Integrating terrestrial and orbital reflectance data improves the soil attribute modeling performance

A comprehensive understanding of soil attributes is crucial for effective environmental management. Geotechnologies offer an alternative to traditional soil surveying methods. This study evaluated the potential of multispectral data from terrestrial and orbital sensors to predict soil attributes of Rhodic Ferralsols in Central Brazil using machine learning algorithms. Physicochemical and spectral attributes of 37 soil samples (0–20 cm depth) were collected and analyzed. Spectral signatures were extracted from visible to shortwave infrared using the ASTER, a satellite-based sensor providing multispectral data, for comparison to laboratory hyperspectral data from Fieldspec Pro 4, and resampled to ASTER bands. Random Forest (RF) and Multiple Linear Regression (MLR) modeled the soil attributes using the spectral libraries, individually and combined. Results showed similar spectral responses between the sensors, indicating that resampling hyperspectral data from terrestrial sensors can be a reliable reference for orbital data. Due to controlled conditions and reduced interference from moisture and atmosphere, the terrestrial sensor and combined approaches had a higher Pearson correlation with soil attributes than the orbital sensor. MLR with combined sensors effectively predicted soil attributes, achieving R² of 0.65 for clay and 0.69 for organic matter. RF showed lower performance, with R² of 0.32 for base saturation and 0.30 for Cation Exchange Capacity, attributed to limited datasets. Combining terrestrial and orbital sensors improves soil attribute modeling, nevertheless, it requires robust sampling, image processing, and sensors testing, datasets, and algorithms. This study highlights the potential of integrating multilevel remote sensing for efficient soil analysis and mapping, contributing to sustainable environmental management.