Summary of the Water Quality Tool
Created by Amy Haase and Toby Carlson
Introduction:
The quality of water in streams and estuaries is becoming a growing concern with the increase in urbanization and other changes in land use. These changes are causing an increase in nutrient and sediment loading from non-point source pollution in watersheds. According to the Pennsylvania Department of Environmental Protection (PaDEP), as of 2002 about 10% of streams throughout the Commonwealth are impaired, 50% which are impaired from sediment and nutrient loading (Sheeder and Evans, 2004). Furthermore, according to the United States Geological Survey (USGS) “nutrient enrichment and the associated hypoxia (reduced oxygen content in the water) are two of the more serious problems facing the Chesapeake Bay” (read the report here). The latest report from the American Rivers organization states the Susquehanna River “tops this year’s America’s Most Endangered Rivers" (dowload the report here) .
Until the creation of our web tool, diagnosing water quality in the multitude of individual streams in the absence of direct measuremens was possible only through the use of sophisticated models. Currently the Environmental Protection Agency (EPA) uses a model called the Generalized Watershed Loading Function (GWLF) which has the capability to estimate sediment and nutrient loading within Pennsylvania watersheds. Though this model is able to account for 90% of the variability in the dependent data set, it requires such information as basin size, precipitation, soil water content, unsaturated storage and a variety of other variables not available in all watersheds.
The United States Geological Survey uses another model called Spatially Referenced Regressions on Watersheds, or SPARROW. Like the GWLF, this model is able to account for 90% of the variability of nutrient loads, but operates with a large amount of required information not available in most watersheds. The fundamental purpose of this research is to create a simple web-based water quality assessment tool that allows users, mainly resource managers, to estimate the impact of land-use on water quality with a reasonable degree of simplicity and accuracy.
We have created a water-quality assessment model which requires far less input but is virtually as accurate as SPARROW in assessing stream health; this program, referred to as WET. WET allows users to interactively investigate the health of rivers, streams and tributaries in the Commonwealth of Pennsylvania or the Chesapeake Bay Watershed with only limited land use information as input. The user simply outlines a desired area on the screen using a cursor to draw a closed boundary around whatever watershed or feature is of interest and clips it out. Various GIS overlays and maps (showing rivers, county boundaries, etc.) are included to help the user define the area of interest. In addition to the basic stream assessment, this program allows users to create scenarios modeling anticipated changes in land-cover to estimate how these changes could impact the water-quality for their watershed. This is illustrated below. WET is solely based on the web (www.sharp.psu.edu) and is part of a larger program which also allows the user to assess surface storm water runoff.. There is no software that needs to be purchased or downloaded for using this model and considerable documentation exists on the web site for the users’ benefit.
Data input
WET functions by relating land use to nutrient and sediment yields. Land use parameters used to create the data base were impervious surface area (ISA) and percent forest cover. These land use data were created by Eric Warner at the Penn State Institutes of the Environment and by Dr. Toby Carlson (private communication, 2004) using Landsat data with 30 m resolution for the year 2000. These maps are also available on The Pennsylvania State University’s digital data clearinghouse, www.pasda.psu.edu. The original algorithms for creating the Nutrient tool were determined from measured nutrient and stream flow data extracted for 42 watersheds in Pennsylvania. Fractional percentages of woodland cover and ISA within the isolated watershed were then calculated and recorded for each of the watersheds. The percentages of woodland and ISA were found to be the main predictors of water quality for the 42 watersheds. Regression equations were developed relating water quality (nutrient yields, N and P, and sediment yields (sedi)) for the 42 watersheds to the land use predictors: percent impervious surface area (%ISA), which is related to urbanization, and fractional woodland cover (% wood). The regression equations were then assumed to pertain to any chosen watershed within Pennsylvania or the Chesapeake Bay basin.
The original water-quality measurements, specifically averaged annual yields of nitrogen, phosphorus and sediment used in this project, were compiled by Dr. Barry M. Evans (Evans et al., 2002) at the Penn State Institutes of the Environment as part of various watershed related projects conducted for the Pennsylvania Department of Environmental Protection (PaDEP) throughout the state. Algorithms involving different predictors were tested to make sure that the equations were the most statistically robust, e.g., had low P scores, a good normalization curve and a high R-squared (about 0.65)
In order to assess the validity of our model, we compared predicted water quality parameters and the measured ones using independent data. For the Chesapeake Bay Basin, annually-averaged nutrient and sediment data published in the September 1998 USGS Report 98-4192 [WRIR 1998] were used for evaluation of the model outside Pennsylvania. Data from this report is referred to as the secondary set of data for this project. Dr. Stephen Prince and Kyle Pittman from the University of Maryland provided ISA and land-cover data for the Chesapeake Bay Basin. The data provided by Dr. Prince and Kyle Pittman were also used in the evaluation process of this resulting web-based model.
We also compared out output with SPARROW Model output; the latter was provided by John Brakebill and Richard B.Alexander from USGS [Preston 1998]. This output of this model was used like data for the purpose of evaluating the Nutrient Tool for the Chesapeake Bay Basin and acted as another source of data for this research project. The SPARROW model output contained nitrogen and phosphorus yields but not sediment.
