Crop area estimates are important for several applications. However, as the crop fields are irregularly distributed, such surveys become a difficult and expensive operation. Remote sensing satellite images, for having a broad coverage and being obtained continuously, can be very helpful. In this paper a simple sampling method for quantifying the main crops in a municipality level is described. A non-exhaustive interpretation of soybean, corn and "non-corn or non-soybean" was performed on a TM/Landsat-5 image, and a random sampling of 521 pixels was selected. The areas of soybean and corn were 26.5% and 3.0% of the municipality, respectively. By simulating sets varying from 10 to 500 sampling points each, and assuming a hypergeometric distribution, it was possible to evaluate the variability of the estimation as a function of the sampling size. It was observed that for a sampling size of 250 pixels there was an overestimation of 3% for soybean and an underestimation of 1% for corn. It is also shown that when a crop occupies 35% of the municipality, for a significance level of 90%, and for 500 sampling pixels, the estimation results ranges between 31.4% and 38.4% of the municipality area. The sampling based on the satellite image structure allows a fast and fairly precise survey of the more important cultivated crop area of a municipality.
remote sensing; agricultural survey; soybean; corn; Landsat; statistics
