Multivariate regressions coupling colorimetric and textural features derived from UAV-based RGB images can trace spatiotemporal variations of LAI well in a deciduous forest
Yi Gan, Quan Wang, Takeshi Matsuzawa, Guangman Song, Atsuhiro Iio
Abstract
The leaf area index (LAI) is a critical structural variable related to a variety of biophysical processes of vegetation, and its efficient and accurate estimation is of importance to understanding ecosystem dynamics. Despite many direct and indirect methods being well developed, in the more recent unmanned aerial vehicle (UAV)-based RGB data have emerged to be a promising alternative source for LAI estimation due to their low-cost, considerable flexibility, and useful applicability. In this study, we have investigated the feasibility of using multivariate regressions combined with different colour indices and textural-based features extracted from the UAV-based RGB images for tracing the dynamical LAI across different phenological stages in a typical temperate deciduous forest. The results demonstrated that the visible atmospherically resistant index (VARI) and red–green ratio index (RGR) were the best when using colour indices (CIs) alone while the difference (D) performed best if using textural information only to estimate the spatiotemporal changes of LAI. More importantly, the multivariate regressions based on the combination of CIs and texture-based features resulted in greater accuracies for LAI estimation, in which the random forest regression (R2 = 0.85, RMSE = 0.56, RPD = 2.52) outperformed other analytical methods. The findings of this study provide a cost-effective yet reliable approach to trace LAI from a commonly and easily obtained data source, with the capability to extend to ever-large scales for quick vegetation monitoring and management.