HYDROLOGIC MODELING UNCERTAINTY RESULTING FROM LAND COVER MISCLASSIFICATION 1806

Authors: MILLER, S. - UNIVERSITY OF WYOMING; GUERTIN, D. - UNIVERSITY OF ARIZONA; Goodrich, David - Dave

Submitted to: Journal of the American Water Resources Association
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/23/2006
Publication Date: 8/15/2007
Citation: Miller, S.N., Guertin, D.P., Goodrich, D.C. 2007. Hydrologic modeling uncertainty resulting from land cover misclassification. Journal of the American Water Resources Association 43(4):1065-1075.

Interpretive Summary: When water quantity or water quality is of interest, watersheds are a natural organizing unit in our landscape. Watersheds gather rainfall, infiltrated water, and runoff and typically discharge that water at a stream location or into a body of water such as a lake or estuary. The pathways and processes that affect runoff generation from a watershed result from a complex interaction of the climate, topography, soils, land cover, and land use. Land cover data over large watersheds is often obtained from satellite images. The digital signals from these images are classified into different land covers such as forest, grass land, urban, etc. by a variety of computer aided methods. When a satellite image pixel is classified, errors in assigning the pixel to the correct land cover class invariably occur. With additional analyses and information obtained from ground-based checks these errors can be quantified. Since land cover influences watershed runoff, errors in the land cover will impact the simulation of watershed runoff when using a watershed computer model. This study developed a method to access the impacts of land cover mis-classification on modeled watershed runoff. This method was tested on over 40 small to large watersheds in the San Pedro basin in southeastern Arizona. Results indicate that as watershed size increases and storm size increases, the impacts of land cover classification errors on runoff decrease. These results provide information on how good a land cover classification must be for watershed runoff modeling. This is also important for water quality as runoff plays a major factor in how sediment, and other chemicals move from the land surface into water bodies.

Technical Abstract: When water quantity or water quality is of interest, watersheds are a natural organizing unit in our landscape. Watersheds gather rainfall, infiltrated water, and runoff and typically discharge that water at a stream location or into a body of water such as a lake or estuary. The pathways and processes that affect runoff generation from a watershed result from a complex interaction of the climate, topography, soils, land cover, and land use. Land cover data over large watersheds is often obtained from satellite images. The digital signals from these images are classified into different land covers such as forest, grass land, urban, etc. by a variety of computer aided methods. When a satellite image pixel is classified, errors in assigning the pixel to the correct land cover class invariably occur. With additional analyses and information obtained from ground-based checks these errors can be quantified. Since land cover influences watershed runoff, errors in the land cover will impact the simulation of watershed runoff when using a watershed computer model. This study developed a method to access the impacts of land cover mis-classification on modeled watershed runoff. This method was tested on over 40 small to large watersheds in the San Pedro basin in southeastern Arizona. Results indicate that as watershed size increases and storm size increases, the impacts of land cover classification errors on runoff decrease. These results provide information on how good a land cover classification must be for watershed runoff modeling. This is also important for water quality as runoff plays a major factor in how sediment, and other chemicals move from the land surface into water bodies.