Development of drought indicators Using Machine Learning Algorithm: A case study of Zambia

Abstract

The overarching objective of this study is to address this problem through the development of a drought monitoring and forecasting system, leveraging the synergistic use of Principal Component Analysis (PCA) and convolutional long short term memory (ConvLSTM) over Zambia. First, the relationships between drought factors (precipitation, temperature, vegetation, soil moisture, and evapotranspiration) were integrated using PCA, and a new cloud-based Multisource Drought Index (CMDI) was constructed. Then, the Spatio-temporal prediction of CMDI on a short-term scale (monthly) was developed using ConvLSTM. The effectiveness of the CMDI in monitoring drought in Zambia was verified by SPI-1 12 based on the IMERG dataset; gross primary production (GPP), and other remote sensing indices that have been used for drought monitoring. The results show that CMDI is well correlated with the SPI and GPP.