Combining Remote Sensing and Field Data for Deriving Unbiased Estimates of Forest Parameters over Large Regions

Abstract

Remote sensing data can be combined with field data to estimate forest variables over large regions. The accuracy of these estimates depends, for example, on how well the field measurements can be linked to satellite images and on how well forest areas can be identified. In practice, it is difficult to delineate forest areas from other land cover classes; this fact might cause biased estimates. In this study, a post-stratification approach was used to combine field data and satellite data to derive unbiased estimates of forest parameters over large regions. Images from Landsat TM and Terra MODIS were used in combination with field data from the National Forest Inventory in Northern Sweden. The results show that the standard deviation for estimates of total stem volume, stem volume for deciduous trees, and dead wood were reduced with 48%, 33%, and 23%, respectively, by using post-stratification based on Landsat TM data instead of field data alone. A significant improvement of the estimation accuracy was obtained also when using MOMS data.