Probabilistic U.S. Drought Monitor Predictions Using Anomalies in Precipitation, Soil Moisture, and Evapotranspiration

Probabilistic forecasts of U.S. Drought Monitor (USDM) intensification over 2, 4 and 8 week periods are developed based on current and past anomalies in 1) precipitation, 2) evapotranspiration from the Evaporative Stress Index (ESI)) and 3) soil moisture from the North American Land Data Assimilation System (NLDAS). These statistical forecasts are based on logistic regression with cross validation. When using precipitation, ESI and soil moisture directly in the logistic regression, almost all of the useful skill is derived from the precipitation field. This is also true when the ESI and soil moisture time tendencies are used. Further skill based on the ESI and soil moisture can be obtained, however, by using these variables to better constrain the "true state" of the USDM. Essentially it is assumed that there exists a hypothetical continuous, normally distributed version of the USDM that is only observed after being artificially discretized based on the 30th, 20th, 10th, 5th, and 2nd percentiles. Using precipitation, ESI, NLDAS anomalies and some reasonable statistical assumptions, a Probability Density Function (PDF) of the continuous version of the USDM is predicted. This PDF can be used to distinguish USDM states that are either far or near from the next higher drought category. For example, when the USDM is in the state "no drought", this PDF determines whether conditions are normal or extremely wet. This information is useful because recent dry anomalies are much more likely to tip the former state into drought than the latter. For the forecasts, the information from this USDM state PDF is combined with the baseline skill from precipitation discussed above. The final USDM intensification forecasts are most skillful in the Midwest where Brier Skill Scores (BSS) average about 0.18. Without the USDM state PDF information the BSS is 0.10. In future work, these forecasts will serve as a baseline for adding improvements derived from the North American Multi-Model Ensemble.