MOD-LSP, MODIS-based parameters for hydrologic modeling of North American land cover change

Earth systems models require gridded land surface properties to compute fluxes of water, energy, and carbon within the landscape and to the atmosphere. However, most parameter sets contain time-invariant properties despite their known variability. Here we present new MODerate Resolution Imaging Spectroradiometer (MODIS)-based land surface parameters (MOD-LSP) formatted for the Variable Infiltration Capacity (VIC) hydrologic model that account for seasonal and interannual variability and longer-term change over the continental United States, Mexico, and southern Canada at 0.0625° spatial resolution and monthly temporal resolution. MOD-LSP improves over previously-available parameter sets via: (1) land cover maps of higher native spatial resolution; (2) multiple versions corresponding to the land cover of years 1992, 2001, and 2011; (3) spatially-explicit mean annual cycles of land surface properties, including leaf area index, canopy fraction, and albedo, derived from 17 years of observations; and (4) additional 17-year time series of these properties. The MOD-LSP parameters are useful as inputs to the VIC model, as an example land surface scheme, to assess the hydrologic impacts of land cover change from interannual to decadal scales; and as stand-alone datasets characterizing the temporal variability of these properties as a function of land cover class.


Background & Summary
Hydrologic and earth system models require specification of gridded land surface properties such as land cover type, leaf area index (LAI), and albedo to simulate the land surface response to meteorological forcings over watershed to global scales, typically at resolutions of less than 10 km. For those properties that can be measured from space, remote sensing products have been useful in providing gridded values at high resolution with global coverage [1][2][3][4][5][6][7] . However, most land surface parameters represent a "snapshot" from a specific time period that remains static in multi-decadal simulations, despite their known temporal variability 8 . Furthermore, for studies at regional to global scales, the effort required to process the large data volume of remote sensing datasets can discourage the updating of these parameters, in some cases leading to the use of land surface parameters that represent time periods outside of the period of simulation.
One such land surface model is the Variable Infiltration Capacity (VIC) model 9,10 . Prior VIC parameters over the United States [11][12][13] , Mexico 14 , and North America 15 share several limitations: (1) they were derived from a coarse-resolution land cover map, based on Advanced Very High-Resolution Radiometer (AVHRR) imagery from the early 1990s 7 ; (2) the annual cycle of monthly LAI values was derived by spatially interpolating a sparse (<100 points over North America) subset of observations from a single year of an AVHRR-based dataset 16 ; (3) in all land cover classes (except bare soil), the vegetation canopy was assumed to have 100% areal coverage; (4) each land cover class was assigned a single spatially (and temporally, for most classes) invariant albedo obtained from literature values; and (5) the urban land cover class was replaced by bare soil.
In addition, the prior efforts have not accounted for land cover variability and change, including: (1) land cover conversions in response to natural or human disturbance; and (2) interannual variations in phenology (seasonally-varying plant characteristics such as LAI, canopy fraction, and albedo) in response to climate fluctuations. Several efforts in the remote sensing community have provided sequential land cover classifications over multiple decades using consistent methodology. For instance, the National Land Cover Database (NLCD) 2 contains consistent classifications over the United States for the years 1992, 2001, 2006, 2011, and 2016; while the Instituto Nacional de Estadística y Geografía (INEGI) Uso del Suelo y Vegetación land cover product 17 contains consistent classifications over Mexico for years 1985, 1993, 2002, 2007, and 2011. Similarly, interannual variability in phenology from the MODerate Resolution Imaging Spectroradiometer (MODIS) has received attention in regional hydrologic modeling studies [18][19][20][21][22][23][24][25][26] .
