Cherie Westbrook


Effects of classification approaches on CRHM model performance

Journal article

Xulin Guo, J. Pomeroy, X. Fang, S. Lowe, Zhaoqin Li, C. Westbrook, A. Minke

Semantic Scholar DOI


Guo, X., Pomeroy, J., Fang, X., Lowe, S., Li, Z., Westbrook, C., & Minke, A. (2012). Effects of classification approaches on CRHM model performance.

Guo, Xulin, J. Pomeroy, X. Fang, S. Lowe, Zhaoqin Li, C. Westbrook, and A. Minke. “Effects of Classification Approaches on CRHM Model Performance” (2012).

Guo, Xulin, et al. Effects of Classification Approaches on CRHM Model Performance. 2012.


The cold regions hydrological model (CRHM) platform, a physically based hydrological model using a modular and object-oriented structure, has been applied for simulating the redistribution of snow by wind, snowmelt, infiltration, evapo-transpiration, soil moisture balance, surface depression storage and run-off routing. Land use and land cover classification is a preprocessing procedure to provide the required parameters for CRHM. Per-pixel-based and object-oriented classifications are the two major classification approaches currently in practice. The objective of this study is to evaluate whether the more complex object-oriented classification method can significantly improve the performance of the CRHM model in a prairie landscape. The study was conducted in the Smith Creek watershed in eastern Saskatchewan. Two Satellite Pour l'Observation de la Terre (SPOT) multispectral images were used to classify the area into seven classes: cropland, fallow, grassland, wetland, water, woodland and town/road. Both pixel-based maximum likelihood supervised classification method and nearest neighbour object-oriented classification approach were applied to the satellite images for the study area. The parameters derived from both classification methods were input into the CRHM to derive the hydrological response unit (HRU), snow water equivalent (SWE) and basin stream flow. Results indicated that classification results influence the model performance slightly. However, no significant improvement from the object-oriented classification was observed for this specific study.