Using LiDAR Point Clouds for predicting Wild Bee Richness at Sub-National Scale: A new methodology for landscape quantification with XYZ points only

Organised by Laboratory of Geo-information Science and Remote Sensing

Mon 30 November 2015 15:00 to 15:30

Venue Gaia, gebouwnummer 101
Room 2

By Bastiaen Boekelo (The Netherlands)


Throughout human history, people have been dependent on nature for the pollination of their cultivated crops. Today, the global economic value of this ecosystem service is estimated at 153 billion euro. Wild bees are major contributors to the world’s crop pollination. Not only do they act as a buffer for possible declining honey bee populations, they increase the pollination quality as well, resulting in higher fruit or crop yields. Knowing what environmental conditions are important drivers for species richness is vital for conservation biologists and decision makers. Species Distribution Models (SDMs) can provide this information by correlating species’ presence with their associated environmental conditions. From the predictions of multiple models species richness maps can be created. The performance and the output of the SDMs depend on the (quality of the) variables describing these environmental conditions. It is common to quantify environmental conditions with (remote-sensing based) land cover variables. In this study, vegetation structure has been quantified from the AHN2 point cloud with a voxel-based classification method for the Southern half of the Netherlands. The predictive performance of SDMs from land variables is compared with SDMs based on vegetation structure variables, using observations of 60 different wild bee species. Area Under the ROC Curve (AUC) evaluation values provide indications that vegetation structure based landscape variables are explaining single wild bee distribution better than land use based variables. Furthermore, wild bee richness is predicted more precisely by landscape variables derived from vegetation structure than from land use variables. In general, the province of Zeeland and the ‘Green Heart’ area of The Netherlands are predicted to be species poor, while the Veluwe, Utrechtse Heuvelrug and the East of The Netherlands are predicted to encompass more wild bee species. Results indicate that the certainty of the prediction is related to the spatial distribution of wild bee observation records. Simple methodological implementations, like the use of different SDM algo-rithms and inclusion of topographical, climatic or other variables might improve SDM performance considerably. Nevertheless, without these adjustments it is shown that point cloud data acquired by airborne LiDAR can contribute significantly to the predictive power of the SDMs as well. Further research is needed to refine and validate the vegetation structure classification and to assess the applicability of this vegetation structure for other (invertebrate) species.

Keywords: Wild bee richness; LiDAR; SDM; ENM; Voxel; AHN2; Vegetation Structure