The Weather Research and Forecasting (WRF) model version 4.0 is used in this study (Powers et al., 2017). WRF was chosen because of its widespread use in the investigation of atmospheric and land processes and contextualization of observations in complex terrain (Rasmussen et al., 2011, 2014). The WRF model is a fully coupled atmospheric and land surface model with a range of user-specific options for subgrid-scale physics schemes. WRF is a regional climate model that requires boundary and initial conditions provided by either global climate model (GCM) outputs or atmospheric reanalyses datasets. Our configuration of the WRF model is designed with three nested domains, with an outer, middle, and inner domain at a grid resolution of 4.5, 1.5, and 0.5 km, respectively, centered around Crested Butte, Colorado, where the East River watershed is located (Fig. 1). All WRF simulations are initialized on 15 September 2018, but we discard the first 15 d of each simulation as spin-up.

While the stand-alone WRF model has been used extensively to advance the understanding of atmospheric processes, it has lower fidelity and applicability to investigate surface-through-subsurface hydrologic processes and consequently is limited as an assessment and modeling tool for understanding integrated mountainous hydrologic cycle. Therefore, in order to provide an estimate of the entire hydrologic budget, we use a one-way coupling between WRF and an integrated hydrologic model, ParFlow-CLM (Maxwell et al., 2015, described in further detail below), that simulates the hydrological response of key variables not otherwise quantifiable in standalone WRF, such as discharge and groundwater storage.

Figure 2 summarizes this approach graphically. It shows that the one-way coupling enables an exploration of sensitivities of modeled hydrologic quantities (many of which can be observed) to combinations of atmospheric, surface, and subsurface process representations. We do not choose a single configuration of WRF or ParFlow-CLM for this one-way coupling but rather explore the uncertainty in representing atmospheric processes for integrated mountainous hydrology by analyzing simulations with multiple, plausible configurations with multiple, plausible meteorological forcings. We recognize that the output from WRF simulations may be dependent on initial conditions, which are inherently difficult to constrain (e.g., Walser and Schär, 2004), but the experimental configuration described here seeks to be insulated from that dependency by running WRF simulations with initial conditions derived from different meteorological forcings.

A major experimental design decision when simulating the integrated mountainous hydrologic cycle is the computational cost associated with the simulations (e.g., simulated years per actual day) that are determined by model horizontal, vertical, and time step resolutions as well as subgrid-scale physics parameterization complexity. The computational expense incurred here to explore the sensitivities of WRF configuration choices was significant: 1 simulated year requires approximately 100 000 CPU hours on Lawrence Berkeley National Laboratory's Lawrencium lr6 supercomputing system. As such, it was highly impractical to simulate the entire configuration space of meteorological forcing and subgrid-scale parameterization choice. A discrete subsample of configurations, as presented here, is used to isolate and systematically determine which combination of subgrid-scale parameterization choice is superior for a given domain such as the ERW. We therefore adopted a parsimonious approach to explore the space of possible WRF configurations, described below.

ParFlow is a physically based surface–subsurface hydrologic model that solves the coupled flow of saturated and variability-saturated groundwater and overland surface water (Ashby and Falgout, 1996; Jones and Woodward, 2001; Maxwell, 2013). The three-dimensional form of Richards' equation is used to solve for lateral and vertical groundwater flow in the subsurface, and the kinematic wave approximation is used to solve two-dimensional overland flow. ParFlow is coupled to the land surface model, the Common Land Model (CLM), which calculates a coupled water energy balance at every surface cell of the domain (Dai et al., 2003) and incorporates spatially distributed vegetative processes by including specified land use types parameterized by the International Geosphere-Biosphere Program standard database. Hourly meteorological forcing derived from WRF drives ParFlow-CLM and includes the following eight variables: precipitation, 2 m surface air temperature, longwave radiation, shortwave radiation, 10 m east–west and south–north wind speeds, atmospheric pressure, and specific humidity. We also forced ParFlow-CLM with PRISM precipitation and temperature fields by evenly distributing daily precipitation and temperature across a diurnal cycle of 24 h within a day.

