Evaluating CMIP5 model agreement for multiple drought metrics.

Global climate models play an important role in quantifying past and projecting future changes in drought. Previous studies have pointed to shortcomings in these models for simulating droughts, but systematic evaluation of their level of agreement has been limited. Here, historical simulations (1950-2004) for 20 models from the latest Coupled Model Intercomparison project (CMIP5) were analysed for a variety of drought metrics and thresholds using a standardised drought index. Model agreement was investigated for different types of drought (precipitation, runoff and soil moisture) and how this varied with drought severity and duration. At the global scale, climate models were shown to agree well on most precipitation drought metrics, but systematically underestimated precipitation drought intensity compared to observations. Conversely, simulated runoff and soil moisture droughts varied significantly across models, particularly for intensity. Differences in precipitation simulations were found to explain model differences in runoff and soil moisture drought metrics over some regions, but predominantly with respect to drought intensity. This suggests it is insufficient to evaluate models for precipitation droughts to increase confidence in model performance for other types of drought. This study shows large, but metric-dependent discrepancies in CMIP5 for modelling different types of droughts that relate strongly to the component models (i.e. atmospheric or land surface scheme) used in the coupled modelling systems. Our results point to a need to consider multiple models in drought impact studies to account for high model uncertainties.

Figure 2: Globally averaged biases in individual CMIP5 models relative to the observed mean for precipitation drought metrics. Duration (months) and frequency (no. of events) for different drought lengths are shown in a), and intensity (mm) in b). The observed mean is noted on top of each column. Metrics were averaged across all drought events (denoted mean) and short- (1-2 month), medium- (3-4) and long-term (5-6) droughts. The global average was calculated for land areas between 65°N and 65°S, excluding grid cells where no droughts were detected in >33% of the models. The models were arranged by their land surface scheme, with models with different schemes separated by horizontal lines (see Table 1).