Information on the Model Used in the Forecasts
The forecasts for the IBI are based on a model created as a part of a grant from the U.S. EPA R824769 from the U.S. EPA/NSF Water and Watersheds Program. Primary work on the model was done by Dr. Sarada Majumder, currently at SBI in Cincinnati, Ohio and Dr. Steven I. Gordon at the City and Regional Planning Department, The Ohio State University in Columbus, Ohio.
 |
The model is based on data from 72 watersheds throughout Ohio that are part of the Eastern Cornbelt Plains Ecoregion.
To explore the ECBP and Ohio in more detail, visit the Interactive Mapping Tool |
Data collected from 1991 through 1994 were used from this region to create an empirical estimate of changes in IBI relating to indicators of the stresses on the ecosystem.
The major variables found to impact IBI levels were:
- The percentage of Dense Urban Land Use - This variable is a proxy or estimator for the amount of runoff with non-point source pollutants being added to the stream. As the percentage of urban land use went up, IBI levels decreased.
- The Substrate and Riparian measures of habitat quality - A decrease in these parameters indicate a decline in the natural aquatic habitat quality. As the rankings for the indicators declined, the IBI levels also declined.
- An estimate of the Point Source Pollution Loads impacting the stream - Here a rather complex index variable was created based on the pollution loads associated with the ammonia, nitrogen and phosphorous concentrations in places upstream of the IBI points, discounted by a distance decay function. This means first that all places were rated according to the levels of these variables moved to a standardized scale. Then, the levels were reduced the further upstream from the IBI reading point that the sources were found to be. The source information was from the Ohio EPA LEAPS database. The distance decay function employed was an inverse linear decay representing the breakdown of these pollutants in the natural environment with time in the stream.
- Indicator of the Soil Conditions in the watershed - This variable of the model was derived from a factor analysis of the general soil groups as derived from the U.S. Natural Resource Conservation Service STATSGO database for Ohio. Principal component analysis was used to summarize the soil properties. The major component weighted heavily on hydrologic soil properties indicating water logging and low pH.
- Stream Order Component - This final variable of the model was a categorical variable to account for the presence of headwater streams. If a sample point was located in a headwater stream, a value of 1 was assigned to that point. All other stream types were coded as 0.
Data were analyzed in a linear regression model where the dependent variable was the IBI level and the independent or explanatory variables were those described above. The final equation derived for this set of scenarios was:
IBI=27.013+0.925(Substrate)+1.161(Pool)-0.636(Urban)-2.946(Soils)-1.571(Point Loads)+5.576(Stream)
The model explained 72% of the variance in the original data. This means that there is about 28% of the system variation that remains unexplained by this model.
Sources of Error
There are several sources of error in this model. As indicated previously, the model explains 72% of the variance. Although this is very good for an empirical environmental model, it does mean that the projected results reflect some inaccuracy. A review of the errors in the model reveal several trends.
Errors are introduced in a number of ways
First, the datasets represent samples at a finite number of places in the streams for a limited number of times. Given the nature of the sampling methods for Ohio EPA, these could not be at the same time. Thus, errors are introduced because data were taken at different times and under somewhat different conditions.
Second, not all tests were performed in all areas that could potentially affect a given watershed. Therefore, water pollutants may be released into certain streams that are going unmeasured, but may in part explain some of the stream deterioration.
Third, there may be a number of additional conditions that varied from place to place over the period the data were collected. For example, the amount of land undergoing development, the nature of storms in a particular year, changes in agricultural practices, and conservation practices put into place on farms or construction sites can all vary significantly at different locations at one point in time.
Also, across all watersheds, the model tends to over predict the IBI levels as conditions worsen. This is to say that although on average gives reasonable estimates of future levels, it tends to give higher levels of IBI than are really found in circumstances where the overall quality of the stream is good to poor. Thus, our projections as shown here may be underestimating the impacts of future urbanization on the streams.
Nevertheless, the model, on average, gives a reasonable estimate of the
nature of the trends in biodiversity even if an exact forecast of the IBI levels is not achieved. The model can then be used to estimate the trends in the overall conditions but not the exact levels that can be expected.
Additionally, there is one basin in which the model was found to give extreme errors - the Mad River watershed. The cause of this is thought to be the large percentage of groundwater feeding this stream which offsets some of the surface pollution impacts found in other watersheds.
