A Framework for Quantifying the Robustness and Uncertainty of Drought Projections

Abstract

Drought may be exacerbated by global warming, but drought projections are largely inconsistent. The existing frameworks cannot adequately quantify the robustness and uncertainty of drought projections. Therefore, this study proposes a framework to solve this problem and verifies it in the Chinese Mainland. This framework consists of three main components: (1) the meteorological drought in the 21st century is projected using an impact propagation modelling chain; (2) the robustness of drought projections is quantified using Identical Trend Percentage (ITP) and Signal-to-Noise Ratio (SNR); (3) the uncertainty of drought projections is investigated using improved multi-way analysis of variance. The study reveals that this framework can include more uncertainty sources, investigate the propagation patterns of uncertainty components and quantify the robustness of drought projections. The results show that drought projections are not robust. Specifically, the mean ITP ranges from 49% to 69%, indicating that nearly half of the projections display trends opposite to those of the mean values. In addition, the mean SNR of drought projections ranges between −0.36 and 0.15, with an absolute value far from 1.0. The dominant uncertainty source is the choice of drought index, of which the mean relative contribution ranges between 47% and 61%. When propagating along with the impact propagation modelling chain, the relative importance among existing uncertainty sources usually remains stable if no new physical quantities are joining in. If the relative importance among the existing uncertainty sources for one particular quantity is different from that for the other quantity, the relative importance among the existing uncertainty sources may be adjusted when the two quantities are pooled together by the newly joined processes. Excluding unreasonable drought indices generally reduces the uncertainty and improves the robustness of drought projections; however, it is insufficient to derive robust drought projections.