The effects of small-scale topography-induced land surface heterogeneity are not well represented in current Earth System Models (ESMs). In this study, a new topography-based subgrid structure referred to as topographic units (TGU) designed to better capture subgrid topographic effects, and methods to downscale atmospheric forcing to the land TGUs have been implemented in the Energy Exascale Earth System Model (E3SM) Land Model (ELM). Effects of the subgrid scheme and downscaling methods on ELM simulated land surface processes are evaluated over the conterminous United States (CONUS). For this purpose, ELM simulations are performed using two configurations without (NoD ELM) and with (D ELM) downscaling, both using TGUs derived for the 0.5-degree grids and the same land surface parameters. Simulations using the two ELM configurations are compared over the CONUS domain, regional levels, and at observational sites (e.g., SNOTEL). The CONUS-level results suggest that D ELM simulates more snowfall and snow water equivalent (SWE), higher runoff, and less ET during spring and summer. Regional-level results suggest more pronounced impacts of downscaling over regions dominated by higher elevation TGUs and regions with maximum precipitation occurring during cool seasons. Results at the SNOTEL sites suggest that D ELM has superior capability of reproducing the observed SWE at 83% of the sites, with more pronounced performance over topographically heterogeneous TGUs with their maximum precipitation occurring during cool seasons. The results highlight the importance of improving representation of small-scale surface heterogeneity in ESMs and motivate future research to understand their effects on land-atmosphere interactions, streamflow, and water resources management over mountainous regions.

Key Points

A topography-based subgrid scheme and downscaling of atmospheric forcing have been implemented in the E3SM Land Model component
These new methods have more pronounced effects in high subgrid elevation areas that receive their major precipitation during cool seasons
The new methods improve model capability to reproduce the observed snow water equivalent at 83% of the SNOTEL sites

Plain Language Summary

Current global Earth System Models (ESMs) used in climate simulation and projection represent the land surface using coarse grids with limited ability to resolve the surface variability within the grid. This study introduced a new subgrid structure and methods to downscale atmospheric variables such as precipitation and temperature from the atmosphere grid to the land subgrid units in the Energy Exascale Earth System Model (E3SM) Land Model (ELM) to improve its ability of representing the effects of topography induced subgrid surface variability on land surface processes. Effects of the new developments on simulations of land surface processes are evaluated using the conterminous United States (CONUS) as a case study. ELM simulations with the new developments generally yield higher snowfall, snow water equivalent and runoff over CONUS. Effects of the new developments are found to be larger over regions dominated by higher elevation landscapes and regions receiving their maximum precipitation during cool seasons. Also, the new developments improved ELM's ability to reproduce observed snow water equivalent at the SNOTEL sites. The new developments have important implications for streamflow modeling and water resources management dependent on hydrologic processes in mountainous regions.

1 Introduction

Land surface heterogeneity, comprising of spatial variations in topography, land use/land cover and soil properties, exerts major control on land surface processes and land-atmosphere interactions through their impacts on the surface energy balance and exchange of water, energy, and other constituents (Giorgi & Avissar, 1997; Wu et al., 2009). As so many characteristics of the Earth's surface are being modified by human activities, land surface heterogeneity plays a pivotal role in determining the dynamics of various processes of the land surface and atmosphere across a wide range of temporal and spatial scales (Bou-Zeid et al., 2020). Despite decades of research efforts to improve its representation in Earth system models (ESMs), it still poses challenges to the ESM community (Giorgi & Avissar, 1997; de Vrese et al., 2016; Bou-Zeid et al., 2020). Motivated by the challenges and the complexity of the processes controlled by surface heterogeneity, many studies have focused on improving representations of the effects of heterogeneity in ESMs (Arthur et al., 2018; Chaney et al., 2016; Gu et al., 2020; Hao et al., 2019, 2021, 2022; Huang, Zhang, et al., 2022, Huang, Ma, et al., 2022; Li et al., 2022; Shi & Xiao, 2022; Tesfa, Li, et al., 2014; Tesfa, Ruby Leung, et al., 2014; Tesfa & Leung, 2017; Tesfa et al., 2020; Wei et al., 2021; Wijngaard et al., 2023, Zorzetto et al., 2023) to enhance their capability to capture the effects of surface heterogeneity on various land surface processes.

Topography, a major driver of land surface heterogeneity, plays a critical role in land surface processes and land-atmosphere interactions through its influence on climate (Arthur et al., 2018; Hao et al., 2021; Tesfa et al., 2020; Zorzetto et al., 2023), the spatial patterns of land cover (Moeslund et al., 2013; Oddershede et al., 2015; Tesfa & Leung, 2017; Wang et al., 2015, 2022) and soil properties (Dessalegn et al., 2014; Kirkpatrick et al., 2014; Tesfa et al., 2009). Because of its influence on atmospheric forcing including surface temperature, precipitation (e.g., orographic precipitation and precipitation partitioning into snowfall and rainfall) and incoming and reflected radiation, topographically heterogeneous regions are characterized by diverse hydroclimatic conditions ((Leung & Ghan, 1995; Tesfa & Leung, 2017) and references there in), which have important implications to land surface processes (e.g., runoff generation). Through its influence on climate variability (precipitation, temperature, solar radiation), topography exerts major control on soil moisture availability, which is the primary driver of spatial and temporal patterns of vegetation growth (Moeslund et al., 2013; Oddershede et al., 2015; Wang et al., 2022). For example, in the Northern Hemisphere, north-facing hillslopes support different vegetation types and density compared to the hillslopes facing south (Moeslund et al., 2013; Hao et al., 2022). Also, as one of the factors of soil formation, topography exerts major control on spatial patterns of soil properties, such as soil depth and soil texture (Fu et al., 2004; Tesfa et al., 2009). Overall, topography-mediated interactions between climate, vegetation, soil, and river flow strongly affect terrestrial ecohydrology and its feedback to the atmosphere (Thompson et al., 2011) and improving representation of the impacts of land surface heterogeneity influenced by topography in ESMs is an ongoing challenge to the macro scale modeling community (Wood et al., 2011). Thus, accurate predictions of Earth system variability and change cannot be achieved without accounting for the impacts of land surface heterogeneity influenced by topography (Hao et al., 2022; Li et al., 2022; Leung & Ghan, 1995; Tesfa & Leung, 2017).

Recognizing the role of land surface processes on climate, food and energy production, water resources and biodiversity and the importance of accurate representation of land surface heterogeneity in macro-scale models, various approaches are used to resolve subgrid variability within the modeling unit (grid cell) to improve representation of the effects of land surface heterogeneity (Chaney et al., 2016; Li et al., 2022) describes four classes of approaches (a) subdividing each modeling unit into multiple tiles to capture subgrid surface properties (Avissar & Pielke, 1989; Swenson et al., 2019; Tesfa & Leung, 2017), (b) representing subgrid heterogeneity within each modeling unit using probability density functions (He et al., 2021), (c) resolving subgrid heterogeneity using parameterizations of subgrid processes (Hao et al., 2021), and, (d) resolving surface heterogeneity by increasing the grid resolution (Kim et al., 2022; Rummukainen, 2016), which includes variable resolution modeling using unstructured meshes for high resolution modeling in selected regions (Wijngaard et al., 2023). In this study, a new topography-based subgrid structure (Tesfa & Leung, 2017) and simple physically based downscaling methods of atmospheric forcing from the atmosphere grid to the topographic units (TGUs) of the land model (Tesfa et al., 2020) have been implemented in the Energy Exascale Earth System Model (E3SM). E3SM is a coupled ESM developed to tackle the grand challenge of actionable predictions of Earth system variability and change (Golaz et al., 2019; Leung et al., 2020).

The default land model component of E3SM (ELM) subdivides each grid into a nested subgrid hierarchy of multiple land units, soil columns and Plant Functional Types (PFTs) representing mainly land surface subgrid heterogeneity due to land cover and land use, assuming homogenous topography and climate within each grid. With the implementation of the new subgrid structure, each grid can have multiple TGUs and each TGU with distinct downscaled atmospheric forcing can have multiple land units, soil columns and PFTs. Downscaling the atmospheric forcing from the atmosphere grid to the TGUs of ELM enables the effects of topographic heterogeneity on atmospheric processes to be captured. Overall, the subgrid scheme improves the capability to capture surface heterogeneity influenced by topography with minimal increase in computational demand by discretizing grids into variable numbers of TGUs per grid depending on the complexity of topography within each grid from high-resolution topographic data (Figure 1). Combined with other parallel improvements to the E3SM model such as representations of human impacts (Zhou et al., 2020), soil erosion (Tan et al., 2018) plant hydraulics (Kennedy et al., 2019), and biogeochemistry (Yang et al., 2023), implementation of the topography-based subgrid structure and methods of downscaling of atmospheric forcings are expected to enhance the capability of ELM to better resolve terrestrial processes in regions of heterogeneous terrain.

Details are in the caption following the image — **Figure 1**
Open in figure viewer PowerPoint

Spatial pattern of the number of topographic units (TGUs) per half degree grid showing the capability of the topography-based subgrid structure to capture topographic heterogeneity over mountainous areas and the associated data reduction metric (drm) which is a measure of the computational efficiency of defining a variable number of TGUs per grid relative to assigning the maximum number of TGUs per grid, which is 12 in this study.

The objectives of this study are: (a) to document the implementation of the topography-based subgrid scheme and methods of downscaling of atmospheric forcing in ELM, and (b) to evaluate the effects of the subgrid scheme and downscaled forcing on simulations of land surface processes, with a greater focus on runoff and snow water equivalent (SWE). For this purpose, TGUs are first derived from high resolution elevation data (90 m) for the half degree grids (Figure 1). TGU-based surface properties (including PFTs, soil texture etc.) input parameters are generated by mapping grid-level values onto the TGUs of each grid. Then offline ELM simulations are performed using two configurations: (a) ELM simulation using mean grid-level forcing, hereafter referred as NoD and, (b) ELM simulation with mean grid-level atmospheric forcing downscaled to the TGU-level, hereafter referred as D; both using the same land surface input parameters. Results simulated by the two ELM configurations are analyzed and compared to evaluate the impacts of the subgrid scheme and downscaling of atmospheric forcing on land surface processes at three levels: (a) using the whole conterminous US (CONUS) domain, (b) at various physically defined regions of the CONUS, and (c) at the Natural Resources Conservation Service (NRCS) Snowpack Telemetry (SNOTEL) observing network sites. For the regional-level evaluation, the CONUS domain is discretized into distinct physically defined regions based on climate and topographic factors. Finally, the effects of the subgrid scheme and downscaling of atmospheric forcing on the capability of ELM to reproduce the observed SWE at the SNOTEL sites are investigated by grouping the SNOTEL sites based on climatic regions.

