An Effective Method for Canopy Chlorophyll Content Estimation of Marsh Vegetation Based on Multiscale Remote Sensing Data

High-precision canopy chlorophyll content (CCC) inversion for marsh vegetation is of great significance for marsh protection and restoration. However, it is difficult to collect the CCC measured data for marsh vegetation that matches the pixel scale of remote sensing image. This article proposes a new method based on unmanned aerial vehicle (UAV) multispectral images to obtain multiscale marsh vegetation CCC sample data. A random forest (RF) regression algorithm was used to evaluate the application performance of GF-1 wide field view (WFV), Landsat-8 Operational Land Imager (OLI), and Sentinel-2 multispectral instrument (MSI) satellite remote sensing data in marsh vegetation CCC inversion. In addition, parameter optimization of the RF regression model was used to construct an optimization algorithm suitable for marsh vegetation, and the importance of input variables was quantitatively evaluated. The results showed that the UAV multispectral images assisted in the acquisition of marsh vegetation CCC sample data, as the method expanded the number of CCC samples while quantifying the CCC sample data collection accuracy [
R
²
≥ 0.86, root mean square error (RMSE) ≤ 6.98 SPAD], which improved the CCC inversion accuracy compared with traditional sampling methods. Extracting pure vegetation pixels through binary classification reduces the uncertainty of the UAV-scale CCC inversion results. Parameter optimization of the RF regression model further improves the CCC inversion accuracy at GF-1 WFV, Landsat-8 OLI, and Sentinel-2 MSI scales. Among the three satellite remote sensing data, Sentinel-2 MSI achieved the highest CCC inversion accuracy for marsh vegetation (
R
²
= 0.79, RMSE = 10.96 SPAD) due to the inclusion of red-edge bands that are more sensitive to vegetation properties. Red-edge Chlorophyll Index (Cl
_red-edge
) and Green Chlorophyll Index (Cl
_green
) have the highest influence on the CCC inversion accuracy among input variables.