Hot, dry and compound hot-dry extremes in decadal predictions

We investigate the representation of hot-dry compound extremes in decadal predictions and their correlations with the corresponding univariate extremes. We use a multi-model ensemble (MME) of 125 members from the CMIP6 Decadal Climate Prediction Project (DCPP) hindcast simulations and compare it with different observational references. Our analysis focuses on the average of forecast years 2 to 5, with the different forecasts initialised every year from 1960 to 2014. We analyse the skill of predicting hot, dry and compound hot-dry events in the MME. Specifically, we select the days above the 90th percentile of the daily maximum temperature for hot events (tx90p). For dry events, we use two indicators, based on the Standardised Precipitation Index (SPI) and the Standardised Precipitation Evapotranspiration Index (SPEI), with accumulation periods of 3, 6 and 12 months, and we consider a dry event a month when the SPI or SPEI value ≤ -1. Finally, we identify days that present both hot and dry conditions according to these criteria as compound hot-dry days (HDSPEI3,-6,-12 and HDSPI3,-6,-12).

For the univariate extremes, results show strong skill for hot extremes, to a large extent driven by the warming trend. On the contrary, dry extremes show less uniform skill, with isolated areas of significant correlations. We find regional skill in some areas of the globe for hot-dry compounds. Through a comparison with historical simulations, we find that initialisation of the predictions leads to additional skill in localised regions, with most areas showing non-significant differences. However, the models still appear to underestimate the connections between compound and univariate extremes, especially between hot-dry compounds and dry conditions. These results show the potential and the limitations of predicting compound events in decadal forecasts, highlighting the pivotal role of dry conditions in assuring a skilful prediction.