Publications des scientifiques de l'IRD

Vaudour E., Gomez Cécile, Lagacherie P., Loiseau T., Baghdadi N., Urbina-Salazar D., Loubet B., Arrouays D. (2021). Temporal mosaicking approaches of Sentinel-2 images for extending topsoil organic carbon content mapping in croplands. International Journal of Applied Earth Observation and Geoinformation, 96, p. 102277 [18 p.]. ISSN 1569-8432.

Titre du document
Temporal mosaicking approaches of Sentinel-2 images for extending topsoil organic carbon content mapping in croplands
Année de publication
2021
Type de document
Article référencé dans le Web of Science WOS:000608483300003
Auteurs
Vaudour E., Gomez Cécile, Lagacherie P., Loiseau T., Baghdadi N., Urbina-Salazar D., Loubet B., Arrouays D.
Source
International Journal of Applied Earth Observation and Geoinformation, 2021, 96, p. 102277 [18 p.] ISSN 1569-8432
The spatial assessment of soil organic carbon (SOC) is a major environmental challenge, notably for evaluating soil carbon stocks. Recent works have shown the capability of Sentinel-2 to predict SOC content over temperate agroecosystems characterized with annual crops. However, because spectral models are only applicable on bare soils, the mapping of SOC is often obtained on limited areas. A possible improvement for increasing the number of pixels on which SOC can be retrieved by inverting bare soil reflectance spectra, consists of using optical images acquired at several dates. This study compares different approaches of Sentinel-2 images temporal mosaicking to produce a composite multi-date bare soil image for predicting SOC content over agricultural topsoils. A first approach for temporal mosaicking was based on a per-pixel selection and was driven by soil surface characteristics: bare soil or dry bare soil with/without removing dry vegetation. A second approach for creating composite images was based on a per-date selection and driven either by the models performance from single date, or by average soil surface indicators of bare soil or dry bare soil. To characterize soil surface, Sentinel-1 (S1)-derived soil moisture and/or spectral indices such as normalized difference vegetation index (NDVI), Normalized Burn Ratio 2 (NBR2), bare soil index (BSI) and a soil surface moisture index (S2WI) were used either separately or in combination. This study highlighted the following results: i) none of the temporal mosaic images improved model performance for SOC prediction compared to the best single-date image; ii) of the per-pixel approaches, temporal mosaics driven by the S1-derived moisture content, and to a lesser extent, by NBR2 index, outperformed the mosaic driven by the BSI index but they did not increase the bare soil area predicted; iii) of the per-date approaches, the best trade-off between predicted area and model performance was achieved from the temporal mosaic driven by the S1-derived moisture content (R-2 similar to 0.5, RPD similar to 1.4, RMSE similar to 3.7 g.kg(-1)) which enabled to more than double (*2.44) the predicted area. This study suggests that a number of bare soil mosaics based on several indicators (moisture, bare soil, roughness...), preferably in combination, might maintain acceptable accuracies for SOC prediction whilst extending over larger areas than single-date images.
Plan de classement
Pédologie [068] ; Sciences du monde végétal [076] ; Télédétection [126]
Description Géographique
FRANCE ; VERSAILLES PLAINE
Localisation
Fonds IRD [F B010080662]
Identifiant IRD
fdi:010080662
Contact