Analyzing Crop Yield Trends and Predictions Using Climatic and Crop Data: A Case Study for Pakistan’s Agricultural Landscape

Abstract

Climate change and fluctuating weather patterns have had a significant impact on crop yields, affecting food security and livelihoods of farmers. As the world population grows, the demand for food which is nutritious and sustainable is also on a rise. Pakistan is one of the countries which has been impacted the most by the global climate change as well as having a high rate of population growth and requires a regular systematic study of how the change in climate is impacting the crop yields to address her growing needs. For this reason, we chose to focus on the staple food crops of Pakistan which are namely wheat, maize, sugarcane and rice. The proposed research aims to predict food crop yields in the Punjab region of Pakistan by analyzing a dataset that includes crop data, climate data, and soil data. The study includes trend analysis to investigate the relationship between crop yields and climate and to identify patterns in these relationships. The data spans the past 15 years and multiple cities of the province of Punjab, allowing for a comprehensive analysis of the trends and patterns that have emerged over time. This research will employ machine learning and deep learning techniques. We applied numerous machine learning and deep learning techniques and chose 4 machine learning techniques which are KNN, Random Forests, SVR and GPR. We saw the best performance by Random Forests with an accuracy ranging between 96% - 99% for each of these crops. We went with 3 deep learning models which were MLP, CNN and LSTM out of which MLP gave the best results, The overall performance of machine learning models was better then deep learning models which can be attributed to the data size we had available. We applied data augmentation techniques to expand our dataset to counter this problem. Our study reveals that modern technological methods can be put to use to improve our agricultural sector and can accuracy predict the crop yields. Our research can be expanded to more regions and to incorporate further advanced techniques for better insights to this rising issue.