This paper evaluates a sampling square method to estimate soybean crop area in Rio Grande do Sul State, Brazil. A soybean thematic map obtained from multitemporal Landsat images classification for the crop year of 2000/01 was used as reference data. The State area was divided into cells of 1 x 1km and stratified into three soybean area densities (0-20, 20-40 and 40-67%) at municipality level. A probabilistic technique was used to determine four sample rates representing 0.06, 0.12, 0.24 and 0.48% of the study area, being each one randomly sampled one hundred times. The soybean area for each sample was evaluated based on the reference data map. The one hundred estimates for each sample rate were then compared with the reference data for the entire study area. Best results were obtained for the highest sample rate with low Coefficient of Variation (5.2%), indicating that this method is not only suitable to accurate estimate soybean crop area, at State level, but it is also an appropriate alternative for early forecast or when cloud free satellite images are not available. The best three samples for the highest sample rate were selected to estimate soybean area over images acquired in the following crop year (2001/02). In this case, an increment between 11.4 and 12.5% in relation to 2000/01 was observed, indicating that the IBGE estimate (8.8%) is underestimated.
remote sensing; agricultural statistics; geographic information system
