Skill of large-scale seasonal drought impact forecasts

Abstract

Forecasting drought impacts is still missing in drought early warning systems that presently do not go beyond hazard forecasting. Therefore, we developed drought impact functions using machine learning approaches (Logistic Regression and Random Forest) to predict drought impacts with a lead-time of 7 months ahead. The skill of the drought impact functions to forecast drought impacts was evaluated using the Brier Skill Score and Relative Operating Characteristic metrics for 5 Cases representing different spatial aggregation and lumping of impacted sectors. For German regions, impact functions developed using Random Forest show a higher discriminative ability to forecast drought impacts than Logistic Regression. Moreover, skill is higher for Cases with higher spatial resolution and less-lumped impacted sectors (Cases 4 and 5), with considerable skill up to 3–4 months ahead.