This study verified the existence ofspatial dependence in a Brazilian savanna fragment and incorporated it to the forest inventory by stratifying, compared the accuracy of systematic sampling (SS) with stratified random sampling (SRS) and verified the accuracy of SRS with reduced sampling intensity. A total of 157 sample plots (area of 1000 m²) were allocated and distributed systematically in the area. The circumference at 1.3m from the soil and the total height of all trees were measured in the plots. The volume ofeach plot was obtained by volume equations generatedfor the physiognomy. It wasfound that the dendrometric characteristic of volume is spatially structured. The inventory error obtainedfor the SS was 11.38% and 6,4 7% for SRS. With the sampling intensity decreased by 60%, the error for SRS was 9.93%. These results showed that even with a marked reduction in the sampling intensity, the estimates by stratified random sampling (SRS) is more accurate than the estimates of systematic sampling (SS). Therefore, the stratification based on the spatial dependence in the characteristic of interest is a very useful tool in improving the quality of forest inventory estimators even with a reduction in sampling intensity.
Forestry inventory; spatial statistic; spatial interpolator; kriging
