From Soil Color to Carbon: A Generative AI and Nix Sensor Approach

Soil Organic Carbon (SOC) is a critical indicator of soil health, yet conventional laboratory-based estimation methods remain costly, time-consuming, and environmentally burdensome. This study evaluates a rapid, low-cost, and environmentally friendly alternative for SOC estimation using high-resolution color information acquired from a Nix Spectro 2 handheld sensor, integrated with machine learning and generative data augmentation approaches. A total of 641 soil samples collected across diverse agro-ecological regions of West Bengal, India, were analyzed using Random Forest, Gradient Boosting, XGBoost, and Artificial Neural Network models. To address data imbalance and limited sample representation at higher SOC ranges, synthetic datasets were generated using Gaussian Mixture Models (GMM), Generative Adversarial Networks (GAN), k-nearest neighbors–based augmentation, and bootstrapping techniques. Among the evaluated models, Random Forest achieved the best baseline performance (R² = 0.71), which further improved with GMM-based data augmentation (R² = 0.77). The results demonstrate the strong potential of combining handheld color sensing with generative artificial intelligence to develop more accurate, robust, and scalable SOC prediction frameworks.