Riparian Land Use Estimation Using GIS and Remote Sensing

Abstract

A land use/land cover classification was merged with buffered areas formed around streams. The resulting data allows biologists and planners to assess individual stream segments based on the percentage of each land use type. The data is being used to identify stream segments with poor corridors for inclusion in stream habitat improvement programs. It is also being used to identify less impacted reaches for protection and for rating streams for Missouri Aquatic GAP.

Background

Researchers have found strong correlations between certain types of land use / land cover (LULC) and stream water quality and or habitat condition. For instance, Omernik (1976) found higher nutrient concentrations in areas where streams drained agricultural land. Wang, et al (1997) found habitat quality was significantly positively correlated with the amount of forested land in watersheds, and negatively correlated with the amount of agricultural and urban land in watersheds. They found similar, though somewhat weaker correlations for LULC classes in stream buffer areas adjacent to each sampling site. However, Gregory, et al (1991) found that riparian vegetation traps sediments and nutrients from runoff, and influences cover through woody debris, water temperature through shading; and primary production through leaf litter inputs. Roell (1994) states that by maintaining a forested buffer of 50-100 feet around the stream channel, landowners and conservation professionals are able to prevent some of the erosion, bank slumping, wetland siltation, stream turbidity, and nutrient loading that result from plowing fields to the stream bank.

The known benefits of vegetated riparian areas have prompted Government agencies to identify riparian re-vegetation as a viable management strategy to help alleviate a host of water quality and habitat problems in streams. The Natural Resources Conservation Service (NRCS) and Missouri Department of Conservation (MDC) have developed a number of programs to promote re-vegetation. Cost-share agreements, such as the Stream Buffer Conservation Reserve Program of NRCS, are provided to landowners to take riparian areas out of row-crop or pasture, and put them into appropriate species of trees, and/or shrubs.

The common practice in public agencies is to give cost share agreements to landowners that request them, or to suggest such a practice to the landowner if the manager can identify that the practice would be suitable. Managers may find that a GIS inventory of the LULC in riparian areas would assist them in this endeavor. They could use this tool to identify those sites where extensive row crop agriculture is occurring in close proximity to the stream channel and try to get those sites into a re-vegetation program. They may also wish to identify areas that had excellent existing riparian vegetation, and use the data to educate landowners about the need for maintaining riparian vegetation in the riparian areas on their property.

GIS riparian LULC data could not only show where these sites occur on a property being reviewed, but could also show the larger distributional pattern of impaired or pristine riparian areas. Identifying all stream reaches that need to be re-vegetated is useful if there are unlimited funds and landowner interest to do so. However, funds are usually finite, and landowner interest does not always match the needs of the resource. It may be useful to identify regions of concern based on the distributional patterns of riparian / LULC. Using a watershed or other type of region to sum the extents of such sites is a useful way to quantify data for assessment. The data could answer a number of questions that would allow a more focused management approach. Where are there clusters of the most degraded riparian areas? Where are there clusters of the most pristine ones? It may be a better use of limited funds, to identify the geographic patterns of the problem and then apportion cost share programs to those regions with the greatest need for intervention.

The pattern of these areas can be quantified and expressed as the total percentage of affected riparian area per some larger geographic unit. However, choosing an appropriate geographic unit to sum riparian land use data may require assessing both the drainage patterns as represented by watersheds, and land use potential as represented by ecological data.

Watershed divides form true boundaries for aquatic systems. Unless influenced by groundwater moving independent of the divides (as in karst topography), most of the variables influencing water quality and stream habitat for a stream will be found within the stream's watershed. Water quality is primarily determined by the sum of all influences that water contacts as it flows within the watershed and then down the main channel to the outlet. For instance, point sources of pollution, such as effluent pipes from a factory can outlet directly to a stream, or non- point sources, such as crop fields sprayed with herbicide, can deliver pollutants to the stream through runoff. Stream habitat can also be influenced by watershed/stream system variables. The amount of woody debris input into the stream system from trees in the riparian zone can determine whether root wads are available to block flow and help form pool habitat for fish. The amount of sediment input to the system from agricultural fields and construction sites can have significant effects on the quality of the substrate, when interstices between stones are filled and invertebrates are buried.

