During the last century, we have observed a warming climate with more intense precipitation extremes in some regions, likely due to increases in the atmosphere's water holding capacity. Traditionally, infrastructure design and rainfall‐triggered landslide models rely on the notion of stationarity, which assumes that the statistics of extremes do not change significantly over time. However, in a warming climate, infrastructures and natural slopes will likely face more severe climatic conditions, with potential human and socioeconomical consequences. Here we outline a framework for quantifying climate change impacts based on the magnitude and frequency of extreme rainfall events using bias corrected historical and multimodel projected precipitation extremes. The approach evaluates changes in rainfall Intensity‐Duration‐Frequency (IDF) curves and their uncertainty bounds using a nonstationary model based on Bayesian inference. We show that highly populated areas across the United States may experience extreme precipitation events up to 20% more intense and twice as frequent, relative to historical records, despite the expectation of unchanged annual mean precipitation. Since IDF curves are widely used for infrastructure design and risk assessment, the proposed framework offers an avenue for assessing resilience of infrastructure and landslide hazard in a warming climate.
Floods are the main natural disaster in Brazil, causing substantial economic damage and loss of life. Studies suggest that some extreme floods result from a causal climate chain. Exceptional rain and floods are determined by large-scale anomalies and persistent patterns in the atmospheric and oceanic circulations, which influence the magnitude, extent, and duration of these extremes. Moreover, floods can result from different generating mechanisms. These factors contradict the assumptions of homogeneity, and often stationarity, in flood frequency analysis. Here we outline a methodological framework based on clustering using self-organizing maps (SOMs) that allows the linkage of large-scale processes to local-scale observations. The methodology is applied to flood data from several sites in the flood-prone Upper Paraná River basin (UPRB) in southern Brazil. The SOM clustering approach is employed to classify the 6-day rainfall field over the UPRB into four categories, which are then used to classify floods into four types based on the spatiotemporal dynamics of the rainfall field prior to the observed flood events. An analysis of the vertically integrated moisture fluxes, vorticity, and high-level atmospheric circulation revealed that these four clusters are related to known tropical and extratropical processes, including the South American low-level jet (SALLJ); extratropical cyclones; and the South Atlantic Convergence Zone (SACZ). Persistent anomalies in the sea surface temperature fields in the Pacific and Atlantic oceans are also found to be associated with these processes. Floods associated with each cluster present different patterns in terms of frequency, magnitude, spatial variability, scaling, and synchronization of events across the sites and subbasins. These insights suggest new directions for flood risk assessment, forecasting, and management.