Sampling Strategy for Digital Soil Mapping in the Thar Desert Region of India: A Conditioned Latin Hypercube Sampling Approach Moharana Pravash Chandra*, Yadav Brijesh1, Meena Roshan Lal1, Nogiya Mahaveer1, Malav Lal Chand1, Tailor Bhagwati Lal1, Biswas Hrittick, Patil Nitin Gorakh ICAR-National Bureau of Soil Survey and Land Use Planning, Nagpur, 440033, Maharashtra, India 1Present address: ICAR-National Bureau of Soil Survey and Land Use Planning, Regional Centre, Udaipur, 313001, Rajasthan, India *Corresponding author (Email: pravashiari@gmail.com)
Online Published on 02 August, 2024. Abstract In soil surveys, appropriate soil sampling technique plays a crucial role in the accurate prediction of soil properties. In recent years, conditioned latin hypercube sampling (cLHS) system has gained prominence in soil surveys. The objective of this work was to develop a sampling strategy in remote areas of Thar Desert region of India based on the cLHS technique and evaluate its operational performance in digital soil mapping. A digital elevation model and its terrain derivatives were the basis for cLHS to determine the sampling points. The cLHS system also required a cost map representing the difficulty of reaching every place in the area. The results showed that this method was able to capture and represent the spatial variation/distribution of the study area in the Thar desert. It was concluded that 80 samples were optimum for digital mapping of soil organic carbon (SOC) content and other soil properties. The cLHS based random forest model predicted the SOC values with R2 (training) 0.961 and R2 (testing) 0.368. The concordance correlation coefficient value varied from 0.160 to 0.418 across the soil depth in the validation data set, suggesting poor agreement between the predicted and observed values. The poor prediction may be attributed to more variability in SOC influenced by soil intrinsic (pedogenic) and extrinsic (land management) factors. The most important variables for predicting SOC variations were the VD, EVI, NDVI, CI, TCA, and TPI. In topsoil (0-5 and 5-15 cm), vegetation index like EVI, was the second crucial covariate. Despite low prediction accuracy, it was evident that the cLHS reduced the time and resources required for the fieldwork. Top Keywords Conditioned latin hypercube sampling, Digital soil mapping, Random forest model, Soil organic carbon, Thar desert region. Top |