Missouri covers an area of transition between ecological regions . The northern part of the state was glaciated, and is flat in some areas, but has mostly rolling land with limited to moderate relief. The Ozarks hills cover most of the southern part of the state, and the topography has more relief and is highly dissected. In the southeast is the Mississippi Embayment, an extremely flat area almost completely given to row crop agriculture. There are also mixing zones between each of these distinct ecological regions. LULC patterns are quite different in each of these regions, and land use around streams also varies between the regions. In the glaciated areas to the north, most of the land is in pasture and some is in crop land. Forested areas are usually directly around streams. In the Ozarks, the opposite can be true; most areas are forested, and much of the crop land and pasture is around streams because the flood plains may be the only areas flat enough for these land use types. In Missouri, the Ecological Classification System project has identified distinct ecological units based on many variables related to topography, geology, soils, vegetation, etc. These ecological units have different percentages of the various LULC types.

Streams often traverse two or more distinct ecological regions, and, at a certain point on a stream, the topography, geology, and thus, many stream habitat variables, such as gradient and substrate texture, can be more related to the local ecological region, than to the sum of influences within the watershed (Omernik and Bailey, 1997). The nature of the topography, geology and soils directly affects land use potential, and thus the extent and frequency of various land uses can be highly correlated to ecological regions. Thus streams within a certain ecological region may require different management than those in the same watershed, but in another ecological region.

However, the aquatic species that stream programs intend to protect are bound by the topographic divides, and they are limited to the watersheds that can be used to define their range. An endangered fish species, such as the Niangua Darter, is not distributed over the whole of the ecological region that covers most of its range; it is confined to certain watersheds. Thus it may be most prudent to assign watersheds to corresponding ecological regions, because the distribution of aquatic species, as well as the distribution of the land use disturbances that affect their habitat, can both be represented in management assessments.

It is possible to identify LULC patterns around streams by producing an inventory of stream buffer areas merged with a LULC classification. Watersheds can also be merged with the LULC classification. The two data sets can then be merged to show watersheds where the LULC percentages around streams are low or high, or how they vary significantly from the non-buffer areas within the watershed. If watershed units are small enough, they can also be assigned to the ecological region which covers the most area within the watershed. In this way, watersheds can be rated for how disturbed or undisturbed the riparian and non-riparian areas within them are, compared to an identifiable averages for those influences within their respective ecological subsections. This rating could then guide various research and management objectives including the apportionment of cost-share monies.

Data

The watershed files are based on the 8-Digit, or sub-basin level of the United States Geological Survey's (USGS) hierarchical watershed system. These files have units that conform to the sub-watershed level as represented by the NRCS 14-Digit Hydrologic Units, which are finer subdivisions of drainages in the hierarchical system. The watersheds are clipped to the state border of Missouri. The sub-watershed units range from 40,000 - 60,000 acres.

The stream data files also conform to 8-Digit HUC river basins,. The line work includes areas outside of Missouri that are the non-Missouri portions of the 8-Digit HUC's. The stream data was acquired by MDC as 1:100,000 scale United States Geological Survey (USGS) hydrography, originally produced by the Geographic Resources Center (GRC), that later was reworked to include center lines for river and lake polygons by the Center for Agricultural Research and Environmental Science (CARES). These streams are not as detailed as the streams on 1:24,000 scale topographic maps. When compared to 24k streams, they represent Strahler second Order streams to their full lengths. So, they include approximately 55% of the stream miles to be found at 24k.

The land cover classification was produced by the Missouri Resource Assessment Partnership (MORAP). Its source is Landsat 30 meter resolution satellite imagery that was collected between 1993 and 1995. The classification included sixteen LULC classes. The original classes were simplified into a set of classes that represent varying runoff potential. There was no need, for instance, to have "Oak-Hickory Forest Mix" and "Oak-Hickory-Cedar Mix" in hydrologically oriented analyses, because the runoff potential is so similar. The resulting classes include the following:

Forest

Wetland

Grassland

Cropland

Urban

Water

The ecological regions were produced by the Missouri Ecological Classification System (ECS) project. It was mapped at a scale of 1:100,000. The classification system is hierarchical, and the subsection level, which includes 32 units for Missouri, was selected as being appropriate for identifying significant land use differences between regions. The next level, the Land Type Association (LTA), has ecological units that are broken at land forms that represent only the headwater areas of the smallest watershed units. This makes assigning the watershed units to an LTA invalid because any hybrid LTA/Watershed unit would not represent an actual drainage area.

