CW3E Launches New S2S Precipitation Forecast Product Using Machine Learning Models

November 12, 2021

The Center for Western Weather and Water Extremes (CW3E), in partnership with the NASA Jet Propulsion Laboratory (NASA JPL), recently launched a new S2S precipitation forecast product that uses machine learning models, along with models from the North American Multi-Model Ensemble (NMME), to predict wintertime seasonal precipitation patterns across the western U.S. This effort was funded and sponsored by the California Department of Water Resources (DWR), including support from the Atmospheric River Program. The research methodology and supporting hindcast skill assessment are described in Gibson et al. 2021.

This new machine learning-based forecast product predicts likely patterns of precipitation anomalies over the western U.S. for November 2021 – January 2022. A second forecast will be made in early January for January 2022 – March 2022. Four different machine learning models (Random Forest, XGBoost, LSTM, and neural networks) were pretrained based on learning relationships between remote ocean and atmospheric patterns and characteristic seasonal precipitation anomalies over the Western United States. Instead of training on observations, the machine learning models were here trained on a large climate model ensemble, which enabled more robust relationships to be established due to the very large sample size in the training phase (Gibson et al. 2021). Seasonal forecast skill is also enhanced by focusing on four characteristic large-scale precipitation patterns across the Western United States.

The top row of the forecast product displays the four possible precipitation anomaly patterns identified in Gibson et al. 2021. The percentage ensemble agreement represents the fraction of machine learning models and NMME models predicting each pattern for the November 2021 – January 2022 period. The middle and bottom rows show the individual machine learning model and NMME model forecasts, respectively. These forecasts are available on the subseasonal to seasonal experimental forecasts webpage.

Gibson, P. B., W. E. Chapman, A. Altinok, L. Delle Monache, M. J. DeFlorio, and D. E. Waliser (2021), Training machine learning models on climate model output yields skillful interpretable seasonal precipitation forecasts. Nature Communications Earth & Environment, 2, 159.