Observational constraints on global climate projections: An original method applied to changes in Indian summer monsoon rainfall

Abstract The lack of consensus in global projections of regional precipitation changes represents a major obstacle for the design of adaptation policies. This persistent spread is mainly arising from modeling uncertainty, i.e., from our limited knowledge in but also plural representation of complex mechanisms in current global climate models. Regardless of further model developments, here we propose an original statistical method in order to make the best use of available projections. Focusing on the Indian Summer Monsoon Rainfall (IMSR), a regional phenomenon of importance for the livelihood of billions of people, we performed a Maximum Covariance Analysis (MCA) to relate the inter-model spread in future precipitation changes to the simulation of recent precipitation linear trends - looking for a present-future nexus in model representation of precipitation. Multiple observations are then used to account for model errors and thus build a revised multi-model ensemble. The robustness of our method is assessed using an out-of-sample validation approach i.e., by checking its ability to reliably predict the future regional precipitation anomalies simulated by a left-out model. Compared to more straightforward emergent constraint approaches, our approach enables a promising reduction in the inter-model spread at the grid-point level, with the benefit of statistically assessing its robustness.