Soils contain the largest terrestrial pool of organic carbon [1
]. A relatively small increase in the soil organic carbon (SOC) stocks could, therefore, play an important role in limiting the net flux of greenhouse gases towards the atmosphere and mitigating climate change. SOC is also a major component of soil fertility and resilience [4
], and increasing SOC content may help climate change adaptation and increase food security, especially in soils having low SOC contents. The societal need for monitoring SOC content has been expressed through the 4p1000 initiative [1
], particularly for topsoil, which is directly impacted by tillage practices [4
] and receives the most organic inputs. More widely, the various effects that agricultural practices have on SOC changes still need to be appraised [5
], which requires updating and monitoring their status and changes in space and time.
The standard method for SOC content measurement consists of collecting soil samples in the field and then preparing 2 mm-sieved air-dried soil samples for their analysis according to standard laboratory determination: it is both time-consuming and expensive. Yet SOC content assessment must be carried out at spatial scales useful for management and for decisions by stakeholders and decision-makers; namely, those at the scale of local, regional, and national or even continental territories. However, until now there is no efficient and straightforward way of monitoring topsoil organic carbon content at such scales. Because soil reflectance in visible near infrared and short wave infrared (Vis-NIR-SWIR, 0.4–2.5 µm) is strongly influenced by SOC and some other soil compounds, such as clay minerals, calcium carbonate and iron oxides, empirical spectral models relating soil reflectance spectra to spectrally-influent properties have been successfully built; e.g., [6
]. However, this was mostly done under controlled lab conditions (e.g., [6
]) and over limited spatial coverages or soil sampling densities. Few studies attempted to spectrally predict SOC contents from Vis-NIR-SWIR image spectra at a regional scale covering tens to some hundreds of km2
: they relied either on airborne hyperspectral images [8
], or on satellite hyperspectral HYPERION data [11
], or even on multispectral satellite data of the former generation, such as SPOT-4 and SPOT-5 [14
]. The key issue regarding whatever airborne acquisition is concerned, is that they are very expensive and not widely available, even more so at regular time intervals, while the above-mentioned satellite images are also not numerous. Moreover, hyperspectral satellite HYPERION has a low signal-to-noise ratio, while multispectral SPOT satellite sensors have low spectral diversity and resolution, limiting prediction performances.
The new generation Sentinel-2 satellites (S2) launched in 2015 (S2A) and then 2017 (S2B), provide time series with frequent revisit every 5 days, which is very promising for the purpose of updating spatial information on SOC content. Indeed, recent works have shown the relevance of the MultiSpectral Instruments (MSIs) aboard Sentinel-2 satellites to predict SOC content over temperate agroecosystems characterized by annual crops in the Czech Republic [15
], Luxemburg, Belgium, Germany [16
], and France [17
]. The MSI has more spectral bands (13) than previous multispectral satellite sensors, covering the Vis-NIR-SWIR spectral range. S2 data are imaged from space over large areas with a 290 km-swath width, at a spatial resolution of either 10 m (bands 2, 3, 4, 8) or 20 m (bands 5, 6, 7, 8A, 11 and 12), which is sufficient for soil surveying.
In detail however, the prediction performance of these studies, as distinguished with cross-validation residual prediction deviation (RPDcv), varied from 1.0 to 1.8 [15
]. Because of crop rotation, the soil areal fraction that is available for the mapping of topsoil properties from remote sensing imagery varies with acquisition date. Furthermore, soil surface conditions change across the available bare soil areas, particularly in terms of soil moisture and roughness according to tillage operations [18
]. The spectral models of SOC content prediction assume that soil organic matter influences the reflectance of bare soil, and such influence operates along the Vis-NIR-SWIR range without unique absorption peaks. Therefore, any factor that disturbs the reflectance spectrum may disturb prediction from this spectrum. Spectral disturbance can be due to: (i) other spectrally influent soil compounds, such as iron, calcium carbonate, coarse fragments [17
], water [19
], the presence of a partial vegetation cover [22
] and crop residues on surface [24
]; (ii) soil surface morphometry (roughness) that generates differential effects of light [26
]; and (iii) an atmospheric disturbance of the at-field reflectance signal [29
], varying with season, clouds, sun azimuth and elevation. None of the recent works have investigated the impact that acquisition date may have on prediction performance for SOC content. Gomez et al. [31
] showed that several single-date S2 data acquired over a same study area may provide different performances of soil texture prediction, but they did not investigate the reasons for these differences. This is the purpose of the present study, considering an area where encouraging performance was previously obtained from a single spring date of March, for the Versailles Plain [17
]: are there optimal dates for predicting topsoil SOC and what are the main factors disturbing the SOC prediction?