Geochemical investigations of 100 superficial soils observed by Sentinel 2 and PRISMA

Geochemical investigations of agricultural soils are fundamental to characterize pedosphere dynamics that sustain ecosystem services linked with agriculture. Parameters like soil moisture, soil organic matter (SOM), and soil organic carbon (SOC) are strong instruments to evaluate carbon sink potential.

Satellite Earth Observation is a significant source of free data that can be linked to soil characteristics and dynamics and employed to produce temporal series. Access to these data is nowadays facilitated by platforms such as ADAM (https://adamplatform.eu), which allow users to quickly search for, visualize and subset data products, greatly reducing the volume of data that end users must handle.

In this work we demonstrate the usefulness of such systems by carrying out a geochemical investigation of 100 superficial (0-15 cm) soil samples collected in the province of Ferrara (North-Eastern Italy) and using the ADAM platform to associate to each a time series of Sentinel 2 data. The samples were collected in October 2021 in fields that were ploughed or mono-cultivated at maize, soybean, rice, and winter vegetables. To obtain the average soil properties over a spatial scale larger than the satellite sensor resolution, we adopted a composite sampling strategy, merging 5 sub-samples collected at the vertexes and at the center of a 30x30 m² area. Soil granulometry was recognized from clay to medium sand, with exception of peat deposits. Soil moisture, and SOM, contents were estimated by loss on ignition (LOI), respectively at 105°C (values from 0.3 to 7.4 wt%), and 550°C (values from 2.1 to 21.0 wt%). SOC contents (values from 0.7 to 9.3 wt%) were determined through DIN19539 analysis performed with an Elementar soliTOC Cube. Using the ADAM platform, we associated a temporal series from 2016 to 2021 of the Sentinel 2 NDVI data product to each sampling location, using a cloud coverage mask to eliminate values taken on cloudy days. Localized phenological cycles for each year are recognizable in the remotely-sensed data. Hence, our database describes for each parcel, geochemical parameters and vegetative temporal series.

In a separate study, we also attempted to train a neural network to predict geochemical properties from the soil spectrum measured by the hyperspectral satellite PRISMA. We used the geochemical properties of our 100 samples as training data, associated with the PRISMA spectra of the sampling locations measured on April 7 2020, when, according to our NDVI data, none was covered in vegetation.