Separate the impact of climate change and non-climatic factors on hydrological drought based on AdaBoost algorithm

Hydrological drought occurs frequently all over the world and has a great impact on human beings. Hydrological drought attribution contributes to a better understanding of the mechanisms of drought occurrence, improves the accuracy of predictions of drought events, and can provide a basis for drought risk reduction. At present, hydrological models which possess physical mechanisms are widely used in attribution analysis. However, this kind of models is complex in calculation, and has very limited time scale. In this study, we developed a hydrological drought attribution method via AdaBoost algorithm. The method divided the study period into natural period unaffected by non-climatic factors and impacted period. Taken the natural period as training period, the impacted period as test, the runoff was obtained to calculate the three-months standardized runoff index (SRI-3). Based on the run-length theory, we calculated average drought characteristics in the impacted period. Finally, the proportion of the average drought characteristics obtained by simulated SRI-3 series to those obtained by observed SRI-3 series is considered as the contribution of the climatic factors to the drought events. We applied this method in the Yangtze River Basin and the results showed that climatic factors are the dominate factors affecting hydrological droughts in this region, with the contributions at all the gauge stations are over 50%. Among all the drought characteristics, average drought severity is the most affected by the climatic factors, the corresponding contributions are all greater than 100%, shown as “excess contributions” (with non-climatic factors shown as negative contributions). Through the applications in various sub-basins of the Yangtze River Basin, the method was shown to provide new ideas for hydrological drought attribution, and the method can also be extended for applications such as meteorological hazards attribution, stock market volatility attribution and so on.