Modeling drought impacts using the SPI and SPEI

Posted by Jim Stagge on October 15, 2015

Our paper, entitled “Modeling drought impact occurrence based on meteorological drought indices in Europe” has been published in this month’s Journal of Hydrology. It shows the feasibility of relating user-provided drought impact reports with meteorological drought indices. The resulting drought impact models predict the likelihood of agricultural, energy, water supply, and ecosystem impacts for 5 European countries.

You can find the full paper (open access) here:

Stagge, J.H., Kohn, I., Tallaksen, L.M., Stahl, K. (2015) "Modeling drought impact occurrence based on meteorological drought indices in Europe" Journal of Hydrology, Vol. 530, Pages 37-50, 10.1016/j.jhydrol.2015.09.039.

Abstract

There is a vital need for research that links meteorological drought indices with drought impacts felt on the ground. Previously, this link has been estimated based on experience or defined based on very narrow impact measures. This study expands on earlier work by showing the feasibility of relating user-provided impact reports with meteorological drought indices, the Standardized Precipitation Index and the Standardized Precipitation-Evapotranspiration Index, through logistic regression, while controlling for seasonal and interannual effects. Analysis includes four impact types, spanning agriculture, energy and industry, public water supply, and freshwater ecosystem across five European countries. Statistically significant climate indices are retained as predictors using step-wise regression and used to compare the most relevant drought indices and accumulation periods across different impact types and regions. Agricultural impacts are explained by 2–12 month anomalies, though anomalies greater than 3 months are likely related to agricultural management practices. Energy and industrial impacts, typically related to hydropower and energy cooling water, respond slower (6–12 months). Public water supply and freshwater ecosystem impacts are explained by a more complex combination of short (1–3 month) and seasonal (6–12 month) anomalies. The resulting drought impact models have both good model fit (pseudo-R2 = 0.225–0.716) and predictive ability, highlighting the feasibility of using such models to predict drought impact likelihood based on meteorological drought indices.

Fig. 5. Reproduced from article.. Agricultural drought impacts: partial logistic regression model for Germany. Predicted impact likelihood is shown in red, while the distribution of SPEI3 is shown for impact months (top) and non-impact months (bottom).

• Next Post →