Publication

Abstract : Soil Organic Carbon Density (SOCD) is a key indicator of intensification of agriculture, land use sustainability, soil fertility, crop yield, and climate change mitigation. SOCD recovery information and prediction to agricultural treatment are required to formulate suitable soil management alternatives. In this research work the data is processed through rigorous preprocessing steps such as data cleaning, feature scaling, data splitting, data reshaping for deep learning models and creation of temporal sequences, prior to using deep learning-based time series modeling. We introduce and contrast some deep neural network models, namely Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model, that can learn the intricate temporal and nonlinear patterns underlying SOCD behavior. The models are cross-validated and trained on stratified time-based split data, and their performance was evaluated using a range of error metrics. This research provides evidence for the potential of deep learning models to provide high-accuracy and scalable SOCD prediction systems. It provides a solid foundation for the use of intelligent decision-support systems in today’s agriculture and environmental monitoring.