Agreement between SHARP and the output of SPARROW showed that the simple SHARP tool yielded nearly as accurate an estimate of the nutrient loads (N and P) as does the SPARROW model for the same watersheds tested.
The Nutrient Model
The equations used in the WET tool are as follows:
Ln (Sedi) = -7.184 + 0.1276 * Ln (%ISA) – 0.01459 * %Wood
Ln (N) = 3.6483 – 0.03115 * %Wood
Ln (P) = -0.375 + 0.20798 *Ln (%ISA) – 0.01556 * % Wood
Here, the units for nitrogen (N), phosphorus (P) and sediment (Sedi) are in kg/ha/yr; Ln refers to the natural log.
In addition to predicting what these yields would be under various land-use/cover conditions based on the equations developed, the nutrient tool was also created to provide qualitative information on the degree of impaction experienced in the watershed based on threshold values for stream impaction that were found at certain elevated levels of nutrient and sediment yields. These threshold values, which expressed a sharp division between impaired and non impaired watersheds, were derived by Sheeder and Evans (2004) from data for 29 of the aforementioned 42 watersheds in Pennsylvania. Impairment of Pennsylvania watersheds was assessed as part of the Clean Water Act. The thresholds were determined from independent biological assessment using an averaging method based on observed annual nutrient and sediment yields and concentrations found within the 29 watersheds.
These thresholds were used to derive a Stream Health Index (SHI), which ranges from -6 for highly impaired watersheds to +6 for very clean watersheds when phosphorus, nitrogen and sediment are all considered. Each parameter, N, P or sediment, contributes to the SHI a value between +2 to -2; A +1 or -1 value represents, respectively, yields within a range between half of the threshold value and 10% below the threshold value or between 10% above threshold and double the threshold value for each constituent. A value of +2 or -2 represent yields less than half or more than double the threshold values for each constituent, respectively. A value of zero corresponds to a watershed within 10% of the threshold value, the small percentage here representing the narrowness of the threshold.
Since the thresholds tended to be very sharp for the original data, a value of plus or minus 1, for example, represents a definitely clean or dirty watershed, respectively Values for each of the three constituents are summed to yield a SHI between -6 and +6, except for comparisons with SPARROW output, in which case the range is between +4 and -4. In general, when one constituent showed a plus or minus value the other two also shared the same sign. Occasionally, one finds a value of one constituent with a plus value and another with a minus value, creating some ambiguity on whether the watershed is impaired or not. However, summing the values for the three constituents generally produces a very reliable indicator of stream health.
Output from SHARP
Below is a sample output from WET. Percent of various land use quantities is tabulated on the left. To the right is shown the calculated nutrient and sediment loads, the variance of the measurements used to create the algorithm and the threshold value for the particular constituent. Size of the watershed, expressed in various dimensions, is shown at the top.
The table at the bottom of the figure shows the percentage of the different land use categories used in the scenarios. Row 1 pertains to the actual watershed land use distribution, the estimated nutrient and sediment yields and the stream health index. Rows 2-4 represent different scenarios created by the user in changing land use values and the resulting nutrient and sediment yields and stream health index. The land use data shown in the vertical column to the left refers only to the last scenario run by the user, scenario, shown in row 4.
This output shows that the stream is quite clean with a good stream health index (+6). Subsequent user-defined increases in urbanization to 20% and concomitant decreases in wooded area to 62% (with a corresponding increase in agricultural area to 15%) are meant to simulate possible future land use distributions. The result is a deterioration of the stream health to zero in the last scenario, which implies marginal impairment of the watershed. (The increase in agricultural land is unimportant; what affects the stream health in this last scenario is a further decrease in woodland.)
The user is also able to click on a button to produce a cut out of the watershed showing a color-coded land use distribution. This product is not shown here.
List of References
Carlson, T.N., 2004. Private Communication.
Evans, B. M., D. W. Lehning, K. J. Corradini, G. W. Petersen, E. Nizeyimana, J. M. Hamlett, P. D. Robillard, and R. Day, 2002: A comprehensive GIS-based modeling approach for predicting nutrient loads in watersheds.Journal of Spatial Hydrology, 2, 1-18.
Preston, S. D., R. A. Smith, G. E. Schwartz, R. R. Alexander, and J. W. Brakebill, 1998: Spatially referenced regression modeling of nutrient loading in the ChesapeakeBay Watershed. Proceedings of the First Federal Interagency Hydrologic
Modeling Conference, Las Vegas, NV, 8 pp.
Sheeder, S. A. and B. M. Evans, 2004: Development of nutrient and sediment threshold
criteria for Pennsylvania TMDL assessment. Journal of the American Water
Resources Association (in press).
Smith,
USGS Report, September, 1998: Yields and trends of the nutrients and total suspended
solids in nontidal areas of the Chesapeake Bay Basin, 1985-1995. Water-
Resources Investigation Report 98-4192. U.S. Department of the Interior, U.S.
Geological Survey, 9 pp.
Figure 1. Sample output from WET