An assessment of hypertemporal Sentinel-2 and hyperspectral PRISMA imagery for indirect soil organic matter and pH mapping in a forested Mediterranean area

Soil mapping over forested areas could benefit significantly from advanced remote sensing approaches, including novel satellite hyperspectral scanners and high-frequency multispectral sensors, together with innovative modelling methods including stacking learning. This study evaluated hypertemporal data derived from Sentinel-2 time series alongside hyperspectral data from PRISMA sensor for soil organic matter (SOM) and pH mapping within a complex forested Mediterranean area. Ninety-two soil samples were collected from Sierra de las Nieves, southern Spain. Two different datasets were utilized for modelling soil properties: a hypertemporal set extracting thirteen phenological features from dense Sentinel-2 time-series (2018–2020), and a hyperspectral set combining raw PRISMA bands with fifty-eight narrow-band indices. Both datasets incorporated terrain features at their native resolutions. The modelling framework employed an ensemble stacking learner based on a random search Feature Selection method. Results highlighted the Middle of Season date as a key feature for SOM modelling, while the Large Integral was crucial for pH in the hypertemporal model. Conversely, the hyperspectral model relied on various spectral bands for SOM and terrain features for pH. The most accurate results were feasible but challenging, being achieved using the hypertemporal dataset for both SOM (RMSE = 5.11, R 2 = 0.36) and pH (RMSE = 0.38, R 2 = 0.42), although the hyperspectral dataset yielded comparable performance, even considering the time acquisition lag. Hence, both approaches provided new insights for soil indirect mapping in forested areas, paving the way to operational soil indirect mapping with hypertemporal and hyperspectral data. • Hypertemporal Sentinel 2 and hyperspectral PRISMA data were evaluated for soil mapping. • An ensemble stacking approach was used for soil organic matter and pH prediction. • Phenological features from S-2 time series and narrow-band hyperspectral indices from PRISMA were extracted. • Middle of Season date, Large Integral and 801 nm band were the most relevant features of the remote sensing subset. • Both remote sensing approaches showed promising results for SOM and pH mapping.