Vegetation Classification of Fish Springs NWR
by
Martha Jean Aiken and Robert Weiss
Introduction
Wetlands are one of the most productive ecosystems in the world. Some of the important uses of wetlands include controlling floodwaters, recharging groundwater, filtering pollutants, habitat for waterfowl and other wildlife, sanctuary for rare and endangered species, and also for educational, recreational, and aesthetic promise (Niering 1989). They are also vital in maintaining biospheric stability by regulating gaseous concentrations of nitrogen, oxygen, and methane.
Classification of wetlands is important to make use of all the available information so there can be some coordination between agencies, different landowners, and other interested parties (Gebhart et al. 1990). There is generally a lack of data on wetlands so all additional research will be useful for future reference (Bridges et al. 1994).
Aerial photography and remote sensing can accelerate and enhance the collection of ground data (Clemmer 1994). Ground cover, bare soil, and acreage can be calculated from aerial photos. Remote sensing photos can also help assess wetland functionality and improve management planning processes like recreation, road construction, mineral activities, etc. Wetland plant communities can be delineated for mapping purposes and generalized vegetation/soil correlations can be made. The aerial photographs can also be transferred over to GIS for further spatial modeling purposes.
Fish Springs NWR lies in a basin and is located on the northern boundary of Juab County, Utah, about 35 miles from the Nevada border. The Great Salt Lake Desert is north of the refuge, while Fish Springs Mountains and rolling dunes surround the rest of the basin. Elevations of the basin range around 4300 ft and annual precipitation averages 7.13 inches. The springs occur at the bed level of the ancient Lake Bonneville. At this point, a hardpan layer underlies the lakebed sediment.
Figure 1: Northern perimeter of Fish Springs, NWR
Three major springs and numerous smaller springs supply the Refuge with water. Slight elevation changes in the lakebed sediments cause water to back up, leading to the formation of marshes. Early settlers relied on these waters for irrigation and developed an elaborate system of dikes and impoundments. Now, the water is managed for wildlife habitat.
The purpose of our study is to classify the vegetation of the marshes along moisture gradients using remote sensing images and ground truthing. Combining ground truthing with carefully selected aerial photo baseline methods is less expensive than just conventional ground methods. We will also delineate habitat for certain wildlife species by GIS. Classification of the wetlands will enable us to see what type of vegetation is out there and in what stage of succession it is in. Looking at successional stages can also help to determine site potential and available wildlife habitat. Each wetland provides requirements for particular species of the plants and animals that live there (Ursin 1975). If plant and animal habitat diversity are kept up, so should a more balanced wetland ecosystem.
A plant ecology paper by Bolen (1964), described the plant communities in the area at that time. The upland desert community had a saltbush (Atriplex sp.) association, while the meadows near the aquatic communities are dominated by saltgrass (Distichlis spicata) (Figure 2). Rush (Juncus sp.) and reedgrass (Phragmites sp.) are also scattered throughout the basin. The emergent plant communities are dominated by bulrush (Scirpus) and cattail (Typha sp.). Emergents are plants that root in the water and have their flowering parts above the water surface. Stonewort (Chara sp.) and ditchgrass (Ruppia sp.) constitute the submersed vegetation in the marshes.
Figure 2 (saltgrass)
Methods
Classifications were developed using Landsat TM data. Because the Refuge is a relatively small area with a high diversity of vegetation types, we chose 30m resolution imagery (Figure 3). The classification was ground-truthed on May 4. Using the data from the ground truthing we mapped the vegetation communities based on two different approaches. Using Erdas Imagine 3.0 we ran an unsupervised classification of the TM data using ISODATA for the first approach. This classification was based on ten classes. The first two of these classes were identified as water and wetland vegetation communities. These classes were masked and then subsetted into five subclasses and three subclasses, respectively.
Figure 3: Landsat TM 30m resolution
The maps from the unsupervised classification also served as a basis for mapping potential habitat of five wildlife species.