The ParFlow-CLM subsurface domain is 30 m deep at 100 m horizontal resolution. The WRF outputs are re-gridded using bilinear interpolation to match the ParFlow-CLM grid cells. The model parameters are based on a variety of geological and soil parameters and calibrated using streamflow measurements. More details can be found in Foster and Maxwell (2019) and Pribulick et al. (2016). The computational expense of ParFlow-CLM is also less substantial than that of WRF for this model configuration but still requires high-performance computing. Excluding the time for a multi-year initial condition spinup, a single water year of the ParFlow-CLM simulations on 64 cores on the NERSC's Cori supercomputing system is approximately 1000 CPU hours.

The Parameter-elevation Relationships on Independent Slopes Model (PRISM) dataset (Daly et al., 2008) was used here as a point of comparison in evaluating model uncertainty across subgrid-scale physical schemes and meteorological forcing datasets for precipitation and temperature. PRISM uses observations from quality-controlled meteorological stations along with a topographic correction method against elevation based on empirical regressions to create daily gridded 800 m total precipitation and daily average, minimum and maximum 2 m surface temperature. Although PRISM was generated using statistical models, it has been widely used for climate and hydrological model assessments (e.g., Lundquist et al., 2019) and associated uncertainty analyses (e.g., Buban et al., 2020). In the assessment of subgrid-scale physics schemes and meteorological conditions, the percent difference in cumulative precipitation is compared against PRISM by calculating by (max - min) / min ⋅ 100, where max and min are the maximum and minimum cumulative precipitation values from the simulations within each group, respectively.

Snowpack Telemetry (SNOTEL) data have been widely used in snowpack assessment (Serreze et al., 1999; Fassnacht et al., 2003), and we use three SNOTEL stations (Butte, Schofield Pass, Upper Taylor) within the WRF inner domain to assess the snowpack simulation skill of each IPM configuration. Significant heterogeneity is sampled by the three SNOTEL stations (within or near the ERW) due to the complex topography. For example, the Butte station is located downstream of the ERW and, on average, receives approximately 0.8 m of precipitation and reaches 0.4 m in maximum snow water equivalent over the year. On the other hand, the Schofield Pass station is located upstream of the ERW and, on average, receives 1.2 m of precipitation and reaches 0.9 m in maximum snow water equivalent. In addition, we use the snow water equivalent product of the Airborne Snow Observatory (ASO; Painter et al., 2016) on 7 April 2019 to evaluate the spatial pattern skill of the snowpack simulation across WRF configurations (Fig. S8 in the Supplement). The raw ASO product has 50 m spatial resolution and is regridded to the same grid resolution as WRF outputs (500 m) for comparison purposes using bilinear interpolation, as documented in Oaida et al. (2019). Since the spatial resolution of ASO data is significantly finer than the WRF outputs, we acknowledge that the underestimation by ASO could be due to the point-to-grid errors (Oaida et al., 2019). Notably, ASO SWE estimates are lower than SNOTEL SWE measurements (ASO: 389 mm at Butte, 938 mm at Schofield Pass; SNOTEL: 490 mm at Butte, 1260 mm at Schofield Pass). In addition to SNOTEL station data, stream gauge measurement of discharge at the pump house, the outlet of the ERW, is used to evaluate the ParFlow-CLM simulation results.

We start by presenting a number of time series of spatial averages over the ERW for WY19. They indicate the gross performance of the IPM across the water year and whether a configuration produces generally reasonable estimates relative to observational products. Figure 3 shows cumulative precipitation, 2 m surface air temperature, and snow water equivalent (SWE) aggregated over the ERW, and the in situ assessments compared against two SNOTEL stations are in Fig. S3. For cumulative precipitation, each configuration produces amounts higher than PRISM (cumulative precipitation of 1201 mm), and the UCD simulates the highest cumulative precipitation. For surface air temperature, the seasonal cycle and daily variability are captured by all configurations; however they exhibit systematic cold biases relative to PRISM (annual average 2 m surface air temperature of 0.6 ∘C). In terms of SWE, all model configurations concur in their representation of the snowpack accumulation season and melt season in late spring and into summer, except UCD, which simulates an earlier peak timing of SWE.

