Spatial Mapping of Ground-Water Quality Using Statistical Methods-Assessing the Risk of Nonpoint-Source Pollution
WRD
PROJECT #: MD127
PROJECT CHIEF: Greene, Earl A.
BEGIN DATE: 01-March-2000
END DATE: 30-September-2002
Customers currently supporting the project:
U.S. Environmental Protection Agency
Problem
The USGS and the US EPA's Regional Vulnerability Assessment Program (ReVA) are teaming together to provide information for a regional-scale, integrated ecological risk assessment study. Logistic regression analysis will be used to determine the relationship between surficial ground-water quality and a number of factors, such as land-use, well depth, geology, and population density. Existing USGS and state agency water-quality data covering the US EPA Region 3 area (Pennsylvania, West Virginia, Virginia, Maryland, D.C., New Jersey, and Delaware) will be examined. Results of the analysis will be used to determine the current condition of ground-water quality at a regional scale and to generate probability maps of aquifer susceptibility and ground-water vulnerability. By applying the groundwater-quality model to cooperating ReVA scientists' alternative future land-use scenarios, ground water-at-risk from land-use changes can be forecasted. The methods employed in this study will be applicable to future regional vulnerability assessments.
Objectives
- Determine the current condition of ground-water quality in the surficial aquifer influenced by non-point sources at the watershed and regional scales and generate surficial aquifer susceptibility and ground-water vulnerability maps at a regional scale.
- Identify areas of greater statistical uncertainty pertaining to the health of ground water for assistance in improving/designing a monitoring program that could be used to forecast ground-water at risk from land-use changes.
Approach
FY00 Activities
Database Development-Water-quality data needed for the study would be obtained form existing USGS data stored in ground water well site files (GWSI) and water-quality data (QWDATA) databases. These databases contain information obtained from NAWQA studies, Statewide ground-water quality networks, and individual projects. Geology and land-use data (explanatory variables) will be obtained for the study area. Other data such as well construction and location and field water-quality parameters (pH, specific conductance, etc) will also be obtained to correlate the presence or absence of selected ground-water quality parameters (response variables) to explanatory variables.
Logistic-Regression Method Development-This method is a statistical analysis used to predict the response (presence/absence) of selected water-quality parameters from independent explanatory variables (land-use, population, field parameters, etc). For this investigation, logistic-regression will be used to develop equations that use explanatory variables to predict the presence of nitrate, pesticides, and/or other water-quality parameters in concentrations above a specified threshold value. The resulting equations will be transformed to predict the probability, P (X), of a water-quality parameter being above a particular threshold value for application in ecological risk assessment studies.
FY01 Activities
Logistic-Regression Analysis continued -It is anticipated additional work will be needed to develop equations that use explanatory variables to predict the presence of water-quality parameters that are of interest to other ReVA researchers.
Determine and Create Uncertainty Maps-From the logistic-regression analysis it will be possible to determine rank correlation between the predicted probabilities of water-quality and observed (sampled) values. Several statistical procedures exist to determine the confidence intervals about correlation coefficients. These intervals will be important in determining how confident we are in the predicted probability values. The larger the interval, the greater the uncertainty. For example it may be possible to obtain a high rank correlation coefficient, but yet have such a large confidence interval that the probability map generated would have little meaning.
Probability Maps - Application of logistic-regression models will be used to develop probability maps of aquifer susceptibility and ground water vulnerability to water-quality parameters. Individual maps will be built for water-quality parameters of interest and will show the spatial probability of the presence of the water-quality parameter being above a designated threshold level.
FY02 Activities
Application of the logistic-regression method to determine regional aquifer susceptibility and mapping groundwater vulnerability at a watershed and local scale.
Forecasting - ground water at risk from land-use changes will be forecasted by applying the model to alternative future land-use scenarios generated by cooperating ReVA scientists.
Compare study results and results from other ReVA scientists' stream-water quality analysis to investigate possible integration of data for assessment of stream and groundwater quality.
