Mitigating Phosphorus in the Connecticut River

Riparian buffers are vegetated land areas near streams that are planted to mitigate the impact of adjacent land uses on the water. This has become a common conservation practice which has been proven to provide a host of environmental benefits from water temperature moderation to floodplain stability. Considering the implications of changing an existing landscape and the cost of planting these riparian buffers, being able to target priority areas to develop and monitor is key.

A detailed report by the VT Agency of Natural Resources on Riparian Buffer Management can be found here.

 
 

With recent reports of phosphorus pollution in Lake Champlain as a result of agricultural runoff, I wanted to further examine phosphorus pollution in Vermont and examine possible relationships with adjacent land cover and land use and stream water quality. Based on previous research (McDonald et al. 2016; Bittman et al. 2017; Elledge et al. 2017), I hypothesized that phosphorus in the water would increase with the increased presence of impervious cover (asphalt roads, buildings, parking lots etc.) and open fields (possibly agricultural land), but decrease with forest cover. 

Working in a group of four, we focused on areas in the West-Williams-Saxtons River Basin that drained into the Connecticut River with the idea that our research could be applied to the wider area of study concerning the longest river in New England.

 
firstmap.jpg
 

I focused on a sub-basin in the West-Williams-Saxtons River Basin to test my hypothesis. Using geographic information systems (GIS) and multivariate statistics, I analyzed land cover imagery derived from VCGI's satellite imagery and water quality data from the EPA. Through this I found suggestions of a positive correlation between the presence of phosphorus and adjacent impervious cover and open fields, as well as a negative correlation with forest cover. However with my limited sample size, the analysis was statistically insignificant. Comparing my results with those of my peers who were focused on the same basin, none of them found statistically significant relationships within the data either and more work would need to be done to overcome the limitations of the data and methodologies used so as to effectively identify areas to develop riparian buffer zones.

Methodology and Limitations

In order to perform an ordinary least squares (OLS) regression model to discern significant relationships between phosphorus in streams and adjacent land cover, I calculated the percentage of each classification of land cover within 100 m of the streams with water quality data. This was done by filtering satellite imagery of the sub-basin and running a supervised classification on the data. This meant that I chose sample areas for each type of land cover (Forest, Impervious and Open Field) that I identified by eye, and based on those samples, the GIS software was able to classify the rest of the data in the sub-basin. I then buffered the streams by 100 m and used this to tabulate the area of the different land cover which I then joined to the data table of the streams with water quality data as seen in the maps above.

The graphs below show prove my hypothesis, where there was an increase in phosphorus with the increase of impervious cover and open fields, but a decrease in phosphorus with the increase of forest cover--proving the effectiveness of riparian buffers.

Graphs showing the OLS regression analysis of Phosphorus (PHOS) against the percentage of impervious cover (PERCENTAGE_IMP), forest (PERCENTAGE_FOR) and open field (PERCENTAGE_OPE) within 100m of the streams in my study area.

Graphs showing the OLS regression analysis of Phosphorus (PHOS) against the percentage of impervious cover (PERCENTAGE_IMP), forest (PERCENTAGE_FOR) and open field (PERCENTAGE_OPE) within 100m of the streams in my study area.

However while my graphs were promising, further inspection of my statistical analysis showed that my data was not statistically significant. This was because the probability value > 0.01 and the r-squared value which indicates how well the linear regression fits the data was ~0.59 where the closer it is to 1 the more accurate the regression is.

tble1.PNG
table2.PNG

There were many other limitations to the way I conducted my analysis. The percentage of land cover types was not accurate as I made buffer zones of 100 m from a vector line representing the rivers, meaning that it did not account for the width of the river in calculating the adjacent land cover. There was also limited water quality data available meaning that only a small portion of the streams could be analyzed. The classification of land cover is also prone to human error as I had to make assumptions of the land cover with the spatial and temporal resolution of the satellite imagery available. This meant that the GIS program read some polluted parts of the river as development  instead of water. 

Questions of scale also play a role in this data analysis as I would need not only more data in my area to make more informed hypotheses about the relationship between land cover and water quality, I would also need to understand the entire network of streams that feed in from other sub-basins in mine (just as mine flows into the Connecticut River). Focusing on a specific stressed stream and following it from the source downstream until the desired length of analysis depending on the purpose of the study.

 

Nicole ChengComment