The spread in cumulative precipitation when comparing across different meteorological forcing datasets is apparent (Fig. 3). Although UCD and NCAR configurations show a greater difference in precipitation forced by ERA5 and CFSR2, the consistency across BSU configurations is notable, which also shows the closest agreement with PRISM. When comparing the relative roles of subgrid-scale physics scheme choices to meteorological forcings, the percent difference of cumulative precipitation, calculated by (maximum - minimum) / minimum ⋅ 100, across BSU-CSFR2, UCD-CSFR2, and NCAR-CSFR2 schemes is nearly 34 % of the minimum cumulative precipitation simulated by BSU-CFSR2, compared to the 4.6 % of the simulations across BSU configurations with different meteorological forcing (CSFR2, ERA5, MERRA2, and NCEP).

BSU simulations are generally in agreement with PRISM. However, the UCD simulations are outliers relative to the other simulations, with cumulative precipitation of 1706 mm, or 42 % higher at the end of the water year, with the most notable differences occurring in March through September. NCAR simulations show general agreement with PRISM and BSU throughout the water year, save for June through September. The 2 m surface air temperature time series reveals that the UCD simulation is systematically colder throughout the winter and spring, regardless of which meteorological forcing dataset is used. The persistent cold bias simulated by the UCD, NCAR, and BSU schemes has been found in previous WRF studies within western US mountain regions (Xu et al., 2018; Rudisill et al., 2021). The SWE time series again shows a similar relationship with precipitation, with the outlier being UCD-ERA5, in terms of the seasonal timing of when snowpack peaks and melts (Fig. S1). Comparing the monthly average between UCD-ERA5 (Fig. S4) and BSU-ERA5 (Fig. S5), the early snowmelt observed in the UCD scheme is likely a result of warmer temperatures in the low-altitude region that melt the snow earlier in the water year. However, the high-altitude regions remain cold enough to maintain snowpack through early summer to midsummer.

In addition to the domain averages, spatial heterogeneity due to land surface cover and topographic effects is shown in Fig. 4. The systematic cold bias simulated throughout the water year appears to be an elevation-dependent phenomenon, with higher elevations exhibiting an enhanced cold bias compared with PRISM. However, the river valley and relatively lower-elevation areas at the southern edge of the ERW, which includes Crested Butte Mountain, stand out as these regions are warmer than the PRISM dataset. Figure 4b shows precipitation in BSU-CFSR2 is wetter in the western regions and drier in the eastern regions of the ERW in comparison to PRISM. Figure S3 shows comparisons between PRISM and the IPM configurations and indicates no biases that are persistent across seasons. During summer, the BSU-CFSR2 simulation consistently produces more precipitation than PRISM.

Although the 2 m surface air temperature bias is evident, it does not vary significantly across either subgrid-scale physics scheme or meteorological forcing. Therefore, subsequent exploration in this study will be focused on precipitation. The bottom row in Fig. 4 shows the grid-cell standard deviation of monthly precipitation across subgrid-scale physics schemes (i.e., UCD, NCAR and BSU simulations with CFSR2 meteorological forcing – bottom left) and BSU simulation driven by different meteorological forcing datasets (ERA5, CFSR2, MERRA2, and NCEP – bottom right). Similar to the conclusions drawn from Fig. 3, Fig. S6 also shows that the monthly spatial standard deviations across subgrid-scale physics schemes are generally greater than meteorological forcing, particularly in regions of higher elevation during the winter season.