The remainder of the paper is organized as follows: Section 2 describes the development of the topography-based subgrid scheme and methods of downscaling of atmospheric forcing implemented in E3SM. The strategy to evaluate the effects of the subgrid scheme and downscaling of atmospheric forcing on land surface processes are discussed in Section 3. Section 4 presents the results and discussion. Section 5 closes with discussion and conclusions.

2 Data and Methodology

2.1 Development of Topography-Based Subgrid Structures

A simple subgrid discretization scheme is used to derive the topography-based subgrid structure implemented in ELM, which uses surface elevation data, elevation threshold value (ETV), maximum number of TGUs per grid (maxNumTGUs), and latitude and longitude values of the edges of the study domain as input parameters. The algorithm can be used to derive subgrid information at various spatial resolutions by varying the values of ETV and maxNumTGUs input parameters. Smaller/larger values of ETV/maxNumTGUs input parameters correspond to finer resolution of the subgrid data.

In this study, subgrid data are developed for the 0.5-degree regular grid units from high resolution (90 m) elevation data (Lehner et al., 2008) using 100 and 12 as the values for ETV and maxNumTGUs, respectively. The algorithm first extracts elevation values based on the boundary of the modeling unit (grid) and discretizes the modeling unit into 12 initial subgrid units using the 12 percentile elevation values calculated to represent the elevation values at each consecutive percentile (10th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, 85th, 90th, 95th and 100th). Then, the 12 values of elevation range are determined using the minimum elevation value within each grid and the corresponding percentile elevation values as class breaks. Furthermore, the 100-m ETV value is used to calculate new values of elevation class break using a recursive algorithm developed in this study. The recursive algorithm (see Figure 2) merges any elevation class with elevation range less than the ETV value to its neighboring class recursively until all the classes with elevation range smaller than the threshold value are removed. This allows the topography-based subgrid scheme to capture the impacts of topographic heterogeneity while minimizing computational demand of the model by varying the number of TGUs per grid depending on the topographic heterogeneity within each modeling unit. This method of representing surface heterogeneity using subgrid units derived using the recursive algorithm differs from that of the variable resolution regional refinement approach (Tang et al., 2023; Wijngaard et al., 2023) in that the number of subgrid units depends on the topographic complexity within the grid whereas the regional refinement approach discretizes a coarse grid into the same number of smaller grids everywhere without considering the topographic heterogeneity within the grids. The regional refinement approach is computationally more demanding because the computational requirement of the atmosphere model scales by a factor of 8 for each doubling of the resolution. On the other hand, the computational requirement of the subgrid TGU approach roughly scales linearly with the number of TGUs per grid and affects only the land surface model which is computationally much less intensive than the atmosphere model. For example, Golaz et al. (2019) showed that the relative computational cost of the land component of E3SM is negligible for fully coupled simulations while that of the atmosphere component is >75%.

The performance of the subgrid scheme in reducing the computational demand can be quantified using the data reduction metric (drm) calculated using the following equation:

\text{drm}=\left(1-\,\frac{\text{Nr}}{\text{Ng}}\right)

(1)

where Nr, and Ng respectively refer to the total number of valid TGUs derived using the recursive algorithm calculated as the sum of the number of valid TGUs per grid over the whole domain, and the total number of subgrid units that would be generated by simply dividing each grid into maxNumTGUs number of TGUs (number of grids in the domain multiplied by maxNumTGUs). Figure 1 shows the spatial pattern of the number of TGUs per grid for the CONUS domain and the performance of the subgrid scheme in reducing computational demand when TGUs are derived for the 0.5-degrees grid by using more TGUs within grid cells over mountainous regions to represent subgrid topographic heterogeneity than the grid cells in topographically homogenous regions. The new subgrid scheme has been implemented in E3SM, allowing each grid to have multiple TGUs and each TGU to have multiple land units, soil columns and PFTs.

2.2 Downscaling Methods of Atmospheric Forcing

In ESMs, downscaling of atmospheric forcing from the atmosphere grid to the TGUs of the land model is needed to capture the topographic influence on atmospheric processes, thus harnessing the benefits of the new subgrid scheme on capturing the impacts of topographic heterogeneity in land surface modeling. Downscaling from grid level of the atmosphere to the TGUs of the land model is expected to provide more accurate atmospheric forcing for modeling land surface processes such as water and energy balance at the TGU-level. Generally, large scale geographical factors such as geographical location, macroscale topography, macroscale overall natural environment and by the local topographic elevation (Tesfa et al., 2020; Zhang et al., 2018) have important effects on atmospheric processes. This study applies simple concepts described in Tesfa et al., 2020 to develop methods of downscaling of atmospheric forcings from the atmosphere model grid to the TGUs of the land model generally as:

{V}_{t}={V}_{g}+{\increment}{V}_{e}

(2)

where V_t and V_g represent atmospheric forcing at each TGU of a grid and atmospheric forcing at the grid level calculated by the atmospheric model, respectively. ∆V_e represents the difference between the grid and TGU level values of atmospheric forcing.

Tesfa et al. (2020) evaluated four simple physically-based methods of precipitation downscaling and identified two methods with superior performance when compared against the widely used high-resolution gridded precipitation data obtained from the Precipitation-elevation Regressions on Independent Slopes Model (PRISM) (Daly et al., 1994). These two methods have been implemented in E3SM: (a) Elevation Range with Maximum elevation Method (ERMM) and (b) Froude Number Method (FNM). The ERMM and FNM downscaling methods differ from the linear model of orographic precipitation downscaling method described in Smith et al. (2005) in its simplicity for implementation in global Earth system models as the more physically-based linear model requires calibration of the cloud delay time using observations that are not globally available. The ERMM method uses only the topographic characteristics of the grid and the TGUs to disaggregate grid-level precipitation to the TGUs of the grid. It can be used in both offline and coupled E3SM model configurations. The FNM method is similar to the ERMM method except that it uses information on height rise of an air parcel encountering mountainous terrain (equivalent to Froude Number), which is calculated based on wind speed and static stability, over topographically heterogenous grids whenever a stable orographic regime is identified within the target grid (Leung & Ghan, 1995, 1998). The FNM method can be used only in a coupled E3SM configuration, as the height rise calculated by the atmosphere model component is needed as input for the downscaling. Thus, only the ERMM method is evaluated in this study using offline ELM simulations. Readers are referred to Tesfa et al., 2020 for more explanations on the ERMM and FNM precipitation downscaling methods.

In the ERMN precipitation downscale method (Tesfa et al., 2020)

{\increment}{V}_{e}

is determined using the following equation:

{\increment}{V}_{e}={V}_{g}\frac{\left({Z}_{t}\,-\,{Z}_{g}\right)}{{Z}_{t\mathrm{m}ax}}

(3)

where Z_t and Z_g are TGU and grid level surface elevation values, respectively, and Z_tmax refers to the maximum TGU level elevation value within the grid. The downscaling methods of precipitation implemented in E3SM also include a simple method of precipitation partitioning into snowfall and rainfall following a method described in Jordan (1991). For more clarifications on the precipitation downscaling methods, readers are referred to Tesfa et al. (2020).

Atmospheric forcing of air temperature, downward longwave radiation, specific humidity, and atmospheric pressure are also downscaled from the atmosphere grid to the TGUs of the land model following the approach used to downscale grid-mean quantities to the elevation classes of ice sheets in ELM. The downscaling of air temperature and downward longwave radiation is based on the assumption of a constant lapse rate Γ_m of 6 K km⁻¹. Therefore, the air temperature at a TGU is obtained by linear interpolation of the deviation of the TGU elevation from the grid cell mean, that is.

{\mathrm{T}}_{t}={T}_{g}-{{\Gamma }}_{m}{\Delta }Z

(4)

where

{\Delta }Z={Z}_{t}\,\mbox{--}\,{Z}_{g}

(5)

For downward longwave radiation, the difference between the TGU and grid cell mean quantities is

{\Delta }{R}_{L}^{{\downarrow}}=\frac{\partial {R}_{L}^{{\downarrow}}}{\partial T}\frac{\partial T}{\partial z}{\increment}Z

(6)

Given that

{R}_{L}^{{\downarrow}}=\varepsilon \sigma {T}^{4}

where ε is the emissivity of the air, σ is the Stefan-Boltzmann constant, and T is temperature,

\frac{\partial {R}_{L}^{{\downarrow}}}{\partial T}=4\varepsilon \sigma {T}^{3}=4\frac{{R}_{L}^{{\downarrow}}}{T}

(7)

Therefore,

{R}_{Lt}^{{\downarrow}}={R}_{\mathit{Lg}}^{{\downarrow}}-4\frac{{{{\Gamma }}_{m}{\Delta }ZR}_{\mathit{Lg}}^{{\downarrow}}}{\overline{T}}

(8)

where

\overline{T}

is the average of the TGU and grid cell mean air temperatures.

The TGU surface pressure is found using the standard hypsometric equation:

{p}_{st}={p}_{sg}{e}^{-{\Delta }Z/H}

(9)

where the scale height

H={R}_{a}\overline{T}/g

, R_a being the gas constant of air and g the gravitational acceleration.

Specific humidity on the TGU is found simply by scaling the grid cell specific humidity by the ratio of the saturated specific humidity values at the TGU and grid cell levels, that is,

{q}_{t}={q}_{g}\frac{{q}_{t}^{\text{sat}}}{{q}_{g}^{\text{sat}}}

(10)

2.3 Input Data and Experimental Approaches

2.3.1 Input Data

ELM simulations require land surface data representing various features of the land surface and atmospheric data to drive the model. The land surface data include land use and land cover information consisting of fractions of urban, lake and glaciers, and plant functional types (PFTs) with leaf area and stem area indices corresponding to each PFT, and soil texture, soil organic matter and soil color data representing the properties of the soil column. The land use and land cover used in this study are obtained from the 0.05 × 0.05° spatial resolution data developed in Ke et al. (2012). Since the main purpose of this study is to evaluate the impacts of the subgrid structure and the downscaling of atmospheric forcing, other surface properties are kept the same between the two ELM configurations by assigning grid-level values of land units, PFTs and soil properties to the TGUs of each grid. Also, the satellite phenology option of ELM is used, so biogeochemical processes are not included in the simulations.

