This repository introduces a hybrid regression model for paddock-level crop yield prediction, combining MAMBA blocks, Transformer attention, and Slot Attention to capture spatial and temporal complexities. By leveraging soil, weather, and Sentinel-2 data, the model improves predictive accuracy. Additionally, the Western Australian (WA) Rainfall Paddocks Dataset, encompassing soil, climate, and satellite data for ~450,000 paddocks over three years, is presented as a resource for high-resolution and large-scale modeling. The proposed model outperforms classical machine learning and ResNet50 in accuracy and inference speed, establishing itself as an efficient solution for precision agriculture.
The code and the dataset will be made available soon.
This research proposes a novel hybrid regression model for paddock-level crop yield prediction, integrating MAMBA blocks, Transformer attention mechanisms, and Slot Attention. This architecture effectively captures spatial and temporal complexities, leveraging diverse data such as soil characteristics, weather patterns, and Sentinel-2 data to enhance predictive accuracy. We also introduce the Western Australian (WA) Rainfall Paddocks Dataset, a comprehensive resource comprising soil, climate, and satellite data for ~450,000 WA paddocks over three years. The dataset supports both high-resolution and large-scale modeling, facilitating diverse research applications. Evaluation against classical machine learning models and ResNet50 demonstrates that our hybrid model significantly improves accuracy while achieving faster inference speeds compared to ResNet50 and some classical models. The results establish the proposed method as a robust and efficient solution for precision agriculture