Accurate and timely crop production information is a critical need in today's economy. During the past decade, satellite remote sensing has been increasingly recognized as a means for crop identification and estimation of crop areas. The Landsat multispectral scanner (MSS) records as a single data point (pixel) a region on the ground about one acre (0.5 ha) in size. When estimates of crop areas are desired for large regions, a statistical sampling scheme is required as it is not feasible to examine all of the pixels in the region of interest. The development of a sampling strategy which is both efficient and cost-effective is thus an important objective.
An extensive experiment, the Large Area Crop Inventory Experiment (LACIE), was conducted by NASA, the USDA, and NOAA from 1974 through 1977 (1). Its data analysis objective was to distinguish small grains from nonsmall grains using Landsat MSS data. Several other investigations have shown that the potential exists for identification and area estimation of corn and soybeans as well.
The LACIE area estimation system was based on analysis of sample segments or cluster samples (each 5 x 6 nm in size) extracted from multidate Landsat data. The selection of this sampling scheme was driven to a large degree by the data registration technology which was available at that time. Registration technology research has made considerable progress toward an operational registration capability for Landsat MSS full frames, and so we are no longer restricted to sampling small geographic regions, each of which has been separately registered. This allows us to examine the sampling efficiencies which may be introduced by using a smaller sampling unit size distributed over a larger geographic area.
One such sampling scheme, described by Bauer et al., separates the functions of sampling for training and sampling for classification and area estimation. Training data were developed by photointerpretation of aerial photography taken along north-south flightlines located at intervals across the area of interest. For classification and crop area estimation, a systematic sample of pixels distributed throughout the region was used. The use of different sampling units for training and classification provides both convenience for the data analyst and high precision of the resulting area estimates.
Date of this Version