CW3E Publication Notice
Seasonal forecasting of precipitation, temperature, and snow mass over the western U.S. by combining ensemble post-processing with empirical ocean-atmosphere teleconnections
July 14, 2023
A new paper entitled “Seasonal forecasting of precipitation, temperature, and snow mass over the western U.S. by combining ensemble post-processing with empirical ocean-atmosphere teleconnections,” was recently published in the AMS Journal of Weather and Forecasting by William Scheftic, Xubin Zeng, and Michael Brunke from the University of Arizona. This paper tests our experimental real-time ensemble seasonal forecasts of snowpack through snow water equivalent (SWE), 2m temperature (T2m) and precipitation (P) across hydrologic subbasins of the western U.S. after post-processing two currently operational state-of-the-art climate models: NCEP CFSv2 and ECMWF SEAS5. The winter forecasts have been hosted internally at CW3E and contribute to DeFlorio et al. (2023; submitted to BAMS and under revision). This work contributes to CW3E’s 2019–2024 Strategic Plan to enable more effective reservoir management through improved weather and water forecasts by demonstrating successful experimental ensemble forecasts of snowpack at the seasonal scale.
This research contributed to our understanding of, and efforts to improve, seasonal climate forecasting over the western U.S. in several ways, as highlighted in Figure 1. First, through our stage 1 adjustment which corrects for prior biases in the forecast distribution and adjusts for errors in spread, significant biases even in current state-of-the-art seasonal climate models were removed from all variables and improved skill over the original model. Second, through our stage 2 adjustments which further adjusts the forecast based on the linear relationship between stage 1 forecast errors and a set of land, oceanic and atmospheric predictor indices, we showed that significant SWE errors arising from poor model snowpack initialization could be further adjusted leading to improved forecasts. The impact of stage 2 adjustments on T2m and P forecasts was more limited, as the residual stage 1 error was not significantly related to the predictor indices. Finally, this study contributed to our understanding of the predictability of SWE relative to T2m and P. Here we saw enhanced skill in SWE, especially during the spring snowmelt period and in regions where snowpack is important but more challenging to initialize and forecast. Further analysis showed that the skill of SWE outperformed, but was strongly related to, the persistence of observed anomalies.
Figure 1: Adapted from Scheftic et al. (2023). The differences between original and Stage 1 forecasts (col 1) and the differences between Stage 1 and 2 forecasts (col 2) of the average of CFSv2 and SEAS5 median RPSS across all seasons for all subbasins in the western U.S. for T2m (top), P (middle), and SWE (bottom) at 1 month lead over all seasons. Subbasins outlined in bold black are 95% significant according to a bootstrap analysis. Skill for super-ensemble (SENS) Stage 2 forecasts for RPSS (right two columns). The 3rd column shows median skill across all seasons for all subbasins at 1 month lead. Subbasins outlined in thick gray show significant improvement (at 95% level) over one of the two models and those in bold black show significant improvement over both models (note: no subbasins were significant over both models). The 4th column shows median skill across all subbasins for each season and lead time. For instance, the value corresponding to “0-mon” and “OND” refers to the median RPSS for early October forecasts of October-December, while the value corresponding to “1-mon” and “NDJ” refers to the median SAC for early October forecasts of November-January. Note: RPSS is the ranked probability skill score which compares the performance of the full ensemble of the forecast to a reference ensemble that uses all historical observed data in the training period acting as a climatological distribution. RPSS of 1 represents perfect skill, RPSS of 0 has the same skill as the reference, and negative RPSS represents forecasts that performed worse than the reference.
DeFlorio, M. J., & Coauthors (2023). The transition from California’s extreme drought to major flooding: Evaluating experimental subseasonal and seasonal forecasts of the onslaught of landfalling atmospheric rivers and associated extreme precipitation during Winter 2022-2023, Bulletin of the American Meteorological Society, in review.
Scheftic, W. D., Zeng, X., & Brunke, M. A. (2023). Seasonal forecasting of precipitation, temperature, and snow mass over the western U.S. by combining ensemble post-processing with empirical ocean-atmosphere teleconnections, Weather and Forecasting (published online ahead of print 2023). https://doi.org/10.1175/WAF-D-22-0099.1