{
 "culture": "en-US",
 "name": "",
 "guid": "",
 "catalogPath": "",
 "snippet": "",
 "description": "This project predicts stream quality in the Kansas City region with 1) field-sampled stream data and 2) regional environmental, land use, and other data. Variables were developed and tested to find a valid regression model that could predict stream quality. A valid model was used to determine variables to predict stream quality throughout Greater Kansas City. Predictions of stream quality may be used to prioritize streams for management and remediation strategies.",
 "summary": "",
 "title": "Stream Health",
 "tags": [
  "stream",
  "Stream",
  "Stream Health"
 ],
 "type": "",
 "typeKeywords": [],
 "thumbnail": "",
 "url": "",
 "minScale": "NaN",
 "maxScale": "NaN",
 "spatialReference": "",
 "accessInformation": "In 2005, Patti Banks Associates (PBA, now known as Vireo) conducted a stream assessment for the City of Kansas City, MO (KCMO). The purpose was to assess and classify the relative condition of all streams within the city, and provide baseline natural resource conditions for sustainable storm water management and land use planning recommendations. Assessment criteria included erosion indicators, bed and bank composition, aquatic habitat features, tree canopy and understory coverage and composition, and indirect water quality indicators. These criteria were assigned individual weighted scores to create a composite score of stream quality at each location and a relative ranking of stream quality throughout the watershed. The assessment was designed to produce generalized results rather than site-specific data. \n289 sample locations were selected along streams in KCMO with about .75 mile between sample locations. Only natural streams were selected for sampling, not channelized or piped streams. Surveys were conducted from May through November, with 5 scoring components in each of four categories: stream stability (STAB_SC_SU in table 1 below), aquatic habitat quality (AQTC_SUM), terrestrial habitat quality (TERR_SUM), and water quality (WQ_SUM). Not all component scores could be determined at each site because of site conditions. Aquatic scoring components were omitted for all ephemeral and dry intermittent streams. Each of the 20 components have a potential score of 10, providing a maximum score of 50 for each of the four categories, with a possible total score (RAW)  of 200. The final stream quality score was calculated by dividing the total site score by the number of components scored (N). The final scores (TOTAL_SC) ranged from 0, indicating poor stream conditions, to 10, indicating optimal stream conditions.  \nStreams were classified by stream type (STR_TYPE) so that Type 2 stream scores fell one standard deviation above or below the mean score, Type 3 stream scores fell within two standard deviations above or below the mean, etc.\u2019 Stream types are described as: Type 1, Highest quality; Type 2, high quality; Type 3, restorable; Type 4, low quality; and Type 5, lowest quality. The classification was assigned relative to the sample population of surveyed streams, rather than applying an absolute score. PBA analysis indicated that terrestrial habitat scores showed the greatest correlation with overall stream quality. The other three general assessment factors (stream stability, aquatic habitat quality, and indirect water quality indicators) did not strongly correlate with overall stream quality. However, when PBA narrowed the analysis to sample locations where all 20 components were scored, the water quality component correlated strongly with overall stream condition. PBA found some high-quality streams in urbanized areas and low-quality streams in agricultural areas. PBA also noted that grouping watersheds into generalized land use classes did not produce useful results, which was consistent with various negative impacts to stream quality observed in urban, agricultural, and mixed use areas. Several variables were derived from land cover data, including impervious land cover, tree cover, and other natural community land cover. Other variables include steep slopes, location of bridges, dams, and facilities that discharge to streams, road density, stream order, erodible soils, and stream bed degradation. Variables were created with 2006 conditions where possible because the KCMO stream samples were collected in 2006.  The final stream quality score was used as the dependent variable in regression models. Additional models were also sought using each of the four scoring categories (stream stability, aquatic habitat quality, terrestrial habitat quality, and water quality) as the dependent variable. More successful models passed tests for model performance, bias, redundancy, and completeness. Terrestrial habitat quality as the dependent variable produced more successful models, but models were found using the final stream quality score that passed all tests except for spatial autocorrelation.\n\nThe Esri Exploratory Regression tool was used to test many combinations of variables to find candidate ordinary least squares regression models that best explain the dependent variable and pass the following search criteria: \n\nMinimum Adjusted R-Squared\t\t> 0.20 \t(Model performance, or the % of score explained by variables, default is .50)\nMaximum coefficient p-value \t\t< 0.05 \t(Coefficients must be significant at the 95% confidence level)\nMaximum VIF value\t\t\t< 7.50\t(Redundancy, or multicollinearity, among variables in model)\nMinimum Jarque-Bera p-value\t\t> 0.10\t(Distribution of model residuals. Low value = residuals are not normally distributed, and model is biased)\nMin. Spatial Autocorrelation p-value\t> 0.05\t(If model residuals are clustered, the model is likely missing key variables. Default is minimum p-value of .1)\n\nThe table below shows which variables were most consistently significant within the many models tested in Exploratory Regression and the variables\u2019 increase (% positive) or decrease (% negative) in value as the stream quality score increases. Any variables not included below were excluded because of low performance, either being significant in fewer than 1% of tested models, not being consistently negative or positive, or if variables had the lowest performance among several redundant or multicollinear variables. Exploratory Regression does not provide full regression models with variable coefficients and all available model performance tests. Groups of variables that perform well in Exploratory Regression are used to create an ordinary least squares regression model. The most successful and useful regression model found is below. The model passed all tests except for the spatial autocorrelation test. Also, the model explains only 22% of the variation in stream health, which is lower than most regression models would expect.\n1.\tUses dependent variable: Total stream score \n2.\tUses independent variables: PCTTREEBUF, ROADDENS06, WSDISH\n3.\tAssess model performance. Multiple R-Squared = .222; Adjusted R-Squared = .214. The model explains 21% to 22% of the variation in stream quality among sample points in KCMO. \n4.\tAssess each explanatory variable in the model.  VIF measures redundancy. Variables with VIF over 7.5 and should be removed, but none have a VIF over 1.14. Since Koenker is not significant (item 4 below), test the significance of each variable using the t-statistic and its probability. If Koenker were significant, use the robust probability. T is not zero for any variables, and all t-statistic probabilities are <.05, so all variables are helping the model. \n5.\tAssess stationarity. Do the variables behave the same everywhere in the study area? The null hypothesis is that variables behave the same everywhere in the study. The Koenker Statistic probability is .14, greater than .05, so there is not statistically significant heteroscedasticity and/or nonstationarity. \n6.\tAssess model significance. Koenker is not significant, so use Joint F-Statistic. Joint F-Statistic probability is .000, which is smaller than .05, so the model is statistically significant.\n7.\tAssess model bias. Are the residuals (the observed/known dependent variable values minus the predicted/estimated values) normally distributed? Jarque-Bera Statistic is 1.9 with a probability of .39 (>.05), so the residuals are normally distributed, and the model is not biased. There may not be missing variables, nonlinear relationships, outliers, or strong heterscedasticity. \n8.\tAssess residual spatial autocorrelation. Run Spatial Autocorrelation tool to determine whether residuals are spatially random. Add OLS14.shp output to SA tool, with Residual as input field. Results (MoransI_Result_OLS14.html): Residuals are spatially clustered. \nPREDICTING REGIONAL STREAM HEALTH \nOnce the best model is created to explain the variation in stream health among samples taken in KCMO, the variables used in the KCMO model are created for the entire region and used to predict stream health throughout the region.",
 "licenseInfo": "NA",
 "portalUrl": ""
}