Here, we describe new MODIS-based land surface parameters for the VIC model (MOD-LSP), covering the continental United States, Mexico, and southern Canada at 0.0625° (6 km) resolution. MOD-LSP is formatted to be compatible with the image driver of VIC model release 5.0 and later 10 . Its spatial domain and grid resolution are compatible with gridded daily meteorological forcings and hourly disaggregation parameters over North America 15,27,28 . MOD-LSP overcomes the limitations of prior parameter sets and facilitate studies of land cover variability and change. To do so, several versions of the MOD-LSP parameters based on NLCD and INEGI classifications from years 1992/3, 2001/2, and 2011 2,17 are available, as well as a land cover classification based on MODIS 1 . In terms of phenology (LAI, canopy fraction, and albedo), MOD-LSP contains the following improvements: (1) canopy fraction (f canopy ) and albedo, which were not derived from remote sensing in prior studies, are

Methods
Several spatial datasets were used in creating and evaluating the parameters at 0.0625° (6 km) resolution. Details of satellite sensors, spatial resolution, and acquisition dates for these datasets can be found in Table 1. Processing of the land surface properties centered around spatial aggregation, i.e., grouping the 500 m MODIS observations by the classes given by the 30 or 500 m land cover pixels within each 6 km grid cell, yielding a separate spatial average for each class within the cell (Fig. 1). Before the spatial aggregation step, the NLCD and INEGI land cover classifications were harmonized to a common legend. After aggregation, data gaps were filled via temporal and spatial interpolation, separately for each land cover class. Details of these procedures are given below.   www.nature.com/scientificdata www.nature.com/scientificdata/ Spatial domain. The spatial domain spans the region bounded by 14.5° and 53°N latitude and 125° and 67°W longitude (Fig. 2). This domain is an extension of the CONUS domain, over which several gridded meteorology and hydrology datasets [11][12][13]29 have been constructed, to include Mexico; hence we will refer to the domain as CONUS_MX hereafter. In those cases for which our analysis omits southern Canada, we will refer to that subset of the domain as USMX.
VIC parameters. The VIC parameter datasets ( Table 2) include "L2015", the parameter set used in Livneh et al. 15 , covering the CONUS_MX domain; and "MOD-LSP", the suite of new parameter sets with phenology based on MODIS observations, covering either CONUS_MX or USMX domains, depending on the underlying land cover classifications.
Land cover classifications. The land cover classification underlying the L2015 dataset was the AVHRR-based University of Maryland (UMD) land cover product 7 , modified for the North American Land Data Assimilation project (NLDAS) 12 to exclude the open water, urban, and perennial ice/snow classes (UMD-NLDAS hereafter). • domain = domain name ("CONUS_MX" or "USMX") • lc_source = land cover source name ("L2015", "MOD_IGBP", or "NLCD_ INEGI") • lc_year = code describing the year(s) of acquisition from which pixels are assigned classes; can be one of ("mode", year, or "s"year) where: ○ mode = class is the most common class from all maps over the period 2000-2013 ○ year = class comes from the land cover map for year year ○ "s"year = same as year, but the "s" signifies that the phenology data were only aggregated over pixels for which land cover class was stable (did not change) between 2001 and 2011. Pixels for which land cover class changed were assigned phenology values via spatial interpolation from stable neighbors of the class to which the pixel belonged in the given lc_year.
• phen_years = years of MODIS products from which phenology values are derived, in the format startyear_endyear. • year = year of phenology to which this file pertains Note: these files must be converted from monthly values to sub-daily values (same time step as the meteorological forcings) via the following scripts, available at https://github.com/tbohn/VIC_Landcover_MODIS_NLCD_INEGI/releases/ tag/v1.6: • disagg_veghist_monthly2hourly_nc.py • wrap_disagg_veghist_monthly2hourly_nc.pl www.nature.com/scientificdata www.nature.com/scientificdata/ MOD-LSP parameter sets are named based on the underlying land cover classification, according to the convention (Table 3): lc_source.lc_year, where lc_source indicates the land cover source and lc_year indicates the year or set of years to which the classification pertains. Values of lc_source include: MOD_IGBP, based on the IGBP classification of the MOD12Q1.005 product 1 over the CONUS_MX domain; and NLCD_INEGI, which covers the USMX domain and is based on the NLCD classification 2 over the United States and the INEGI classification 17 over Mexico. To combine the NLCD and INEGI products into a single product, the more numerous INEGI land cover classes were mapped to the 16 NLCD 2011 classes (e.g., all shrub classes were mapped to shrublands, etc.). Details of the procedure can be found in the NLCD_INEGI GitHub archive (https://github.com/tbohn/ NLCD_INEGI/releases/tag/v1.6) 30 .
