A multiple linear regression model for the prediction of summer rainfall in the northwestern Peruvian Amazon using large-scale indices

Abstract

The northwestern Peruvian Amazon (NWPA) basin (78.4–75.8° W, 7.9–5.4° S) is an important region for coffee and rice production in Peru. Currently, no prediction models are available for estimating rainfall in advance during the wet season (January–February–March, JFM). Hence, we developed multiple linear regression (MLR) models using predictors derived from sea surface temperature (SST) indices of the Pacific, Atlantic, and Indian Oceans, including central El Niño (C), eastern El Niño (E), tropical South Atlantic (tSATL), tropical North Atlantic (tNATL), extratropical North Atlantic (eNATL), and Indian Ocean basin-wide with E and C removed (IOBW*) indices. Additionally, we utilized large-scale convection indices, namely, the eastern Pacific intertropical convergence zone (ITCZe) and South American Monsoon System (SAMSi) indices, for the 1981–2018 period. Rainfall in the lowland NWPA exhibits a bimodal annual cycle, whereas rainfall in the highland NWPA exhibits a unimodal annual cycle. The MLR model can be used to accurately capture the interannual variability during the wet season in the highland NWPA by utilizing predictors derived from the C and SAMSi indices. In contrast, regarding rainfall in the lowland NWPA, the Pacific SST variability, SAMS and tropical North Atlantic index were relevant. For long lead times, the MLR model provided reliable forecasts of JFM rainfall anomalies in the highlands (R3, approximately 2700 m asl) as these regions are governed by Pacific variability. However, the MLR model exhibited limitations in accurately estimating the wettest JFM season in the highlands due to the absence of a predictor for the amplified effect of the Madden–Julian Oscillation on rainfall.