Submitted to: Carotgraphy and Geographic Information Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/2/2008
Publication Date: 12/1/2008
Citation: Wu, S., Finn, M.P., Usery, E.L., Bosch, D.D. 2008. Effect of cell sizes on spatial statistics of AGNPS-Simulated Runoff. Carotgraphy and Geographic Information Science. 35(4) pp 265-278. Interpretive Summary: Watershed modeling is a frequently used tool for water resource planning and management. These models require input data which represent the conditions on the land. Frequently, subjective decisions must be made regarding the resolution or accuracy of these input data. This research uses statistical methods to provide guidance on this resolution. The results indicate that the optimum input resolution which provides accurate assessments of simulated surface runoff while minimizing the effort involved with generating the input characterization is 210 m. Guidance regarding the expectations for using lower resolution input as well as investigations into variability across the landscape is also provided.
Technical Abstract: The purpose of this study is to investigate how the properties and statistics of predicted runoff from the Agricultural Non-Point Source (AGNPS) pollution model change with model input data at eight different cell sizes (30 m, 60 m, 120 m, 210 m, 240 m, 480 m, 960 m, and 1920 m). The Little River Watershed (Georgia, USA) is used as a case study area. The overland cell runoff, total cell runoff volumes, overall runoff clustering degree, runoff spatial clusters, and runoff hot spots at different data resolutions are studied. The results show total cell runoff, the product of cell area and runoff depth, decreases as data resolution becomes coarser. The clustering degree of watershed runoff decreases with increasing cell size, and the clustering becomes of no statistical insignificance at a data resolution of >1920 m. The general locations of the simulated clusters and hot spots within the watershed do not change with increasing cell size while the size of clusters and hot spots decreases. A data resolution of < 960 m is considered the limit for detecting spatial clusters, while hot spots calculated at this data resolution are still relatively definable. A geogstatisical neighborhood-distance parameter was used to examine the scope of spatial interaction between simulated runoff hot spots. Hot spots become more fragmented at smaller neighborhood distance. The suggested neighborhood distance for detecting hot spots is between 960 m and 4800 m.