For analyzing wildlife habitat and diversity at Fish Springs, five species of wildlife were chosen that frequent wetlands, and optimum habitat was estimated from literature and personnel at Fish Springs. My criteria for habitat was to select breeding or nesting habitat and areas where the wildlife species could obtain food. First, we used GIS OVERLAY to group classes 1-10 and the subclasses from Classes 1 and 2 all together. We ended up with eighteen classes. For each species we selected three different habitat types: optimum (3), marginal (2), and no habitat (1). We then took these numbers 3,2,and 1 and matched them up against the eighteen classes for all five wildlife species using GIS RECODE. Using MODEL MAKER, we established images of optimum habitat and optimum+marginal habitat (Figure 4).
Figure 4
The five species we chose to do habitat on were the frog (Rana sp.), which can be found near springs or standing water with emergent vegetation during the breeding season (Wernett 1982). The muskrat (Ondatra zibethicus) houses in cattail, reed, and grass structures, and is tied to water for most of its life (Armstrong 1982). The northern harrier (Circus cyaneus) is a bird of prey that nests among low bushes or among reeds in or near a marsh (Sutton and Walton 1994). The last two species are migrating birds and include the sandhill crane (Grus canadensis) and the Canada goose (Branta canadensis). The sandhill crane (Figure 5) nests in open areas with shallow water and dense vegetation like sedge, grass, reed, and rushes (Wernett 1982; Walkinshaw 1949). The information for nesting areas for the Canada goose we obtained from personnel at Fish Springs. They prefer islands of saltgrass vegetation surrounded by moderately deep water. Sandhill cranes will also nest on islands similar to this.
Figure 5 (sandhill crane)
Approach 2
The second approach mapped the communities based on supervised training of the data set. Mapping the vegetation communities based on Landsat TM imagery was complicated by the presence of standing water. Water levels within vegetation communities are regulated to meet wildlife management objectives. Because standing water absorbs almost all spectral radiation, an emergent vegetation community may have lower pixel values than the same community in non-emergent conditions. To minimize the confounding influences of standing water, an NDVI image was developed for the area (Figure 6,7). This image was used in the supervised classification. Data collected at twenty field sites formed the training set used in the supervised classification. The classification was run non-parametrically.
Figure 6: NDVI Model
Figure 7: NDVI image
Patterns of soil moisture and standing water influence species distribution in most marshes. Given the lack of information for the conditions at the time the image was taken, it is not possible to develop a supervised classification for the moisture gradients. Instead, a relative moisture gradient was developed using unsupervised classification for 40 classes. The classification was run using only Band 6.
Results
Approach 1
The outcome of our unsupervised classification (ISODATA) and ground truthing show some good correlations of wetness and greenness. We ended up only classifying the first six classes out of the original ten (Figure 7). The spectral signatures for the last four classes were indicative of more bare soil or rock, because of higher average reflectance values throughout all six layers. This is clearly shown in Figure 8.
Figure 8
Figure 9 is an image of the first six classes out of the ten we originally did and table 1 consists of rough descriptions of the different vegetation types. It was these six classes that we did the ground truthing on.
Figure 9: Classes 1-6
Table 1
Classes one and two, which are water and marsh vegetation, are located in the wetland area on the valley floor. The light blue and light green areas on the hillside to the left of the marsh wetlands are of a different cover type than that of water or marsh vegetation. Class three is mainly on the hillside to the west of Fish Springs and the dominant vegetation is rabbitbrush (Chrysothamnus sp.) and shadscale (Atriplex sp.). There are also grasses, cliffrose (Cowania mexicana) sagebrush (Artemisia sp.) and spiny desert brush. Class four is located in the northern end of Fish Springs and the lower slopes of the hillside. Dominant vegetation is greasewood (Sarcobatus vermiculatus) rabbitbrush, and saltgrass. Class five is on the fringes of the marsh on the valley floor, and is characterized by greasewood, saltgrass, and cracked, dried mud. Class six mainly surrounds standing water in the northern portion of the valley and is characterized by shallow standing water with extensive coverage of saltgrass.