Analysis

Analyses were conducted to identify LULC percentages for watersheds, as well as for stream segment buffer areas. A final analysis combines and compares the information from the resulting watershed and stream files into a watershed file that can be used to show LULC trends over the state.

The "shedland.aml" program is used to convert the 14-Digit units to raster and then merge them with the land cover classification. The program then derives the percentage of each land cover class within each 14-Digit unit. The program joins this information back to the original vector watershed file. The user is then able to select watersheds that contain more than 50% Forest, or more than 25% Cropland, etc.

The term "Riparian Estimator" (R.E.) has been coined for the results of a program called "strmland.aml". The strmland.aml program is more complex than shedland.aml. Its purpose is to identify potentially deleterious (or benign) land use types around streams at very specific sites for management purposes. There is a tendency for such sites to not be identified because their percentages are "watered down" by being part of a large buffer zone around a long stream segment. To avoid this problem, the program splits the lines with the DENSIFY ARC command such that no line segment can be more than 400 meters (1/4 mile) in length. It then converts the stream network to raster format in GRID and buffers it by one pixel on either side creating a 3 pixel (three 30 meter pixels = 90 meters) wide swath about the channel. This width includes some areas around headwater streams that are not riparian, and excludes some areas around larger ordered streams that are considered to be riparian, but is still useful for stream management purposes. Uniquely coded buffer areas are formed for each segment, and the buffer area grid is merged with the land cover classification. Finally, percentages are derived for each land cover type for each unique stream segment's buffer area. The result is joined back to the original vector stream file, which can be used to show where various types of surrounding land use intersect with streams. Stream reaches can be grouped into classes describing their LULC, such as "75-100% Forest", or "Greater than 50% Cropland". The short segments' buffer zones show high percentages of deleterious land uses, such as crop land, where these land uses occur, allowing managers to identify the sources of potential water quality and stream habitat problems.

The "shedcalc.aml" and "streamcalc.aml" were written to compare the land use in the riparian area to the land use in the non-riparian areas of watershed units. The FREQUENCY command was used to sum the land cover classification class counts for the buffer areas around streams in each watershed unit, and then the percentages for each class were derived for each watershed unit. To make a comparison between riparian and non-riparian areas within each watershed, the counts for these riparian area land cover classes were then subtracted from the counts for each class for each watershed unit in the watershed coverage. This was done by both 8- Digit and 14-Digit hydrologic units. Comparison percentage items were added to both the stream and watershed files such that for any watershed unit, the user can see the difference between the Forest percent in the 8-Digit basin and the 14-Digit sub-watershed, see the difference between riparian Forest percent and sub-watershed Forest percent, and see the difference between the Forest percent around a certain segment compared to the Forest percent around streams in the sub- watershed unit, etc.

Several operations were carried out to test the validity of merging watershed and ecological units. A program was written to assign each 14-Digit watershed unit to the ecological subsection which represented the majority of its area. The FREQUENCY command was used to sum pixel counts for the various LULC classes by watershed unit as assigned to ecological subsection. The ecological units were merged with the LULC classification using a modified version of the shedland.aml. Center points for each 400 meter stream segments of the Riparian Estimator data were collected into a state-wide file from the sixty-six 8-Digit basin stream files. These 461,587 points were run through IDENTITY against the ecological subsection layer to attach the value of the subsection item. FREQUENCY was then used to sum the stream segments' LULC values by subsection boundary. This allowed a more direct comparison between the LULC percentages for ecological subsections and the LULC percentages for watershed units as assigned to those ecological subsections.

