Skill and lead time of vegetation drought impact forecasts based on soil moisture observations

Abstract

Timely and skilful forecasts of vegetation drought impacts should enable more proactive drought preparedness, management, and mitigation. Numerous previous studies found a temporal correlation between soil moisture and vegetation condition. However, a correlation across the full range of soil moisture and vegetation condition does not automatically translate into skill in forecasting infrequent events such as agricultural droughts. Here, we develop a threshold- or impact-based forecasting framework to assess early warning capability (EWC). We analysed the skill and lead time achieved using soil moisture observations at multiple depths as predictors of subsequent vegetation drought impacts inferred from MODIS satellite observations at 93 sites across the United States. Forecast thresholds were expressed in terms of seasonally-adjusted standardised anomalies (z-scores) to distinguish climate-related impacts from any seasonal vegetation cycle. Different combinations of soil moisture integration depth, satellite vegetation indices and threshold levels were tested. Near-Infrared Reflectance of vegetation (NIRv) yielded a marginally better EWC than other indicators of vegetation drought impact. The optimal soil moisture integration depth varied between land cover types, from 0 to 10 cm for cropping systems to 0–100 cm for grassland and savanna. The greatest skill improvements were achieved using similar z-score threshold values for the soil moisture trigger and vegetation impact, producing typical lead times of two to four weeks. Further research is recommended to combine the framework developed here with spatially continuous soil moisture analyses or forecasts available from remote sensing and land surface models.