@article{fdi:010076683,
  title = {{S}atellite data integration for soil clay content modelling at a national scale},
  author = {{L}oiseau, {T}. and {C}hen, {S}. and {M}ulder, {V}. {L}. and {D}obarco, {M}. {R}. and {R}icher-de-{F}orges, {A}. {C}. and {L}ehmann, {S}. and {B}ourennane, {H}. and {S}aby, {N}. {P}. {A}. and {M}artin, {M}. {P}. and {V}audour, {E}. and {G}omez, {C}{\'e}cile and {L}agacherie, {P}. and {A}rrouays, {D}.},
  editor = {},
  language = {{ENG}},
  abstract = {{S}oil clay content is a key parameter that influences many other soil properties and processes. {T}he potential of adding new and contemporary satellite data for soil property mapping in {F}rance is assessed in this study. {T}he soil property maps used for this analysis were produced within the framework of {G}lobal{S}oil{M}ap, which was created to deliver global fine grids of soil properties and associated uncertainties using existing soil information and ancillary data to predict these properties based on digital soil mapping techniques. {I}n this study, we evaluate the added value of {M}oderate {R}esolution {I}maging {S}pectroradiometer ({MODIS}), {P}roject for {O}n-{B}oard {A}utonomy-{V}egetation ({PROBA}-{V}), and {S}entinel-2 ({S}2) data for predicting the soil clay content at 90 m resolution for mainland {F}rance. {T}he rationale behind adding these data is that satellite images and derived products may enable the biogeo-chemical characteristics of the earth's surface to be captured more effectively, which in turn enables more precise predictions of the soil clay content. {F}or this methodology, we i) create composite bare soil mosaics and derive the spectral indices from {S}2 data acquired during sowing periods from 2016 to 2017, ii) extract the first three principal components of harmonized {MODIS} and {PROBA}-{V} normalized difference vegetation index ({NDVI}) time series acquired in 2003 and 2016 to represent vegetation changes, and iii) test whether the complementary datasets are able to improve the soil clay information compared to a benchmark value. {T}he soil clay content is obtained by using quantile regression forest ({QRF}) for each {G}lobol{S}oil{M}ap depth interval of 0-5 cm, 5-15 cm, 15-30 cm, 30-60 cm, 60-100 cm, and 100-200 cm along with a 10-fold cross-validation having 10 replicates. {T}he results show that the complementary satellite data improve the clay content estimation on bare soil for the topsoil layers (e.g., 0-30 cm) by increasing the {R}-2 and decreasing the bias at averages of 0.05 and 1 g kg(-1), respectively. {M}oreover, the first principal component of the harmonized {NDVI} data is shown to be the second most important variable for estimating the clay content, as indicated by the {QRF} models. {H}owever, the use of only the satellite data and products as input for the {QRF} does not yield a satisfactory estimate of the clay content. {F}inally, this work provides a reference for embedding new remote sensing data in existing national soil inventories and national soil information systems. {F}urther research should incorporate new techniques for considering the spatial-temporal variability of the earth's surface parameters such as soil moisture and roughness.},
  keywords = {{R}emote sensing ; {D}igital soil mapping ; {S}entinel-2 ; {N}ational scale ; {Q}uantile random forest ; {FRANCE}},
  booktitle = {},
  journal = {{I}nternational {J}ournal of {A}pplied {E}arth {O}bservation and {G}eoinformation},
  volume = {82},
  numero = {},
  pages = {art. 101905 [16 ]},
  ISSN = {0303-2434},
  year = {2019},
  DOI = {10.1016/j.jag.2019.101905},
  URL = {https://www.documentation.ird.fr/hor/fdi:010076683},
}