The atmospheric forcing created by Qian et al. (2006) (commonly referred to as the QIAN forcing data) are used to drive ELM simulations. ELM inherited these data sets from the Community Land Model version 4.5 (CLM4.5 Oleson et al., 2013) as default forcings for offline simulations. The QIAN data sets were created from observed monthly precipitation, surface air temperature, cloud cover, and satellite solar radiation to constrain the 6-hourly values from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis. The spatial resolution is determined by the NCEP T62 Gaussian gridding system (Li, Leung, Getirana, et al., 2015; Li, Leung, Tesfa, et al., 2015). The QIAN data set used to drive ELM includes precipitation, air temperature, incoming solar radiation, wind speed, surface pressure, and specific humidity. In this study, ELM simulations are run from 1980 to 2004.

2.3.2 Experimental Approaches

The experimental approach designed to evaluate the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing on land surface processes involves ELM simulations using two model configurations: (a) ELM offline simulations with all TGU-level values of atmospheric forcing mapped from the grid-level average values (NoD ELM); and, (b) ELM offline simulations with the TGU-level values of precipitation, air temperature, longwave radiation, specific humidity, and atmospheric pressure downscaled from the grid average values using the downscaling methods described in Section 2.2 (D ELM). In both model configurations, ELM is run on a half degree grid, using the same surface properties and the QIAN atmospheric forcing (Qian et al., 2006). In both configurations, ELM is first run for 250 years until the models are well-spun up as determined by the convergence of the soil moisture and groundwater table, followed by a final 25-year model run to be used to evaluate the effects of the subgrid scheme and atmospheric forcing downscaled to the TGUs on ELM simulated land surface process. The strategy of our experimental approach is to keep all model inputs the same, except for the downscaling of the atmospheric forcing. We focus on the final 25 years of the simulations to better coincide with the availability of observation data.

ELM simulations are performed globally, but for the purpose of this study, the results over the CONUS spatial domain are extracted for the final 25-year simulations of the two model configurations to be analyzed and compared at the CONUS domain level as well as its subregions derived based on various approaches. The CONUS region is selected due to its diverse climatic and topographic characteristics (Tercek et al., 2021) as well as the freely available observation data required for evaluation of model performance (Daly et al., 1994; Thornton et al., 1997), making it an ideal region for evaluation of the impacts of the topography-based subgrid structure and downscaling of atmospheric forcing.

3 Evaluation Strategies

The evaluation strategy involves three main levels of comparisons of results from the simulations of the D and NoD configurations of ELM: (a) comparisons over the whole CONUS level to explore the overall impacts of the topography-based subgrid scheme and the downscaling of atmospheric forcing on ELM simulated land surface fluxes; (b) regional-level comparisons to investigate how the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing vary across different regions of the CONUS domain; and, (c) comparisons using observation data to evaluate the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing on the capability of ELM to reproduce the observed data.

3.1 Evaluation Over the CONUS

Simulated results from the D and NoD ELM configurations are compared at the CONUS level to explore the overall effects of the topography-based subgrid scheme and downscaling methods of atmospheric forcing on ELM simulated land surface fluxes. For this purpose, results from the two ELM configurations are first compared using maps of differences in snowfall fraction, total runoff and SWE to explore the spatial patterns of their differences. The long term monthly average values of total runoff, runoff fraction, soil moisture, snow melt, snowfall fraction and SWE from D ELM configuration are then compared against to those of the NoD ELM configuration. Furthermore, the SWE of April 1st simulated by the D configuration of ELM is compared against that of the NoD as well as their differences in simulated SWE against long term winter temperature and topographic heterogeneity at the CONUS level. The CONUS level comparisons serve as the basis to direct the regional level evaluation in selecting regionalization methods and simulated fields for more detailed comparisons.

3.2 Regional-Level Evaluation

Following insights learned from the CONUS level evaluation, the regional level evaluation is intended to investigate which approaches of discretization and regions of CONUS show the greatest impacts, and how the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing vary across different regions of CONUS in a more detailed manner. For this purpose, the CONUS domain is discretized into (a) regions of long-term annual snowfall fraction; (b) regions of water limited and energy limited climate; (c) regions of season of maximum precipitation; and (d) regions of topography-based characteristics. Hydrologic fluxes simulated using the D and NoD configurations of ELM are compared across different discretization approaches and CONUS regions.

3.2.1 Evaluation Over Snowfall Fraction Regions

The fraction of snowfall in precipitation is highly dependent on topographic elevation. Given that the TGUs are derived using topographic elevation as well as the atmospheric forcings downscaled based on topographic information, one can expect more pronounced differences between variables simulated using the D and NoD configurations of ELM over the regions of CONUS that are derived based on the fraction of snowfall. To investigate this, the CONUS domain is represented using three snowfall fraction regions (SNR): (a) areas with snowfall fraction greater than 0.3, hereafter referred as SN1; (b) areas with snowfall fraction values between 0.3 and 0.1, hereafter referred as SN2; and (c) areas with values of snowfall fraction less than 0.1, hereafter referred as SN3. The snowfall fraction regions were generated by first classifying the globe into five regions based on the values of snowfall fraction of long-term annual precipitation using the Natural Breaks (Jenks) classification method. Then the five regions were reclassified into three regions by merging the regions with small areal coverage to derive three regions with similar areal coverage. Finally, the regions for the CONUS domain were extracted. Hydrologic fluxes simulated by the D and NoD configurations of ELM are compared across the SNR regions to evaluate the impacts of the new subgrid structure and downscaling of atmospheric forcing on ELM simulations.

3.2.2 Evaluation Over Water- and Energy-Limited Regions

The effects of the topography-based subgrid structure and downscaling of atmospheric forcing are expected to manifest differently in water- and energy-limited regions. For example, according to Li et al. (2022), water-limited areas are more sensitive to precipitation than those of energy-limited regions. To investigate this, the CONUS domain is discretized into water- and energy-limited climate regions (WELR). For this purpose, ELM simulated annual potential evapotranspiration, annual actual evapotranspiration and annual precipitation data averaged over 25 years are used to classify the CONUS domain into water- and energy-limited regions, hereafter referred as WL and EL regions, respectively. Hydrologic fluxes simulated by the D and NoD configurations of ELM are compared across the WELR regions to evaluate how the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing vary between the WL and EL regions of the CONUS domain.

3.2.3 Evaluation Over Maximum Precipitation Season Regions

Specific sources/types of precipitation dominate different regions of the CONUS domain during different seasons of the year. For example, precipitation during wintertime comes from large frontal systems, while precipitation during summertime generally results from local convective systems; and areas on the west facing slopes of the Cascade mountains receive their maximum precipitation during winter season, dominantly coming from the frontal system (Thornton et al., 1997). Such variations lead to different regions of the CONUS domain receiving their maximum precipitation in different seasons of the year. In this part of the study, the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing on ELM simulations across different CONUS regions are examined by discretizing the CONUS domain into four regions of season of maximum precipitation (MPSR) (Figure 3) determined as the season with the maximum long-term seasonal mean precipitation within the annual cycle, hereafter referred as SON, DJF, MAM and JJA regions. Hydrologic fluxes simulated by the D and NoD configurations of ELM are compared to evaluate how the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing vary across the MPSR regions of CONUS.

3.2.4 Evaluation Over Topography-Based Regions

Generally, topographic heterogeneity exerts major control on climatic variability (e.g., orographic precipitation and temperature) (Tesfa et al., 2020). Within CONUS, for instance, areas along the western slope of the Cascade mountains, which receive frequent heavy precipitation, exhibit increasing annual total precipitation with increasing topographic elevation (Thornton et al., 1997). The effects of the topography-based subgrid scheme and downscaling of atmospheric forcing on ELM simulations across different regions of the CONUS domain are further examined by discretizing the domain into topographic regions (TR) using a value introduced to represent the relative dominance of the subgrid high elevation areas versus the low-lying areas, hereafter referred to as Topographic Ratio (TPR). For each grid cell, the TPR value is calculated as follows:

\text{TPR}=\frac{{T}_{h}}{{T}_{t}}

(11)

where, TPR, T_h, and T_t refer to the Topographic Ratio, the number of TGUs with elevation values higher than the mean elevation of the grid cell, and the total number of TGUs in the grid, respectively. Grids with TPR >0.5 are dominated locally by high elevation areas relative to the grid mean value, while grids with TPR <0.5 are dominated locally by low lying areas relative to the grid mean value (Figure 4). On the other hand, grids with TPR = 0.5 are characterized by similar fractions of lower and higher local elevation areas. Using TPR values, the CONUS domain is classified into four TR regions: TPR <0.5, TPR = 0.5, 0.5 < TPR <1, and TPR = 1, hereafter referred as TR1, TR2, TR3, and TR4, respectively. Hydrologic fluxes simulated by the D and NoD ELM configurations are compared across the TR1, TR2 and TR3 regions only as there is only one TGU per grid in the TR4 (TPR = 1) region, hence simulations from the two ELM configurations in TR4 region are the same.

3.2.5 Evaluation Over TR Regions Within MPSRs

Topography exerts major control on precipitation and temperature in the western US (Daly et al., 1994). Most of the precipitation in this region particularly in the high elevation mountains occur during cool seasons between October and March (Hamlet et al., 2007). For more detailed investigation of how the effects of the topography-based subgrid structure and downscaling of atmospheric forcing on ELM simulations vary in different topography-based regions of the MPSR regions, each MPSR region is further discretized into the topography-based regions (TRs). Then, the two ELM configurations (D and NoD) are compared using simulated runoff, runoff ratio, snowfall fraction and SWE calculated in each TR region of the MPSR regions.

3.3 Evaluation Against Observations

Finally, TGU-level snow water equivalent (SWE) simulated using the D and NoD configurations of ELM are compared against the SWE observed at the SNOTEL sites. For this purpose, SNOTEL sites are first mapped onto the TGUs of each grid using the latitude and longitude of the sites and surface elevation at the location, where the latitude and longitude enable locating each site to the corresponding grid cell and the elevation of the site is used to identify the corresponding TGU within the grid. Statistical metrics of coefficient of determination (R²) and Root Mean Squared Error (RMSE) are then calculated between the SWE data observed at each SNOTEL site and the TGU-level simulated SWE calculated at the corresponding TGU for both the D and NoD configurations of ELM. The statistical metrics are compared using scatter plots to examine the impact of the topography-based subgrid scheme and downscaling methods of atmospheric forcing on the capability of ELM to predict the observed SWE. To further investigate how the capability of ELM to predict observed SWE varies across different regions, comparisons are conducted by categorizing the SNOTEL sites according to the MPSR regions of their location.

