Randomized trial shows satellite-based soil models improve accuracy in delivering soil data for stakeholders, suggesting a cost-effective strategy for precision agriculture.
Key Points
This study aims to develop a satellite-based soil spectral library to provide soil information effectively to stakeholders.
Constructed a satellite-based soil spectral service (Sat4-Service) using Landsat-8 and Sentinel-2 data.
Compared bare-soil reflectance with laboratory spectra from 7711 topsoil samples to validate accuracy.
Simulated a provider expanding the dataset and retraining models as more clients and soil data were added.
Model accuracy for soil organic carbon (SOC) reached R2 = 0.63 and for clay reached R2 = 0.79 after ~50 clients.
A national model based on over 60,000 samples had lower accuracy (R2 = 0.43 for SOC, 0.65 for clay) on the same farms.
Findings indicate that regionally focused libraries outperform broader continental-scale models for soil estimation.