The forest stand margin area in the interpretation of growing stock using Landsat TM imagery

In this study the reliability of the interpretation of growing stock volume from a satellite image in a forest stand margin area was investigated. The accuracy of the estimation result was lower close to the stand edge (R-2 = 0.118) than inside the stands (R-2 = 0.510). Neighbouring areas affect the pixel reflectance values. When the field data of edge sample plots were derived from the closest neighbouring stand, the volume estimate error was the smallest. Three different edge-detection methods were tested. The Canny operator performed better than Haralick's correlation or local adaptive binarization. It was able to produce edge differences between non-edge plots and plots where a sharp edge is closer than 50 m from the plot. If a slight edge is closer than 30 m from the plot, edge operators still produced a similar response to non-edge plots. The accuracy of edge detection algorithms was not sufficient to improve the final interpretation result. On the other hand, if all sample plots in the forest margin area were ignored in the training data, the results were biased. Thus, a field data set for forest inventory based on satellite image interpretation should also include forest margin plots.