4 Results and Discussion

Following the evaluation strategy, results comparing the two configurations of ELM (D and NoD) over the CONUS domain and its regions are used to evaluate the effects of the topography-based subgrid scheme and downscaling of atmospheric forcing on ELM simulations of land surface processes.

4.1 Results Using the US CONUS Domain

In Figure 5, the long-term seasonal snowfall fraction from the D and NoD configurations of ELM are compared over the CONUS domain during different seasons of the year: SON (Figures 5a and 5b), DJF (Figures 5c and 5d), and MAM (Figures 5e and 5f). The differences maps (Figures 5b, 5d and 5f) calculated by subtracting the NoD simulation from the D simulation show that the D ELM configuration generally simulates more snowfall over the mountainous regions of the Northwest compared to the NoD ELM configuration. The difference in snowfall fraction is mainly driven by the precipitation partitioning method implemented as part of the precipitation downscaling in the model. The downscaling implemented in ELM partitions the TGU-level precipitation into snowfall and rainfall components based on the TGU-level temperature, which generates more/less snowfall over the TGUs with higher/lower elevation. The results also suggest that more of the differences between the D and NoD configurations of ELM occur during SON and DJF in the topographically heterogeneous high precipitation regions, especially the western slopes of the Cascade Mountains and the Coast Mountains.

Figure 6 compares the fractions of long-term seasonal total runoff of the SON (Figures 6a and 6b), MAM (Figures 6c and 6d), and JJA (Figures 6e and 6f) seasons simulated by the two ELM configurations (D and NoD). The results generally show that the D configuration of ELM generates more runoff compared to that of the NoD over the mountainous regions of CONUS during SON, MAM and JJA. The differences between the D and NoD configurations are more pronounced during the JJA season. The higher differences during JJA are caused by the differences in snowfall fraction (Figure 5) between the two ELM configurations. The D configuration of ELM generates more snow melt driven runoff than the NoD ELM because more snowfall is accumulated during SON and DJF, which is melted during the warm seasons to produce more runoff.

In Figure 7, the two ELM configurations are compared using their long-term seasonal SWE over the CONUS domain. The results suggest that generally the D ELM generates more SWE than the NoD ELM over the mountainous regions of CONUS especially during DJF and MAM. This can be explained by the higher snowfall fraction that the D ELM generates over the Northwest mountainous areas (Figure 5).

Figure 8 compares the D and NoD ELM configurations using the long-term monthly total runoff (Figure 8a), fraction of total runoff (Figure 8b), soil moisture at the topmost soil layer (Figure 8c), snow melt (Figure 8d), fraction of snowfall (Figure 8e), and SWE (Figure 8f) averaged over the CONUS domain grid cells with the number of TGUs per grid greater than 1. Results suggest that the D ELM generates more runoff during spring and summer months driven by its higher snowfall fraction and snow melt compared to NoD ELM (Figure 8a). Similarly, the comparison using fraction of total runoff (Figure 8b) suggests that higher runoff fraction is generated by the D ELM during summer months than the NoD ELM, implying that the NoD ELM generally simulates higher ET than the D ELM configuration. On the other hand, the NoD ELM simulates generally higher soil moisture than D ELM during winter months (Figure 8c), which can be explained by the lower fraction of snowfall, because higher rainfall fraction in NoD ELM allows more water to infiltrate into the soil.

In Figure 9, the two ELM configurations are compared using the simulated April 1st SWE, which approximately corresponds to the date of maximum snow accumulation over the mountainous areas of the western US (Hamlet et al., 2005). Figure 9b shows that the D ELM generally simulates more SWE on April 1st compared to NoD ELM, particularly for the mountainous areas of the northwestern US (see Figure 1). The scatter plot in Figure 9c shows a comparison of the difference in their April 1st simulated SWE against the winter (DJF) temperature exploring how the difference between the two ELM configurations in SWE is related to the seasonal temperature of winter months (DJF) and topographic heterogeneity. The results suggest that the difference between the April 1st SWE simulated by the two ELM configurations is generally more pronounced over mountainous areas particularly in regions with near-freezing long-term winter temperature. These results show that areas with near freezing winter temperature are more sensitive to the subgrid scheme and downscaling of atmospheric forcing. The results also suggest that downscaling of atmospheric forcing tends to generate more/less snow in regions with warmer/colder winter temperature. Thus, subsequent analyses and discussions will focus on comparisons of the two ELM configurations across different CONUS regions.

4.2 Results Over CONUS Regions

In Figure 10, the D and NoD ELM configurations are compared using simulated long-term monthly total runoff over regions of the CONUS domain delineated using the approaches described in Section 3. Figure 10a, compares the D and NoD ELM configurations using simulated total runoff over the three CONUS regions derived from long term annual snowfall fraction: (a) snowfall fraction >0.3 (SN1), (b) 0.3 ≥ snowfall fraction >0.1 (SN2), and (c) snowfall fraction ≤0.1 (SN3). By comparing the hydrographs of the three SNR regions, one can easily see that regions SN1 and SN2 have bimodal runoff hydrographs (peaking in both fall and spring seasons), while region SN3 has a single peak runoff value, showing the importance of snowmelt in SN1 and SN2. Figure 10a also shows larger difference between the two ELM configurations in simulating total runoff over SN1 and SN2 particularly in spring and summer, while the difference is minimal in SN3. Hence the impacts of the topography-based subgrid scheme and downscaling of atmospheric forcing on runoff are larger in spring and summer in regions that have higher snowfall fraction. This has important implications to the water resources management of the region given that more than 70% of the runoff in the western US is generated from snowmelt (Hamlet et al., 2005).

Figure 10b compares the simulated long-term monthly total runoff from the two ELM configurations over water and energy limited regions of CONUS. The results show generally larger difference in the energy limited region than the water limited region. Generally, in both ELM configurations, the results show that simulated runoff is much higher over the energy limited region, and in both water and energy limited regions, the D ELM configuration generates more runoff in spring and summer than the NoD ELM. Also, in the energy limited region, runoff peaks in fall and spring in both ELM configurations, similar to those of the SN2 region, indicating that the energy limited areas of CONUS receive a large portion of precipitation as snowfall (Thornton et al., 1997), and snowmelt during the warm season reduces the water limitation on evapotranspiration.

In Figure 10c, the two configurations of ELM are compared using simulated long-term monthly total runoff over CONUS regions delineated based on the season of maximum precipitation (SON, DJF, MAM and JJA) as described in Section 3.2.3. The results show that the difference between the two ELM configurations is larger in the SON and DJF regions particularly in spring and summer, while the differences in the MAM and JJA regions are minimal. In the CONUS domain, regions that receive their maximum precipitation during cool seasons (SON and DJF) tend to have higher amount of snowfall compared to regions that receive their maximum precipitation during MAM and JJA. The larger differences in simulated runoff between the D and NoD ELM configurations in the SON and DJF regions suggest that the effects of downscaling and the subgrid scheme are more pronounced in areas receiving higher snowfall. Also, Figure 10c shows that both ELM configurations generate higher amount of runoff in the SON and DJF regions compared to the MAM and JJA regions. This can be explained by the fact that the SON and DJF regions, as shown in Figure 3, encompass the mountainous regions of the western US, which receive generally higher annual precipitation (Thornton et al., 1997) than the MAM and JJA regions. However, the results also show that the two regions are characterized by different runoff hydrographs. The SON region has a bimodal hydrograph peaking in late fall and early spring, while the DJF region runoff peaks in early spring only. Comparing the hydrographs of the SON and DJF regions against those of the snowfall-based regions in Figure 10a suggests that there is similarity between the SON and SN1 regions in that both regions have their runoff peaking in late fall and early spring and similarity between the DJF and SN3 regions in that in both regions feature a single runoff peak in early spring with comparable values.

Figure 10d compares the two ELM configurations using simulated total runoff over the three topographic regions (TR1, TR2, and TR3) generated based on the ratio of the number of TGUs with elevation higher than the grid cell mean to the total number of TGUs in that grid cell (Figure 4) as described in Section 3.2.4. Results show exceptionally higher difference between the monthly runoff values simulated by the two ELM configurations in TR1, especially during spring and summer, while smaller differences are shown in TR2 and TR3. The results also show that the total runoff generated by both ELM configurations decreases from TR1 to TR3, suggesting that the topography-based classification approach can capture the changes in precipitation with increasing surface elevation (Thornton et al., 1997). The change of the runoff hydrographs from bimodal toward monomodal from TR1 to TR3 suggests that the topography-based regions are capturing the increase of snowfall fraction with increasing surface elevation. Figure 10d emphasizes the capability of a simple topography-based index to capture areas of larger effects of the subgrid scheme and downscaling of atmospheric forcing on ELM simulated hydrologic fluxes. Similar comparisons of the two ELM configurations are also shown in Figure S1 (Supporting Information S1) using long term monthly mean snowfall fraction and SWE calculated at the four types of CONUS regions (SNR, WELR, MPSR, and TR). Overall, the results over the SNR and WELR regions (Figures S1a, S1b, S1c, and S1d in Supporting Information S1) show that the differences between the D and NoD ELM configurations in snowfall fraction and SWE are generally small and similar across the regions. Results over the MPSR regions (Figures S1e and S1f in Supporting Information S1) show that the differences between the D and NoD ELM in snowfall fraction and SWE are larger over regions that receive their maximum precipitation during SON and DJF compared to those of the MAM and JJA regions. The results over the MAM region show that the difference in snowfall fraction is comparable to those of the SON and DJF regions, but that is not translated to similar differences in SWE (Figures S1e and S1f in Supporting Information S1) because snowfall in the MAM season tends to melt quickly compared to the SON and DJF seasons. The results over the TR regions show that the differences in both snowfall fraction and SWE between the two ELM configurations are larger in the TR3 (TPR >0.5) region compared to the TR1 and TR2 regions (Figure S1g and S1h in Supporting Information S1). Generally, the comparisons of the two ELM configurations using the MPSR and TR regions show more pronounced differences, suggesting their superior ability to capture areas with larger effects of the subgrid scheme and downscaling of atmospheric forcing. Hence, subsequent analyses and discussions will further focus on those regions only.

