137-2 Comparison of Different Prediction Techniques for Digital Soil Mapping of Soil Clay Content In An Area of Diverse Landscape Types In Denmark.

Poster Number 1530

See more from this Division: S05 Pedology
See more from this Session: New Challenges for Digital Soil Mapping: II
Monday, October 22, 2012
Duke Energy Convention Center, Exhibit Hall AB, Level 1
Share |

Kabindra Adhikari1, Rania B. Kheir1, Peder K. BÝcher2, Mette B. Greve1 and Mogens H. Greve1, (1)Department of Agroecology, Aarhus University, Tjele, Denmark
(2)Department of Biosciences, Aarhus University, Aarhus, Denmark
Most soils and environmental research and modeling studies need soil data as primary inputs where soil property maps have been considered as a main source of soil information. Different prediction techniques are available to map soil properties at different scales; however, choosing  the most appropriate technique for a given site primarily depends on many factors such as available soil and environmental data and their relationship, prediction models and their  performance in defining such relationship, spatial autocorrelation of the soil properties and to some extent the scale of operation. We studied the prediction performance of Ordinary kriging (OK), Stratified OK (OKstrata), Regression Trees (RT) and Regression Rules (RR) for digital mapping of soil clay content using approximately 6920 topsoil (0-30cm) samples from a 45km wide east-west transect in mid-Jutland in Denmark which represents all major Danish landscape types. We used 80% data for the prediction model building and the rest for validation studies. Existing landscape classes were considered for stratification in OKstrata whereas several land surface parameters extracted from LiDAR DEM together with soil and landscape types, land use and geology were used for regression modeling in RT and RR. Predicted and measured clay contents at 20% validation points were compared and R2, RMSE, and Residual prediction deviation (RPD) values were calculated to check the model performance. The mean and CV of clay content in our study area were 6.82 and 71.39% respectively. Among all the prediction methods, highest R2 (0.74) and lowest RMSE (0.27) were associated with the RR with RPD value of 2.24 suggesting RR as the best, most stable and an effective prediction method. The performances of OK and OKstrata were similar but lower than that of RT and RR. So we suggest RR the most suitable prediction method to map topsoil clay content in our study area and we recommend to use this method for further soil mapping activities in Denmark.
See more from this Division: S05 Pedology
See more from this Session: New Challenges for Digital Soil Mapping: II