Improving Weeks 1-2 Temperature Forecasts in the Sierra Nevada Region Using Analog Ensemble Post-Processing with Implications for Better Prediction of Snowmelt, Water Storage, and Streamflow

October 1, 2025

A new study demonstrates that the Analog Ensemble (AnEn) post-processing method substantially improves Week-1 and Week-2 temperature forecasts at high spatial resolution (4 km) over California’s Sierra Nevada during the spring snowmelt season (April–July). This research is detailed in the paper titled “Improving Weeks 1-2 Temperature Forecasts in the Sierra Nevada Region Using Analog Ensemble Post-Processing with Implications for Better Prediction of Snowmelt, Water Storage, and Streamflow”, recently published in the Journal of Hydrometeorology. The study was conducted by Zhiqi Yang (CW3E), Weiming Hu (UGA), Agniv Sengupta (CW3E), Luca Delle Monache (CW3E), Michael J. DeFlorio (CW3E), Mohammadvaghef Ghazvinian (Lynker), Mu Xiao (CW3E), Ming Pan (CW3E), Jacob Kollen (CA DWR), Andrew Reising (CA DWR), Angelique Fabbiani-Leon (CA DWR), David Rizzardo (CA DWR), and Julie Kalansky (CW3E). This research was supported by the California Department of Water Resources (CA DWR) Atmospheric River Program. This work supports the Advanced Precipitation and Streamflow Prediction priority in CW3E’s 2025-2029 Strategic Plan by advancing our understanding of potential postprocessing methods that could be implemented in near real-time over California to improve the skill of subseasonal temperature prediction in the vicinity of B120 watersheds.

California relies on Sierra Nevada spring snowmelt for 60% of its water, serving 23 million people. Forecasting this snowmelt is vital for water supply planning and is a key task of the California Department of Water Resources’ Bulletin-120 (Cuthbertson et al. 2014). Accurate predictions rely on subseasonal 2-m temperature (T2m) forecasts, especially at high elevations where snowpack and runoff contributions are greatest. Current systems like the California Nevada River Forecast Center’s (CNRFC) Hydrologic Ensemble Forecast Service (HEFS) have identified T2m forecasts as a key uncertainty source (https://www.cnrfc.noaa.gov/documentation/hefsAtCnrfc.pdf). Replacing the current T2m product with a higher-accuracy dataset offers a straightforward and effective pathway for enhancing snowmelt and flood risk predictions. Collaborating with CA DWR to facilitate the integration of such advancements into regional applications could provide significant value and offer a foundation for real-time operational forecasting.

This study uses the Parameter-elevation Regressions on Independent Slopes Model (PRISM) dataset as ground truth and National Oceanic and Atmospheric Administration (NOAA) Global Ensemble Forecast System (GEFS) reforecasts to apply AnEn post-processing (Delle Monache et al. 2013), producing high-resolution (4-km) daily T2m forecasts for the Sierra Nevada. Since spring snowmelt and its streamflow contribution are elevation-dependent (Hunsaker et al. 2012; Musselman et al. 2017), we also evaluate AnEn performance across elevation bands from 0 to 3500 m in 500 m increments.

We find that during the spring snowmelt season (April–July), AnEn post-processing significantly improves T2m forecasts by reducing RMSE by up to 1°C (60% at 1-day leads and 20% at 15-day leads), lowering CRPS by 1.1°C across all lead times, increasing correlation by 11%, and extending predictive skill by up to one week beyond the dynamical benchmark GEFS (Figure 1). It also demonstrates clear advantages over a basic bias correction method (the Ensemble Model Output Statistics, EMOS), particularly during week 1, where EMOS performance is limited by growing spatial noise and is less spatially consistent. Moreover, The AnEn method demonstrates a substantial improvement in skill across all elevation bands, particularly at higher elevations (e.g., 3000–3500 m), where it yields even more notable enhancements (Figure 2). Specifically, RMSE decreases by approximately 4°C, CRPS reduces by approximately 4.5°C, the correlation increases from 0.1 to 0.9 over 3000–3500 m, and the skill period extends by two weeks. While EMOS also improves RMSE and CRPS, its performance is consistently weaker than AnEn’s, particularly during the first week. Since a 5°C shift can significantly influence snowmelt timing, by correcting the bias in temperature forecasts, AnEn could improve the accuracy of snowmelt timing predictions by aligning them more closely with observed snowmelt dates, which could potentially enhance the reliability of water resource management models that rely on snowmelt timing for streamflow and water storage predictions. Furthermore, from each station’s perspective, the AnEn method increases skill across much of the Sierra Nevada watershed domain, showing significantly reduced RMSE and CRPS, more than 2°C, in most stations (Figure 3).