Impacts of the subgrid scheme and downscaling of atmospheric forcing on snow season length and days of peak SWE are evaluated by comparing the two ELM configurations over the regions of different maximum precipitation seasons and the topographic regions (Table 1). The snow season lengths are determined as the number of days between the dates of 10% of SWE accumulation and of 90% of SWE melting, relative to the annual maximum SWE. The index of the day of peak SWE is determined by counting from the beginning of the water year on 1st September. Both the snow season length and index of day of peak SWE are calculated over regions of different seasons of maximum precipitation and topographic regions. Results (Table 1) show that the two ELM configurations are generally similar in the day of peak SWE, while the comparison using snow season length shows that the D ELM generally simulates longer snow season compared to that of the NoD ELM. The differences in snow season lengths between the two ELM configurations vary from region to region. The comparison over the regions of maximum precipitation seasons suggests that the difference in snow season length increases from the warm season precipitation dominated areas to the cold season precipitation areas, with the minimum/maximum difference occurring in the JJA/DJF regions, respectively (Table 1). Comparison in the topography regions shows that the largest difference between the two ELM configurations occurs at the subgrid or local high elevation region (TR3) (Table 1).

Table 1. The Day of Peak SWE and Snow Season Length for ELM Configurations With (D) and Without (NoD) Downscaling of Atmospheric Forcing Over Regions of Different Topographic Ratios and the Maximum Precipitation Seasons of CONUS

CONUS regions		Day of peak SWE (index)			Snow season length (days)
CONUS regions		D	NoD	D-NoD	D	NoD	D-NoD
Regions of maximum precipitation season	SON	193	193	0	187	167	20
	DJF	167	167	0	158	119	39
	MAM	167	167	0	84	74	10
	JJA	192	191	1	176	172	4
Topographic ratio regions	TGU1	167	167	0	188	179	9
	TGU2	190	190	0	199	190	9
	TGU3	190	190	0	218	194	24

Further comparison of the two ELM configurations is performed by extracting the topography-based regions (TRs) within each maximum precipitation season region (MPSR). Figure 11 compares the long-term total runoff simulated by the D and NoD configurations over the TR regions of each MPSR region. Results show that the difference in total runoff simulated by the two ELM configurations is exceptionally high in the TR3 region of both the SON and DJF regions of MPSR particularly during spring and summer compared to all the other TR regions. The results also show that the differences between the two ELM configurations in the TR1 and TR2 regions of the SON and DJF regions (Figures 11a and 11b) are generally larger than those of the MAM and JJA regions (Figures 11c and 11d). The results over the MAM region show negligible differences between the two ELM configurations. This can be explained by the fact the mountainous areas of the MAM region are mostly located further inland, which are generally drier than the coastal mountain ranges of the western CONUS (see Figure 3). For the same reason, the results in the MAM region also show that the TR3 region (larger fraction of higher elevation areas) generates in both model configurations much less runoff than the other two topographic regions (TR1 and TR2), which is explained by the fact that the TR3 region generally receives lower precipitation than TR1 and TR2 (Figure not shown). The comparison of the two ELM configurations using the topographic regions within the JJA region show slightly larger difference in the region of TR3 than the regions of TR1 and TR2. Given the negligible difference in total runoff of the JJA region compared to Figures 10c and 11d suggests that the topography-based regions are important in capturing the effects of the subgrid scheme and downscaled atmospheric forcing even in MPSR regions that do not show large differences.

Figure 12 compares the two ELM configurations using long term monthly mean fraction of snowfall and SWE over the topographic regions (TRs) of each MPSR region. Results show that larger differences in snowfall fraction and SWE between the two ELM configurations in the TR3 region of all the MPSR regions. However, the difference in snowfall fraction and SWE is particularly high in the TR3 region of the SON and DJF MPSR regions (Figures 12c and 12d), which explains the exceptionally large differences in spring and summer total runoff of the TR3 regions in the SON and DJF MPSR regions shown in Figures 12a and 12b. The results also show that in the three MPSR regions (SON, DJF, and JJA) snowfall fraction and SWE in both ELM configurations decreases from TR3 (larger fraction of high local elevation) to TR1 (larger fraction of low local elevation) indicating that snowfall increases with surface elevation. In the MAM region, however, the TR2 region has the lowest snowfall fraction and SWE instead of TR1 (low-lying areas), which may be explained by the fact that the MAM region is mostly located in the midwestern US, where precipitation is generally less dependent on topography. Results also show that the differences in snowfall fraction and SWE between the two ELM configurations is minimal in the JJA region (Figures 12g and 12h). The minimal difference between the D and NoD ELM configurations in both snowfall fraction and SWE can be explained by the fact that the JJA region mostly encompasses areas with flat topography, despite receiving snowfall (e.g., the northern portion of midwestern US), and topographically heterogeneous areas with no or minimal snowfall (e.g., mountainous regions in the northern Mexico) (see also Figures 1, 3 and 4).

Figure 13 compares the two ELM configurations over the topographic regions (TRs) delineated within each MPSR region using snow season length. Results show that the D ELM generally simulates longer snow season suggesting earlier/delayed snow accumulation/snow melt than those of the NoD ELM configuration, respectively. The difference between the two ELM configurations in snow season length is more pronounced over the local high elevation regions (TR3) of the cold season maximum precipitation regions (DJF and SON). Furthermore, Table S1 in Supporting Information S1 compares the two ELM configurations over the topographic regions of the MPSR regions using day of peak SWE. Results show the D ELM generally simulates delayed peak of SWE compared to the those of the NoD ELM. The maximum delay in the day of peak SWE (2 weeks) occurs in the topographic regions of local high elevation (TR3) within the region dominated by the coldest season precipitation (DJF), emphasizing the importance of the subgrid scheme and downscaling over mountainous regions, which receive their major precipitation during cold seasons.

Furthermore, to understand the significance of the impacts of the subgrid scheme and downscaling of atmospheric forcing on simulated snow fraction, SWE, and total runoff by the two ELM configurations p-values from a non-parametric statistical significance test (Wilcoxon test) at the topographic ratio (TR) and season of maximum precipitation regions are compared in Table 2. Using a confidence level of 95%, results show (a) significant differences in snow fraction and SWE across all the TR regions and regions of seasons of maximum precipitation; (b) significant differences in total runoff in the SON and DJF regions of maximum precipitation season and all the TR regions; and (c) insignificant differences in total runoff over the MAM and JJA regions of maximum precipitation season. Further investigation of statistical significance of their differences in snow fraction, SWE and total runoff suggest over the TR regions within each region of the season of maximum precipitation (Table S2 in Supporting Information S1) shows (a) significant differences in snow fraction over the TR regions of all the regions of maximum precipitation season; (b) significant differences in SWE in all the TR regions of all the regions of maximum precipitation season except the TR2 region of the JJA region; (c) significance differences in total runoff over all the TR regions of the SON region and TR2 and TR3 regions of the DJF region; and (d) insignificant difference in total runoff over all the TR regions of the MAM and JJA regions and the TR1 region of the DJF region. The significance difference in runoff over the regions receiving their maximum precipitation during cool seasons and topographic ratio regions dominated with high subgrid elevation areas is driven mainly the differences in snow fraction. Overall, the significance test results generally suggest that the effects of the subgrid scheme and downscaling of atmospheric forcing are more pronounced over the mountainous regions receiving their maximum precipitation during cool seasons.

Table 2. Significance Test (Wilcoxon Test) Results (P-Values) on Differences of Snow Fraction, SWE and Total Runoff Simulated by the Two ELM Configurations Over the Regions of the Topographic Ratio and Season of Maximum Precipitation

CONUS region		P-values
CONUS region		Snow fraction	SWE	Total runoff
Regions of maximum precipitation season	SON	3.07E−04	6.06E−05	9.42E−03
	DJF	5.81E−11	2.59E−12	2.74E−02
	MAM	1.07E−03	3.45E−04	2.87E−01
	JJA	1.97E−02	3.82E−02	2.11E−01
Topographic ratio regions	TGU1	9.69E−04	5.48E−04	2.34E−02
	TGU2	3.01E−03	1.03E−03	3.34E−02
	TGU3	5.39E−05	3.75E−05	2.40E−03

4.3 Comparison to Observation Data

Improving representation of impacts of small-scale topographic heterogeneity in land surface models is expected to improve modeling of surface water components such as soil moisture and SWE (Singh et al., 2015). Thus, the effects of the subgrid scheme and downscaling of atmospheric forcing are evaluated using SWE measured at the SNOTEL sites. Figure 14 shows scatter plots comparing the R2 (Figure 14a) and RMSE (Figure 14b) statistical metrics calculated between the observed SWE at SNOTEL sites and the TGU-level SWE simulated by the NoD and D ELM configurations at the TGUs where the SNOTEL sites are located. The color represents the topographic heterogeneity of the TGUs using the standard deviation in elevation calculated within each TGU where a SNOTEL site is located. Results show that the values of R2/RMSE of the D ELM are larger/smaller at 83% of the SNOTEL sites compared to those of the NoD ELM configuration. In both scatter plots, the color scale suggests that the advantages of the D ELM configuration are more pronounced at the TGUs with more pronounced topographic heterogeneity. The results generally suggest that the topography-based subgrid scheme and downscaling of atmospheric forcing resulted in superior performance of the simulations of the D ELM configuration compared to the NoD ELM, which has important implications for modeling the impacts of climate change on water resources (e.g., streamflow and stream temperature modeling) (Li, Leung, Getirana, et al., 2015; Li, Leung, Tesfa, et al., 2015).

For more detailed investigation of the effects of the subgrid scheme and downscaling of atmospheric forcing on simulations of land surface processes, the SNOTEL sites were divided into four groups according to their locations in the MPSR regions and the comparisons shown in Figure 15 are repeated and shown in Figure 15. Results show that the values of R2/RMSE of the D ELM are larger/smaller at 100%, 87.13%, 78.57% and 80.24% of the SNOTEL sites in the SON, DJF, MAM and JJA regions, respectively, compared to those of the NoD ELM configuration. The D ELM configuration performed better than the NoD ELM configuration generally in all the MPSR regions, with more pronounced performance improvement in the SON and DJF regions and over areas with more topographic heterogeneity. The more pronounced performance over the SON and DJF regions emphasizes the advantages of the subgrid scheme and downscaling of atmospheric forcing in improving the capability of ELM to reproduce the observed SWE, as shown in Figures 10-12 for those regions with higher differences between the simulations of the D and NoD ELM configurations. Given that about 70% of runoff in the western US comes from snowmelt (Hamlet et al., 2005), a more pronounced capability to reproduce the observed SWE over the SNOTEL sites located in the SON and DJF regions has important implications to water resource management.

