Predictive soil nutrient modeling with spectral data and machine learning in four major Indonesian Provinces located on the island of java

Abstract This study provides a detailed analysis of predicting soil nutrient content using spectral data and machine learning techniques in four Indonesian provinces: West Java, Central Java, Yogyakarta (DIY), and East Java. The research collected 145 soil samples to predict various key soil nutrients, such as N Total, NH4, NO3, P Total, P Available, K Total, K Available, C Organic, and pH. The study used linear regression (LR) and deep neural networks (DNN) with a deep cross-network (DCN) architecture to model the relationships between soil spectral data and nutrient content. LR was used as a baseline model to understand linear relationships between spectral features and soil properties and identify the most influential spectral frequencies in predicting soil nutrient levels. On the other hand, the DNN model captured complex, non-linear patterns within the data. Results showed that while the DNN model displayed advanced capabilities, the LR model generally outperformed it in predictive accuracy, particularly for nutrients like N-Total, P-Total, and K-Total. The findings highlight the potential of combining spectral data with advanced machine-learning techniques for precise soil nutrient estimation, which could significantly enhance agricultural productivity and soil management practices in Indonesia.