Evaluating the Potentials of PLSR and SVR Models for Soil Properties Prediction Using Field Imaging, Laboratory VNIR Spectroscopy and Their Combination

Pedo-spectroscopy has the potential to provide valuable information about soil physical, chemical, and biological properties. Nowadays, we may predict properties using VNIR field imaging spectra (IS) such as Prisma satellite data or laboratory (LS). The primary goal of this study is investigate machine learning models namely Partial Least Squares Regression (PLSR) Support Vector (SVR) for prediction several properties, including clay, sand, silt, organic matter, nitrate NO3-, calcium carbonate CaCO3, five dataset combinations (% IS, % LS) follows: C1 (0% 100% LS), C2 (20% 80% C3 (50% 50% C4 (80% 20% C5 (100% 0% LS). Soil samples were collected at bare soils upper (0–30 cm) layer. set been split into a training (n = 248) validation 61). proposed PLSR SVR trained then tested each combination. According our results, outperforms both: For Organic Matter (SOM) prediction, it achieves (R2 0.79%, RMSE 1.42%) 0.76%, 1.3%), respectively. fusion improved property prediction. highest improvement was obtained SOM 0.80%, 1.39) when model applying second Combination IS