Because the MOD12Q1.005 product provided 13 separate annual maps over 2001-2013, during which many pixels changed class multiple times, a single map representative of the entire period was created for the MOD_ IGBP parameter set by assigning to each pixel the class it most frequently had over the period. This has been denoted by setting lc_year to "mode". The NLCD_INEGI datasets used the NLCD maps from 1992, 2001, and 2011 2,31,32 over the US and the INEGI maps from 1993, 2002, and 2011 17 over Mexico. It should be noted that although both the NLCD and INEGI products were derived from Landsat imagery, they were created with different methods: NLCD was classified automatically on a pixel-by-pixel basis, while INEGI products were delineated Time-varying surface properties. VIC requires specification of a repeating annual cycle of 12 monthly values of time-varying surface properties for each land cover class of each grid cell: LAI, albedo, roughness length, and displacement height. An additional annual cycle of canopy fraction (f canopy ) is optional 24 , and was included in the MOD-LSP parameter sets. For roughness length and displacement height, all VIC parameter datasets used a spatially-uniform repeating annual cycle of monthly values for each land cover class, taken from literature 33 .
The L2015 parameter set contained a unique annual cycle of monthly LAI values for each land cover class in each 6 km grid cell, derived from an AVHRR-based product 34 . This annual cycle was derived from only a single year (1992-1993), and from a sparse (<50) set of pixels in each land cover class, which were extended over the CONUS_MX domain by spatial interpolation. For albedo, L2015 used a constant, spatially-uniform value for each land cover class, taken from literature 33   www.nature.com/scientificdata www.nature.com/scientificdata/ Difference Vegetation Index (NDVI) of the 16-day MOD13A1.006 product 4 for 2000-2016. For consistency with LAI, only those albedo observations corresponding to the 8-day LAI schedule were used. For those 8-day intervals when the 16-day NDVI data were unavailable, NDVI values were treated as missing data.
Phenology variables were aggregated from 500 m to the 6 km grid by computing a separate spatial average value for each land cover class in each 6 km grid cell: where x is a phenology variable, k l is an index of the set of N l (c) land cover pixels of class c within the 6 km cell, k p (k l ) is the index within the cell of the MODIS pixel containing land cover pixel k l , t is the time index (at 8-day intervals), and i and j are the row and column indices of the grid cell. For the MOD_IGBP dataset, land cover pixels were defined on the same 500 m sinusoidal grid as phenology pixels, so that each phenology pixel corresponded to exactly one land cover pixel. For the NLCD_INEGI datasets, the smaller size of land cover pixels (30 m) led to multiple land cover pixels of potentially many different classes corresponding to a single phenology pixel and taking on the same value of x. This may cause some contamination of phenology values by other classes in regions where classes occur in patches smaller than the 500 m MODIS resolution.
Pixels that were contaminated by poor retrievals, clouds, or snow, as indicated by the FparLAI_QC and FparExtra_QC variables included in the MCD15A2H.006 product, were excluded from the spatial average. In some cases, this led to entire grid cells missing data for one or more dates, particularly for cells north of 50°N in winter ( Supplementary Fig. 1). These missing values were filled by interpolation. Even after filtering out pixels that were flagged for snow, observations immediately before and after the flagged observations often exhibited albedo values much larger than the typical range of snow-free variability, indicating at least partial snow coverage. Therefore, after the aggregation step, we removed all data for which the albedo exceeded four standard deviations www.nature.com/scientificdata www.nature.com/scientificdata/ above the long-term mean (this was the smallest threshold that avoided throwing away substantial amounts of valid snow-free data).
The MCD15A2H.006 product did not provide estimates of LAI over pixels classified in the MOD12Q1.005 product as open water, barren, perennial snow/ice, urban, or unclassified. With the exception of urban pixels, LAI was set to 0 in these cases. For urban pixels, LAI was estimated via an empirical relation estimated by sampling pixels of shrubland, grassland, and forest classes surrounding Phoenix, Los Angeles, San Francisco, Portland, and Seattle: LAI = 8(NDVI − NDVI min ) 2 , where NDVI min = 0.1.
f canopy was derived from NDVI as: where NDVI max = 0.8. Gaps in the 8-day data were filled as illustrated in Supplementary Fig. 1: At each grid cell, standardized anomalies were computed, following: x x where x′ = standardized anomaly, c = land cover class index, y = year, d = day of year, x = phenological variable, and μ x and σ x = climatological mean and standard deviation of x for class c and day of year d. All values of μ x and σ x for which fewer than 5 observations were available on a given day of the year were set to "missing". In the anomaly time series, missing values either on the first or last time steps or for which the nearest valid data points were more than 2 intervals away were set to 0, and remaining gaps in the anomaly time series were linearly www.nature.com/scientificdata www.nature.com/scientificdata/ interpolated. Temporal gaps in μ x and σ x were filled by linear interpolation, treating the climatological cycle as periodic. Subsequently, anomalies were recombined with μ x and σ x to assemble the final gap-filled time series. Any remaining gaps (cells for which no observations were available at any time) were filled with spatial interpolation from values from the same land cover class in neighboring cells, using a Gaussian kernel with σ = 1 cell. Thus, the final gap-filled data are estimates of snow-and cloud-free values. In the case of albedo, a land surface model such as VIC will replace the input albedo with a simulated snow albedo when snow is present.
