A hybrid Granger TCN framework for generating climate analogues and determining the future of agricultural practices

The accurate forecasting of weather parameter facilitates effective analysis of crop yield, and ensures the sustainable allocation of crop usage in domestic purpose. Also, the climate change and human activities have increased the complexity of weather systems, making accurate prediction of weather parameter more challenging, thus impacting yield. Therefore, this work proposes a hybrid approach that includes deep learning model Temporal Convolution Network (TCN) with Grangers Causality Test. The approach involves capturing a single historic weather parameter i.e. the maximum temperature for four Indian Cities: Coimbatore, Gwalior, Ludhiana and Kashmir. The study focuses on finding out the specific future years while which the maximum temperature of Kashmir (output region), will match the current or history of maximum temperature of aforementioned other cities (input regions) from south to north. The result has shown a significant mapping till present years and given the possibility of match of maximum temperature in future years. The standard deviation of RMSE across 75 years of mapping is 0.0271 with variance of 0.0007 for 1st rank match. The study will help in maize crop planning and will be highly useful for automation purpose that requires good time for infrastructure setup like a well advance planning on making availability of irrigation sources.