The subclasses from class 1 and 2 are shown here in Figures 10 and 12 and their accompanied classified habitat in Tables 2,3.
Figure 10: Subclasses of Class 1
Table 2 (Subclasses of Class 1)
The green portion (subclass 11) is deep standing water one or more meters. The dark blue section (subclass 12) indicates shallow water up to one meter. Yellow (subclass 13) is also standing water, but includes small islands of vegetation within it. Bulrush is dominant on the pink portion (subclass 14), along with saltgrass and occasional cattails. The red portion (subclass 15) is predominately shallow water with saltgrass growing in parts of it.
Figure 11
Our signature plots for subclasses from Class one in Figure 11 clearly show the pink line (bulrush) and yellow line with the greatest increase between layer three and four, which indicates greenness in this otherwise water dominated class.
Figure 12 (Subclasses of Class 2)
Table 3 (Subclasses of Class 2)
The red section (subclass 16) in Figure 12 is associated with the wetter part of the vegetation in the marsh as indicated by layer 6 (moisture indicator layer) in Figure 13 and is dominated by reedgrass, rushes, and saltgrass, with minor amounts of bulrush.
Figure 13
The green section (subclass 18) has slightly less soil moisture than red and is characterized by rushes with lesser amounts of reedgrass and saltgrass. The signature plot (Figure 13) indicates greeness for reedgrass subclass 16 and rush subclass 18. The blue section (subclass 17) is mainly on the ridges high above the valley and is dominated by bluebunch wheatgrass (Agropyron spicata)and big sagebrush (Artemisia tridentata).
The results from the habitat diversity models show two different images below. The green color in Figure 14 shows subclasses 13,14,and 15 from Table 2 and depicts optimum and marginal habitat where all five species may be found. It includes bulrush, saltgrass, cattails, rushes, shallow to medium depth water, and dispersed islands of saltgrass and other vegetation.
Figure 14
The next image (Figure 15) shows subclass 13 from Table 2 and is optimal habitat for four of the species and marginal habitat for northern harrier. It is mainly characterized by standing water, with islands of saltgrass and fringes of cattail and bulrush at the edges of the water.
Figure 15
Results for Approach 2
Wetland vegetation communities cover approximately 2364 ha of the Refuge. The supervised classification delineated these communities (Figure 16) at a finer scale than the unsupervised classification. This complex mosaic of communities is consistent with the patterns Bolen (1964) described as "unequivocally variable and...[becoming] increasingly complex as one proceeds inward from the periphery of the basin's confines".
Figure 16 and Figure 16a: Wetland vegetation communities.
The unsupervised classification of soil moisture into 40 classes produced a continuous map grading from blue, indicating standing water, to brown, indicating the wettest soils, and then to white at the driest end of the gradient (Figure 17). Comparing the vegetation classification with the relative moisture gradients suggests trends for only a few of the communities.
Figure 17: Moisture index.
Data from the summary analysis is presented in Appendix 1. Saltgrass-dominated communities (Figure 18) account for 45% (1074 ha) of the marsh area (excluding the probable commission error in assignment to Class 1). These communities occur in 75% of the moisture gradient classes. The number of occurences decreases as the soil becomes drier. Some communities are dominated by both saltgrass and rush. These communites occur over a wide range in the moisture gradient (75% of the classes) and account for 19% (460.89) of the marsh vegetation.
Figure 18: Saltgrass community.
Pure stands of rush, stands of reedgrass and mixtures of these species are restricted to the wettest classes on the moisture gradient. Each of the pure stands accounts for approximately 9% (216-236 ha) of the wetland vegetation. Mixed communities account for approximately 16% (376.2) of the marsh.
Commission and omission errors are evident in the classification. It is unlikely, for example, that greasewood will occur at the center of a pond. These areas are probably aquatic vegetation. Successive iterations of the supervised classification failed to correct these errors, and often resulted in larger commission/omission errors. To rectify these discrepancies requires further ground-truthing with greater attention to delimiting the drier community types and submerged aquatic communities.