The requisite repeating annual cycle of monthly phenology was computed from different sets of years for different MOD-LSP parameter sets. For the MOD_IGBP.mode parameter set, the gap-filled 8-day timeseries of MODIS phenology spanning 2000-2016 were aggregated to a single year of climatological mean monthly values. For the NLCD_INEGI parameter sets, the annual cycle was derived in two alternate ways: (1) taken from the single year corresponding to the land cover map, using data from all pixels; or (2) computed as the climatological mean annual cycle over 2000-2016, using only those 30-m land cover pixels that had a stable land cover class between the 2001 and 2011 maps, thereby providing a climatological cycle free of the impacts of land cover change (in which case lc_year in the filename was prepended with "s"). In all cases, the string phen_years in the filename indicates the years from which the annual cycle was derived.  www.nature.com/scientificdata www.nature.com/scientificdata/ Non-MODIS surface properties. All parameter sets obtained grid cell elevations from the USGS GTOPO30 digital elevation model 6 . For the MOD-LSP parameter sets, GTOPO30 was aggregated to 0.0625° resolution, but L2015 was sub-sampled from earlier 0.125° (12 km) spatial resolution datasets; thus their land masks differ along the coastlines. All parameter datasets used the soil properties of L2015, which were obtained from the FAO-UNESCO Digital Soil Map of the World 35 . Similarly, the MOD-LSP datasets used the conceptual soil parameter values (e.g., b infilt , Dsmax, and soil layer thicknesses) of L2015, which were derived via calibration in previous studies 11,14 .
Time-and space-invariant surface properties for each land cover class (e.g., presence of overstory, stomatal and canopy architectural resistance parameters) were taken from the L2015 dataset, which in turn obtained these values from literature 33 . Because the new parameter datasets used different land cover classification schemes than the L2015 dataset, these time-and space-invariant parameters were mapped to classes of MOD_IGBP and NLCD_INEGI from the most appropriate class of the UMD-NLDAS classification as shown in Supplementary  Tables 1 and 2, respectively. For the unvegetated open water, perennial ice and snow, and barren soil classes, which were not present in UMD-NLDAS, grassland was chosen as the source class. The impact of this assignment was minimal because the values of LAI, f canopy , and albedo observed by MODIS for these classes corresponded to those of extremely sparse vegetation, essentially rendering the other vegetation parameters (e.g., stomatal and canopy structural resistances) inactive in the VIC model.

Data Records
All parameter sets discussed herein are available for download at Zenodo 36 . The parameter sets are stored in NetCDF files, formatted for input to the VIC model, version 5.0 and later 10 , in image mode. There are three types of files included in the MOD-LSP project ( Table 1): (1) VIC parameter files, which contain grids of land cover class area fractions ("Cv"), spatially-varying annual cycles of monthly phenology variables ("LAI", "fcanopy", "albedo") for all classes, soil properties, and vegetation structural properties that are invariant in time and space; (2) "veg_hist" compressed tar files, each containing 17 yearly files of monthly time-varying phenology variables; and (3) a template ("global_param.template") for the global parameter file, which specifies the names, locations, and contents of all input and output files, and simulation parameters such as model time step, start and end dates, and model physics options (Supplementary Table 3). Detailed information on parameter file variables and formatting can be found in the VIC model code archive on GitHub (https://github.com/UW-Hydro/VIC).

technical Validation
Evaluation of MOD-LSP phenology. Because MOD-LSP LAI and albedo were only spatially aggregated from their source datasets, which have been extensively validated elsewhere 3,5,16,34,37,38 , a comprehensive validation of those values against observations has not been performed. However, we have evaluated natural f canopy over much of the United States; and urban LAI and f canopy over Santa Barbara, CA, USA.
