Using Accumulation Based Network Identification Methods to Identify Hill Slope Scale Drainage Networks in a Raster GIS
The simple accumulation-based network identification method (ANIM) in a raster Geographic Information System (GIS) posed by O'Callaghan and Mark (1984) has been criticized for producing a spatially uniform drainage density (Tarboton 2002) at the watershed scale. This criticism casts doubt on the use of ANIMs for deriving properties such as overland flow length for nonpoint source pollution models, without calibrating the accumulation threshold value. However, the basic assumption that underlies ANIMs is that convergent topography will yield a more rapid accumulation of cells, and thus, more extensive flow networks, with divergent, or planar terrain yielding sparser networks. Previous studies have focused on networks that are coarser than the hill-slope scale, and have relied upon visual inspection of drainage networks to suggest that ANIMs lack the ability to produce diverse networks. In this study overland flow lengths were calculated on a sub-watershed basis, with standard deviation, and range calculated for sub-watershed populations as a means of quantifying the diversity of overland flow lengths produced by ANIM at the hill slope scale. Linear regression and Spearman ranking analyses were used to determine if the methods represented trends in overland flow length as suggested by manual delineation of contour lines. Three ANIMs were analyzed: the flow accumulation method (O'Callaghan and Mark, 1984), the terrain curvature method (Tarboton, 2000) and the ridge accumulation method (introduced in this study). All three methods were shown to produce non-zero standard deviations and ranges using a single support area threshold, with the terrain curvature method producing the most diverse networks, followed by the ridge accumulation method, and then the flow accumulation method. At an analysis unit size of 20 ha, the terrain curvature method produced a standard deviation that was most similar to those suggested by the contour crenulations, -13.5%, followed by the ridge accumulation method, -21.5%, and the flow accumulation method, -61.6%. The ridge accumulation produced the most similar range, -19.1%, followed by terrain curvature, -24.9%, and flow accumulation, -65.4%. While the flow accumulation networks had a much narrower range of predicted flow lengths, it had the highest Spearman ranking coefficient, Rs=0.722, and linear regression coefficient, R2=0.602. The terrain curvature method was second, Rs=0.641, R2=0.469, and then ridge accumulation, Rs=0.602, R2=0.490. For all methods, as threshold values were varied, areas of dissimilar morphology (as evidenced by the common stream metric stream frequency) experienced changes in overland flow lengths at different rates. This results in an inconsistency in ranking of sub-watersheds at different thresholds. When thresholds were varied to produce average overland flow lengths from 75 m to 150 m, the terrain curvature method showed the lowest incidence of rank change, 16.05%, followed by the ridge accumulation method, 16.73%, then flow accumulation, 25.18%. The results of this investigation suggest that for all three methods, a causal relationship exists between threshold area, underlying morphology, and predicted overland flow length. This causal relationship enables ANIMs to represent contour network trends in overland flow length with a single threshold value, but also results in the introduction of rank change error as threshold values are varied. Calibration of threshold value (varying threshold in order to better match observed overland flow lengths) is an effective means of increasing the accuracy of ANIM predictions, and may be necessary when comparing areas with different stream frequencies. It was shown that the flow accumulation method produces less diverse networks than the terrain curvature and ridge accumulation methods. However, the results of rank and regression analyses suggest that further investigation is required to determine if these more diverse ANIM are in fact more accurate than the flow accumulation method.