Discussion
Our vegetation classification shows that different vegetative covers have different greeness and wetness characteristics on the signature profile. Looking at Figure 11 and the differences between layer 3 and 4, subclasses 14 (bulrush) and 13 (water with island vegetation) are excellent indicators of greeness. The other three subclasses in this figure clearly indicate standing water by looking at layer 6. Subclass 16 (reedgrass) and subclass 18 (rush) in Figure 13 are also excellent indicators of greeness. These two figures show that bulrush (subclass 14) and reedgrass (subclass 16) which are normally the tallest marsh plants at Fish Springs, also are the best indicators for healthy green vegetation. Overall, classes 1 and 2 in Figure 8 show the highest levels of greeness, with classes 3, 4, and 5 coming close behind. Classes 1, 2, and 6 indicate more moisture and wetness. When saltgrass and water are combined, like in Class 6 and subclass 15, wetness prevails. As bulrush gives way to reed, rush or saltgrass, greeness and wetness decrease.
We also did a TASSLED CAP TRANSFORMATION comparing wetness vs. greeness for all 18 classes. Figure 19 indicates that Classes 1 and 6 are the wettest, while Classes 1 and 2 are the greenest. Figure 20 shows subclasses 14 and 16 being the greenest, with subclass 15 being the wettest. The Tassled Cap data correlates well with the greeness and wetness information we analyzed from the spectral signatures.
Figure 19
Figure 20
Seasonal natural fluctuations of water at Fish Springs are not of magnitude enough to cause large amounts of emergent marsh to a dried and decadent state or to kill a lot of vegetation due to long durations of unfavorable inundation (Bowen 1964). Man-made flooding or draining, however, can induce large amounts of vegetation senescence, change bare ground and water cover, and change reflectance values for the landscape.
Changes in vegetative cover can also disturb wildlife habitat. As discussed earlier, wildlife habitat is tied to certain combinations of vegetative cover and water in the marsh.
Figure 21 (muskrat)
Figure 22 (Canada Goose)
Figures 21 and 22 show optimal and marginal habitat for muskrat and Canada Goose, respectively. The yellow color in both figures indicates optimal habitat and is directly related to sublass 13 (water and island vegetation). Significant decreases in water levels may greatly reduce favorable nesting and denning habitat. Knowing this, careful management practices should be followed as to avoid any large changes in ecosystem structure.
Classifying wetland vegetation can also lead to assessing soils, erosion, sedimentation, nutrients, water quality, and functionality (Bridges et al. 1994). Photo image data is currently lacking for soils in Juab County and would be helpful for future assessments. Combining remote sensing, GIS, and ground data can further fill some gaps and provide for better managment practices.
References
Armstrong, D.M. 1982. Mammals of the canyon country. Canyonlands Nat. His. Assoc. 263p.
Bolen, E.G. 1964. Plant ecology of spring-fed salt marshes in western Utah. Ecol. Mono. 34:143-166.
Bridges, C., R. Krapf, S. Leonard, W. Hagenbuck. 1994. Riparian area management process for assessing proper functioning condition for lentic riparian-wetland areas. US Dept. Int. Tech. Ref. 1737-11. 37p.
Clemmer, P. 1994. Riparian area management the use of aerial photography to manage riparian-wetland areas. US Dept. Int. Tech. Ref. 1737-10. 54p.
Gebhart, K., S. Leonard, G. Staidl, D. Prichard. 1990. Riparian area management riparian and wetland classification review. US Dept. Int. Tech. Ref. 1737-5. 56p.
Niering, W.A. 1989. Wetlands. Chanticleer Press, NY. 638p.
Sutton, C. and R.K. Walton. 1994. North American birds of prey. Chantileer Press. 191p.
Ursin, M.J. 1975. Life in and around freshwater wetlands. T.Y. Crowell Co. 116p.
Walkinshaw, L.H. 1949. The sandhill cranes. Cranbrook Ins. of Sci. 202p.
Wernett, S.J. 1982. North American wildlife. Readers Digest Assoc. Inc. 559p.
