Landscape-level vegetation classification and fractional woody and herbaceous vegetation cover estimation over the dryland ecosystems by unmanned aerial vehicle platform

The change of fraction vegetation cover (FVC) is the key ecological index for vegetation dynamics of dryland ecosystem. However, it is difficult to directly map woody vegetation and herbaceous vegetation in the dryland from the satellite images due to the mixture of their distribution at small scale. Emerging UAV remote sensing provides a good opportunity to capture and quantify the distribution of the sparse vegetation in the drylands ecosystem. In this study, we proposed a new method to classify woody vegetation and herbaceous vegetation and calculate their FVC based on the high-resolution orthomosaic generated from UAV images by the machine learning algorithm of classification and regression tree (CART). This proposed method was validated and evaluated by visual interpretation, the detailed ground measurement dataset of 4832 trees and 18,798 shrubs and three popular machine learning algorithms of Support Vector Machine(SVM), Random Forest(RF), Gradient Boosting Decision Tree(GBDT). The overall assessments showed good overall accuracy (0.78), average accuracy (0.76), and the Kappa coefficient (0.64). The FVC of woody vegetation calculated from orthomosaic agreed well with that estimated from ground measurements. Both group of FVC have a stable linear relationship over different spatial scales. The proposed method showed higher efficiency of 166%, 111% and 290% than SVM, RF, GBDT respectively. A new optimized model was developed to reduce the workload of vegetation investigation and to design more efficient sampling strategies. The proposed method was incorporated into an interactive web-based software "UAV-High Resolution imagery Analysis Platform" (UAV-HiRAP, http://www.uav-hirap.org). Our study demonstrates that UAV-HiRAP combined with UAV platform can be a powerful tool to classify woody vegetation and herbaceous vegetation and calculate their FVC for sparse vegetation in the drylands. The new optimization model will inspire researchers to design more effective sampling strategies for future field investigation.