Comparison of stand volume predictions based on airborne laser scanning data versus aerial stereo images

Frankfurt am Main

2016

S. 1-13

Artikel aus einer ZeitschriftUnselbständiges Werk

200200155

Practical forest management requires information on dendrometrical forest parameters in a high spatial resolution, particularly interesting is the timber volume. Nearest neighbor techniques and the random forest approach were employed in this study to predict timber volume per hectare (total stem volume and stem volume of large beech trees, DBH ≥ 60 cm) at forest stand level. The predictions were based on sample plot data from a regional forest inventory and selected sets of auxiliary variables derived from two different remote sensing data sources – airborne laser scanning (ALS) data and aerial stereo images (ASI) – to quantify and compare prediction precision. Existing studies conclude that ALS data provide more precise height information, but also that acquisition of ALS data is more expensive than of ASI data, which are often already available from other monitoring projects. Currently the cost of ASI data is about a half to a third of ALS data. To make spatial predictions we compared two frequently used methods for imputation: random forest and k-most similar neighbors. For both methods, the prediction precisions (RMSE) were similar. Most promising was the fact that the two different sources of auxiliary variables resulted in predictions of almost the same precision. The similarity between ASI and ALS predictions suggest that ASI may serve as a lower-cost alternative to ALS data for estimating many forest stand-level variables.
Stand volume prediction; spatial prediction; k-nearest neighbour; random forest; regional inventory; airborne laser scanning; aerial stereo images.
Schätzung des Bestandesvorrats; räumliche Vorhersage; „ k-Nächste-Nachbarn; Random Forest; regionale Inventur; luftgestütztes Laserscanning; Stereo-Luftbilder.
In der Forstpraxis werden dendrometrische Informationen in einer hohen räumlichen Auflösung benötigt. Von besonderem Interesse ist der Holzvorrat. In dieser Studie werden Nächste-Nachbarn_Methoden und der Random Forest Ansatz verwendet, um den Vorrat je Hektar (Gesamtvorrat, Vorrat dicker Buchen, BHD> 60 cm) auf Bestandesebene zu schätzen.