To evaluate the values of f canopy , two independent datasets were used: the NLCD US Forest Service Percent Tree Canopy Cartographic product 36 (NLCD-Forest hereafter) and the NLCD Shrubland product 37 (NLCD-Shrub hereafter). These datasets were aggregated from 30-m to 0.0625° resolution over the NLCD_INEGI.2011 land cover classification in a similar fashion to the MODIS observations. The MOD-LSP estimates of f canopy substantially outperformed L2015 in reproducing the statistics of the NLCD-Shrub and NLCD-Forest products (Table 4) (Table 4, Fig. 3 and Supplementary Fig. 2), f canopy exhibited a strong seasonal cycle, with a minimum in winter and a maximum in spring or summer. Over shrublands and grasslands, f canopy both underestimated and showed less spatial variation than the NLCD-Shrub values in all seasons, leading to worse underestimation where NLCD-Shrub coverage values were higher, e.g., at higher elevations (Fig. 3). Over forests, f canopy underestimated NLCD-Forest values in winter, overestimated them in summer, and was relatively less biased in spring and fall, particularly in the eastern United States (Fig. 3). The seasonality is due to both to real phenology (leaf growth and loss in deciduous vegetation) and to biases inherent in Eq. (2) (variation of the solar angle and canopy shading) 39 . Nevertheless, a prior study 24 found that VIC simulations using f canopy values derived from MODIS NDVI via Eq. (2) resulted in substantial improvements over using f canopy = 1.0 in simulated ET at 60+ shrubland and forest AMERIFLUX 40 eddy covariance tower sites.
A comprehensive evaluation of urban LAI and f canopy from MOD-LSP has not been undertaken. However, over Santa Barbara, CA, July values of LAI and f canopy compare favorably to 71 randomly-sampled field observations from a prior study 41  Because the MOD-LSP LAI timeseries changed source datasets (from MOD15A2H.006, Terra sensor only; to MCD15A2H.006, Terra + Aqua sensors) after the first 1.5 years of record, changes in statistics are a potential issue. Prior studies 37 have found that the two products yielded generally similar values, particularly over forests and grasslands, but exhibited different temporal mean and variance over highly variable biomes such as croplands. However, we expect the impact on the climatological mean values to be small, as the time span derived from the MOD15A2H.006 product comprised less than 10% of the record.
www.nature.com/scientificdata www.nature.com/scientificdata/ Comparison of land cover distributions. The L2015, MOD_IGBP, and NLCD_INEGI.2011 land cover classifications generally agree on the geographic distributions of major land cover categories, although there are notable differences (Fig. 4). For the "forest" category, all three land cover datasets show high coverage in eastern Canada and northeastern US, the southeastern US, southern Mexico, mountains west of 100° W, and the northern Pacific coast (Fig. 4a-c). However, L2015 has higher forest coverage around the fringes of these regions than MOD_IGBP, while NLCD_INEGI has lower coverage. This might be due in part to the spatial resolutions of the underlying land cover products (1 km for UMD, 500 m for MOD12Q1.005, and 30 m for NLCD and INEGI). In particular, NLCD does not include classes for mixes of trees and grass. For shrublands (Fig. 4d-f) and grasslands (Fig. 4g-i), the classifications agree on the overall locations of these classes but differ on the boundary between shrubland and grassland. L2015 is more similar to NLCD_INEGI in its differentiation between shrublands and grasslands, with shrublands extending into the northwestern US, while in MOD12Q1.005 shrublands are confined to the southwestern US and western Mexico. The products agree on agricultural areas (Fig. 4j-l), but MOD_IGBP has higher coverage in most of the US and Canada, while NLCD_INEGI exhibits greater extent in Texas and Mexico.
Comparison of phenology distributions. The improvements of MOD-LSP phenology relative to L2015 are evident when comparing distributions over North America. For the "mixed forest" class ( Supplementary  Fig. 3), the July LAI of L2015 contains regions of homogeneous values with abrupt boundaries, while MOD_IGBP. mode exhibits a physically reasonable distribution. The L2015 July f canopy is 1.0 everywhere, while MOD_IGBP. mode varies from 0.5 to 1.0 and shows a similar spatial distribution to LAI. Similarly, the L2015 July albedo is 0.18 everywhere, while that of MOD_IGBP.mode ranges from 0.10 to 0.18. Similar problems are evident in the L2015 cropland class, relative to that of MOD_IGBP.mode (Supplementary Fig. 4).
