Variability in Rainfall Prediction Quality Over Time

Jayalath Bandara Ekanayake, Thanuja Dananjali


Many rainfall prediction models have been proposed. The common methodology followed by those models is that the model is trained using the data prior to the target and tested the model in one or few time points and claimed that the model is generalized. However, this project shows that the above procedure is not sufficient to generalize rainfall prediction models as in some target periods the models fail to achieve a decent prediction quality. The models--Multilayer Perceptron (MLP), M5P and Linear Regression--were trained and tested in all possible combinations of targets and training periods from the weather data collected between the year 2002 and 2015 from the station located at Badulla, Sri Lanka. The prediction quality of the models was measured using Mean Absolute Error (MAE) and visualize them in heat-maps to show that the prediction quality varies over the targets and length of the training periods. This indicates that testing models in one or a few time points is not sufficient to generalize the models. Further, the reasons for such drastic changes in prediction quality will be investigated in our future projects.

Full Text:


DOI: 10.17700/jai.2020.11.1.565

Journal of Agricultural Informatics