Comparing hyperspectral spaceborne EnMAP, PRISMA and DESIS missions for soil salinity mapping in a high-saline and semi-arid area

Novel hyperspectral spaceborne missions might play an important role in monitoring environmental stressors such as soil salinity for rice crops. This study aimed at comparing the performance of three hyperspectral spaceborne sensors (DESIS, EnMAP and PRISMA) for soil salinity mapping using electrical conductivity (EC) as proxy. A DESIS image was acquired in May 2022, whereas the EnMAP and PRISMA images were acquired in March 2023 in Bajo Guadalquivir (southern Spain). Raw bands and soil salinity indices (SSI) were evaluated for each image. Two machine learning methods, Random Forest (RF) and Support Vector Regression (SVR) were used together with a Sequential Feature Selection method. The results showed similar performance for EC using both sensors and methods. EnMAP models reached the highest accuracy using RF (R
²
= 0.14, RMSE = 3.14), but SVR performed worse than RF (R
²
= -0.37 and RMSE = 3.38). PRISMA models were very similar. RF model achieved a R
²
= 0.01 and RMSE of 3.25, whereas SVR model got a R
²
= 0.08 and RMSE = 3.24. DESIS models performed worse than those with EnMAP and PRISMA data, with an (R
²
= -0.1 and -0.03 and RMSE = 3.56 and 3.67 for RF and SVR respectively). Both EnMAP models selected the same two features, two SSI, 756/871 nm, and the 972/1234 nm. PRISMA models selected one feature only, RF the 592/2268 nm SSI pair, while the SVR model selected the SSI composed of 453/2400 nm bands. The RF DESIS model selected the 647.4/979.9 SSI, whereas SVR did the 811.5/860.2 SSI.