Rainfall extremes, weather and climate drivers in complex terrain: A data-driven approach based on signal enhancement methods and EV modeling

Abstract

Weather and climatic characterization of rainfall extremes is both of scientific and societal value for hydrometeorological risk management, yet discrimination of local and large-scale forcing remains challenging in data-scarce environments. Here, we present an analysis framework that separates weather regime and climate controls using data-driven process identification. The approach is based on signal-to-noise separation methods and explanatory extreme value (EV) modeling of multisite rainfall extremes. The EV models integrate the temporal component of the weather/climate driver using semi-automatic parameter identification. At weather scale, the EV models are combined with a state-based Markov model to represent the spatiotemporal structure of rainfall as weather states. At climate scale, the EV models are used to search for drivers leading to the shift of weather patterns. The drivers are brought out in a climate-to-weather signal subspace, built via dimension reduction of climate model reconstructions.We apply the framework to a complex terrain region: the Western Andean Ridge in Ecuador and Peru (0–6°S) using ground data from the second half of the 20th century. Overall, we show that the framework, which does not make any prior assumption on the explanatory power of the weather and climate drivers, allows identification of well-known and new features of the regional climate in a purely data-driven fashion. Thus, the approach shows potential to identify weather controls on precipitation extremes in data-scarce and orographically complex regions in which model reconstructions are the only climate proxies.