Results

The percentages of LULC classes in the buffer area surrounding Missouri streams is different than percentages of LULC classes in the non-buffer areas of the state. See Table 1.

Over the state of Missouri, there is more Forest in the buffer area around streams than in the non-buffer area. Steep banks and flooding potential can limit land uses and favor Forest in the stream buffer area. Consequently, there is less Grassland and Cropland in the buffer area. Compared to the state-wide LULC percentages, the Urban class percentage is slightly lower, and the Water and Wetland class percentages are higher in the stream buffer area. This may be due to the fact that urban development is easier in upland areas, and streams are coincident with open water and wetlands.

Watershed LULC percentages varied widely depending on where each watershed lay in the varied topography of the state of Missouri. Missouri has four ecological units at the section level, and LULC percentages are markedly different between them. Note the great difference between the Forest percentage for the Ozarks, and the rest of the Sections. The Ozark unit has more than three times the percentage of Forest compared to the next highest unit's percentage of Forest. Percentages varied little between the percentages of LULC as summed for the Section by Section, or by Section as represented by watersheds assigned to a Section. See Table 2.

The LULC percentages vary even more widely by subsections, of which there are 32 for Missouri. See Table 3. In general, there is more of the Forest class in the buffer area around streams than in the non-buffer areas, but this is not always the case. There are subsections that represent regions with higher relief and percentage of Forest, such as in the Ozarks, where there is less Forest in the buffer area around streams than in the non-buffer area. This can be seen in the first record in the table, which lists values for the "Black River Ozark Border" subsection. This is because Forest has been replaced by other LULC classes, such as Cropland and Grassland, which can be deleterious to both water quality and stream habitat. Forest percentages are considerably higher in the buffer area around streams than in the non-buffer area, in the flatter, glaciated areas in northern Missouri. However, the Forest percentage around streams in those areas is low compared to other areas in the state, such as in the Ozarks. An example of this is the "Deep Loess Hills" subsection, a glaciated unit, as compared to the "Elk River Hills" subsection, an Ozark unit.

Watersheds were fairly well contained by subsections. Of all watershed units, 61.40% fell completely into a subsection. The mean containment was 89.83% in the subsection the watershed was assigned to. The minimum containment was 39.04%, and the standard deviation for containment was 14.94%.

LULC percentages varied little between the percentages as summed for the Subsection by Subsection, or by Subsection as represented by watersheds assigned to a Subsection. See Table 2. See Table 3. Compare the corresponding items on either side of the double line boundary. For instance, for the last record in Table 3., "Wyaconda River Dissected Till Plains", the "Subsection Forest" value, 16.39%, is similar to the "Subsection Watershed Forest" value, 15.95%.

Conclusions

The Riparian Estimator (RE)stream buffer LULC data shows local patterns of LULC in a way that should help conservationists manage streams. A simple method of showing stream / Cropland interaction is to use the "Natural Breaks function in the Legend Editor of ArcView to make a five class graduated color legend for the RE data using the "Sstbuf-crp-per" (stream buffer Cropland percent) item in the RE data's table. The lowest valued classes can be colored green, and the highest, red. The conservationist can then use the data with the LULC classification or DOQQ's in the background as he/she reviews a property under question. Ideally, the data would allow a proactive approach. The conservationist would look for areas needing rehabilitation, and then contact landowners to offer assistance.

Summing the data by watershed unit and ecological subsection allows managers to better understand the resource and thus better manage it by focusing their time and monies on areas of the greatest concern. The areas of concern can be identified within this data or through overlays with other GIS data and then displayed as a separate layer that can be used to make management decisions. For instance, the watersheds with the best and/or worst condition based on the average condition of the ecological subsection could be selected. Specifically, this might be effected thus: watersheds with more than one standard deviation of the watershed's assigned ecological subsection's mean stream buffer Cropland percentage could be flagged and prioritized to receive a higher percentage of the cost share money available for a stream buffer re-vegetation program. Similarly, watersheds with the highest percentages of Forest in both buffer and non-buffer areas as compared to the subsection's corresponding mean Forest percentages could be flagged to be protected.

