Resolving regional frequency analysis of precipitation at large and complex scales using a bottom-up approach: The Latin America and the Caribbean Drought Atlas

• Published in: Journal of hydrology (Amsterdam) 2016-07, Vol.538, p.515-538
• Main Authors: Núñez, J., Hallack-Alegría, M., Cadena, M.
• Format: Article
• Language: English
• Subjects: Algorithms; Assessments; Drought; Drought Atlas; Droughts; Hazard assessment; Homogeneity; Hydrology; L-moments; Latin America; Proposals; Random forest; Regional analysis; Regional frequency analysis
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2016.04.025

•We improve L-moments based regional frequency analysis for large and complex scales.•We develop a new framework for RFA-LM based on a bottom-up approach.•We propose regionalization efficiency metrics.•We propose a new algorithm for precipitation regional frequency analysis.•We propose a new approach for frequency estimation and mapping at ungauged sites. Hydrologic frequency analysis is a statistical technique for the assessment of natural hazards, particularly the so-called water hazards caused exclusively by extreme hydrologic events. In particular, L-moments based regional frequency analysis (RFA-LM) has being adopted as the standard method for hydrologic frequency analysis in many parts of the world and for many other applications relating to hydrological extremes. However, despite the widespread use of RFA-LM, its application at large network and high and complex spatial scale conditions (LNHCSSC) has been poorly studied. The lack of studies about RFA-LM under such conditions and its use preferentially at subnational scales and areas with low number of rain gauge stations, or based on grid data, may be explained by the fact that the most difficult, less robust, and most subjective stage of RFA-LM is the delineation of homogeneous regions. This work proposes an integral procedure for the application of RFA-LM under LNHCSSC. The proposed method is applied to the study of drought event frequency in three case studies from Latin America, and incorporates innovating aspects compared to the state-of-the-art RFA-LM. These aspects are specifically (a) the decoupling of the cause of homogeneity from the regionalization stage; (b) the proposal of regionalization efficiency metrics; (c) the development of a regionalization algorithm; and (d) the development of a frequency estimation and mapping method for ungauged sites.