Land Use and Water Quality

Evaluating Land Cover in the Black River Headwaters Watershed of Vermont

Impact of Riparian Buffers on Water Quality

Overview of Project

This project explores the relationship between water quality in the Southern portion of the Black River Headwaters Watershed near Ludlow, Vermont (just north of the Okemo Ski Mountain) and the land covers in the surrounding area. I will investigate whether certain land covers (developed, open field/agriculture, forest) contribute to worse water quality in streams and rivers in this watershed, and whether any management implications can be drawn from the results. Specifically, I will be focusing on land covers in 100-meter buffer-zones around streams and rivers in this watershed, otherwise known as riparian areas, explained in detail below.

Importance of Riparian Buffers

Riparian areas, the transition zones between bodies of water and the surrounding uplands, are very important areas of concern for environmental organizations. These zones have an enormous impact on water quality, species habitats and diversity, and other ecological factors. A majority of human-caused damage to ecosystems near bodies of water and water quality is due to encroachment into these zones (Vermont Department of Conservation). For this reason, public and private environmental groups advocate for so-called riparian buffers, vegetated areas along bodies of water that protect from excess run-off, reduce erosion, and provide healthy habitats for numerous species that live near or in the water (Vermont Agency of Natural Resources).

Managing pollution from excess run-off is the most important of these functions. Pollutants in water primarily come from two types of sources: 1) point sources, like sewage discharge and storm-water, and 2) non-point sources, like agricultural or urban run-off (Sliva and Williams, 2001). For management of non-point sources, which are the more commonly hazardous of the two, riparian forest buffers have been proven to be very effective. They have been shown to "significantly reduce nitrate concentrations" and are "very effective in reducing phosphorus concentrations" (Bongard and Wyatt, 2010). For this project, I will be analyzing these metrics to assess water quality.

Diagram explaining riparian buffers along bodies of water.

http://www.irwp.org/assets/_resampled/resizedimage600298-ChesagraphicColor.png

Hypothesis and Supporting Evidence

In the past two decades, there has been significant research on the relationship between land use/land cover and water quality using GIS (geographic information systems) techniques. This research has yielded important information regarding different approaches to assessing the problem, which land uses/land covers tend to have the greatest effect on water quality, and ways that conservationists can manage these ecosystems more effectively. Two papers have influenced my hypothesis and workflow for this analysis: 1) Snyder et al. explore stream health using land cover based on satellite imagery, and 2) Sliva and Williams investigate the impact of land use on water quality and whether studying the entire watershed or just the buffer zone is more effective.

Both papers reveal interesting results that informed the creation of my hypothesis. Snyder et al. find that the area of impervious surfaces is the highest indicator of stream health, while tree cover within and outside the buffer zone also has a significant impact. In their analysis, agriculture, or percent of crop cover, did not have a clear correlation with stream health. These findings were especially informative because Snyder et al. use land cover metrics derived from aerial imagery using a similar workflow to the one employed in this study, which is explained in detail below (Snyder et al., 2005). Similarly, Sliva and Williams find that urban land use has the "greatest influence" on water quality and forested land had a negative correlation with poor water quality, although in their analysis agricultural land did not have a strong influence either way (Sliva and Williams, 2001). They also find that the entire watershed or catchment has a greater influence on water quality than a 100-meter buffer. My analysis will explore only the 100-meter buffer, but with more time and processing power, it may be a future project to compare it with the entire watershed. Sliva and Williams' analysis also included comparisons across seasons, through which they found that fall and spring samples contained more pollutants than those taken during the summer (Sliva and Williams, 2001). I incorporated some of the results from Sliva and Williams, as well as Snyder et al. to predict my results.

My hypothesis was that developed land would have the most positive correlation with poor water quality, agricultural/open land would have a slightly positive correlation, while forested areas would have a negative correlation. Since my study area of Southeastern Vermont is primarily forested and open land with minimal developed space, I thought developed land might have less of an impact than I would anticipate and agriculture would have more. Within a 100-meter buffer though, I thought you may not be able to assess the impact of agricultural land, since run-off might come from much farther away.

Data and Methods

1. Pre-Processing

The downloaded aerial imagery from VCGI is in the form of small square files that must be stitched together to cover the watershed boundary. Once I had a single file of aerial imagery for my watershed area, I cropped this file to the boundaries of my watershed to decrease processing time and focus on the necessary data for my analysis.

