (Aboveground Net Primary Productivity) by both the model and RAP 450 at the Jornada site in the Chihuahuan Desert (Figure 6) could be linked to the high spatial 451 variability of ANPP observed in desert grasslands, as illustrated in Figure 2 by Muldavin et al. 452 (2008). This evidence highlights the need for long-term simulations in this critical ecosystem to 453 better quantify the accuracy of high-quality remote sensing products like RAP in capturing this 454 high spatial variability (Smith et al., 2019). 455 The simulated DayCent-UV and the RAP ANPP estimates are intended to serve as a valuable 456 tool for validation or calibration of various models that aim to capture accurate productivity 457 dynamics across diverse grassland ecosystems. Examples includes models that uses machine 458 learning for ANPP estimation (Sun et al. 2021 and Wiley et al. 2016), Land Surface Models 459 performance against ANPP estimations derived from vegetation optical depth (VOD) (Fawcett et 460 al. 2022), data assimilitation approaches in the Community Land Model (Fox et al. 2018), and 461 process-based model driven by remote sensing observations like the Rangeland Carbon Tracking 462 and Monitoring (Xia et al. 2024). 463 6 Conclusions 464 This study presented a long-term simulation of grassland productivity across the western and 465 midwestern United States using the DayCent-UV model. The challenge to define the site 466 phenology across such a broad region was addressed by leveraging the MODIS MCD12-Q2 467 product. The model exhibited strong temporal performance in most ecosystems compared to a 468 newly proposed modification of the RAP ANPP estimate. This was particularly evident across 469 the Great Plains and in specific regions like the Great Basin, Palouse Steppe, Montana Valley, 470 and Wyoming Steppe. The simulated grid-level productivity was further evaluated against 471 available site-level and county-level ANPP observations. This comparison revealed good 472 temporal and spatial correlation, although some bias was observed at the site-level likely due to 473 specific landscape features. The public availability of both the simulated dataset and the 474 improved RAP ANPP estimate presents a valuable resource for supporting rangeland 475 management practices and validating, calibrating, and performing data assimilation within 476 various ecosystem models that simulate these important ecosystems. These model and validation 477 improvements will be crucial for future assessments of climate change impacts on grassland 478 ecosystems. 479 480 481 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Acknowledgments 482 This study is supported by the U.S. Department of Agriculture (USDA) UV-B Monitoring and 483 Research Program, Colorado State University, under USDA National Institute of Food and 484 Agriculture Grant 2022-34263-38472. This research is also partially supported by funds from 485 USDA Grass-Cast Award 58-3012-0-021. 486 487 Open Research 488 The grasslands grid cell distribution, the daily and annual DayCent-UV simulated ANPP, along 489 with the ANPP estimates from the RAP product, are publicly available on Zenodo 490 (https://zenodo.org/records/11165665) 491 492 493 494 495 496 497 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal

498 499 Abatzoglou, J. T. (2013). Development of gridded surface meteorological data for ecological applications and modelling. 500 International Journal of Climatology, 33(1), 121-131. 501 502 Bardgett, R. D., Bullock, J. M., Lavorel, S., Manning, P., Schaffner, U., Ostle, N., ... & Shi, H. (2021). Combatting global 503 grassland degradation. Nature Reviews Earth & Environment, 2(10), 720-735. 504 505 Blair, J. and J. Nippert. (2024). PAB01 Aboveground net primary productivity of tallgrass prairie based on accumulated plant 506 biomass on core LTER watersheds (001d, 004b, 020b) ver 17. Environmental Data Initiative. 507 508 Browning, D. M., Snyder, K. A., & Herrick, J. E. (2019). Plant phenology: taking the pulse of rangelands. Rangelands, 41(3), 509 129-134. 510 511 Burke, I.C., Yonker, C.M., Parton, W.J., Cole, C.V., Flach, K. and Schimel, D.S. (1989), Texture, Climate, and Cultivation 512 Effects on Soil Organic Matter Content in U.S. Grassland Soils. Soil Science Society of America Journal, 53: 800-805. 