Mapping Invasive Lupinus polyphyllus Lindl. in Semi-natural Grasslands Using Object-Based Image Analysis of UAV-borne Images

Abstract

Knowledge on the spatio-temporal distribution of invasive plant species is vital to maintain biodiversity in grasslands which are threatened by the invasion of such plants and to evaluate the effect of control activities conducted. Manual digitising of aerial images with field verification is the standard method to create maps of the invasive Lupinus polyphyllus Lindl. (Lupine) in semi-natural grasslands of the UNESCO biosphere reserve “Rhön”. As the standard method is labour-intensive, a workflow was developed to map lupine coverage using an unmanned aerial vehicle (UAV)-borne remote sensing (RS) along with object-based image analysis (OBIA). UAV-borne red, green, blue and thermal imaging, as well as photogrammetric canopy height modelling (CHM) were applied. Images were segmented by unsupervised parameter optimisation into image objects representing lupine plants and grass vegetation. Image objects obtained were classified using random forest classification modelling based on objects’ attributes. The classification model was employed to create lupine distribution maps of test areas, and predicted data were compared with manually digitised lupine coverage maps. The classification models yielded a mean prediction accuracy of 89%. The maximum difference in lupine area between classified and digitised lupine maps was 5%. Moreover, the pixel-wise map comparison showed that 88% of all pixels matched between classified and digitised maps. Our results indicated that lupine coverage mapping using UAV-borne RS data and OBIA provides similar results as the standard manual digitising method and, thus, offers a valuable tool to map invasive lupine on grasslands.

Mapping Invasive Plant Species with Hyperspectral Data Based on Iterative Accuracy Assessment Techniques

“…As a result, spatial distribution of the S. altissima with user's and producer's accuracies ranged from 80.0% to 83.3%, with an overall accuracy of 81.7% [9]. Research on the identification of invasive and expansive plant species conducted in recent years was focused mainly on machine learning methods, such as support vector machines (SVMs) [13][14][15] and random forest [16][17][18]. For example, The SVM algorithm and the recursive feature elimination (SVM-RFE) approach were used to map the species Solanum mauritianum in KwaZulu Natal (eastern parts of South Africa) with an overall accuracy (OA) of 93% on selected 17 hyperspectral bands of the AISA Eagle image [19].…”

Remote sensing data fusion as a tool for biomass prediction in extensive grasslands invaded by L. polyphyllus

“…The effect of lupine on model accuracy needs further investigation. The potential of an additional classification step as described by (Wijesingha et al, 2020) before biomass prediction could be helpful to extrapolate the biomass prediction models in time and space.…”

Remote Sensing in Invasive Species Detection and Monitoring

“…Acacia longifolia [11,12] Hakea sericea [10] Heracleum mantegazzianum, Fallopia japonica, Fallopia sachalinensis, Fallopia bohemica [8] Acacia dealbata [13] Iris pseudacorus [14] Lupinus polyphyllus [15] Surveying using Manned Aerial Systems Digital aerial sketch mapping (DASM) Ailanthus altissima [16]…”