Prediction of soil depth using a soil-landscape regression model: a case study on forest soils in southern Taiwan.

Techniques for conventional forest soil surveys in Taiwan need to be further developed in order to save time and money. Although some soil-landscape regression models have been developed to describe and predict soil properties and depths, they have seldom been studied in Taiwan. This study establishes linear soil-landscape regression models related to soil depths and landscape factors found in the forest soils of southern Taiwan. These models were evaluated by validating the models according to their mean errors and root mean square errors. The study was carried out at the 60,000 ha Chishan Forest Working Circle. About 310 soil pedons were collected. The landscape factors included elevation, slope, aspect, and surface stone contents. Sixty percent of the total field samples were used to establish the soil-landscape regression models, and forty % were used for validation. The sampling strategy indicated that each representative pedon covers an area of about 147 ha. The number of samples was appropriate considering the available time and budget. The single variate and/or multivariate linear regression soil-landscape models were successfully established. Those models revealed significant inter-relations among the soil depths of the B and B+BC horizons, solum thickness, and landscape factors, including slope and surface stone contents (p < 0.003). The mean errors in the validation of the soil-landscape model were low and acceptable for this case study. In addition, the slope data derived from the DEM (digital elevation model) database in this case study were used to predict the soil depths of the B, B+BC horizons and the solum thickness without carrying out a field survey. Surface stone should be collected in a field soil survey to increase the precision of soil depth prediction of the B and B+BC horizons, and the solum thickness.