513 514 Burke, M. W., Rundquist, B. C., & Caparó Bellido, A. (2024). Modelling vegetation phenology at six field stations within the US 515 Great Plains: constructing a 38-year timeseries of GCC, VCI, NDVI, and EVI2 using PhenoCam imagery and DAYMET 516 meteorological records. Theoretical and Applied Climatology, 1-17. 517 518 Chen, M., Parton, W. J., Adair, E. C., Asao, S., Hartman, M. D., & Gao, W. (2016). Simulation of the effects of photodecay on 519 long‐term litter decay using DayCent. Ecosphere, 7(12), e01631. 520 521 Chen, M., W.J. Parton, S.J. Del Grosso, M.D. Hartman, K.A. Day, C.J. Tucker, J.D. Derner, A.K. Knapp, W.K. Smith, D.S. 522 Ojima, and W. Gao. (2017). The Signature of Sea Surface Temperature Anomalies on the Dynamics of Semi-arid Grassland 523 Productivity. Ecosphere 8(12):e02069. 10.1002/ecs2.2069. 524 525 Chen, M., Parton, W. J., Hartman, M. D., Del Grosso, S. J., Smith, W. K., Knapp, A. K., ... & Gao, W. (2019). Assessing 526 precipitation, evapotranspiration, and NDVI as controls of US Great Plains plant production. Ecosphere, 10(10), e02889. 527 528 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Cui, T., Martz, L., Lamb, E. G., Zhao, L., & Guo, X. (2019). Comparison of grassland phenology derived from MODIS satellite 529 and PhenoCam near-surface remote sensing in North America. Canadian Journal of Remote Sensing, 45(5), 707-722. 530 531 Currier, C. M., & Sala, O. E. (2022). Precipitation versus temperature as phenology controls in drylands. Ecology, 103(11), 532 e3793. 533 534 Dai, Y., Zeng, X., Dickinson, R. E., Baker, I., Bonan, G. B., Bosilovich, M. G., ... & Yang, Z. L. (2003). The common land 535 model. Bulletin of the American Meteorological Society, 84(8), 1013-1024. 536 537 Del Grosso, S. J., Parton, W. J., Keough, C. A., & Reyes‐Fox, M. (2011). Special features of the DayCent modeling package and 538 additional procedures for parameterization, calibration, validation, and applications. Methods of introducing system models into 539 agricultural research, 2, 155-176. 540 541 Dorich, Christopher D.; Derner, Justin; Torell, Greg; Volesky, Jerry; Brennan, Jameson; Archer, David; Blair, John; Knapp, 542 Alan; Nippert, Jesse; Hartnett, David; McClaran, Mitchel; Maurer, Greg; Moore, Douglas; Clark, Pat; Parton, William; Peck, 543 Dannele; Kramer, Lauren; Smith, William Kolby; Elias, Emile; Fuchs, Brian; Schacht, Walter H.; Hendrickson, John; Harmoney, 544 Keith; Collins, Scott; Baur, Lauren; Porensky, Lauren; Vermeire, Lance; Wilcox, Kevin (2021). Grass-Cast Database - Data on 545 aboveground net primary productivity (ANPP), climate data, NDVI, and cattle weight gain for Western U.S. rangelands. Ag Data 546 Commons. https://doi.org/10.15482/USDA.ADC/1521120 547 548 Fawcett, D., Cunliffe, A. M., Sitch, S., O’sullivan, M., Anderson, K., Brazier, R. E., ... & Zaehle, S. (2022). Assessing model 549 predictions of carbon dynamics in global drylands. Frontiers in Environmental Science, 10, 790200. 550 551 Fox, A. M., Hoar, T. J., Anderson, J. L., Arellano, A. F., Smith, W. K., Litvak, M. E., ... & Moore, D. J. (2018). Evaluation of a 552 data assimilation system for land surface models using CLM4. 5. Journal of Advances in Modeling Earth Systems, 10(10), 2471-553 2494. 554 555 Friedl, M., Gray, J., Sulla-Menashe, D. (2022). MODIS/Terra+Aqua Land Cover Dynamics Yearly L3 Global 500m SIN Grid 556 V061 [Data set]. NASA EOSDIS Land Processes Distributed Active Archive Center. 557 558 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Friedl, M., Sulla-Menashe, D. (2022). MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 500m SIN Grid V061 [Data set]. 559 NASA EOSDIS Land Processes Distributed Active Archive Center. 560 561 Gherardi, L. A., & Sala, O. E. (2020). Global patterns and climatic controls of belowground net carbon fixation. Proceedings of 562 the National Academy of Sciences, 117(33), 20038-20043. 563 564 Hajek, O. L., & Knapp, A. K. (2022). Shifting seasonal patterns of water availability: ecosystem responses to an unappreciated 565 dimension of climate change. New Phytologist, 233(1), 119-125. 566 567 He, Y., Jaiswal, D., Liang, X. Z., Sun, C., & Long, S. P. (2022). Perennial biomass crops on marginal land improve both regional 568 climate and agricultural productivity. GCB Bioenergy, 14(5), 558-571. 