Deep Learning Weather Prediction on Seasonal Precipitation Forecasting for Ethiopia

Rainfall is the most important weather and climate variable for most of the world and particularly more so for the developing world. From intuitive applications in planting and harvesting crops to advanced practices in the likes of resource planning and optimization for socioeconomic prosperity, every civilization and generation after another has tried to predict the weather in some form. Currently, the seasonal forecast in Ethiopia is at best analogue based, exploiting the statistical correlation between rainfall amount and sea surface temperatures (SSTs) at designated predictor locations in the preceding months, with expected and noticeable error in long-range accuracy and skill. This research is a work on Ethiopia’s seasonal rainfall prediction capitalizing on the current industry-wide progress in observing, understanding, and predicting weather from both numerical weather prediction tools based on massive computer simulations of atmospheric physics and AI based methods that learns and recognizes complex weather patterns from a wealth of historical data on a global grid. From previous research works, Ethiopia’s weather, global circulation patterns and global teleconnection mechanisms have indicated relevant oceanic regions as seasonal predictors and are used as the bases for this study. Historical weather and sea surface data from the past 50 years on a global grid with land, sea and atmospheric variables are used to train, validate and test the deep-learning weather prediction model. The result of this study, expected in a few weeks, will compare the test seasonal rainfall data with the LSTM and Transformer RNN-based seasonal rainfall predictions through a comparison of root mean square errors (RMSE), mean absolute errors (MAE) and correlation coefficients. Further, a better understanding is expected to be extracted on the relationship between seasons, climatic regimes and large-scale global predictors for most of Ethiopia. The AI on DLWP can thus be used to produce realistic seasonal rainfall forecasting throughout Ethiopia for a period of three months into the future with better accuracy, prediction skill and interpretability while becoming a valuable source of vital rainfall data with a longer lead time giving communities and critical sectors such as public health, water management, energy, and agriculture more time to prepare for opportunity or mitigate potential disasters and going as far as contributing to the global movement in protecting the environment and the global effort in reaching a level of climate resilience for the future of our planet.