Mapping the Kansas Grasslands: A Multiseasonal Approach
Stephen Egbert, Chris Lauver, Clayton Blodgett, Kevin
Price, and Ed Martinko
The extensive grasslands of Kansas dominate the states natural vegetation. To the west, in the lee of the Rocky Mountains, sparse rainfall generates arid shortgrass prairies, while increased rainfall in the central part of the state yields mixed-grass prairies. To the east, sufficient precipitation occurs to support tallgrass prairie that mixes with oak-hickory deciduous forest in the far eastern part of the state. Most of the grasslands in the western two-thirds of the state are native, having never been plowed, and are primarily used for grazing domestic livestock. In the tallgrass prairie region, grazing is also prevalent, but many grasslands (both tame and native) are managed for hay production. Kansas also contains large acreages of former cropland that are now covered with native and non-native grasses as part of the USDA Conservation Reserve Program (CRP).
According to a recent map of land cover patterns in Kansas (Whistler et al. 1997), the distribution of grasslands is often associated with the moderately sloping terrain of major and minor drainages, whereas the alluvial river valleys and level-to-gentle upland slopes are used for crop production. However, there are several regions in Kansas that contain relatively intact grassland ecosystems, mainly because of high topographic relief and rocky or sandy soils. These include the famed Flint Hills region with its rolling tallgrass prairie in eastern Kansas that stretches from near the northern border into the Osage Hills in Oklahoma. In the south-central region, Red Hills mixed prairie is found on gypsum hills in a scenic landscape dotted with red cedar trees and caves. Another grassland type, chalkflat mixed prairie, occurs in west-central Kansas along Hackberry Creek and the valleys of the Smoky Hill River. This region is famous for its beautiful erosional remnants of Niobrara chalk. Sand prairie and sandsage shrubland occur in the southwest along extensive sand dunes to the south of the Arkansas and Cimarron Rivers.
Development of a Grassland Classification System for Kansas
In 1989, the Kansas Natural Heritage Inventory of the Kansas Biological Survey (KBS) developed a preliminary statewide vegetation classification to identify and plan protection for exemplary occurrences of Kansas ecological communities. The classification was based on examining Küchlers (1974) potential natural vegetation map in relation to the geology, soils, and physiographic provinces of Kansas. Vegetation types were identified based on variations in physical features (e.g. climate, soils, and topography) that contributed to differences in plant species composition. For example, although sharing the same dominant species, a "northeastern" and "southeastern tallgrass prairie" were formed because of known differences in soil development (i.e. glaciation) and the floristic composition of communities in these areas.
The present grassland classification system used in the Kansas GAP Project and by KBS is a conversion of the 1989 version into the vegetation classification system developed by The Nature Conservancy in cooperation with state, federal, and academic partners (The Nature Conservancy Ecology Working Group 1997). The new classification of the natural vegetation of Kansas (Lauver et al. in prep.) contains 23 grassland community types under 13 different alliances.
Problems Inherent in Mapping Grasslands
Several problems are inherent in attempting to map grassland types using satellite imagery. The first one is the nature of grasses themselves. Individual grass plants are much smaller than trees and shrubs and are below the resolving ability of any commonly used digital or photographic system. Closely related to this is the frequent spatial variation in cover composition within a given grassland type. Unlike crop fields, grassland vegetation is rarely homogeneous unless it has been planted and managed. Each grassland type consists of mixtures of grasses, forbs, and even shrubs. In addition, patches of bare soil often enter the picture, particularly in arid regions. All of these factors create an environment where "pure" pixels are a rarity, and where considerable spectral heterogeneity can be found within a single grassland parcel.
Another issue is that most of the grassland in Kansas is actively managed for agricultural uses, including grazing and haying. Intensity and seasonality or timing of use, especially for grazing, vary widely, depending on the practices of the owner or manager, climatic conditions, and grassland health. Visual inspection of satellite images or air photos often shows clear delineations among land parcels because of differences in grazing intensity. In addition, grassland used for grazing is often burned in the spring to stimulate production. Hayed grasslands, whether natural or planted, also present challenges because haying practices vary by land owner. From the standpoint of land cover mapping, the biggest concern arising from the intensive human management of grasslands is that spectral variations due to management practices may mask the variations among grassland types and cause unacceptable confusion in land cover maps.