569 570 Hermance, J. F., Augustine, D. J., & Derner, J. D. (2015). Quantifying characteristic growth dynamics in a semi-arid grassland 571 ecosystem by predicting short-term NDVI phenology from daily rainfall: a simple four parameter coupled-reservoir model. 572 International Journal of Remote Sensing, 36(22), 5637-5663. 573 574 Hoover, D. L., Lauenroth, W. K., Milchunas, D. G., Porensky, L. M., Augustine, D. J., & Derner, J. D. (2021). Sensitivity of 575 productivity to precipitation amount and pattern varies by topographic position in a semiarid grassland. Ecosphere, 12(2), 576 e03376. 577 578 Hui, D., & Jackson, R. B. (2006). Geographical and interannual variability in biomass partitioning in grassland ecosystems: a 579 synthesis of field data. New phytologist, 169(1), 85-93. 580 581 Irisarri, J.G.N., Derner, J.D., Porensky, L.M., Augustine, D.J., Reeves, J. and Mueller, K.E. (2016), Grazing intensity 582 differentially regulates ANPP response to precipitation in North American semiarid grasslands. Ecol Appl, 26: 1370-1380. 583 https://doi.org/10.1890/15-1332 584 585 Jones, M. O., Robinson, N. P., Naugle, D. E., Maestas, J. D., Reeves, M. C., Lankston, R. W., & Allred, B. W. (2021). Annual 586 and 16-day rangeland production estimates for the western United States. Rangeland Ecology & Management, 77, 112-117. 587 588 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Knapp, A. K., Beier, C., Briske, D. D., Classen, A. T., Luo, Y., Reichstein, M., ... & Weng, E. (2008). Consequences of more 589 extreme precipitation regimes for terrestrial ecosystems. Bioscience, 58(9), 811-821. 590 591 Lauenroth, W.K. and Sala, O.E. (1992), Long-Term Forage Production of North American Shortgrass Steppe. Ecological 592 Applications, 2: 397-403. 593 594 Li, X., Melaas, E., Carrillo, C. M., Ault, T., Richardson, A. D., Lawrence, P., ... & Young, A. M. (2022). A comparison of land 595 surface phenology in the Northern Hemisphere derived from satellite remote sensing and the Community Land Model. Journal of 596 Hydrometeorology, 23(6), 859-873. 597 598 Liang, X. Z., Choi, H. I., Kunkel, K. E., Dai, Y., Joseph, E., Wang, J. X., & Kumar, P. (2005). Surface boundary conditions for 599 mesoscale regional climate models. Earth interactions, 9(18), 1-28. 600 601 Liu, Lingling, and Xiaoyang Zhang. "Effects of temperature variability and extremes on spring phenology across the contiguous 602 United States from 1982 to 2016." Scientific Reports 10.1 (2020): 17952. 603 604 McCulley, R. L., Burke, I. C., & Lauenroth, W. K. (2009). Conservation of nitrogen increases with precipitation across a major 605 grassland gradient in the Central Great Plains of North America. Oecologia, 159, 571-581. 606 607 Milchunas, D. G., & Lauenroth, W. K. (2001). Belowground primary production by carbon isotope decay and long-term root 608 biomass dynamics. Ecosystems, 4, 139-150. 609 610 Morgan, J. A., Parton, W., Derner, J. D., Gilmanov, T. G., & Smith, D. P. (2016). Importance of early season conditions and 611 grazing on carbon dioxide fluxes in Colorado shortgrass steppe. Rangeland ecology & management, 69(5), 342-350. 612 613 Muldavin, E. H., Moore, D. I., Collins, S. L., Wetherill, K. R., & Lightfoot, D. C. (2008). Aboveground net primary production 614 dynamics in a northern Chihuahuan Desert ecosystem. Oecologia, 155, 123-132. 615 616 Nippert, J. B., Ocheltree, T. W., Skibbe, A. M., Kangas, L. C., Ham, J. M., Shonkwiler Arnold, K. B., & Brunsell, N. A. (2011). 617 Linking plant growth responses across topographic gradients in tallgrass prairie. Oecologia, 166, 1131-1142. 618 619 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal NCEP, N. (2005). National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR). 620 Boulder, CO: Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems 621 Laboratory 622 623 Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Powell, G. V. N., Underwood, E. C., D'Amico, J. A., Itoua, 624 I., Strand, H. E., Morrison, J. C., Loucks, C. J., Allnutt, T. F., Ricketts, T. H., Kura, Y., Lamoreux, J. F., Wettengel, W. W., 625 Hedao, P., Kassem, K. R. (2001). Terrestrial ecoregions of the world: a new map of life on Earth. Bioscience 51(11):933-938. 