MOD-LSP and L2015 parameter sets also differ in total "scene" phenology, or the area-weighted average phenology across all land cover classes (Fig. 5). In January, the "scene" LAI of L2015 (Fig. 5a) is substantially higher than that of MOD_IGBP.mode (Fig. 5g), particularly in the northern forests. This could be due to L2015 holding LAI constant year-round for evergreen forest and possible residual impacts from clouds and snow in the MOD-LSP processing. In July, however, the LAI of L2015 (Fig. 5d) and MOD_IGBP.mode (Fig. 5j) look more similar. f canopy remains 1.0 year-round in vegetated cells in L2015 (Fig. 5b,e) but displays a seasonal cycle in MOD_IGBP.mode (Fig. 5h,k). For albedo, spatial variations are evident in L2015, ranging from 0.12 to 0.20, but there are no temporal variations except over crops (Fig. 5c,f). The MOD_IGBP.mode albedo exhibits much larger spatial variability than L2015, ranging from 0.10 to 0.30 (Fig. 5i,l).

Impacts on hydrology. Differences in land cover distributions and phenology between L2015 and
MOD-LSP impacted water and energy fluxes obtained from the VIC simulations. Holding f canopy constant at 1.0 (Fig. 6), differences in mean annual evapotranspiration (ET) and runoff (Q) between simulations using the MOD_IGBP and L2015 parameter sets were positively correlated in the case of ET and negatively correlated in the case of Q with the difference in mean annual LAI between the two parameter sets. In the warmer or wetter regions of the domain (south and east of the dashed lines), ET and Q exhibit sensitivities to differences in LAI of about +/−200 mm y −1 per unit change in LAI. However, the differences diminish in the cooler or drier portions of the domain (to the north and west of the dashed lines).
Similarly, allowing f canopy to vary resulted in changes of about −/+50 mm y −1 per unit change in f canopy in the warmer and wetter portions of the domain, due to reduced canopy evaporation being outweighed by the increased soil evaporation and transpiration from increased throughfall ( Supplementary Fig. 5). Sensitivities to changes in f canopy diminished as climate became drier, but where annual precipitation (P) fell primarily as snow, the sensitivities of ET and Q to changes in f canopy changed sign, due to reduced snow storage in the forest canopy, where aerodynamic resistance is lower than at the surface of the ground snow pack. The total difference between MOD_IGBP with varying f canopy and L2015 was dominated by the difference due to LAI, but f canopy differences reinforced the LAI impacts in snowy mountains.
The NLCD_INEGI products show substantial land cover change between 2001 and 2011 ( Supplementary  Fig. 6). Forests experienced substantial losses in the southeastern US, the northern Pacific coast, and much of Mexico. Shrublands and grasslands expanded in the southeastern and northwestern US, but shrank in northern Mexico. Agriculture expanded substantially in Mexico and cities expanded everywhere. At smaller scales (boxes 1-4), the NLCD_INEGI products show the loss of forest to shrub/grassland in response to drought and fire in the Rocky Mountains 42 (box 1); a mix of forest-shrub, crop-forest, and crop-urban conversion around Lake Michigan (box 2); forest-shrub, forest-crop, and shrub-crop conversion in northwestern Mexico due to pasture clearing and agricultural expansion 43 (box 3); and forest-shrub and forest-urban conversion around Atlanta (box 4).
These changes in land cover, coupled with interannual variations in phenology (via the NLCD_ INEGI.2001.2001_2001 and NLCD_INEGI.2011.2011_2011 parameter sets), led to changes in hydrologic fluxes (Fig. 7). Forest-shrub conversion in the Rocky Mountains (box 1) and forest-agriculture conversion in northwestern Mexico (box 3) were accompanied by reductions in LAI and f canopy , leading to reduced ET and increased Q. However, crop-urban conversion around Lake Michigan (box 2) and forest-urban conversion around Atlanta (box 4) did not lead to reductions in LAI or f canopy and had minimal impacts on hydrology. Furthermore, interannual fluctuations in LAI and f canopy in eastern Texas (box 5) led to substantial changes in hydrology without being accompanied by substantial land cover change.

Usage Notes
To run VIC with the parameter files described herein requires corresponding domain files, available on Zenodo 44 . Instructions for configuring VIC to use these parameter files are available in the MOD-LSP User Guide included with the parameter files 36 .