Mapping Soil Characteristics: Spatio-Temporal Comparison of Land Use Regression and Ordinary Kriging in an Arid Environment

• Published in: Journal of the Indian Society of Remote Sensing 2024, Vol.52 (1), p.79-93
• Main Authors: pirestani, Niloofar, Ahmadi Nadoushan, Mozhgan, Abolhasani, Mohammad Hadi, Zamani Ahmadmahmoudi, Rasool
• Format: Article
• Language: English
• Subjects: Arid environments; Climate change; Distribution patterns; Earth and Environmental Science; Earth Sciences; Independent variables; Land use; Landsat; Organic carbon; Random sampling; Remote sensing; Remote Sensing/Photogrammetry; Research Article; Soil mapping; Summer; Support vector machines; Surface temperature; Topsoil; Vegetation index
• ISSN: 0255-660X; 0974-3006
• DOI: 10.1007/s12524-023-01804-y

This study compared the performance of three interpolation models: Ordinary Kriging (OK) and two Land Use Regression (LUR) models trained with Random Forest (RF) and Support Vector Machine (SVM) models in the distribution prediction of organic carbon (OC), total nitrogen (TN) and pH in central Iran. Soil characteristics were measured from 60 samples collected through a systematic random sampling approach. An array of 12 independent variables, divided into three groups of terrain characteristics (elevation, slope and aspect), Landsat 8-derived remote sensing indices (NDVI, EVI, NDWI, MNDWI, TVI, TVI and MSAVI) and climatic variables (Seasonal mean surface temperature and rain), were used to train LUR models. The best performance was obtained by SVM-LUR in the prediction of TN (RMSE range of 0.011–0.027). In the study area, pH values were found to be independent of human activities. In comparison with the pH distribution pattern, topsoil OC and TN stocks had a high variability across the region. The highest OC and TN percentage were measured in summer and distributed along the Zayandeh-rood River in which intense agricultural activities are present, especially in the summer season. DVI and MSAVI as vegetation indices had a significant influence on the performance of distribution prediction.