Quantitative statistics of the aggregated domain-average precipitation and temperature simulations for the WRF simulation across subgrid-scale physical schemes and large-scale meteorological forcings are presented in Table 3. Although NCAR-CFSR has a higher R2 than other simulations, NCAR-ERA5 has a very low R2. The BSU simulations provide a closer approximation of cumulative precipitation to PRISM.  Specifically, BSU does better in simulating extreme precipitation events (i.e., 95th percentile). Therefore, we conclude that BSU WRF subgrid-scale physics schemes outperform the UCD and NCAR WRF subgrid-scale physics schemes in simulating both precipitation and temperature. On the other hand, the differences in precipitation and 2 m surface air temperatures across the four meteorological forcings are not statistically significant, and their standard deviations are much smaller than the differences in simulations across subgrid-scale physical schemes. While there are many metrics of model skill when selecting a meteorological forcing to simulate the hydrological processes in the ERW, we choose BSU-CFSR for the topographic radiation study in the next subsection due to its better match with PRISM, using our skill measures, in simulating both precipitation and 2 m surface air temperature.

Based on the assessment of simulated precipitation and 2 m surface air temperature compared with PRISM, the BSU-CFSR2 configuration is selected as a baseline to further explore the influence of topographic radiation scheme effects. The difference caused by turning on and off the 3D topographic radiation effects is similar in other WRF configurations; therefore, only the BSU-CFSR is presented. Figure 5 shows daily ERW spatial average time series over the water year for the major mountainous water and energy budget variables. By isolating the impacts of subgrid-scale physics schemes and meteorological forcings across IPM simulations, it is easier to systematically intercompare cause and effect across different topographic radiation options. Consistent with previous findings, all configurations still overestimate cumulative precipitation and are too cold relative to PRISM.

Figure 6 shows the seasonally resolved shortwave radiation, 2 m surface air temperature, latent heat flux, and SWE for different configurations of shortwave radiation in the simulation with and without topo_shading and slope_rad options in the inner domain. While no3DRad does not adjust the SWdown (incoming shortwave radiation), the 3DRad simulation recalculates the SWdown based on the shadows cast by nearby topography. Spatial differences in IPM-simulated shortwave radiation (Fig. 6b) are seen in the northeast and western portions of the ERW, when the topographic effect of shortwave radiation is included. As a result, a corresponding change in the spatial pattern of simulated 2 m surface air temperature and latent heat flux is seen, driven by the change in downwelling shortwave radiation with topographic shading (Fig. 6a and d). Topographic shading makes a difference locally in latent heat (LH) flux, by redistributing the energy flux and thus affecting LH flux spatial distribution. Nevertheless, the domain-average LH flux remains unchanged between cases. The resulting pattern change in SWE (Fig. 6c) shows that the northern and northeastern sections of the ERW, where snowpack is concentrated, are sensitive to shortwave radiation. This is expected and consistent with previous findings that included topographic effects in shortwave radiation and found distinct spatial patterns of hydrologic variable sensitivity due to both shadows and surface reflection that produce time-varying effects on net surface radiation (Lee et al., 2015; Palazzi et al., 2019; Gu et al., 2020; Hao et al., 2021).

Although Fig. 5 shows that realistic shortwave radiation produces small effects on the seasonal cycle of the surface energy and mass budgets when averaged over the entire watershed, including annual average SWE (Fig. 5c), Fig. 6c shows that mountains and valleys have different amounts of SWE. Furthermore, seasonal patterns show simulated latent heat is diminished at lower elevations from March to May, when snowmelt occurs in the valley, and the remaining snowpack in the mountains and late snowmelt in the 3DRad simulation cause the lower latent heat flux shown in July (Fig. S7). The 3D radiation shading scheme does not significantly affect the total water balance but rather the spatial distribution of radiation fluxes. Thus, despite having minimal impacts on water impacting on the water balance, the scheme does have important localized impacts on SWE and surface energy budget spatial patterns. The 3DRad simulation has less SWE in the valleys during the accumulation season but more SWE at higher elevations during the melt season, which is a direct result of the differences in shortwave radiation redistribution. Figure S5 also shows that the latent heat differences in north-facing and south-facing sides are most apparent in the snowmelt and warm seasons. This is consistent with previous findings (Lee et al., 2015; Palazzi et al., 2019; Gu et al., 2020; Hao et al., 2021), that a more realistic treatment of shortwave radiation, which includes shadows and projected insolation on sloped surfaces, results in lower shortwave insolation on the surface at this time of year. The lower shortwave radiation should, in turn, decrease the energy available for the IPM to produce snowmelt. In summary, the simulations show that, while local spatial differences in surface radiation with and without realistic topography are apparent in Fig. 6, the domain spatial averages (even for SWE) are the same between shaded and non-shaded formulations. This suggests that while localized differences may be highlighted when shading is included, the impact of topographic shading on the entire water balance over a spatial domain like the ERW is negligible.

