Probabilistic stratified point sampling to estimate soybean crop area

Abstract

The objective of this work was to evaluate the performance of a probabilistic sampling model stratified by points and to define an appropriate sample size to estimate the cultivated soybean area in the state of Rio Grande do Sul, Brazil. The area was stratified according to the percentage of soybean cultivated in each state municipality: less than 20, from 20 to 40 and more than 40%. Estimates were evaluated based on six sample sizes, resulting from the combination of three significance levels (10, 5 and 1%) and two sampling errors (5 and 2,5%), choosing 400 random samples for each sample size. The estimates were compared to a reference soybean thematic map available for the crop year 2000/2001 that was derived from a careful automatic and visual classification of multitemporal TM/Landsat-5 and ETM+/Landsat-7 images. The soybean area in Rio Grande do Sul State can be estimated through a probabilistic sampling model stratified by points with best estimates obtained for the largest sample size (1,990 points), which differed -0.14% in relation to the estimate of the reference map and had a coefficient of variation of 6.98%.