626 627 Parton, W. J., Schimel, D. S., Cole, C. V., & Ojima, D. S. (1987). Analysis of factors controlling soil organic matter levels in 628 Great Plains grasslands. Soil Science Society of America Journal, 51(5), 1173-1179. 629 630 Parton, W. J., Hartman, M., Ojima, D., & Schimel, D. (1998). DAYCENT and its land surface submodel: description and testing. 631 Global and planetary Change, 19(1-4), 35-48. 632 633 Paruelo, J.M., Epstein, H.E., Lauenroth, W.K. and Burke, I.C. (1997), ANPP ESTIMATES FROM NDVI FOR THE CENTRAL 634 GRASSLAND REGION OF THE UNITED STATES. Ecology, 78: 953-958. 635 636 Petrie, M. D., N. A. Brunsell, R. Vargas, S. L. Collins, L. B. Flanagan, N. P. Hanan, M. E. Litvak, and A. E. Suyker (2016), The 637 sensitivity of carbon exchanges in Great Plains grasslands to precipitation variability, J. Geophys. Res. Biogeosci., 121, 280–294, 638 doi:10.1002/2015JG003205. 639 640 Post, A. K., Hufkens, K., & Richardson, A. D. (2022). Predicting spring green-up across diverse North American grasslands. 641 Agricultural and Forest Meteorology, 327, 109204. 642 643 Ren, S., Chen, X., Lang, W., & Schwartz, M. D. (2018). Climatic controls of the spatial patterns of vegetation phenology in 644 midlatitude grasslands of the Northern Hemisphere. Journal of Geophysical Research: Biogeosciences, 123(8), 2323-2336. 645 646 Reeves, M. C., Hanberry, B. B., Wilmer, H., Kaplan, N. E., & Lauenroth, W. K. (2021). An assessment of production trends on 647 the Great Plains from 1984 to 2017. Rangeland Ecology & Management, 78, 165-179. 648 649 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Ryals, R., M.D. Hartman, W.J. Parton, M.S. DeLonge, and Silver, and W.L. Silver. (2015). Long-term climate change mitigation 650 potential with organic matter management on grasslands. Ecological Applications, 25(2), pp. 531–545. 651 (https://doi.org/10.1890/13-2126.1) 652 653 Sala, O. E., Parton, W. J., Joyce, L. A., & Lauenroth, W. K. (1988). Primary production of the central grassland region of the 654 United States. Ecology, 69(1), 40-45. 655 656 Schwarz, G. E., & Alexander, R. B. (1995). State soil geographic (STATSGO) data base for the conterminous United States (No. 657 95-449). 658 659 Seyednasrollah, B., Young, A. M., Hufkens, K., Milliman, T., Friedl, M. A., Frolking, S., & Richardson, A. D. (2019). Tracking 660 vegetation phenology across diverse biomes using Version 2.0 of the PhenoCam Dataset. Scientific data, 6(1), 222. 661 662 Smith, J. T., Allred, B. W., Boyd, C. S., Davies, K. W., Jones, M. O., Kleinhesselink, A. R., ... & Naugle, D. E. (2022). The 663 elevational ascent and spread of exotic annual grass dominance in the Great Basin, USA. Diversity and Distributions, 28(1), 83-664 96. 665 666 Smith, W. K., Dannenberg, M. P., Yan, D., Herrmann, S., Barnes, M. L., Barron-Gafford, G. A., ... & Yang, J. (2019). Remote 667 sensing of dryland ecosystem structure and function: Progress, challenges, and opportunities. Remote sensing of environment, 668 233, 111401. 669 670 Sun, Y., Feng, Y., Wang, Y., Zhao, X., Yang, Y., Tang, Z., ... & Fang, J. (2021). Field‐based estimation of net primary 671 productivity and its above‐and belowground partitioning in global grasslands. Journal of Geophysical Research: Biogeosciences, 672 126(11), e2021JG006472. 673 674 White, Robin, Siobhan Murray, and Mark Rohweder, (2000). Pilot Analysis of Global Ecosystems: Grassland Ecosystems, 675 World Resources Institute, Washington D.C. November 2000. ISBN 1-56973-461-5. 676 677 Wylie, B., Howard, D., Dahal, D., Gilmanov, T., Ji, L., Zhang, L., & Smith, K. (2016). Grassland and cropland net ecosystem 678 production of the US Great Plains: Regression tree model development and comparative analysis. Remote Sensing, 8(11), 944. 679 680 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint manuscript submitted to replace this text with name of AGU journal Xia, Y., Sanderman, J., Watts, J., … & Billesbach, D. (2024). Coupling Remote Sensing with a Process Model for the Simulation 681 of Rangeland Carbon Dynamics. ESS Open Archive. (DOI: 10.22541/essoar.171269303.34921116/v1) 682 683 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 2, 2024. ; https://doi.org/10.1101/2024.06.28.601224doi: bioRxiv preprint