Mapping Aboveground Biomass of Four Typical Vegetation Types in the Poyang Lake Wetlands Based on Random Forest Modelling and Landsat Images.

Xiaolong Wang,Rongrong Wan,Peng Wang,Xue Dai,Xin Yao

doi:10.3389/fpls.2019.01281

Abstract

Wetland biomass is an important indicator of wetland ecosystem health. In this study, four dominant vegetation communities (Carex cinerascens, Phalaris arundinacea, Artemisia selengensis, and Miscanthus sacchariflorus) in the Poyang Lake wetland from 2010 to 2016 were classified from Landsat images using spectral information divergence (SID). We combined aboveground biomass (AGB) field measurements and remote sensing data to establish a suitable model for estimating wetland AGB in Poyang Lake, which is on the Ramsar Convention’s list of Wetlands of International Importance. The results showed that (1) overall, the classification accuracy for vegetation pixels across 5 years ranged from 59.1% to 73.7% and (2) the inter-annual and spatial variations in the AGB of the four vegetation types were clear. C. cinerascens had an average AGB density value of 1.28 kg m−2 in Poyang Lake from 2010 to 2016; M. sacchariflorus had the highest AGB density with an average value of 1.39 kg m−2; A. selengensis had almost the same level at 1.26 kg m−2; and P. arundinacea had the lowest AGB density at 0.64 kg m−2. This study provides useful experience for estimating carbon sequestration of vegetation in freshwater wetlands.

Highlights

Wetland vegetation, as an important component of wetland ecosystems, plays an important role in maintaining ecosystem structure and function (Nilsson and Keddy, 1988; Pinay et al, 2002)
Researchers have proposed modified vegetation indices (VIs), including soil-adjusted vegetation index (SAVI) (Huete, 1988), modified SAVI (MSAVI) (Qi et al, 1994), and enhanced vegetation index (EVI) (Liu and Huete, 1995), to overcome the weaknesses of normal difference vegetation index (NDVI); namely, that it is affected by the atmosphere, soil composition, and heavy saturation in dense vegetation
On the basis of root mean square error (RMSE), R2, and mean absolute error (MAE), we evaluated the accuracy of the random forests (RFs) model in predicting aboveground biomass (AGB) on the training set and test set

Summary

Introduction

As an important component of wetland ecosystems, plays an important role in maintaining ecosystem structure and function (Nilsson and Keddy, 1988; Pinay et al, 2002). With the help of remote sensing technology, combined with field investigations in typical areas, it is possible to conduct research on long-term, dynamic, and fine spatial-scale observations of the vegetation biomass of typical wetland ecosystems (Shen et al, 2015; Zhu et al, 2015). Compared with the obvious stratification and zoning phenomena of dry land ecosystems, the plant unit of the wetland ecosystem environment usually presents a transient land–water interface, which makes the plant’s spectral characteristics and spatial distribution characteristics highly different, increasing the difficulty of large-scale AGB assessment of freshwater wetland vegetation (Adam and Mutanga, 2009; Zomer et al, 2009). Many vegetation indices have been used widely in biomass estimation, resource surveys, vegetation dynamic monitoring, assessment of landscape structure and function, and global change research in recent years

Methods

Results

Discussion

Conclusion