Estimating mangrove leaf area index based on red-edge vegetation indices: A comparison among UAV, WorldView-2 and Sentinel-2 imagery
Xianxian Guo, Mao Wang, Mingming Jia, Wenqing Wang
Abstract
Accurate estimation of mangrove leaf area index (LAI) is fundamental for effective mangrove ecosystem management and protection. Remote sensing technology has showed its powerful potential in accurately retrieving mangrove LAI. The generic estimation model combining vegetation indices (VIs) with physically-based law, simplified as LAI-VIs model, has successfully estimated crop LAI. However, the capacity of estimating mangrove LAI using this model, so far, is unclear. Moreover, some studies have proved that estimation accuracy of terrestrial forests and crops LAI can be ameliorated with VIs based on red-edge band (VIs_RE) because of less affecting by canopy structure. However, little literature explores the ability of VIs_RE, especially, from different multispectral sensors, for estimating mangrove LAI. Therefore, our main purpose is to evaluate the robustness and sensitivity of the LAI-VIs_RE model from Sentinel-2, WorldView-2 (WV-2) and Unmanned Aerial Vehicle (UAV) multispectral imagery for estimating mangrove LAI. The estimation models with input variables of NDVI, NDVI_RE1 (band combination from red-edge and visible band), NDVI_RE2 (band combination from red-edge and near-infrared reflectance) from three types of multispectral imagery are used to calculate mangrove LAI of 99 plots. The results showed that the WV-2 imagery acquires the best estimation accuracy (R 2 = 0.72, RMSE = 0.414), followed by Sentinel-2 imagery (R 2 = 0.68, RMSE = 0.440), and UAV multispectral imagery (R 2 = 0.48, RMSE = 0.570). The analyses display the good results of the LAI-NDVI model and LAI-NDVI_RE1 model from WV-2 and Sentinel-2 imagery with the range of R 2 from 0.57 to 0.72, and the discrepant consequences of LAI-NDVI_RE2 model from UAV imagery with R 2 of 0.15, WV-2 imagery with R 2 of 0.67 and Sentienl-2 imagery with R 2 of 0.65, 0.18 and 0.12. This study proves that the generic estimation model and NDVI_RE1 derived from WV-2 and Sentinel-2 multispectral imagery could be deemed as a basic method and input variables of mapping mangrove LAI.