Classification and Mapping of Agricultural Land for National Water-Quality Assessment

By Robert J. Gilliom and Gail P. Thelin

U.S. Geological Survey Circular 1131

EXISTING CLASSIFICATION SYSTEMS AND DATA SOURCES

Two general types of classification systems and associated geographic data are available for characterizing agricultural land uses: (1) General land-use and land-cover classifications mapped from areal photography and remotely sensed data, and (2) agricultural activities and production statistics gathered in farm-by-farm censuses and surveys. General land-use and land-cover maps have the advantage of showing the actual geographic or spatial distribution of general categories of agricultural land in relation to other land uses, irrespective of political boundaries, but have the disadvantage of not being specific on such information as crop types and management practices. Classification of agricultural land use based on activities and production statistics from the national Census of Agriculture (U.S. Department of Agriculture, 1950) or similar data presents the opposite problems. The basic census unit is generally the county or census tract, and the distribution of land uses within the unit is not addressed. However, the activities and production statistics are very detailed on such factors as the total area of each crop harvested within a county, the number of poultry or livestock, irrigation, chemical and fertilizer expenses, and many other features potentially important for water- quality assessment.

General Land-Use Classification

The highest resolution, nationally consistent classification of land use and land cover currently available for the United States is the U.S. Geological Survey Land Use and Land Cover (LULC) data stored in the Geographic Information Retrieval and Analysis System (GIRAS). The national LULC data were compiled primarily from color-infrared aerial photography acquired during the mid-1970's (Anderson and others, 1976). From this photography, polygons of land use and land cover were manually interpreted and delineated on 1:250,000-scale maps, and in a few cases, 1:100,000-scale maps for the entire United States. Land cover classes were based on Anderson's "Land Use and Land Cover Classification System for Use with Remote Sensor Data" (Anderson and others, 1976). The Anderson classification system is a two-level hierarchy (table 1). The minimum mapping unit for the national LULC data is 10 acres for urban uses, as well as for some other categories such as confined feeding operations and strip mines, and 40 acres for all other classes. Land-use maps have been digitized for most of the nation and are available as digital files. An example of LULC data for the Lower Susquehanna NAWQA Study Unit is shown in figure 2.

The U.S. Geological Survey national LULC data have adequate spatial resolution for regional water-quality assessment in most parts of the nation. Its primary disadvantages for water- quality assessment are: (1) lack of specificity on land-use characteristics, such as dominant crops grown and irrigation, (2) it is almost 20 years old, and (3) it is difficult to produce comparable updates using currently available remote-sensing data.

In addition to the national LULC data stored in GIRAS, there is also a much more generalized characterization of Major Land Uses (MLU) of the United States (U.S. Geological Survey, 1970). The MLU map, which is also available in digital form, was interpreted from a variety of information sources, generally representing conditions in the 1960's. Although more general and with much lower resolution compared to the national LULC data, this characterization of major land use patterns is useful for national- scale evaluations.

Agricultural Land-Use Classification

The U.S. Department of Agriculture (USDA) has been classifying agricultural land in the United States since the early 1900's. Studies conducted from 1908 to 1950 were generally referred to as "type of farming studies" (Spillman, 1908; Smith and others, 1916; Baker, 1921; Elliot, 1933; and U.S. Department of Agriculture, 1950). These studies generally progressed from division of the nation into 10 or 12 major agricultural provinces based upon crop or livestock dominance and physiographic conditions (Smith and others, 1916; Baker, 1921), to much more detailed subdivisions based primarily on farm income sources (Elliot, 1933; U.S. Department of Agriculture, 1950). Elliot differentiated 514 major types of farming areas and many additional subareas, providing a very detailed interpretive view of agricultural patterns in 1930. Elliot's work was updated and generalized by U.S. Department of Agriculture (1950), resulting in 165 types of farming areas, 61 subregions, and 9 major regions. These early studies, although now out-dated in many respects, contain thorough descriptions of the environmental and economic factors that affect the distribution and nature of agriculture in different parts of the nation that are still valid today.

After many years of little activity in the area of agricultural land classification, Smith and Hines (1988) and Sommer and Hines (1991) used cluster analysis of county-based data from the 1980 Census of Population and 1987 Census of Agriculture, primarily related to farm income sources, to derive an economics-based division of the nation into agricultural categories representing distinct farm-sector characteristics. Sommer and Hines (1991) identified 12 major clusters (fig. 3) of counties with relatively similar agricultural economic environments with respect to farm enterprise, farm resources, and farm-nonfarm linkages.

With varying degrees of detail and rigor, all early and recent attempts to classify agricultural land have relied primarily on county- based data from the Census of Agriculture for characterizing crop production, income sources, expenses, and other factors. These data have the advantages of being collected regularly (currently every 5 years), and comparable at the county, state, and national levels. Primary disadvantages are the limitation of spatial resolution to county-aggregated statistics and, closely related, the lack of relation to the geographic areas in which the agricultural activities actually take place within counties.

In some respects, the agricultural classifications developed by Elliott (1933) and U.S. Department of Agriculture (1950), represent the type of classification needed for national water-quality assessment. They integrated farm income, crop dominance, and physiographic features in classifying areas and determining boundaries. A significant problem with the approach taken in these studies, however, is that the criteria and rationale for determining classes and boundaries were general and subjective, leading to an analysis that would be impossible to update over time in a comparable fashion. For example, the relative influence of variations in physiographic features and agricultural activities on these classifications is not always clear and there are few quantitative criteria used. In addition, because of the emphasis on farm income, high-income activities have a dominant influence compared to extent of land area in particular crops.

Recent efforts employing cluster analysis (Smith and Hines, 1988; Sommer and Hines, 1991; fig. 3) are a reproducible statistical approach derived directly from Census of Agriculture data. The advantage of this approach compared to historical descriptive approaches is offset by the low resolution of the classification (12 clusters) and the problem that cluster definitions will necessarily change with the changing statistical characteristics of each new census data set. In addition, the analysis is based primarily on income-producing activities to emphasize economic patterns, rather than on the areal extent of agricultural land uses, which may be more important to regional water-quality conditions.

Next - Classification of Agricultural Land by Crop

Back to:
Table of Contents

Pesticide National Synthesis Project