A simple Mahanalobis distance in the kNN method using aerial photograph for predicting stand table in a boreal mixed forest

Nowadays new digital aerial photographs are seen as potential complement of expensive ground plot inventory. The purpose of this study was to assess the reliability of a simple Mahanalobis distance in the kNN method for estimating stand table with new digital aerial photographs. The study area was a natural boreal mixed forest with stands of differing species composition, structure, and density. The 927 georeferenced 400 m 2 -circular ground plots were distributed randomly throughout a 1753-km 2 study area. Stand table was aggregated into three classes: small timber (dbh 9-15 cm), medium timber (dbh 15.1 -25 cm) and large timber (dbh 25.1 cm and more). The near infrared/red ratio was calculated from digital aerial photographs, and was used to obtain eight textural variables (mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment and correlation. These textural variables were used as predictors, and they were based on co-occurrence matrix for capturing the pixel texture resulting from the forest stand structure. Stand table predictions were obtained with RMSE% of 97.8, 67.1, 84.9 for small, medium and large timber of conifer, and with RMSE% of 118.1, 119.6, 173.7 for small, medium and large timber of deciduous. All biases are relatively small varying from 0.7 to 13.1 for conifer and -1.5 to -0.9% for deciduous.