Precipitation forecasting using machine learning in the region of Beni Mellal-Khenifra

Abstract

Agriculture in the region of Beni Mellal-Khenifra, Morocco relies on irrigation from rain and dams, but recently there has been a lack of precipitation which may negatively affect crop growth. This has made accurate precipitation forecasts even more important for farmers, as they need this information to make informed decisions about their crops. However, a lack of data-driven research utilizing past data presents a challenge for the development of such research and leaves farmers relying solely on weather forecasts from TV, which cannot relied upon in systems such as irrigation. The objective of this paper is to propose various approaches for forecasting precipitation in the region of Beni Mellal-Khenifra using big data analytics and machine learning techniques. The study made use of Apache Spark, a big data analytics tool, and five machine-learning algorithms: Lasso regression, ridge regression, elastic net, auto regressive integrated moving average, and random forest. These algorithms were applied on dataset of daily rainfall from 2000 to 2015 to forecast the amount of precipitation in the region. The results of the study showed that the random forest algorithm had the lowest mean absolute error, making it the most effective at forecasting precipitation in the region.