Seasonal Analysis and Multivariate Forecasting of Monthly Precipitation in Jipijapa, Manabí, Using ARIMA, ETS, NNAR, and Prophet Models (2000–2024)

Abstract

The present study analyzes the variability, seasonality, and trend of monthly precipitation in Jipijapa Canton (Manabí, Ecuador), a region particularly vulnerable to climate change and water scarcity. This research is justified by the lack of detailed studies on local rainfall dynamics, despite their importance for agricultural planning and water resource management. The main objective was to evaluate the historical behavior of precipitation from 2000 to 2024 and implement comparative forecasting models. Satellite data from NASA’s POWER platform were used, and the analysis was carried out entirely in RStudio using statistical and computational modeling techniques. Four time series models were applied—SARIMA, ETS, NNAR, and Prophet—and their forecasting accuracy was assessed using MAE, RMSE, and MASE metrics, as well as  rolling-origin  cross-validation  and  the  Diebold-Mariano  test. The results revealed strong seasonality, with peak rainfall concentrated between January and March, and years of extreme precipitation linked to El Niño events (e.g., years 2012 and 2017). The Seasonal Mann-Kendall test indicated no statistically significant trend (p = 0.866), suggesting long-term stability in rainfall volumes. Among the evaluated models, ARIMA demonstrated superior predictive performance over NNAR (p = 0.045). It is concluded that precipitation in Jipijapa follows a highly seasonal regime, strongly influenced by ocean-atmospheric phenomena. Therefore, the use of multivariate forecasting models with climate predictors is appropriate for strengthening local climate adaptation strategies and territorial planning.