Problems Associated with Comparing In Situ Water Quality Measurements to Pollution Model Output for Geographic Analyses Presentation to the Annual Meeting of the Association of American Geographers March , Chicago, IL Michael P. Finn
Authors Michael P. Finn a David D. Bosch b E. Lynn Usery a Austin D. Hartman a a U. S. Geological Survey, National Geospatial Technical Operation Center b U. S. D. A., Agricultural Research Service (ARS), Southeast Watershed Research Laboratory
Background Agricultural Non-Point Source (AGNPS) Pollution Model Usery et al., (2004) showed utility of using GIS databases to automatically generate input parameters for AGNPS Automated input and output visualization through the ADGen program (described in Finn et al., 2006) Current research is focused on quantifying model output as generated by ADGen –Today’s topic is one of three parallel investigations of the current research
Objective Use output of previous research to quantify significance of various resolutions of spatial parameters on the model output values –More specifically, determine the accuracy of output values relative to in situ measurements over a range of spatial resolutions and identify threshold of diminishing returns (shoulder in the curve)
Shoulder (or Knee) in Curve Example: Synthetic Data
Study Area Little River Watershed, Georgia ARS benchmark watershed for tillage management, pesticide management, and riparian restoration issues Agricultural areas with some woodland, wetlands, and small urban areas
AGNPS Output A non- point source (“.nps”) file –ASCII file (tabular, numeric)
ADGen Output ADGen Image of Phosphorous Output for the Little River, Georgia. Single Band: band 4, Total soluble phosphorous
Shoulder (or Knee) in Curve Example: Synthetic Data X Resolution Y Accuracy
In Situ Water Quality Measurements Hydrologic “field” data Sources: literature, spreadsheets, websites, archived files, tables in docs, etc. Monthly Water Quality Averages - in spreadsheet
In Situ Water Quality Measurements Hydrologic “field” data Sources: literature, spreadsheets, websites, archived files, tables in docs, etc. Daily Output - by sub-basins - text file
In Situ Water Quality Measurements Hydrologic “field” data Sources: literature, spreadsheets, websites, archived files, tables in docs, etc. Streamflow archives - from webpage
In Situ Water Quality Measurements Hydrologic “field” data Sources: literature, spreadsheets, websites, archived files, tables in docs, etc. Verification runs - random point capture and comparison
Problems Comparing Field Data to Model Output Convoluted matching of field and model values Wide, disparate sets of data sources Only two parameters match directly (with unit conversions) and an additional 6 indirectly Direct match
Measures of Accuracy 51 model output values of sediment, nitrogen, phosphorus, and other nutrients Curves representing the accuracy via mathematical means –N th order polynomials, cubic splines or logistic regression Identify shoulder in curve -> quantify spatial resolution threshold
Accuracy Curve 6 th Order Polynomial Example 1st: y' = -1.2E-10x x x x x y' = x = , , , nd: y'' = -6E-10x x x x y'' = x = , , rd: y''' = -2.4E-9x x x y''' = x = ,
Accuracy Curve Cubic Splines Example
Accuracy Curve Logistic Regression Example
Overcoming these problems Quantification of spatial threshold –Provide insight to the role of spatial resolution on the variance of output values –Ultimately, insight into geographic analyses of water-quality investigations
Conclusions Resolution affects model results Some promising techniques to determine the accuracy of output values relative to in situ measurements for the purpose of identifying threshold of diminishing returns Issues remain in building “Accuracy Graphs” for 51 model output parameters
Problems Associated with Comparing In Situ Water Quality Measurements to Pollution Model Output for Geographic Analyses Presentation to the Annual Meeting of the Association of American Geographers March , Chicago, IL Michael P. Finn