We have evaluated the aforementioned WRF configurations' subgrid-scale physical scheme and large-scale meteorological forcings in representing precipitation, temperature, snowpack, and radiation fluxes and their impacts on the integrated water budget within the ParFlow-CLM. We also evaluated the simulated discharge from ParFlow-CLM forced by PRISM as a comparison with WRF forcings. Figure 7 shows the simulated hydrologic output from the ParFlow-CLM model for watershed outlet discharge (top row) and watershed-average groundwater storage (bottom row). Discharge at the watershed outlet (see exact location in Fig. 1) shows a different timing across the various WRF subgrid-scale physics scheme configurations and large-scale meteorological forcings that lead to a temporal shift in simulated streamflow, where the daily averaged time series (left) shows only minor differences through time. However, cumulative discharge by the year-end reveals substantial differences (right), especially after peak snowmelt where estimates of cumulative discharge begin to diverge. Differences across the WRF configurations are especially large; the difference across the three subgrid-scale physics scheme configurations with ERA5 (UCD, NCAR, and BSU) varies by 26 % by the year-end. Differences across meteorological forcing (using the BSU physics configuration as a control, shown in green) are also noteworthy, although smaller, approximately 6 %. These results are consistent with the variation of simulated precipitation in WRF described earlier, confirming that for this basin, meteorological forcing drives less variance on the hydrologic response than the subgrid-scale physics scheme configuration.

In addition to the variance of cumulative discharge with WRF simulations across different conditions, a comparison to observed discharge is also shown in Fig. 7, which for all scenarios suggests a delayed snowmelt response in the IPM. While the objective of this study is not to replicate the observations, but rather determine sensitivity across IPM configuration choice, the mismatch in streamflow response suggests a systematic cold bias from the WRF input into ParFlow-CLM, which is consistent with the discussion surrounding Fig. 3 in relationship to PRISM. An early-fall peak in simulated discharge is also seen in all WRF simulations, and not in observed discharge, although a significant increase in SNOTEL precipitation was measured in October of that year (see Figs. S1, S2). This further supports a temperature bias, albeit opposite that of the cold bias discussed previously, where precipitation around that October storm event falling as rain (as opposed to snow) leads to a sharp increase in discharge. A sensitivity analysis of the BSU-ERA5 model run for a lower precipitation year (water year 2018, which was nearly half the precipitation of 2019) showed better agreement with observed discharge, which suggests the bias in timing may be a function of accumulated precipitation and/or snowmelt, and this is reserved for future studies (not shown).

Basin-average groundwater storage, shown in Fig. 7c in area-normalized units, shows a strong annual signal for all WRF configurations with minimal differences across IPM configurations. Here all groundwater, inclusive of saturated or unsaturated storage, is considered. The cumulative, area-normalized annual groundwater storage, when accounting for only vadose zone storage (Fig. 7d), which is most responsive to sub-annual differences in precipitation inputs, is meaningful in this context because it relates a cumulative impact on near-surface groundwater storage due to IPM configuration. Similar to the year-end cumulative discharge, year-end departures in vadose zone groundwater storage across the different simulations are evident. Differences across the IPM configurations of subgrid-scale physics schemes are larger than the difference across the forcing simulations (4 % versus 2 %, respectively). While the differences in groundwater signals are not as pronounced as the discharge signals, streamflow signals are very reactive and noisy and change quickly, whereas groundwater signals are the product of slower processes via infiltration and vadose zone dynamics, often at longer timescales, which result in very different temporal signals as compared to streamflow.

