"... In PAGE 4: ...1 Watershed - based segmentation algorithm 1. Free parameters and sensitivity analysis Table1 shows the free parameters in the watershed algorithm and how sensitive the performance of the algorithm is to each. Here sensitivity represents a subjective measure: low sensitivity indicates that the parameter is highly robust and can be used for a variety of images, while high sensitivity indicates that the parame ter is highly sensitive and may need to be adjusted for various images.... ..."

"... In PAGE 4: ...1 Watershed - based segmentation algorithm 1. Free parameters and sensitivity analysis Table1 shows the free parameters in the watershed algorithm and how sensitive the performance of the algorithm is to each. Here sensitivity represents a subjective measure: low sensitivity indicates that the parameter is highly robust and can be used for a variety of images, while high sensitivity indicates that the parame ter is highly sensitive and may need to be adjusted for various images.... ..."

"... In PAGE 7: ...7 analysis based on the following criteria: (1) the entire watershed is contained within the study area; (2) the drainage area is less than 300 km2; (3) there are more than 20 years of flow records; and (4) the gauges are not affected by hydraulic constraints (dams, reservoirs, etc) that were officially stated in the gauge report. Table 1 summarizes the twenty flow gauges selected for this study and Table2 summarizes selected characteristics of the streams and watersheds above these gauges. These watersheds drain 1616 km2 and range in size from 4.... In PAGE 15: ...7% among the 20 watersheds examined in this paper. Thirteen of the 20 watersheds listed in Table2 are relatively pristine as evidenced by percent impervious surface values of less than 5%. Three watersheds (#14, #34, and #31) have experienced moderate development (e.... In PAGE 15: ...mperviousness values of 45.0%, 15.3 %, and 6.6% respectively ( Table2 ). The four remaining watersheds (#15, #25, #27, and #28) are dominated by residential, commercial, and transportation land uses and percent impervious surface values are greater than 50%.... In PAGE 18: ...ainfall events than non-urban (e.g. forested) land surfaces. The high values reported for these mountainous watersheds occurred in landscape with bare rock surfaces (Jennings 1977), steep slopes ( Table2... ..."

"... In PAGE 23: ... Watershed analysis is a method to evaluate the cumulative effects of forestry management on habitat conditions for salmonids and other public resources. The watershed analysis method defines a suite of habitat parameters and provides a numerical value index for each of these parameters to create ratings of good, fair, and poor conditions ( Table6 ). While watershed analysis was developed specifically for forestry, the resource condition indices are appropriate and applicable in other land use situations.... ..."

"... In PAGE 8: ...antly correlated (p lt; .05, Table 5). The strongest correlations were observed between FLOW and LOADPCl for each landscape analysis method. A significant amount of variation in the level of sediment metals and organics was explained by the regression models ( Table6 ). Coefficients of multi- ple determination (R2) ranged from 0.... ..."

"... In PAGE 2: ... In the research of [35] special attention was given to hypsometric analysis of basins at Nanga Parbat (Figure 2). The results are presented in Table1 . The study showed that Nanga Parbat basins can be divided into three groups, based on their dominate surface processes and unique topography [35].... ..."

"... In PAGE 12: ... The non-linear response of dominance (top) and edge density (bottom) are illus- trated. the broadest range of values across sub-catchments ( Table2 ). Dominance and edge density at both the landscape-level and the class-level (Fig.... In PAGE 12: ... However, significant responses of pattern variables to variations in proportion of land use/cover were less frequent when forest or wetland was used as the predictor variable. The proportions of these land use/cover types occur over a relatively small range in values within our data set ( Table2 ), which limits the opportunity to display more complex relationships. Class-level edge density showed a tight relationship with changes in the proportion of the corresponding land cover type (Fig.... ..."

"... In PAGE 6: ...ere digitized from U.S.G.S. 1:24,000 topographic maps. Land use and wetland types were measured using the ERDAS GIs, and expressed as both a per- centage of the lake fringe area (a 200 m band or 4 ha pixel width surrounding each lake) and total watershed area ( Table2... In PAGE 7: ...extract individual watersheds from the regional data files, and to compute average soil (pH, availa- ble soil P) and topographic variables (maximum elevation difference: Table2 ) for each watershed. Streams flowing into each lake were divided into 1 km increments from the inlet to the headwaters us- ing a map wheel.... In PAGE 7: ...4 Statistical analyses To reduce the large number of watershed charac- teristics to a smaller number of variables, we per- formed a principal components analysis (PCA) without rotation on selected watershed variables (Norusis 1988). The original watershed variables were not expressed in common units and thus had a wide range of variances ( Table2 ). Therefore, ... In PAGE 9: ... The presence of linearly related variables will produce a singular matrix (and computer overflow errors) or an ill-conditioned matrix with associated round-off errors (Draper and Smith 1981). To resolve this problem, we chose a subset of 27 watershed variables to eliminate redundant combi- nations and gave preference to variables of known importance to water quality ( Table2 ). For exam- ple, when summary variables (e.... In PAGE 9: ...al Lake watershed was most highly developed (e.g. 83% urbadresidential), while Chub Lake (36% forest) and Golden Lake watersheds (41% wet- lands) were the least developed watersheds. Total watershed area ranged from 2 to 50 kmz, with an average of 17 km2 ( Table2 ). Because lakes had been selected in a stratified fashion to cover the range of lake sizes found in the Twin Cities metropolitan area, lakes in our subsample were larger on average ... ..."

"... In PAGE 13: ...5. Multiple regression and partial correlation analyses Multiple regression analysis demonstrated that trophic state variables were correlated with watershed-scale (PCl), land-use (PC2, PC5), and wetland (PC4) components ( Table8 ). Positive ... ..."