Information System

Permanent URI for this collection

https://etd.aau.edu.et/handle/123456789/4134

Browse

Browsing Information System by Subject "Deep Learning"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Rainfall Prediction using Combined Satellite and Station Data: Adeep Learning Approach
    (Addis Ababa University, 2023-07-01) Mubarek Jemal; Melkamu Beyene (PhD)
    Currently, Ethiopia has high rainfall variance, which is a result of global climate change that has an influence on the environment, property values, and human lives. Accurate rainfall prediction is highly important to smart agriculture practices for developing countries. For rainfall prediction, using station data alone often lacks the required accuracy and spatial coverage, and satellite data has spatial coverage but cannot predict rainfall as accurately as station data. The objective of this research is to develop a model for rainfall prediction using deep learning approaches by combining weather station and satellite data. A design science research methodology was used to develop a rainfall prediction model with 30 years (1990 - 2020) of daily weather station data from the National Meteorological Agency Ethiopian and satellite data from TAMSAT v3.1 and JRA-55 climate models. In data engineering, missing values were handled using mean imputation by dividing the dataset based on the three seasons of Ethiopia. Deep learning approach that includes multi-layer perceptron (MLP), Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and Bidirectional Long Short Term Memory (BiLSTM) were experimentally evaluated to predict rainfall for selected areas. Lastly, we proposed a model using Bidirectional Long Short Term Memory (BiLSTM) architecture that capable of forecasting daily rainfall for Ethiopia. The performance of the model is evaluated using the state of the art performance evaluation metrics such as; Root Mean Square Error (RMSE), Mean Square Error (MSE), and Mean Absolute Error (MAE), and the results were 0.0472, 0.0025, and 0.021 respectively. We also compared the proposed model with other deep learning approaches like MLP, CNN, and LSTM. The proposed BiLSTM model outperformed LSTM with an RMSE of 0.0015; CNN with RMSE of 0.0023, and MLP with RMSE of 0.0025. The experimental results show that the Bidirectional Long Short Term Memory (BiLSTM) model has a lower RMSE, MSE, and MAE.