Statistical and Machine Learning Models as a Tool for Forecasting Extreme Summer Precipitation Driven by ENSO in the Northern Coast of Peru (1981-2018)

The northern coast of Peru is frequently affected by extreme summer precipitation during events such as El Niño phenomenon. For instance, the 2017 event was not classified by NOAA as a global ENSO, however, the precipitation occurred was too heavy in the coast of Peru and Ecuador. The main goal of this project is to improve current forecasting techniques comparing the use of two different models in the Piura river basin.

Two types of modelling approach were used: Statistical using Canonical correlation analysis (CCA) and Machine Learning using Clustering analysis. In both, for the predictive variables (sea surface temperature, geopotential height, sea level pressure, tropospheric temperature), we considered 3- El Niño zones (NIÑO 1+2, NIÑO 3.4, West Pacific-Tw) through october-november-december quarter (OND) from 1981-2016. For the predictor, we used accumulated precipitation anomalies through january-february-march quarter (JFM) from 1982-2017 period.

Correlation coefficients between forecast and actual precipitation data showed a better correlation for Clustering (~0.6) than CCA (~0.4). In both techniques, the use of gridded dataset revealed spatial characterization of precipitation patterns in three delimited regions in the Piura river basin(upper, middle and lower). In addition it was possible to know which atmospheric variables are related to the generation of precipitation. The best result was obtained with cluster analysis, since this tool classified satisfactorily the years into El Niño, La Niña and Extraordinary events. Both techniques coincide in considering the low tropospheric temperature (TLT) and sea surface temperature (SST) as the main predictor variables in the Tw and NIÑO1+2 zones. Although both techniques coincide in the identification of the peaks of extraordinary events, we consider clustering as a promised tool for the forecast of these extreme events on the piura river basin.