A Multiseasonal Approach
To address the problems outlined above, we have elected to use a multiseasonal two-stage approach to land cover mapping. Using a multiseasonal approach in Finney County in southwest Kansas, we earlier produced excellent results in separating grasslands from croplands and in mapping individual crop types (Egbert et al. 1995). Based on that study, we decided to apply a similar approach to mapping grasslands. For each scene center in Kansas, we acquire three dates of Landsat Thematic Mapper imagery over the growing season: spring, summer, and fall. Our rationale for using this approach is that seasonal differences in plant development vary by species, and using multiple dates of imagery will increase the likelihood of sensing the differences among vegetation types. For example, we have found that when we use July images to classify vegetation in western Kansas, croplands like corn and milo are often spectrally confused with some riparian vegetation types, such as cattail and bulrush marshes. When a spring scene is added, however, the differentiation among the classes is simplified because the corn and milo fields are bare soil at that time of year.
Our methodology employs both unsupervised and supervised classification. Unsupervised classification separates cropland from natural vegetation, while supervised classification is used to map vegetation alliances. In initial processing, the images are georectified and registered to each other. The three images are then combined to form a single multidate image. To reduce the volume of data, only bands 3, 4, 5, and 7 are used from each image, resulting in a 12-band image. The 12-band multidate image then undergoes unsupervised classification using the ISODATA clustering algorithm and a maximum likelihood classifier, creating 100 raw classes. Analysts examine the raw classes and assign them to one of two categories: cropland or non-cropland vegetation. Classes with large percentages of pixels in both categories are placed in a third, confused class. The confused classes are isolated and undergo a second unsupervised classification in a "cluster-busting" technique (Jensen et al. 1987); the resultant new classes are then assigned to either the cropland or non-cropland categories. The result of the unsupervised classification phase is a map of cropland and noncropland land cover. This map is used to create an image mask containing only the noncropland pixels, which is retained for further processing.
Supervised classification is applied to the masked noncropland pixels to create a map of GAP land cover categories. Representative field sites are collected and labeled by grassland ecologists using images, maps, and GPS receivers. Two-thirds of the sites are used for training the classifier, while the remainder are used for verification. In the accuracy assessment process, the verification sites are used to create contingency tables, and to calculate users and producers accuracies, along with KAPPA.
Land cover mapping for Kansas GAP is being conducted by the Kansas Applied Remote Sensing Program and the Kansas Biological Survey at the University of Kansas in Lawrence, Kansas. Principal investigators for Kansas-GAP land cover mapping are Edward A. Martinko and Kevin Price. Researchers and staff members contributing to Kansas-GAP include Stephen Egbert, Chris Lauver, Clayton Blodgett, Miguel Ortega-Huerta, Ellen Ellis, Aimee Stewart, and Ryan Boyce.
A poster showing a map of the Kansas GAP pilot project and current land cover mapping status can be viewed on the National GAP home page at http://www.gap.uidaho.edu/gap/AnnualMeetings/1997/Posters/index.htm.
Egbert, S.L., K.P. Price, M.D. Nellis, and R. Lee. 1995. Developing a land cover modelling protocol for the high plains using multi-seasonal Thematic Mapper imagery. Proceedings, ACSM/ASPRS 95 Annual Convention and Exposition, Charlotte, North Carolina 3:836-845.
Jensen, J.R., E.W. Ramsey, H.E. Mackey, E.J. Christensen, and R.R. Shartiz. 1987.Inland wetland change detection using aircraft MSS data. Photogrammetric Engineering and Remote Sensing 53(5): 521-529
Küchler, A.W. 1974. A new vegetation map of Kansas. Ecology 55: 586-604.
Lauver, C.L., K. Kindscher, D. Faber-Langendoen, & R. Schneider. In prep. A classification of the natural vegetation of Kansas. (to be submitted to The Southwestern Naturalist).
The Nature Conservancy Ecology Working Group. 1997. International classification of ecological communities: terrestrial vegetation of the United States. The Nature Conservancy, Arlington, VA. (in prep.)
UNESCO (United Nations Educational, Scientific and Cultural Organization). 1973. International classification and mapping of vegetation. Paris, France. 35 pp.
Whistler, J.L., M.E. Jakubauskas, S.L. Egbert, E.A. Martinko, D.W. Baumgartner, and R.Y. Lee. 1997. Kansas Land Cover Patterns. Map available from Kansas Geological Survey, Publications Division, 1930 Constant Ave., Lawrence, KS 66047.