Evaluation of methods to detect and quantify the bimodal precipitation over Central America and Mexico

Abstract

AbstractBimodal precipitation is a globally observed and regionally significant event that has a significant influence on the agriculture, public health, and insurance needs of associated regions. Many studies have focused on the mechanisms behind the generation and development of this event; however, little research into its characteristics exists due to a lack of a widely accepted method for accurate detection and quantification. Using a function collection containing various methods, different methods can be compared in terms of their performance in the detection and quantification of bimodal precipitation signals, allowing the proposal of appropriate criteria for method choice in various study types. Five methods (Mosiño and García, 1966; Curtis, 2002; Angeles et al., 2010; Karnauskas et al., 2013; Zhao et al., 2020) are adapted to the Climate Prediction Centre data during 1979–2017 in the domain of southern Mexico and Central America, and their performances are evaluated and compared. While outputs from the five methods reach general consistence for strong bimodal features over the Pacific side of Central America and Yucatán Peninsula, some biases are identified, specifically shown by the fact that methods using monthly climatological data demonstrates bimodal precipitation over the Caribbean side of Central America, while those using daily annual data indicate the existence of bimodal precipitation over the Pacific side of southern Mexico. By comparing two typical algorithms, we determined that this bias was induced by the limitation of temporal resolution in monthly climatological data and the nature of algorithms applying daily annual data. As part of a case study, a cluster algorithm was applied to outputs from an algorithm using daily annual precipitation, and a classification algorithm was used to test clustering performance. The resultant general high accuracy shows that annual bimodal signals offer good adaption to cluster and other potential machine learning algorithms.