Cyanobacteria in Lake Iroquois

Methods

I obtained remotely sensed data from VCGI for the area of Steven's River watershed, and used that imagery (6 tiles, stitched together with the mosaic tool in Arc GIS) and a supervised classification method to develop a land cover layer. I used the Maximum Likelihood Classification tool to assign pixels to one of four categories of land use: water, forested areas, agricultural areas, and impervious cover or urban areas. Then with my water quality and streams layer, I selected the Steven's River watershed streams with non-zero phosphorous or nitrogen values. I created a 100 m buffer around those streams to focus the analysis on riparian areas, and intersected my land cover layer with the water quality/streams layer. With the intersected layer, I performed an Areal Weighted Re-aggregation to determine the land cover composition of the 100 m buffers around the streams. With that information, I ran an Ordinary Least Squares regression analysis with levels of phosphorous as a dependent variable, in order to find the relationship between riparian land cover type and nutrient pollution.

Land Cover and Water Quality in the Steven's River Watershed, Vermont

Riparian areas, or buffers, refer to the area of land immediately adjacent to a body of water. While often difficult to define, riparian areas are generally characterized by higher levels of soil moisture, frequency of flooding, and unique ecosystems at the water's edge. The type of land cover that makes up riparian areas can have an impact on the water quality of rivers and streams, and studies have shown that forested riparian buffers can filter nitrogen and phosphorous from agricultural runoff and reduce nutrient pollution (Klapproth and Johnson, 2009).

In this analysis, I used remotely sensed imagery for the area of the Steven's River Watershed, water quality data from the EPA, a streams layer and an impaired streams layer from the Vermont Center for Geospatial Information (VCGI) to quantify the relationship between four categories of land cover and water quality of nearby streams. A 2001 study by Sliva and Williams found urban land use in riparian buffer zones to have the greatest impact on overall water quality, but found no correlation between nitrogen and phosphorous levels and riparian buffer land cover characteristics (Sliva and Williams, 2001). I expected to find a positive correlation between agricultural riparian land cover and the level of nitrogen and phosphorous, since northern Vermont has more forested and cultivated land than it does urban land cover.

Results and Implications

The Ordinary Least Squares analysis showed a significant strong negative correlation between field land cover type and nitrate levels (-94.6%; p = 0.032757), and the model explains 12.01% of the difference in nitrate. There was a statistically significant but small positive correlation between forest land cover and phosphate levels (5.1%; p = 0.000007), with the model explaining 76.5% of the variation in phosphate. I expected urban land cover to have had a more significant impact on water quality, but the lack of significant correlation makes sense due to the relatively small area of urban land cover in the Steven's River Watershed, and northern Vermont in general.

Limitations

Supervised classification is useful in cases where ArcGIS has difficulty correctly classifying land cover types, and the categories can be determined based on samples of pixels with known identity. The Maximum Likelihood Classification, the most commonly used supervised classification, is based on the assumption that each category of spectral data can be described by a normal distribution (ArcGIS). There are two types of errors that are common limitations of this method. Errors of comission, where pixels are incorrectly included in a category, and errors of omission, where pixels are incorrectly left out of a category. Examples of comission errors can be seen in the Steven's River Watershed.

Parts of northern Lake Harvey were incorrectly classified as urban land cover: