Improving species-specific plot volume estimates based on airborne laser scanning and image data using alpha shape metrics and balanced field data

Abstract Airborne laser scanning (ALS) data with sparse point densities are increasingly used for forest growing stock estimations. The area-level point distributions derived by ALS are not considered informative on tree species, however, and the required information is typically produced using an additional data source, such as spectral images. We developed new volumetric and structural features (so called alpha shape metrics), hypothesizing that these could produce additional information on the species‐size variation as compared to the previously used ALS and image features. These metrics were tested in the prediction of species-specific, plot-level volume using the Most Similar Neighbor imputation method and a data set consisting of altogether 426 training and 142 validation field plots. The considered forest area was dominated by Scots pine, while Norway spruce and deciduous trees formed the other two species groups to be distinguished. The developed metrics improved the species-specific estimates by 13–30 or 2–4 percentage points compared to features based on ALS data alone or a combination of the ALS and image features, respectively. The metrics had a higher importance when the reference data insufficiently covered the species‐size variation within the area. Although the estimates produced using a combination of the ALS and image data had a superior accuracy compared to those produced by the ALS data alone, the results indicate that species-specific estimates may be further improved by developing computational features based on ALS data.