Student project

Rainfall Forecasting in Burkina Faso Using Bayesian-Wavelet Neural Networks

This work aims to forecast rain locally in Tambarga, Burkina Faso, to be able to fight against a worm inducing the disease called schistosomiasis. The chosen approach relies on a machine-leaning technique called Artificial Neural Networks, which simulates the synapses of a brain, with climatic parameters as inputs, activation functions and outputs in the form of rain prediction. A special case of Neural Networks using Bayesian Computations is used, along with as a transform allowing to capture the changes in climatic conditions, called Wavelet Transform. The precipitation is forecasted in different manners: binary forecast on the presence or absence of rain, linear forecast on the daily and weekly intensity, as well as a rain-class forecast. The most successful predictions have been found to be the binary forecast, as well as the weekly windowed cumulative rain forecast. The daily cumulative rain, as well has the classes forecast have not produced satisfying results, mainly because of the high temporal variability of the observations, as well as the very unequal distribution of observations in the different rain classes. In the end, it has been shown that it is possible to use Bayesian Networks to forecast precipitation in some extent, and that the wavelet transform of the inputs has a positive impact on the accuracy of the prediction.

    Keywords: Rainfall Forecasting


    • EPFL-STUDENT-213510

    Record created on 2015-11-06, modified on 2016-08-09

Related material