Integrating satellite radar vegetation indices and environmental descriptors with visible-infrared soil spectroscopy improved organic carbon prediction in soils of semi-arid Brazil
Erli Pinto dos Santos, Michel Castro Moreira, Elpídio Inácio Fernandes Filho, José Alexandre Melo Demattê, Uemeson José dos Santos, Jean Michel Moura-Bueno, Renata Ranielly Pedroza Cruz, Demétrius David da Silva, Everardo Valadares de Sá Barretto Sampaio
Abstract
• We modeled soil organic carbon (SOC) in semi-arid Brazil. • Integrated satellite indices and spectroscopy improved prediction accuracy. • Radar vegetation indices reflected biomass and enhanced SOC estimation. • Environmental covariables added precision to Vis-NIR spectral models. • Models’ transparency and interpretability was prioritized for SOC prediction. Soil Organic Carbon (SOC) is a paramount soil attribute for climate regulation, soil fertility, and agricultural productivity. The global demand for SOC testing came in response to expanding soil management practices aimed at ensuring soil health. This study explores enhanced accuracy in predicting SOC using soil spectroscopy (proximal sensing). A Soil Spectral Library (SSL), made from 127 soil profiles in Northeast Brazil, mainly by using soils from a semi-arid region, was used. Four modeling scenarios were employed, incorporating distinct covariable sets: 1) diffuse reflectance from laboratory spectroscopy (SSL); 2) diffuse reflectance and radar vegetation indices from all-weather and globally available Sentinel-1 satellite data; 3) diffuse reflectance and environmental factors; 4) all covariables. Integration of radar vegetation indices and environmental factors significantly improved SOC estimates by soil spectroscopy. Predicting SOC solely from SSL reflectance data yielded an average RMSE of 4.54 g kg −1 and R 2 of 0.62. However, by using all covariables significantly reduced RMSE by approximately 13 % (to 3.94 g kg −1 ) and increased R 2 by 14 % (to 0.71). This comprehensive approach, combining SSL, satellite radar vegetation indices, and environmental variables, substantially advances SOC spectroscopic prediction accuracy, offering valuable insights for applications in agriculture and environmental monitoring. These findings contribute to the reliability of proximal and remote sensing methodologies in soil testing.