Prediction of Soil Organic Carbon based on Sentinel-2 NDVI data in Sardinia, Italy
Autore/i: A. Frongia, D. Manca, M.L. Biggio, M. Cocco, M. Melis, V.A. Marrone, R. Puddu, S. FanniAnno di pubblicazione: 2023
Lingua: Inglese
Abstract:
Major drivers of gains or losses in soil organic carbon (SOC) include land management, land use change, and climate change (Beillouin at al., 2022). Environmental variables are widely used in SOC prediction. However, it is still difficult to determine which methods and variables are effective for SOC mapping (Chen at al., 2022). While knowledge of the spatial distribution of soil SOC content and its mapping is closely linked to its conservation (Zeraatpisheh at al., 2021), a spatially explicit map of it on the Sardinian Island is non-existent. The peculiarity of the latter is a great variety of landscapes owing to the different parent material, climatic conditions, topography, vegetation, and geomorphology. The interaction of these features has produced many soil types, such as Leptosols, Regosols, Luvisols, Cambisols and Umbrisols, with large differences in SOC content. Therefore, the aim of this study was to quantify the spatial distribution of SOC stocks and associated uncertainties to a target depth of 0–30 cm based on a multiple linear regressions (MLR) approach for Pedological Units to fill this knowledge gap. Data for 4032 georeferenced topsoils, extracted from Sardinian Soils Database (DBSS), were divided into calibration (n=3000) and validation (n=1032) dataset. Environmental variables including temperature, precipitation, elevation, slope, distance from the coast and Normalized Difference Vegetation Index (NDVI) data have been explored and included as independent variables to establish the model and estimate the SOC stock. We selected all S-2 cloud-free images covering the Sardinian Island (by averaging the monthly values of the most recent images) obtaining an average annual NDVI value. This study presente an effective method to overcome the selection of auxiliary variables for digital soil mapping in Sardinian Island and indicate that NDVI was conducive for predicting SOC. The results and the method will show exhaustively in the Poster.