Generation of digital soil mapping for Coimbatore districts using multinomial logistic regression approach

Abstract

Digital soil mapping (DSM) is a significant advancement in soil mapping systems, enabling efficient mapping of soil patterns across different temporal and spatial scales. This computer-assisted method surpasses the traditional soil mapping techniques in terms of both compatibility and accuracy. This study employed multinomial logistic regression to map the soil subgroup levels in the Coimbatore district. Primary sample points and Natural Resource Information System (NRIS) database points serve as the dependent variables, while significant covariate layers act as independent variables. The accuracy assessment showed an overall mapping accuracy of 52.58%, with a kappa statistic of 0.50. Additionally, the calculated disagreement measures, including quantity and allocation disagreements, were 21.50% and 25.92%, respectively. The approach provides spatial soil maps at 30 m resolution and was extended for the Coimbatore district of Tamil Nadu, considering the lack of organized high resolution soil maps for operational use. The area statistics calculated from the digital soil map showed that the soil orders Vertisols cover the largest area, accounting for approximately 25.97% (122,630.38 ha) of the total land area. Soil subgroups like Ultic Haplustalfs and Vertic Ustorthents occupy substantial portions of the land, accounting for 9.95% and 9.62% of the total area, respectively. The total land area classified by the map accounts for 427,432.10 ha, i.e., 90.53% of the total land area, of which 44,696.46 ha (9.467%) remains unclassified. The study also presents the statistics on soil order at the block level. These findings provide valuable insights into soil classification, offering a comprehensive understanding of soil distribution and characteristics that support effective decision-making for sustainable land management and agricultural practices.