Improving soil organic carbon predictions from a Sentinel–2 soil composite by assessing surface conditions and uncertainties

Abstract

Soil organic carbon (SOC) prediction from remote sensing is often hindered by disturbing factors at the soil surface, such as photosynthetic active and non–photosynthetic active vegetation, variation in soil moisture or surface roughness. With the increasing amount of freely available satellite data, recent studies have focused on stabilizing the soil reflectance by building reflectance composites using time series of images. Although composite imagery has demonstrated its potential in SOC prediction, it is still not well established if the resulting composite spectra mirror the reflectance fingerprint of the optimal conditions to predict topsoil properties (i.e. a smooth, dry and bare soil).

We have collected 303 photos of soil surfaces in the Belgian loam belt where five main classes of surface conditions were distinguished: smooth seeded soils, soil crusts, partial cover by a growing crop, moist soils and crop residue cover. Reflectance spectra were then extracted from the Sentinel–2 images coinciding with the date of the photos. After the growing crop was removed by an NDVI < 0.25, the Normalized Burn Ratio (NBR2) was calculated to characterize the soil surface, and a threshold of NBR2 < 0.05 was found to be able to separate dry bare soils from soils in unfavorable conditions i.e. wet soils and soils covered by crop residues. Additionally, we found that normalizing the spectra (i.e. dividing the reflectance of each band by the mean reflectance of all spectral bands) allows for cancelling the albedo shift between soil crusts and smooth soils in seed–bed conditions. We then built the exposed soil composite from Sentinel–2 imagery for southern Belgium and part of Noord-Holland and Flevoland in the Netherlands (covering the spring periods of 2016–2021). We used the mean spectra per pixel to predict SOC content by means of a Partial Least Squares Regression Model (PLSR) with 10–fold cross–validation. The uncertainty of the models was assessed via the prediction interval ratio (PIR). The cross validation of the model gave satisfactory results (mean of 100 bootstraps: model efficiency coefficient (MEC) = 0.48 ± 0.07, RMSE = 3.5 ± 0.3 g C kg–1, RPD = 1.4 ± 0.1 and RPIQ = 1.9 ± 0.3). The resulting SOC prediction maps show that the uncertainty of prediction decreases when the number of scenes per pixel increases, and reaches a minimum when at least six scenes per pixel are used (mean PIR of all pixels is 12.4 g C kg–1, while mean SOC predicted is 14.1 g C kg–1). The results of a validation against an independent data set showed a median difference of 0.5 g C kg–1 ± 2.8 g C kg–1 SOC between the measured (average SOC content 13.5 g C kg–1) and predicted SOC contents at field scale. Overall, this compositing method shows both realistic within field and regional SOC patterns.