Figure 8 shows maps of standard deviations in annual total evapotranspiration (ET) simulated by ParFlow-CLM across IPM configurations (top row) and the cell-binned relationship of those standard deviations of annual ET with land use and cover type, as well as elevation (bottom). Consistent with variations shown in the simulated discharge and groundwater storage, ParFlow-CLM simulates greater variations of ET under WRF configurations driven by different subgrid-scale physics schemes (Fig. 8a), compared to the simulations conducted with different meteorological forcings (Fig. 8b). These results suggest that locations populated by high-water-demanding vegetation (namely evergreen and deciduous forests) at mid-elevations result in the highest ET variability across IPM configurations. Conversely, low-water-demanding vegetation (barren/sparsely vegetated land and grasses), which resides across a range of elevations in the study domain, results in the lowest variability in annual ET across IPM configurations. These differences in water demand essentially magnify any differences in atmospheric conditions.

In spite of previous efforts to characterize the sensitivity of WRF simulations to model configuration choices, the mountain climate and hydrology scientific community has not sufficiently explored the implications of those choices for surface and subsurface hydrology in high-altitude complex terrain. Here, we used an IPM with one-way feedbacks from WRF to ParFlow-CLM to assess the hydrometeorology of the ERW, which is characterized by strong hydrological gradients indicative of mountain environments of the UCRB.

In this paper, we present a number of numerical experiment results that are informative for the scientific community to better understand atmosphere-through-bedrock process interactions and the uncertainties of those interactions between climate and hydrological model experimental setup choices. First, the uncertainties associated with meteorological forcing choice are less important than subgrid-scale physics scheme choice, at least in the ERW. This finding has important implications for IPM in complex terrain, since it reveals that the differences in reanalysis products are less consequential for initializing and forcing IPMs than atmospheric configurations and that efforts to advance IPMs such as collecting observations and using them to evaluate physical process parameterizations at the sub-HUC-8 (hydrologic unit code) scale could help to better constrain model performance. This result also shows that the large-scale meteorological forcing of the IPM simulation is less important in driving the magnitude and spatial variability of key hydrometeorological variables than the details in choosing and optimizing atmospheric subgrid-scale physics schemes (e.g., microphysics or boundary layer turbulence). Ultimately, we used the BSU-CFSR2 configuration to recreate WY2019 in the ERW, which allows researchers, in this case, to prioritize process studies and the development of associated observational constraints within the ERW. However, further investigation is needed to evaluate the systemic cold bias across IPM configurations, particularly at higher elevations, and the consequence of delayed snowmelt and timing of discharge peaks.

We recognize that numerous works in meteorological disciplines have demonstrated that “physical parameterization is much more important than lateral or initial conditions” (e.g., Solman and Pessacg, 2012; Pohl et al., 2011). However, our findings are not redundant with the published literature, as those references either evaluated large-scale meteorological processes or did not focus on high-altitude complex terrain regions, which are central to our study. Additionally, most IPM studies to date do not show how the range of reasonable IPM configurations (based on configurations that have been presented in the published literature) affects water-management-relevant processes such as discharge, ET, and subsurface hydrology. With our set of one-way atmosphere-through-bedrock process modeling results, we now show how choices in atmospheric process model configurations impact the surface and subsurface hydrology. Specifically, we evaluate and quantify the sensitivity of discharge, ET, and subsurface hydrology to IPM configurations, and we also address how 3D topographic radiation schemes affect both the spatial distribution and spatial average aspects of the mountainous hydrologic budget.

