Drought Forecasting Using Remote Sensing And Meteorological Point Data For Drought-Prone Areas Of Ethiopia

Abstract

Drought is a natural phenomenon that is caused by shortage in rainfall, which affects people’s health and well-being adversely as well as impacting the society’s economy and politics with far-reaching consequences. Researchers in the area of drought forecasting conduct their researches either using remote-sensing or meteorological point data independently. However, due to the complex and dynamic nature of drought, and also it has many factors that causes for the occurrences of drought, there is a lack of exact prediction of drought for prevention and mitigation purposes. Therefore, this study aims to design a more sophisticated drought forecasting model by integrating those independent remote-sensing and meteorological point data-based drought indices together. In this study, we applied both traditional models such as auto regressive integrated moving average (ARIMA), Markov Chain, ANN, SVR and also modern models like long short-term memory (LSTM) to forecast drought conditions based on both time series remote sensing data and meteorological point data simultaneously. And we found that long short-term memory (LSTM) model is better than other models, and forecasting is performed using this model. After forecasting is done independently, we applied the decision tree (DT) algorithm to integrate and classify the severity of drought. To this end, standardized precipitation index (SPI) and normalized difference vegetation index (NDVI) was used as a measure of drought severity. Normalized difference vegetation index (NDVI) was extracted from NOAA-AVHRR satellite images and standardized precipitation index (SPI) were derived from meteorological data such as Precipitation, Temperature, Pressure, Humidity and Wind using multivariate LSTM Model. Then the model received these features as input and outputted the severity of the drought conditions using decision tree (DT) algorithm. Applying the model to the drought-prone stations of Ethiopia, we showed that it could forecast the drought condition with an average accuracy of 99.76%. It is understood that this work will give new views for ways in drought forecasting at near real-time.