5 Discussion and Conclusions

This study introduces a topography based subgrid scheme and methods to downscale atmospheric forcing from the atmosphere grid to the land subgrid units in the Energy Exascale Earth System Model (E3SM) Land Model (ELM) and evaluates their effects on simulated hydrologic fluxes, with a greater focus on runoff and SWE. The subgrid scheme is designed to improve the capability of ELM to capture land surface heterogeneity influenced by topography with modest increase in computational demand compared to methods that divide modeling units (grids) into equal number of smaller grids everywhere, while downscaling of atmospheric forcing is performed using simple physically based methods that disaggregate atmospheric forcing following the topographic variation of the land surface. To evaluate the effects of the subgrid scheme and downscaling of atmospheric forcing, ELM simulations are performed using two offline configurations: (a) ELM simulations using mean grid-level forcing (NoD ELM), and, (b) ELM simulations with the mean grid-level atmospheric forcing downscaled to the TGUs (D ELM); with both configurations using TGUs derived from 0.5-degree grids and the same TGU-level land surface input parameters.

Simulations from the two ELM configurations are first compared using runoff, soil moisture, snow melt, snowfall fraction and SWE at the CONUS level to explore the overall impacts of the subgrid scheme and downscaling of atmospheric forcing on ELM simulations. Further comparison of the two ELM configurations conducted at regional level is intended to investigate how different regionalization approaches identify more pronounced effects of the subgrid scheme and downscaling of atmospheric forcing on runoff, snowfall fraction and SWE. For this purpose, CONUS is discretized into 3, 2, 4, and 3 regions using snowfall fraction, water versus energy limited state, season of maximum precipitation, and topographic index, respectively. Also, a similar comparison is performed by discretizing the regions of maximum precipitation season further into subregions of topographic index. Finally, the two ELM configurations are compared to evaluate how the effects of the subgrid scheme and downscaled atmospheric forcing impact the capability of ELM to capture the observed SWE at the SNOTEL sites.

The CONUS scale results generally suggest that the D ELM configuration produces more snowfall and SWE, drier soil, and higher/less runoff/ET during the spring and summer seasons. The regional level results suggest that the regions derived using TR and MPSR discretization can identify areas with more pronounced effects of downscaling of atmospheric forcing better than the regions derived based on the SNR and WELR discretization approaches. Combining the TR and MPSR regions show that more pronounced impact of downscaled atmospheric forcing occurs in regions that receive their maximum precipitation during cool seasons (SON and DJF) and dominated by subgrid or local high elevation topography. The simple topography-based discretization approach is shown to be the most effective method for identifying areas that are highly impacted by the subgrid scheme and downscaling of the atmospheric forcing. Also, the superior performance of the D ELM configuration in reproducing the observed SWE at 83% of the SNOTEL sites suggests that the subgrid scheme and downscaling of atmospheric forcing greatly enhanced the capability of ELM for simulating SWE. Furthermore, the advantages of the subgrid scheme and downscaling of atmospheric forcing in reproducing the observed data are more pronounced over topographically heterogeneous regions that are dominated by cool season precipitation. Overall, the results in this study highlight the importance of improving the representation of effects of small-scale land surface heterogeneity in ESMs and motivate future research to understand the effects small-scale surface heterogeneity on land-atmosphere interactions using coupled land atmosphere modeling over mountainous regions.

This study demonstrated that by discretizing the land surface following topographic heterogeneity and downscaling atmospheric forcing using simple physically based methods, improved prediction of hydrologic fluxes could be achieved in ESM simulations, particularly in topographically heterogeneous regions dominated by cool season precipitation. Even though the formulations and parameters that govern hydrologic fluxes were tuned for the NoD ELM configuration, it is encouraging that the D ELM configuration performed better with no changes to the ELM parameterizations. Given that about 70% of runoff in such regions comes from snowmelt (Hamlet et al., 2005), the improvement in the capability of the model to predict SWE has important implications to water resource management. Furthermore, this study demonstrated the importance of simple regionalization methods in identifying regions with more pronounced impacts of the subgrid scheme and downscaling of atmospheric forcing on model simulations. It should also be noted that the improvements shown in our results are achieved with minimal increase in computational demand compared to other approaches such as hyper-resolution modeling (Wood et al., 2011).

The topography-based subgrid scheme and downscaling of atmospheric forcing can have important implications to simulating surface energy fluxes in coupled land-atmosphere modeling and to improving modeling of the hydrologic impacts of climate change such as streamflow and stream temperature modeling (Li, Leung, Getirana, et al., 2015; Li, Leung, Tesfa, et al., 2015). Thus, the results of our systematic analyses motivate future research to further investigate the impacts of the subgrid scheme and downscaling of atmospheric forcing on: (a) land-atmosphere interactions in coupled land-atmosphere modeling, and (b) streamflow simulations and implications to water resources management. Evaluation of the impacts of the subgrid scheme and downscaled atmospheric forcing on streamflow simulation and water resource management using observation data such as USGS streamflow data, streamflow data from MOPEX basins and dams and reservoirs data would be useful to advance our understanding. Since, there is substantial difference between precipitation data sets over the mountainous regions of the CONUS domain (He et al., 2019), evaluation of the performance of the two ELM configurations using different forcings would help to reveal how the advantages of downscaling would vary across different forcings. While the advantages of the subgrid scheme and downscaling methods presented in this study are expected to hold across mountainous regions, it would be useful to evaluate how the advantages would vary in other mountainous regions such as South American Andes and European Alps. Although our results already demonstrate substantial improvements in modeling snowpack using the subgrid topographic units and downscaling of atmospheric forcing, Jennings et al. (2018) and Wang et al. (2019) showed that the partitioning of precipitation into snowfall and rainfall depends on both temperature and relative humidity. Testing such a method of precipitation partitioning may also prove useful to further improve the modeling of snow accumulation in mountain regions. Finally, it would be useful to evaluate the relative importance of the effects of the subgrid scheme and downscaling of atmospheric forcing compared to other sources of land surface heterogeneity such as soil properties and land use and land cover on the variability of simulated water and energy fluxes.

Acknowledgments

This research was supported by the Office of Science of the U.S. Department of Energy Biological and Environmental Research through the Earth System Model Development program area as part of the Energy Exascale Earth System Model (E3SM) project. The Pacific Northwest National Laboratory is operated for the US Department of Energy by Battelle Memorial Institute under contract DE-AC05-76RL01830. The reported research used the DOE's Biological and Environmental Research Earth System Modeling program's COMPY computing cluster located at Pacific Northwest National Laboratory. The authors thank the three reviewers for their insightful comments, questions, and suggestions.

Notice: Manuscript Authored by Battelle Memorial Institute Under Contract Number DE-AC05-76RL01830 with the US Department of Energy. The US Government retains and the publisher, by accepting this article for publication, acknowledges that the US Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so for US Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan: (http://energy.gov/downloads/doe-public-access-plan)

Open Research

Data Availability Statement

The model used in this study (E3SM model) has been described in Golaz et al. (2022). The data sets utilized in ELM simulations including surface data file (subgrid-level), model configuration parameter files, model restart files, post-processed grid and TGU level model output files, data files used to create topographic and climatic regions are documented in Tesfa et al. (2024). The atmospheric forcing data used in this study are described in Qian et al. (2006). The observed snow water equivalent data from the SNOTEL sites that have been used to evaluate performance of ELM with and without downscaling have been described in Tesfa et al. (2024).