2. Creating Land Cover layer

To semi-manually create a land cover layer with aerial imagery, I utilized the Image Classification toolbar in ArcGIS, using the Supervised Classification tool. This tool allows the user to draw polygons on the image that correspond to a certain land use, group the similar ones together, all to "train" the computer to recognize certain light reflectances and patterns as developed land, forest, etc. The benefits of Supervised Classification include the control the user has over creating the classes and assessing the accuracy, and it is better for analyzing difficult imagery than the computer (Dreiss). For this project, I used four classes: developed land, water, agriculture/open field, and forest. I decided on these classes after visually analyzing the imagery and seeing what land uses were common in my study area. Once the computer is trained to recognize land covers, I used the Maximum Likelihood Classification Tool, which analyzes each pixel in the image and assigns it one of the classes I defined. At this point, I had a raster file of land cover in my study area, which I then converted into vector format, meaning that each grouping of pixels was its own distinct polygon. My data after this step is shown below:

3. Creating the 100-meter buffer zones around streams

Given a layer of streams for the state of Vermont, my first step was to clip the streams to the borders of my study area to speed up processing time. Next, I used the Buffer tool in ArcGIS to create a data layer of streams with a 100-meter "buffer" on both sides, representing the riparian areas I discussed above. Next, I calculated the area of each of these stream buffers, an important step that will be necessary for the Areal-Weighted Reaggregation below.

4. Areal-Weighted Re-aggregation

Areal-Weighted Re-aggregation is a fancy term for re-aggregating data in one areal unit by the boundaries of another areal unit when the two units are not nested. For this analysis, I was re-aggregating the land cover data by the boundaries of the riparian areas (100-meter buffers). The first step in this workflow is to intersect the land cover layer and my buffered streams, meaning that I am left with only the data that overlaps in each layer, basically clipped land cover data to the buffered streams boundaries. Next, I created a new field in the attribute table of the data for each land use in my analysis and calculated the area of each in the new intersected layer. Then, using the Dissolve tool in ArcGIS, I re-aggregated the area of each land use type by the buffered streams. The final step involved creating new fields in this newly aggregated layer for the proportion of each land use area to the area of each buffered stream. At this point, I had the percent of land in each buffered stream area that was developed land, open field/agriculture, water, and forest.

5. Regression Analysis

To complete my analysis and test the influence of each of these land uses on water quality, I ran a regression analysis in ArcGIS using the OLS Regression tool. As proxies for water quality, I used the amount of phosphorus and nitrogen in the water. This analysis analyzed the relationship between the proportion of each land use in the riparian buffer zone and the water quality.

The analysis for this project was done entirely in ArcGIS, software that allows you to work with spatial data and perform complex analyses. The data for this project included:

Water quality data for every stream in Vermont, including total phosphorous (ug/l), total nitrogen (mg/l), and total suspended solids (mg/l), from EPA STORET
2014 Color and Infrared Statewide 1m aerial imagery from the National Agricultural Imagery Program (NAIP), accessed from the Vermont Center for Geographic Information (VCGI)
Watershed boundary, created by myself in a previous project using ArcGIS HydroTools and Digital Elevation Model data. This workflow involved working with the Digital Elevation Model to establish flow direction of streams, where streams would accumulate, the order of streams in the system, and eventually delineating my own watersheds based on this analysis. The watershed my analysis created ended up being the Southern portion of the Black River Headwaters Watershed.

The workflow for this analysis was extensive, but allowed us to get as accurate an assessment as possible given the data and software:

Results

Overall, my results do not show a clear influence by any of the land uses on water quality, positive or negative. The graphs shown to the right are one of the outputs of the OLS Regression tool. The scatter plots show which variables were the best predictors of worse water quality. There is a slight positive correlation with all three land uses and phosphorous, and only forest and open field with nitrogen. In fact, developed land has a slightly negative correlation with higher levels of nitrogen. However, these results are not conclusive and given the sample size of the data (n=12), it is not smart to make conclusions based on these results.

Below are the Multiple R-squared values for this analysis, which show how much variation in the levels of nitrogen and phosphorous can be explained by the model. These are measured from -1 to 1, with positive numbers meaning more significant.

Nitrogen

Forest: 0.163
Developed: 0.039
Open/Agriculture: 0.224

Phosphorous

Forest: 0.048
Developed: 0.029
Open/Agriculture: 0.121

These values show that while there was some positive correlation, the model was not strong enough to show a definitive result or prove my hypothesis. However, the results can inform our understanding of which land uses may have a greater impact on water quality in this specific area. Based on this analysis, open field/agriculture adjacent to water seems to have an impact on increased nitrogen levels and phosphorous levels, and forested lands tend to increase nitrogen levels. Yet, as stated, these results cannot be seen as definitive, both because of the low values, and the many limitations of the methods/data outlined below.