In the investigation of topographical and slope gradient effects on shortwave radiation, our study shows those considerations in WRF are essential in redistributing radiation flux over regions of complex terrain, even though the differences in spatial-average performance over ERW are minimal. This is because the spatial redistribution of shortwave radiation leads to approximately ±30 W m-2 difference in the east-/west-facing slopes that lead to ±1 K difference in 2 m surface air temperature in August and September (when snowpack is nonexistent). Throughout most of the water year when snowpack exists, the spatial heterogeneity of temperature differences is less apparent than for shortwave radiation. Latent heat buffers the differences of the shortwave radiation contribution to the radiation budget and causes early snowmelt in the high-elevation mountains in those simulations with topographical and slope gradient shortwave radiation effects turned on. At the same time, the systemic cold bias and limitations of one-way feedback in this study are potentially indicative of challenges in extrapolating findings from one mountainous watershed to another. If atmospheric process details are significant for surface and subsurface hydrological modeling, and if the findings regarding atmospheric processes in one study area are marginally or completely irrelevant to other mountainous watersheds, then additional fieldwork would be needed in mountainous hydrology research to address this issue, given that the extrapolation of fieldwork results remains a central challenge for field-based research and modeling activities.

A limitation of our study, given the computational constraints of running IPMs, is that it was infeasible to explore the full parameter spaces of WRF and ParFlow-CLM exhaustively; thus, our conclusions are limited to the selected subgrid-scale physics schemes and meteorological forcing datasets analyzed. Additional work is needed to improve the systemic cold bias in 2 m surface air temperature throughout all experiments as this may have been the major driver in the delayed snowmelt and peak discharge simulated by the IPM.

Another methodological constraint is that our WRF and ParFlow-CLM experiments were only one-way instead of two-way feedbacks, which ignores potentially important feedbacks from the subsurface hydrology to the atmosphere via ET and the radiation budget. For example, Givati et al. (2016) reported that simulated precipitation was improved with two-way coupling in WRF-Hydro compared to WRF-only, and Forrester et al. (2018) showed that boundary layer dynamics were impacted in IPM simulations in regions where shallow water tables exist. On the other hand, ParFlow-CLM is essential in our experiment for quantifying hydrological responses, including streamflow and groundwater storage. Although another fully coupled integrated hydrology model (i.e., WRF-Hydro) provides some insights into streamflow, it still uses a simplified and prescribed stream network. Groundwater storage in WRF-Hydro is also highly simplified, using a bucket model, while ParFlow-CLM simulates the full 3D continuum of variable saturation in three dimensions. Importantly, in a similar fashion to the hierarchy of climate models approach oft used in the climate community (Jeevanjee et al., 2017), we would also like to assess one-way coupling performance of our IPM prior to assessing two-way coupling IPM performance.

The East River watershed is already highly instrumented due to the presence of the long-standing Rocky Mountain Biological Laboratory (RMBL), the SNOTEL network, the United States Geological Survey's Next Generation Water Observing System (NGWOS), the National Science Foundation's Sublimation of Snow (SOS) project, and the DOE Watershed Science Focus Area project, which has been adding instrumentation to the watershed over the last ∼7 years. While these observations focus primarily on surface and subsurface processes, the East River watershed has become even more instrumented in recent years (2021–2023) through the support of the US DOE (SAIL campaign) and US NOAA (SPLASH campaign) deployments of a comprehensive set of atmospheric instrumentation (e.g., radar and radiation measurements). Future work will include integration of data, either indirectly through IPM benchmarking or directly through data assimilation into the IPM, from the SAIL campaign. SAIL is collecting a wide array of observations with the intent to advance understanding of precipitation, snow, aerosol, aerosol–cloud interaction, and radiation processes in complex terrain and establish the minimum but sufficient level of process understanding to develop a robust predictive understanding of seasonal surface water and energy budgets in the ERW (Feldman et al., 2021). SAIL aims to develop a wide range of hydrometeorological datasets to constrain atmosphere, surface, and subsurface processes simultaneously. Together, these resources are contributing to the establishment of a highly instrumented and in-depth-studied UCRB watershed. We look forward to building upon the knowledge learned from this paper to compare the most appropriately configured IPM to SAIL and SPLASH campaign observations. Our study highlights that the benchmarking provided by these data collections will be critical in addressing the systemic IPM cold bias by providing a more constrained estimate of radiation budgets in complex terrain that ultimately shape snowmelt and discharge.