Supporting Information

References

Arthur, R. S., Lundquist, K. A., Mirocha, J. D., & Chow, F. K. (2018). Topographic effects on radiation in the WRF model with the immersed boundary method: Implementation, validation, and application to complex terrain. Monthly Weather Review, 146(10), 3277–3292. https://doi.org/10.1175/mwr-d-18-0108.1
10.1175/MWR-D-18-0108.1
ADSWeb of Science®Google Scholar
Avissar, R., & Pielke, R. A. (1989). A parameterization of heterogeneous land surfaces for atmospheric numerical models and its impact on regional meteorology. Monthly Weather Review, 117(10), 2113–2136. https://doi.org/10.1175/1520-0493(1989)117<2113:apohls>2.0.co;2
10.1175/1520-0493(1989)117<2113:APOHLS>2.0.CO;2
ADSWeb of Science®Google Scholar
Bou-Zeid, E., Anderson, W., Katul, G. G., & Mahrt, L. (2020). The persistent challenge of surface heterogeneity in boundary-layer meteorology: A review. Boundary-Layer Meteorology, 177(2), 227–245. https://doi.org/10.1007/s10546-020-00551-8
10.1007/s10546-020-00551-8
ADSWeb of Science®Google Scholar
Chaney, N. W., Metcalfe, P., & Wood, E. F. (2016). HydroBlocks: A field-scale resolving land surface model for application over continental extents. Hydrological Processes, 30(20), 3543–3559. https://doi.org/10.1002/hyp.10891
10.1002/hyp.10891
ADSWeb of Science®Google Scholar
Daly, C., Neilson, R. P., & Phillips, D. L. (1994). A statistical-topographic model for mapping climatological precipitation over mountainous terrain. Journal of Applied Meteorology, 33(2), 140–158. https://doi.org/10.1175/1520-0450(1994)033<0140:astmfm>2.0.co;2
10.1175/1520-0450(1994)033<0140:ASTMFM>2.0.CO;2
ADSWeb of Science®Google Scholar
Dessalegn, D., Beyene, S., Ram, N., Walley, F., & Gala, T. S. (2014). Effects of topography and land use on soil characteristics along the toposequence of Ele watershed in southern Ethiopia. Catena, 115, 47–54. https://doi.org/10.1016/j.catena.2013.11.007
10.1016/j.catena.2013.11.007
CASWeb of Science®Google Scholar
de Vrese, P., Schulz, J. P., & Hagemann, S. (2016). On the representation of heterogeneity in land-surface–atmosphere coupling. Boundary-Layer Meteorology, 160(1), 157–183. https://doi.org/10.1007/s10546-016-0133-1
10.1007/s10546-016-0133-1
ADSWeb of Science®Google Scholar
Fu, B. J., Liu, S., Ma, K., & Zhu, Y. (2004). Relationships between soil characteristics, topography and plant diversity in a heterogeneous deciduous broad-leaved forest near Beijing, China. Plant and Soil, 261(1), 47–54. https://doi.org/10.1023/b:plso.0000035567.97093.48
10.1023/B:PLSO.0000035567.97093.48
CASGoogle Scholar
Giorgi, F., & Avissar, R. (1997). Representation of heterogeneity effects in Earth system modeling: Experience from land surface modeling. Reviews of Geophysics, 35(4), 413–437. https://doi.org/10.1029/97rg01754
10.1029/97RG01754
CASADSWeb of Science®Google Scholar
Golaz, J.-C., Caldwell, P. M., Van Roekel, L. P., Petersen, M. R., Tang, Q., Wolfe, J. D., et al. (2019). The DOE E3SM coupled model version 1: Overview and evaluation at standard resolution. Journal of Advances in Modeling Earth Systems, 11(7), 2089–2129. https://doi.org/10.1029/2018ms001603
10.1029/2018MS001603
ADSWeb of Science®Google Scholar
Golaz, J.-C., Van Roekel, L. P., Zheng, X., Roberts, A. F., Wolfe, J. D., Lin, W., et al. (2022). The DOE E3SM model version 2: Overview of the physical model and initial model evaluation. Journal of Advances in Modeling Earth Systems, 14(12), e2022MS003156. https://doi.org/10.1029/2022ms003156
10.1029/2022MS003156
ADSWeb of Science®Google Scholar
Gu, C., Huang, A., Wu, Y., Yang, B., Mu, X., Zhang, X., & Cai, S. (2020). Effects of subgrid terrain radiative forcing on the ability of RegCM4.1 in the simulation of summer precipitation over China. Journal of Geophysical Research: Atmospheres, 125(12), e2019JD032215. https://doi.org/10.1029/2019jd032215
10.1029/2019JD032215
ADSWeb of Science®Google Scholar
Hamlet, A. F., Mote, P. W., Clark, M. P., & Lettenmaier, D. P. (2005). Effects of temperature and precipitation variability on snowpack trends in the western United States. Journal of Climate, 18(21), 4545–4561. https://doi.org/10.1175/jcli3538.1
10.1175/JCLI3538.1
ADSWeb of Science®Google Scholar
Hamlet, A. F., Mote, P. W., Clark, M. P., & Lettenmaier, D. P. (2007). Twentieth-century trends in runoff, evapotranspiration, and soil moisture in the western United States. Journal of Climate, 20(8), 1468–1486. https://doi.org/10.1175/jcli4051.1
10.1175/JCLI4051.1
ADSWeb of Science®Google Scholar
Hao, D., Bisht, G., Gu, Y., Lee, W. L., Liou, K. N., & Leung, L. R. (2021). A parameterization of sub-grid topographical effects on solar radiation in the E3SM land model (version 1.0): Implementation and evaluation over the Tibetan Plateau. Geoscientific Model Development, 14(10), 6273–6289. https://doi.org/10.5194/gmd-14-6273-2021
10.5194/gmd-14-6273-2021
ADSWeb of Science®Google Scholar
Hao, D., Bisht, G., Huang, M., Ma, P., Tesfa, T., Lee, W., et al. (2022). Impacts of sub-grid topographic representations on surface energy balance and boundary conditions in the E3SM land model: A case study in sierra Nevada. Journal of Advances in Modeling Earth Systems, 14(4), e2021MS002862. https://doi.org/10.1029/2021ms002862
10.1029/2021MS002862
ADSWeb of Science®Google Scholar
Hao, D., Wen, J., Xiao, Q., Lin, X., You, D., Tang, Y., et al. (2019). Sensitivity of coarse-scale snow-free land surface shortwave albedo to topography. Journal of Geophysical Research: Atmospheres, 124(16), 9028–9045. https://doi.org/10.1029/2019jd030660
10.1029/2019JD030660
ADSWeb of Science®Google Scholar
He, C., Chen, F., Barlage, M., Liu, C., Newman, A., Tang, W., et al. (2019). Can convection-permitting modeling provide decent precipitation for offline high-resolution snowpack simulations over mountains? Journal of Geophysical Research: Atmospheres, 124(23), 12631–12654. https://doi.org/10.1029/2019jd030823
10.1029/2019JD030823
ADSWeb of Science®Google Scholar
He, S., Smirnova, T. G., & Benjamin, S. G. (2021). Single-Column validation of a snow subgrid parameterization in the rapid update cycle land-surface model (RUC LSM). Water Resources Research, 57(8), e2021WR029955. https://doi.org/10.1029/2021wr029955
10.1029/2021WR029955
ADSWeb of Science®Google Scholar
Huang, L., Zhang, S., Niu, G., Wei, N., Yuan, H., Wei, Z., et al. (2022). A catchment-based hierarchical spatial tessellation approach to a better representation of land heterogeneity for hyper-resolution land surface modeling. Water Resources Research, 58(5), e2021WR031589. https://doi.org/10.1029/2021wr031589
10.1029/2021WR031589
ADSWeb of Science®Google Scholar
Huang, M., Ma, P. L., Chaney, N. W., Hao, D., Bisht, G., Fowler, M. D., et al. (2022). Representing surface heterogeneity in land–atmosphere coupling in E3SMv1 single-column model over ARM SGP during summertime. Geoscientific Model Development, 15(16), 6371–6384. https://doi.org/10.5194/gmd-15-6371-2022
10.5194/gmd-15-6371-2022
ADSWeb of Science®Google Scholar
Jennings, K. S., Winchell, T. S., Livneh, B., & Molotch, N. P. (2018). Spatial variation of the rain–snow temperature threshold across the Northern Hemisphere. Nature Communications, 9(1), 1148. https://doi.org/10.1038/s41467-018-03629-7
10.1038/s41467-018-03629-7
ADSPubMedWeb of Science®Google Scholar
Jordan, R. (1991). A one-dimensional temperature model for a snow cover. US Army CRREL.
Google Scholar
Ke, Y., Leung, L. R., Huang, M., Coleman, A. M., Li, H., & Wigmosta, M. S. (2012). Development of high resolution land surface parameters for the Community Land Model. Geoscientific Model Development, 5(6), 1341–1362. https://doi.org/10.5194/gmd-5-1341-2012
10.5194/gmd-5-1341-2012
ADSWeb of Science®Google Scholar
Kennedy, D., Swenson, S., Oleson, K. W., Lawrence, D. M., Fisher, R., Lola da Costa, A. C., & Gentine, P. (2019). Implementing plant hydraulics in the community land model, version 5. Journal of Advances in Modeling Earth Systems, 11(2), 485–513. https://doi.org/10.1029/2018ms001500
10.1029/2018MS001500
ADSWeb of Science®Google Scholar
Kim, G., Kim, J., & Cha, D. H. (2022). Added value of high-resolution regional climate model in simulating precipitation based on the changes in kinetic energy. Geoscience Letters, 9(1), 38. https://doi.org/10.1186/s40562-022-00247-6
10.1186/s40562-022-00247-6
ADSGoogle Scholar
Kirkpatrick, J. B., Green, K., Bridle, K. L., & Venn, S. E. (2014). Patterns of variation in Australian alpine soils and their relationships to parent material, vegetation formation, climate and topography. Catena, 121, 186–194. https://doi.org/10.1016/j.catena.2014.05.005
10.1016/j.catena.2014.05.005
Web of Science®Google Scholar
Lehner, B., Verdin, K., & Jarvis, A. (2008). New global hydrography derived from spaceborne elevation data. Eos, Transactions American Geophysical Union, 89(10), 93–94. https://doi.org/10.1029/2008eo100001
10.1029/2008EO100001
ADSGoogle Scholar
Leung, L. R., Bader, D. C., Taylor, M. A., & McCoy, R. B. (2020). An introduction to the E3SM special collection: Goals, science drivers, development, and analysis. Journal of Advances in Modeling Earth Systems, 12(11), e2019MS001821. https://doi.org/10.1029/2019ms001821
10.1029/2019MS001821
ADSWeb of Science®Google Scholar
Leung, L. R., & Ghan, S. J. (1995). A subgrid parameterization of orographic precipitation. Theoretical and Applied Climatology, 52(1), 95–118. https://doi.org/10.1007/bf00865510
10.1007/BF00865510
ADSWeb of Science®Google Scholar
Leung, L. R., & Ghan, S. J. (1998). Parameterizing subgrid orographic precipitation and surface cover in climate models. Monthly Weather Review, 126(12), 3271–3291. https://doi.org/10.1175/1520-0493(1998)126<3271:psopas>2.0.co;2
10.1175/1520-0493(1998)126<3271:PSOPAS>2.0.CO;2
ADSWeb of Science®Google Scholar
Li, H.-Y., Leung, L. R., Getirana, A., Huang, M., Wu, H., Xu, Y., et al. (2015). Evaluating global streamflow simulations by a physically based routing model coupled with the community land model. Journal of Hydrometeorology, 16(2), 948–971. https://doi.org/10.1175/jhm-d-14-0079.1
10.1175/JHM-D-14-0079.1
ADSWeb of Science®Google Scholar
Li, H.-Y., Leung, L. R., Tesfa, T., Voisin, N., Hejazi, M., Liu, L., et al. (2015). Modeling stream temperature in the Anthropocene: An earth system modeling approach. Journal of Advances in Modeling Earth Systems, 7(4), 1661–1679. https://doi.org/10.1002/2015ms000471