Regression analysis results, using phosphorous in water as the dependent variable.

Regression analysis results, using nitrogen in water as the dependent variable.

Limitations and Sources of Error

Due to the limitations of this approach, there are a number of significant sources of error that impacted the results:

The human interpretation aspect of the supervised classification tool most likely caused some errors in my analysis. When I was classifying the imagery, I could not tell the difference between agricultural land and open fields, so I ended up classifying them as the same thing. This generalized the results and made the analysis less applicable to future work. In addition, the computer interpretation step clearly had some significant flaws. The classified image shows that the computer thought there was developed land where clearly in the image there was water, and repeatedly classified forest as open field. These sources of error are a combination of the limitations of the tools, and human error. These errors result in what is called omission and commission. Omission refers to pixels being left out of a certain category, or "under-classification." Commission, on the other hand, refers to pixels being improperly added to the incorrect class, leading to "over-classification." Both of these phenomenon occurred during my analysis.
Another significant source of error was the amount of water quality data that I had available for the streams in my study area. After removing the streams that had no data, I was left with only 12 samples with which to perform my analysis. This number of samples is much too low and it is nearly impossible to generalize anything based on that.
Finally, the water quality data I analyzed was not provided with a date for the testing sample. As a result, it is not known whether these values were affected by the time of year, weather conditions, or other factors.

Management Implications

Based on my analysis, but more broadly the supporting literature and environmental conservation organizations, it is clear that states like Vermont must take initiative to address the impact various land uses/land covers have on water quality. There are a number of organizations in Vermont that have done important work in bringing this issue to the public and addressing the need for healthy riparian zones.

The Vermont Department of Environmental Conservation has a division focused on Watershed Management that focuses on "protecting, maintaining, enhancing, and restoring" all of Vermont's water resources. Part of this work involves establishing guidelines and best practices for river corridor protection through the River Corridor Easement Program. This program offers financial incentives to landowners in exchange for land rights along bodies of water, with which the Department of Conservation can implement river corridor plans that manage erosion and pollutants. The Vermont DEC also has published their Water Quality Standards, which recognize the importance of riparian buffers to improve water quality. Additionally, the University of Vermont Watershed Alliance has published an interactive map and other information online that is focused on increasing awareness of and knowledge about watersheds in Vermont. These steps by the State of Vermont to educate the public about watersheds, financially invest in river corridor protection, and establish guidelines surrounding water quality standards are necessary steps along to road to clean water in Vermont.

Riparian forest buffer along a river.

https://gf.nd.gov/gnf/private-lands/images/crep.jpg

I would recommend several strategies Vermont could implement to improve the management of the water quality of its lakes, rivers, and streams:

Invest in infrastructure and research such that landscape-scale studies can be conducted of land use/land cover impact on water quality. This includes taking samples of water quality at numerous spatial and temporal scales, higher quality imaging with more land cover and land use classifications, and, overall, investing in databases that are developed for the sole purpose of investigating the effect of land covers on water quality (Sliva and Williams, 2001).
Infrastructure improvement projects to implement riparian buffers and other natural forms of erosion/overflow control.
Increased focus on agricultural land and impervious surfaces as areas of improvement. For agricultural land, irrigation management and controlling grazing and livestock is essential for minimizing run-off into streams and lakes. For impervious surfaces, important steps would include investment in green infrastructure and storm water management infrastructure.

Sources

Bongard, Phyllis, and Gary Wyatt. Benefits of Riparian Forest Buffers. University of Minnesota Extension, 2010, www.extension.umn.edu/environment/agroforestry/riparian-forest-buffers-series/benefits-of-riparian-forest-buffers/.

Sliva, L., Williams, D.D., 2001. Buffer zone versus whole catch- ment approaches to studying land use impact on river water quality. Water Res. 35.

Snyder, M.N., Goetz, S.J., Wright, R.K. (2005), “Stream health rankings predicted by satellite derived land cover metrics”, Journal of the American Water Resources Association, 41(3)

Vermont Agency of Natural Resources, December 2015. Riparian Management Guidelines for Agency of Natural Resources Lands. http://fpr.vermont.gov/sites/fpr/files/About_the_Department/Rules_and_Regulations/Library/Riparian%20Final%20Guidelines%20%28signed%20copy%29_resized.pdf

Vermont Department of Conservation. Values of Riparian Buffers. http://dec.vermont.gov/sites/dec/files/wsm/rivers/docs/rv_riparianvalues.pdf