A Preliminary Assessment of the Effciency of Using Drones in Land Cover Mapping

Authors: Andrea Francesca Bellia and Sandro Lanfranco

Corresponding: Sandro Lanfranco (sandro.lanfranco@um.edu.mt)

Keywords: drone imagery, land cover mapping, veget- ation mapping, image analysis, k-means clustering

Doi: http://dx.medra.org/10.7423/XJENZA.2019.1.02

Issue: Xjenza Online Vol. 7 Iss. 1 - September 2019

Abstract:
This study represents a preliminary assessment of the efficiency of drones in surveying land cover at both large (c: 10 ha) and smaller (1m2) spatial scales. A DJI Mavic 2 drone was used to image the entire area of study and an orthomosaic was produced. This was converted into a land cover map through k-means clustering, with k = 3, where `Vegetation', `Bedrock' and `Bare soil' corresponded to the land cover categories. Regions of interest (ROIs) were selected and sub- sequently surveyed from close range. The correspondence between predicted land cover (pLC) and observed land cover (oLC) was then assessed. On a large spatial scale, absolute correspondence was present between pLC and oLC. In terms of relative representation of land cover categories, `Vegetation' was the only significantly correlated category across pLC and oLC, whilst the analogous correlations for `Bedrock' and `Bare soil' were weaker. The lower correspondence between pLC and oLC for `Bedrock' and `Bare soil' was due to the low value of k = 3 in the k-means clustering algorithm. This constrains a mixture of land covers into just one land cover category, with consequent reduction of the correlation between pLC and oLC. The method's accuracy and cost-effectiveness were compared to that of standard methods for land cover surveying. The entire process, including verification and orthomosaic land cover map processing times, approximated 32 hours. Consequently, this method is much shorter than standard surveys, which take days or weeks, and also requires less manpower.

Download Article: