The Impact of climate change on crop yield in water scare region through deep learning

Implications for Food Security in a Changing Climate

Abstract

Climate change poses a severe threat to agricultural sustainability, particularly in water-scarce regions such as Multan, Pakistan, which receives an annual average rainfall of only 186mm and experiences frequent drought conditions (Pakistan Meteorological Department, 2022). The increasing depletion of groundwater resources further exacerbates agricultural challenges in the region. It is a rising threat to agriculture, a fundamental component of food security worldwide, and due to this, new ways of formulating crop estimate models are needed. Most traditional approaches to forecasting ATTR might not accurately capture the complex and non-linear or even non-additive mechanisms that connect climate parameters and crop yields. In this work, we employ Long Short-Term Memory (LSTM networks) a state-of-art deep learning method to forecast crop yields based on temperature, rainfall, and crop production data. As a result, the LSTM model can process the sequential data and identify the temporal dependence pattern as the best model for this task. The main input data included climate and yield data from a particular year in a specific region and some preprocessing was done to handle missing values, scale inputs, and group rainfall data. The model had a MAPE of 5.36%, an MAE of 1136.70, and an RMSE of 1136.70 giving the model a prediction accuracy of 94.64%. This work shows the efficiency of the proposed model and confirms that this approach is more effective than traditional statistical methods. These predictions are highly accurate and provide valuable information for different users including farmers, policy-makers, and researchers. This work elucidates how LSTM-based models can help solve the future challenges of agricultural management and production. When able to yield reliable predictions of the yields, such models can also help in proper resource planning, managing the effects of climate change, and improving food security all over the world. As a result, it is evident that machine learning critically occupies the subject area in constructing a long-lasting agricultural future resistant to dramatic changes.