Document Type : Research Article

Authors

Abstract

This study was conducted to evaluate using terrain attributes derived from digital elevation model (DEM) as ancillary data to predict soil organic carbon (SOC) by implementing different statistical and geostatistical techniques. A linear regression model (LR), Artificial Neural Network model (ANN), ordinary kriging (OK), ordinary co-kriging (OCK), regression kriging (RK) and kriging with an external drift (KED) were performed to predict spatial distribution of SOC in an area of 2400 km2 in mashhad, iran. The SOC was measured for 200 soil samples of the study area and their corresponding Terrain attributes value was extracted from derived from 10-m resolution DEM. correlation between measured SOC and individual terrain attributes was determined, the number of 160 data were used for model development and 40 as validation data set. Resulting maps of different interpolation methods were compared to evaluate map quality using MAE and R2 criteria calculated from plotting measured versus estimated data. The results showed that there is a significant but not strong correlation between SOC and terrain attributes. The comparison of estimation techniques showed that the KED technique with wetness index as ancillary data has the best performance (MAE=0.18 %, R2=0.67) of all, but no significant difference with RK. There were modest differences between maps created with geostaistical technique but sensible difference with LR and ANN ones. The results of this study propose that although there is a significant correlation between SOC and terrain attributes therefore It can be use for enhancing the quality of map, but it is not able to express the spatial variability of SOC as it is necessary for detailed soil map. Because there is other factors controlling SOC spatial distribution

Keywords

CAPTCHA Image