Assigning the watersheds to ecological subsections appears to be valid, despite a number of Missouri's original 14-Digit watershed units being bifurcated by the subsection lines. This intersection of watersheds / ecological subsections is being improved. There is an ongoing project within Missouri to "refine" the existing 14-digit watersheds, making them more hydrologically pure and implementing new USGS guidelines. This is being undertaken by the author (Mark Caldwell, Missouri Dept. of Conservation (MDC)) and Terry Barney (NRCS). A part of this refinement process is the inclusion of watershed breaks at confluences that are at or near major ecological breaks. This part of refinement is being done in concert with the ongoing Missouri Aquatic GAP Project, headed by Scott Sowa and Gust Annis of MORAP. In that project, the results of the co-refinement process is a series of Ecological Drainage Units (EDU's), which are major drainage areas (usually at the 4-Digit to 6-Digit level) that contain an area of similar ecological regions. These are further subdivided into Aquatic Ecological Systems (AES) which are assemblages of 14-digit units with similar ecology.

Methodological Concerns

There are some concerns about the systematic use of the three pixel buffer area in the Riparian Estimator (R.E.) stream data. One concern is that the buffer width should be varied according to the size of the stream. The buffer width exceeds the riparian area around headwater streams, and represents open water areas on large rivers and impoundments, rather than their riparian area. Ideally, it would be varied by stream order, or simply include the whole flood plain of the stream as a buffer area. It is maintained for three reasons. The first reason is one of accuracy. The data with the coarsest resolution (and, thus, positional accuracy) is the land cover classification at 30 meters. The stream data is converted to raster format with the same resolution and grid as the land cover classification prior to merging. If a one-pixel representation of the streams were used, there is a potential for those pixels to not lie in the same grid cell as the position of the stream in the land cover classification. Therefore, the pixel could have a completely different and incorrect class. For most streams, using the three pixel wide swath about the channel will certainly include the stream as well as areas on either side of it considering all of the possibilities of error, such as: 1) greater than USGS map accuracy error for streams (they are not included in map accuracy), 2) digitizing error for both stream and watershed files, 3) positional and classification error for the land cover classification, etc. The second reason is the lack of appropriate layers. Missouri has no statewide flood plain layer. Stream Order is recently available, but channel width for the various stream orders is variable and difficult to estimate because of extensive karst geology. National Wetlands Inventory (NWI) data is available and features 1:24,000 scale river polygons. These could be buffered, but then the polygons (or their outlines) would have to be segmented so that significant localized deleterious LULC classes would have a high enough percentage to be recognized. This has been done as a pilot, but it requires programming and editing and has not been done for the whole state as yet. The third reason is one of practicality; though this buffer width of 90 meters (295 feet) seems very wide, it locates problem and/or pristine areas with an acceptable accuracy for the management of the riparian area of streams. This can be verified using Digital Ortho Quarter Quads (DOQQ's) or other air photos. River channels large enough to cause the three pixel buffer width to include significant numbers of water pixels, are nearly always leveed. The levee provides the function of trapping sediments and reducing bank erosion, and thus these riparian areas are not valid for stream corridor programs. Large impoundments have unique management issues that require different analytic approaches and are not covered by stream programs.

References

Gregory, Stanley V., Frederick J. Swanson, W. Arthur McKee, and Kenneth W. Cummins. 1991. An ecosystem perspective of riparian zones. Bioscience. Vol. 41, NO. 8.

Omernik, J.M. 1976. The influence of land use on stream nutrient levels. Environmental Protection Agency Ecological Research Series EPA-600/3-76-014, U.S. Environmental Protection Agency, Environmental Research Laboratory, Corvallis, OR.

Omernik, J.M. 1997. Distinguishing between watersheds and ecoregions. Journal of the American Water Resources Association. Vol. 33, NO. 5.

Roell, M. J. 1994. Considerations for recommending streamside protection zones in Missouri. Missouri Department of Conservation, Jefferson City.

Wang, Lizhu, John Lyons, Paul Kanehl, and Ronald Gatti. 1997. Influences of watershed land use on habitat quality and biotic integrity in Wisconsin Streams. Fisheries. Vol. 22, NO. 6.

Contact