Figure 1. (Figure 2 from Yang et al. 2025). Spatial correlations, RMSE (unit: °C), and CRPS (unit: °C) of GEFS ensemble mean (black lines), EMOS (orange lines), and AnEn (blue lines) forecasts compared to observation (PRISM) for 1-day to 15-day lead across the entire domain.

Figure 2. (Figure 3 from Yang et al. 2025). Spatial correlations (left column), RMSE (unit: °C, middle column), and CRPS (unit: °C, right column) of GEFS ensemble mean (black lines), EMOS (orange lines), and AnEn (blue lines) forecasts compared to observation (PRISM) for 1-day to 15-day lead, evaluated across different elevation ranges (0–500 m, 1000–1500 m, 2000–2500 m, and 3000–3500 m).

Figure 3. (Figure 6 from Yang et al. 2025). Temporal correlations, RMSE (unit: °C), and CRPS (unit: °C) of AnEn forecasts minus temporal correlations, RMSE (unit: °C), and CRPS (unit: °C) of GEFS in each station at 1-day to 5-day lead (left column), 6-day to 10-day lead (middle column), and 11-day to 15-day lead (right column).

This study presents a simple yet effective post-processing approach that improves daily weather-to-subseasonal forecasts, outperforming basic bias correction and extending skill beyond traditional weather timescales, an underexplored area, particularly at regional scales with fine spatial resolution. The approach provides a practical pathway for post-processing other key variables, such as precipitation, snow depth, and snow water equivalent, across broader regions and establishes a framework for extending AnEn to longer lead times (e.g., 14–35 days). Moreover, our findings contribute to advancing subseasonal-to-seasonal hydrological forecasts in the Sierra Nevada, aligning with California’s water management priorities and fostering the development of high-resolution dynamic models.

Cuthbertson, A., Lynn, E., Anderson, M., & Redmond, K. (2014). Estimating historical California precipitation phase trends using gridded precipitation, precipitation phase, and elevation data. California Department of Water Resources, Sacramento, California.

Delle Monache, L., Eckel, F. A., Rife, D. L., Nagarajan, B., & Searight, K. (2013). Probabilistic weather prediction with an analog ensemble. Monthly Weather Review, 141(10), 3498-3516. https://doi.org/10.1175/MWR-D-12-00281.1

Hunsaker, C. T., Whitaker, T. W., & Bales, R. C. (2012). Snowmelt runoff and water yield along elevation and temperature gradients in California’s Southern Sierra Nevada. JAWRA Journal of the American Water Resources Association, 48(4), 667-678. https://doi.org/10.1111/j.1752-1688.2012.00641.x

Musselman, K. N., Molotch, N. P., & Margulis, S. A. (2017). Snowmelt response to simulated warming across a large elevation gradient, southern Sierra Nevada, California. The Cryosphere, 11(6), 2847-2866. https://doi.org/10.5194/tc-11-2847-2017

Yang, Z., Hu, W., Sengupta, A., Monache, L. D., DeFlorio, M. J., Ghazvinian, M., Xiao, M., Pan, M., Kollen, J., Reising, A., Fabbiani-Leon, A., Rizzardo, D., & Kalansky, J. (2025). Improving Weeks 1-2 Temperature Forecasts in the Sierra Nevada Region Using Analog Ensemble Post-Processing with Implications for Better Prediction of Snowmelt, Water Storage, and Streamflow. Journal of Hydrometeorology (published online ahead of print 2025). https://doi.org/10.1175/JHM-D-25-0012.1