AI in agriculture : crop yield forecasting

Abstract

Advancements in Machine Learning (ML) are transforming agricultural practices by enhancing the accuracy of crop yield forecasts in the face of challenges such as climate change and decreasing arable land. This research employs a diverse array of ML models, including Long Short-Term Memory (LSTM) networks, Temporal Convolutional Network (TCN), Random Forests (RF), Gradient Boosted Decision Tree (GBDT), and Temporal Fusion Transformer (TFT). These models were trained on robust datasets integrating historical crop yield records with environmental variables like temperature and precipitation, sourced from global repositories such as FAOSTAT and the Climate Change Knowledge Portal. The models’ efficacy was rigorously tested, highlighting the strengths of LSTMs in recognising temporal patterns crucial for predicting cyclical crop production and the capability of TCNs in processing spatiotemporal information, significantly outperforming traditional naive methods. Further exploration into cutting-edge Artificial Intelligence (AI) techniques such as Zero-Shot Learning (ZSL) and transformer-based models demonstrated their potential to substantially enhance prediction accuracy across various agricultural contexts by adeptly combining empirical data with comprehensive domain knowledge. The results emphasise the transformative impact advanced ML models can have in agriculture. Accurate predictions lead to optimised resource management, reduced waste, and more sustainable farming practices worldwide. These insights could significantly contribute to the agricultural sciences, showcasing the broad potential of sophisticated ML models to revolutionise global farming practices, thereby boosting productivity and sustainability.