Integrating meteorological forcings and satellite-derived observations to emulate snow water equivalent in the Colorado Rocky Mountains using a spatiotemporal deep learning model
A deep learning–based emulator is developed to reproduce SWE simulated by Noah-MP using a Convolutional Long Short-Term Memory (ConvLSTM) network trained on meteorological forcings from the Weather Research and Forecasting model and augmented with static physiographic variables and remotely sensed snow cover and leaf area index data.
Abstract
Accurate estimates of snow water equivalent (SWE) are critical for hydrologic forecasting and water resource management in snow-dominated regions. While in situ observations, such as those from the United States Department of Agriculture snow telemetry (SNOTEL) network, provide direct measurements, their sparse spatial coverage limits their broader applicability. Numerical models such as Noah Multi-Parameterization (Noah-MP) provide spatially continuous SWE estimates but are computationally demanding. This study develops a deep learning–based emulator to reproduce SWE simulated by Noah-MP using a Convolutional Long Short-Term Memory (ConvLSTM) network trained on meteorological forcings from the Weather Research and Forecasting model and augmented with static physiographic variables and remotely sensed snow cover and leaf area index data. Incorporating spatial coupling substantially improves performance relative to a pixel-wise Long Short-Term Memory baseline, increasing the fraction of grid cells with non-negative Nash–Sutcliffe Efficiency (NSE) by 15–27% during the test period. The emulator reproduces large-scale SWE patterns consistent with both Noah-MP and Snow Data Assimilation System (SNODAS) and maintains competitive NSE and root mean square error performance relative to SNOTEL observations. Seasonal sensitivity experiments yield high
R
2
values up to 0.96 during accumulation and 0.97 during melt. Once trained, the emulator generates SWE predictions faster than the parent physics-based simulation, enabling rapid, high-resolution estimates that support real-time monitoring and ensemble forecasting.
This study presents an integrated framework combining water vapor (WV) imagery, simulated water vapor (SWV), and potential vorticity (PV) diagnostics to systematically identify errors in numerical weather prediction (NWP). The analysis employs a radiance-space comparison among Geo-Kompsat-2A (GK2A) WV observations, model-simulated WV fields, and dynamically coherent PV structures. Model-equivalent WV images were generated from outputs of the Korean Integrated Model (KIM) using the Radiative Transfer for TOVS (RTTOV) model under all-sky conditions. Phase-displacement vectors and phase-corrected brightness-temperature difference fields were derived using a variational echo-tracking technique built upon the McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation (MAPLE). PV fields derived from KIM were composited with WV imagery to establish a dynamically consistent reference framework linking upper-tropospheric moisture structures with tropopause-level flow features. Quantitative validation over East Asia during a 1-year period (January 2025–January 2026) showed systematic reductions in root-mean-square error (RMSE) and increases in spatial correlation after phase correction across all WV channels and forecast lead times, confirming the reliability of the displacement vectors. The framework was further evaluated through two midlatitude weather events representing different manifestations of upper-level dynamical development, demonstrating its capability to diagnose forecast-error characteristics using physically consistent WV–PV–SWV relationships.
Jun-Hyung Heo, Boram Kim, Minsang Kim et al.· Asia-Pacific Journal of Atmo...· 0 citations
Snow density influences Arctic climate, ecosystems, and surface energy exchange, yet spatially continuous observations remain limited. This study constructed an ERA5-supervised snow-density dataset for 60–90° N by collocating Advanced Microwave Scanning Radiometer 2 (AMSR-2) Level-1R brightness temperatures with ECMWF Reanalysis v5 (ERA5) snow density, Soil Moisture Active Passive (SMAP) surface roughness, and auxiliary variables. Ten models were evaluated using 29 observation days spanning September 2022–February 2023 under a chronological training–validation–test split. Extra Trees achieved the best overall performance, with a root mean square error of 18.54 kg m−3 and an R2 of 0.87, while the bidirectional gated recurrent unit (BiGRU) was the strongest deep-learning model. Feature-attribution and ablation analyses showed that microwave brightness temperatures contained predictive information, although geographic and auxiliary variables also contributed substantially. The evaluated models could reproduce ERA5-referenced Arctic snow-density patterns, but their performance partly reflected regional information. Moreover, ERA5 showed limited consistency with station-based Northern Hemisphere Snow Water Equivalent estimates. Consequently, the reported metrics quantify agreement with ERA5 rather than accuracy against independently observed snow density. Temporally coincident and spatially independent field validation remains necessary in the future.
Jianjun Zhang, Wentao Zhou, Shuhu Yang et al.· Journal of Marine Science an...· 0 citations
Accurately estimating Snow Water Equivalent (SWE) in glacierized regions is critical for glacier modeling and mass balance analysis yet remains challenging due to sparse observations and uncertainties in precipitation products. The objectives are twofold: (a) assess how well three widely used precipitation datasets—Stage IV (a radar–gauge composite product), Integrated Multi‐satellite Retrievals for GPM version 07 (IMERG V07, a satellite‐based global product), and ERA5 (a global atmospheric reanalysis)—represent end‐of‐season SWE over seven Alaskan glaciers across two winter seasons, using airborne radar SWE retrievals (ARBN) from NASA's Operation IceBridge campaign as the reference, and (b) determine whether combining precipitation with auxiliary variables via machine learning (here XGBoost) modeling can reliably predict SWE beyond airborne coverage. Results reveal systematic biases: Stage IV generally overestimates SWE, IMERG V07 underestimates it, and ERA5 aligns most closely with ARBN. Model performance depends strongly on training data representativeness, with significant discrepancies arising when training and testing SWE distributions diverge. The most effective framework integrates IMERG V07, ERA5 and key auxiliary variables–snowfall fraction, total precipitable water and 2‐m air temperature–with IMERG V07 and ERA5 found to be the dominant predictors. Nevertheless, accuracy is constrained by sparse in situ measurements and the paucity of high‐resolution data in complex terrain. These findings underscore the need for high‐quality targeted observations and integrated strategies that leverage diverse precipitation data and representative training data sets to advance SWE estimation and assessment in cryospheric environments.