10.1002/2015MS000471
ADSWeb of Science®Google Scholar
Li, L., Bisht, G., & Leung, L. R. (2022). Spatial heterogeneity effects on land surface modeling of water and energy partitioning. Geoscientific Model Development, 15(14), 5489–5510. https://doi.org/10.5194/gmd-15-5489-2022
10.5194/gmd-15-5489-2022
ADSWeb of Science®Google Scholar
Moeslund, J. E., Arge, L., Bøcher, P. K., Dalgaard, T., Odgaard, M. V., Nygaard, B., & Svenning, J. C. (2013). Topographically controlled soil moisture is the primary driver of local vegetation patterns across a lowland region. Ecosphere, 4(7). https://doi.org/10.1890/es13-00134.1
10.1890/ES13-00134.1
Google Scholar
Oddershede, A., Svenning, J. C., & Damgaard, C. (2015). Topographically determined water availability shapes functional patterns of plant communities within and across habitat types. Plant Ecology, 216(9), 1231–1242. https://doi.org/10.1007/s11258-015-0504-6
10.1007/s11258-015-0504-6
Web of Science®Google Scholar
Oleson, K., Lawrence, D. M., Bonan, G. B., Drewniak, B., Huang, M., Koven, C. D., et al. (2013). Technical description of version 4.5 of the community land model (CLM).
Google Scholar
Qian, T., Dai, A., Trenberth, K. E., & Oleson, K. W. (2006). Simulation of global land surface conditions from 1948 to 2004. Part I: Forcing data and evaluations. Journal of Hydrometeorology, 7(5), 953–975. https://doi.org/10.1175/jhm540.1
10.1175/JHM540.1
ADSWeb of Science®Google Scholar
Rummukainen, M. (2016). Added value in regional climate modeling. WIREs Climate Change, 7(1), 145–159. https://doi.org/10.1002/wcc.378
10.1002/wcc.378
Web of Science®Google Scholar
Shi, H., & Xiao, Z. (2022). Exploring topographic effects on surface parameters over rugged terrains at various spatial scales. IEEE Transactions on Geoscience and Remote Sensing, 60, 1–16. https://doi.org/10.1109/tgrs.2021.3098607
10.1109/tgrs.2021.3098607
Google Scholar
Singh, R. S., Reager, J. T., Miller, N. L., & Famiglietti, J. S. (2015). Toward hyper-resolution land-surface modeling: The effects of fine-scale topography and soil texture on CLM4.0 simulations over the Southwestern U.S. Water Resources Research, 51(4), 2648–2667. https://doi.org/10.1002/2014wr015686
10.1002/2014WR015686
ADSWeb of Science®Google Scholar
Smith, R. B., Barstad, I., & Bonneau, L. (2005). Orographic precipitation and Oregon’s climate transition. Journal of the Atmospheric Sciences, 62(1), 177–191. https://doi.org/10.1175/jas-3376.1
10.1175/JAS-3376.1
ADSWeb of Science®Google Scholar
Swenson, S. C., Clark, M., Fan, Y., Lawrence, D. M., & Perket, J. (2019). Representing intrahillslope lateral subsurface flow in the community land model. Journal of Advances in Modeling Earth Systems, 11(12), 4044–4065. https://doi.org/10.1029/2019ms001833
10.1029/2019MS001833
ADSWeb of Science®Google Scholar
Tan, Z., Leung, L. R., Li, H., & Tesfa, T. (2018). Modeling sediment yield in land surface and earth system models: Model comparison, development, and evaluation. Journal of Advances in Modeling Earth Systems, 10(9), 2192–2213. https://doi.org/10.1029/2017ms001270
10.1029/2017MS001270
ADSWeb of Science®Google Scholar
Tang, Q., Golaz, J. C., Van Roekel, L. P., Taylor, M. A., Lin, W., Hillman, B. R., et al. (2023). The fully coupled regionally refined model of E3SM version 2: Overview of the atmosphere, land, and river results. Geoscientific Model Development, 16(13), 3953–3995. https://doi.org/10.5194/gmd-16-3953-2023
10.5194/gmd-16-3953-2023
CASADSGoogle Scholar
Tercek, M. T., Thoma, D., Gross, J. E., Sherrill, K., Kagone, S., & Senay, G. (2021). Historical changes in plant water use and need in the continental United States. PLoS One, 16(9), e0256586. https://doi.org/10.1371/journal.pone.0256586
10.1371/journal.pone.0256586
CASPubMedWeb of Science®Google Scholar
Tesfa, T. K., Leung, L. Y. R., Thornton, P., Brunke, M. A., & Duan, Z. (2024). Data for impacts of topography-based subgrid scheme and downscaling of atmospheric forcing on modeling land surface processes in the conterminous US. [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.12594204
10.5281/zenodo.12594204
Google Scholar
Tesfa, T. K., Leung, L. R., & Ghan, S. J. (2020). Exploring topography-based methods for downscaling subgrid precipitation for use in Earth System Models. Journal of Geophysical Research: Atmospheres, 125(5), e2019JD031456. https://doi.org/10.1029/2019jd031456
10.1029/2019JD031456
ADSWeb of Science®Google Scholar
Tesfa, T. K., & Leung, L. Y. R. (2017). Exploring new topography-based subgrid spatial structures for improving land surface modeling. Geoscientific Model Development, 10(2), 873–888. https://doi.org/10.5194/gmd-10-873-2017
10.5194/gmd-10-873-2017
ADSWeb of Science®Google Scholar
Tesfa, T. K., Li, H. Y., Leung, L. R., Huang, M., Ke, Y., Sun, Y., & Liu, Y. (2014). A subbasin-based framework to represent land surface processes in an Earth system model. Geoscientific Model Development, 7(3), 947–963. https://doi.org/10.5194/gmd-7-947-2014
10.5194/gmd-7-947-2014
ADSWeb of Science®Google Scholar
Tesfa, T. K., Ruby Leung, L., Huang, M., Li, H. Y., Voisin, N., & Wigmosta, M. S. (2014). Scalability of grid- and subbasin-based land surface modeling approaches for hydrologic simulations. Journal of Geophysical Research: Atmospheres, 119(6), 3166–3184. https://doi.org/10.1002/2013jd020493
10.1002/2013JD020493
ADSWeb of Science®Google Scholar
Tesfa, T. K., Tarboton, D. G., Chandler, D. G., & McNamara, J. P. (2009). Modeling soil depth based upon topographic and landscape attributes. Water Resources Research, 45(10). https://doi.org/10.1029/2008wr007474
10.1029/2008WR007474
Google Scholar
Thompson, S. E., Harman, C. J., Troch, P. A., Brooks, P. D., & Sivapalan, M. (2011). Spatial scale dependence of ecohydrologically mediated water balance partitioning: A synthesis framework for catchment ecohydrology. Water Resources Research, 47(10). https://doi.org/10.1029/2010wr009998
10.1029/2010WR009998
PubMedWeb of Science®Google Scholar
Thornton, P. E., Running, S. W., & White, M. A. (1997). Generating surfaces of daily meteorological variables over large regions of complex terrain. Journal of Hydrology, 190(3), 214–251. https://doi.org/10.1016/s0022-1694(96)03128-9
10.1016/S0022-1694(96)03128-9
ADSWeb of Science®Google Scholar
Wang, B., Zhang, G., & Duan, J. (2015). Relationship between topography and the distribution of understory vegetation in a Pinus massoniana forest in Southern China. International Soil and Water Conservation Research, 3(4), 291–304. https://doi.org/10.1016/j.iswcr.2015.10.002
10.1016/j.iswcr.2015.10.002
Google Scholar
Wang, C., Wang, J., Naudiyal, N., Wu, N., Cui, X., Wei, Y., & Chen, Q. (2022). Multiple effects of topographic factors on spatio-temporal variations of vegetation patterns in the three parallel rivers region, southeast Qinghai-Tibet plateau. Remote Sensing, 14(1), 151. https://doi.org/10.3390/rs14010151
10.3390/rs14010151
ADSGoogle Scholar
Wang, Y.-H., Broxton, P., Fang, Y., Behrangi, A., Barlage, M., Zeng, X., & Niu, G. (2019). A wet-bulb temperature-based rain-snow partitioning scheme improves snowpack prediction over the drier western United States. Geophysical Research Letters, 46(23), 13825–13835. https://doi.org/10.1029/2019gl085722
10.1029/2019GL085722
ADSWeb of Science®Google Scholar
Wei, L., Wang, Y., Liu, S., Zhang, G. J., & Wang, B. (2021). Distinct roles of land cover in regulating spatial variabilities of temperature responses to radiative effects of aerosols and clouds. Environmental Research Letters, 16(12), 124070. https://doi.org/10.1088/1748-9326/ac3f04
10.1088/1748-9326/ac3f04
ADSGoogle Scholar
Wijngaard, R. R., Herrington, A. R., Lipscomb, W. H., Leguy, G. R., & An, S. I. (2023). Exploring the ability of the variable-resolution community earth system model to simulate cryospheric–hydrological variables in high mountain Asia. The Cryosphere, 17(9), 3803–3828. https://doi.org/10.5194/tc-17-3803-2023
10.5194/tc-17-3803-2023
ADSGoogle Scholar
Wood, E. F., Roundy, J. K., Troy, T. J., van Beek, L. P. H., Bierkens, M. F. P., Blyth, E., et al. (2011). Hyperresolution global land surface modeling: Meeting a grand challenge for monitoring Earth's terrestrial water. Water Resources Research, 47(5). https://doi.org/10.1029/2010wr010090
10.1029/2010WR010090
PubMedWeb of Science®Google Scholar
Wu, Y., Nair, U. S., Pielke, R. A., Sr., McNider, R. T., Christopher, S. A., & Anantharaj, V. G. (2009). Impact of land surface heterogeneity on mesoscale atmospheric dispersion. Boundary-Layer Meteorology, 133(3), 367–389. https://doi.org/10.1007/s10546-009-9415-1
10.1007/s10546-009-9415-1
ADSGoogle Scholar
Yang, X., Thornton, P., Ricciuto, D., Wang, Y., & Hoffman, F. (2023). Global evaluation of terrestrial biogeochemistry in the Energy Exascale Earth System Model (E3SM) and the role of the phosphorus cycle in the historical terrestrial carbon balance. Biogeosciences, 20(14), 2813–2836. https://doi.org/10.5194/bg-20-2813-2023
10.5194/bg-20-2813-2023
CASADSGoogle Scholar
Zhang, T., Li, B., Yuan, Y., Gao, X., Sun, Q., Xu, L., & Jiang, Y. (2018). Spatial downscaling of TRMM precipitation data considering the impacts of macro-geographical factors and local elevation in the Three-River Headwaters Region. Remote Sensing of Environment, 215, 109–127. https://doi.org/10.1016/j.rse.2018.06.004
10.1016/j.rse.2018.06.004
ADSWeb of Science®Google Scholar
Zhou, T., Leung, L. R., Leng, G., Voisin, N., Li, H., Craig, A. P., et al. (2020). Global irrigation characteristics and effects simulated by fully coupled land surface, river, and water management models in E3SM. Journal of Advances in Modeling Earth Systems, 12(10), e2020MS002069. https://doi.org/10.1029/2020ms002069
10.1029/2020MS002069
ADSWeb of Science®Google Scholar
Zorzetto, E., Malyshev, S., Chaney, N., Paynter, D., Menzel, R., & Shevliakova, E. (2023). Effects of complex terrain on the shortwave radiative balance: A sub-grid-scale parameterization for the GFDL earth system model version 4.1. Geoscientific Model Development, 16(7), 1937–1960. https://doi.org/10.5194/gmd-16-1937-2023
10.5194/gmd-16-1937-2023
ADSGoogle Scholar

Volume16, Issue8

August 2024

e2023MS004064