Document Type : Research Paper
Authors
1
Department of Soil Science, Faculty of Agriculture, Bu-Ali Sina University, Hamedan, Iran
2
Department of Soil Science, Faculty of Agriculture, University of Kurdistan, Sanandaj, Iran
Abstract
Soil water retention characteristics, such as field capacity (FC) and permanent wilting point (PWP), are vital for efficient water management, yet their direct measurement is often challenging. This study aimed to estimate FC and PWP using visible-near infrared (Vis-NIR) spectral data and soil physicochemical properties through random forest (RF) and multiple linear regression (MLR) models. A total of 130 soil samples were collected from five provinces in Iran. Spectral and physicochemical properties were analyzed, and the dataset was divided into training (90) and testing (40) subsets. Eleven pedotransfer functions (PTFs) were developed using three modeling steps. Spectral preprocessing methods, including multiplicative scatter correction (MSC), first and second derivatives with Savitzky–Golay filtering (FD–SG, FD–SG2), and standard normal variate (SNV), were compared with no-preprocessing (NP). The RF model (RMSE = 0.050) outperformed MLR (RMSE = 0.057) for FC prediction. For PWP, RF produced slightly better results across most PTFs, with significant improvement for PTF2 (AIC = −264.3). During training, PTF11 achieved the best performance for FC (AIC = −540.2), while PTF7 showed the highest accuracy for PWP (AIC = −612.4). PTF3, based on sand, clay, and organic matter, was the most accurate estimator of FC (AIC = −553.3), and PTF6, using sand, clay, organic matter, and total porosity, was most effective for PWP (AIC = −616.2). Principal component analysis identified key wavelengths at 409 nm for FC and 1414, 1912, and 2150 nm for PWP. Integrating spectral and soil data with machine learning improved prediction accuracy over spectral-only models.
Keywords
Main Subjects
)