Modelling Crop Yield with Deep CNN-LSTM for Spatiotemporal Data Analysis

Abstract

Agriculture is a fundamental component of human civilization. It strengthens the economy while providing sustenance. Plant foliage and crops are susceptible to several diseases in agricultural techniques. Precise forecasting of crop yield is crucial for effective agricultural resource management. Elements such as meteorological conditions, soil hydration, and temperature significantly influence agricultural productivity, making accurate forecasting essential. Traditional crop recommendation approaches often rely on heuristic principles or expert knowledge, which may lack accuracy and adaptability to changing environmental and market circumstances. Consequently, a data-driven methodology is required to use existing knowledge on soil, weather, and crops to provide best agricultural recommendations to farmers. This study introduces a novel approach for predicting agricultural output via a hybrid model. Our study introduces a hybrid model that improves crop production forecasting by combining a 1D Convolutional Neural Network (CNN) with a Long Short-Term Memory (LSTM) network and an attention mechanism. The model is primarily used for wheat and rice, which are principal crops in India. The model transforms into a CNN-LSTM hybrid, aimed at enhancing prediction accuracy via adjustments such as multi-head attention and a multiplication skip connection. This proposed system offers substantial decision-making support for farmers and agricultural stakeholders, facilitating informed, data-driven choices that enhance sustainability and efficiency in agriculture.