Yang Song, A. Behrangi· Earth and Space Science· 0 citations
Quantifying the spatiotemporal distribution of mountain snowpack is critical for hydrology, as it represents a vital freshwater resource for over a billion people. While C-band synthetic aperture radar (SAR) from Sentinel-1 provides high-resolution estimates over mountains, its use for daily snow monitoring is hindered by infrequent revisits and signal attenuation in wet snow. To address these limitations, this work proposes the adaptive snow-phenology shape model fitting (ASP-SMF), a physically constrained fusion algorithm that generates all-weather continuous daily 500-m snow-depth maps for Northern Hemisphere mountains. Designed for data-scarce regions, ASP-SMF integrates Sentinel-1, ERA5-Land, and Interactive Multisensor Snow and Ice Mapping System (IMS) data independently of in situ calibration. The algorithm first adaptively segments the highly dynamic snow time series into three distinct phenological phases using physical rules of snowfall and snowmelt. Subsequently, it utilizes Sentinel-1 retrievals and IMS snow-free constraints to correct the temporal, amplitude, and offset biases of the reanalysis data via a robust reduced major axis (RMA) regression and uncertainty-aware sequential least-squares programming (SLSQP) optimization. Comprehensive validation against 565 sites and 55 airborne LiDAR/photogrammetry surveys (2017–2021) demonstrates that ASP-SMF: 1) reduces errors relative to Sentinel-1 under dry snow conditions, reducing root-mean-square error (RMSE) by 26% (from 0.27 to 0.20 m) and increasing R2 from 0.49 to 0.72; 2) exhibits pronounced improvement under wet snow conditions, reducing RMSE by 23% (from 0.61 to 0.47 m) and boosting R2 by 145% (from 0.22 to 0.54); and 3) shows comparable performance to the SNOw Data Assimilation System (SNODAS) product without in situ data assimilation, with a higher Kling–Gupta efficiency (KGE) (0.57 versus 0.37) against Airborne Snow Observatory (ASO) LiDAR in the Rocky Mountains. Critically, this independence from in situ inputs enables ASP-SMF to deliver spatiotemporally consistent snow monitoring in ungauged regions, providing a useful framework for high-resolution assessment of mountain snow water resources.
Seasonal snow plays a crucial role in hydrology and soil–plant systems, and accurate estimation of snow water equivalent (SWE) is essential for sustainable water management and climate-smart agriculture. This study compares two approaches for generating monthly SWE estimates in the South Central Region (SCR) of Bulgaria for 2020: (i) process-based simulations using the Community Land Model (CLM 3), and (ii) archived SWE data from the Copernicus Global Land Component, derived from integrated in situ measurements and satellite observations.
The comparison shows good agreement between the two approaches in lowland areas, while satellite-based SWE estimates are unreliable in mountainous terrain, which represents about 40% of the study area.
The results obtained highlight the importance of integrating hydrological modeling with satellite data for estimating the water equivalent of snow cover in regions with limited observation capabilities and confirm the applicability of the Copernicus Global Land Component product for hydrological and hydrogeological forecasting, agricultural planning and climate change adaptation in Bulgaria. The assessment of the snow cover depth and its water equivalent during the winter months is important for predicting and managing the risk of hazardous geological and hydrological processes in the spring such as deep landslides and floods.
O. Nitcheva, Donka Shopova, Z. Mateeva et al.· Engineering Geology and Hydr...· 0 citations
Coastal and inland communities are increasingly vulnerable to tropical cyclones (TCs), yet accurately assessing flood risk remains challenging due to uncertainties in precipitation inputs. This study evaluates how three gridded precipitation products; gauge-based (GCOOP:Gridded Cooperative Observer Program), radar-based (Stage IV) and satellite-based (IMERG:Integrated Multi-satellitE Retrievals for GPM) influence surface hydrology, specifically total runoff, streamflow volume and timing of peak flow through a case study of TC Beryl 2012 in the South Atlantic-Gulf region. Using the Variable Infiltration Capacity (VIC) model and a GIS routing model, we evaluate how biases in daily precipitation estimates propagate into streamflow outputs. While precipitation differences of up to 186 mm in storm totals among datasets had minimal effect on peak streamflow timing, however, the peak discharge magnitudes varied with relative errors of 24% and 12% for IMERG compared to GCOOP and Stage IV, respectively. Results indicate that, surface runoff showed high sensitivity to precipitation variability, while baseflow and evapotranspiration exhibited buffering effects. The watershed-scale integration dampened local precipitation variability, resulting in a 2.4 cm/day streamflow difference between GCOOP and IMERG, which is acceptable for regional flood assessment. These findings show that, for this drought-preconditioned watershed and TC event, satellite precipitation products provide reasonable streamflow estimates, particularly when peak timing is more critical than absolute magnitude. Further evaluation across diverse TC events and watershed conditions is needed before broader operational conclusions can be drawn.
Alka Tiwari, K. Cherkauer· Journal of Hydrometeorology· 0 citations
