Exploring Bamboo Forest Aboveground Biomass Estimation Using Sentinel-2 Data

Abstract

Bamboo forests, due to rapid growth and short harvest rotation, play an important role in carbon cycling and local economic development. Accurate estimation of bamboo forest aboveground biomass (AGB) has garnered increasing attention during the past two decades. However, remote sensing-based AGB estimation for bamboo forests is challenging due to poor understanding of the mechanisms between bamboo forest growth characteristics and remote sensing data. The objective of this research is to examine the remote sensing characteristics of on-year and off-year bamboo forests at different dates and their AGB estimation performance. This research used multiple Sentinel-2 data to explore AGB estimation of bamboo forests in Zhejiang Province, China, by taking into account the unique characteristics of on-year and off-year bamboo forest growth features. Combining field survey data and Sentinel-2 spectral responses (spectral bands and vegetation indices) and textural images, random forest was used to identify key variables for AGB estimation. The results show that (1) the on-year and off-year bamboo forests have considerably different spectral signatures, especially in the wavelengths between red edge 2 and near-infrared wavelength (NIR2) (740-865 nm), making it possible to separate on-year and off-year bamboo forests; (2) on-year bamboo forests have similar spectral signatures although AGB increases from as small as 40 Mgha −1 to as high as 90 Mgha −1 , implying that optical sensor data cannot effectively model on-year bamboo AGB; (3) off-year bamboo AGB has significant relationships with red and shortwave infrared (SWIR) spectral bands in the April image and with red edge 2 in the July image, but the AGB saturation problem yields poor estimation accuracy; (4) stratification considerably improved off-year bamboo AGB estimation but not on-year, non-stratification using the April image is recommended; and (5) Sentinel-2 data cannot solve the bamboo AGB data saturation problem when AGB is greater than 70 Mgha −1 , similar to other optical sensor data such as Landsat. More research should be conducted in the future to integrate multiple sources-remotely sensed data (e.g., lidar, optical sensor data) and ancillary data (e.g., soil, topography)-into AGB modeling to improve the estimation. The use of very high spatial resolution images that can effectively extract tree density information may improve bamboo AGB estimation and yield new insights.

Stratification-Based Forest Aboveground Biomass Estimation in a Subtropical Region Using Airborne Lidar Data

“…Therefore, for each validation, almost all samples were used as training data, which produced reliable prediction, avoiding the impact from random factors. The accuracy of models was assessed by R 2 , RMSE, and RMSEr [64]. By comparing the performances of those models developed using LR and RF algorithms under four stratification scenarios, the best model for the corresponding scenario was selected and applied to predict the forest AGB of the entire study area.…”

Integration of ZiYuan-3 Multispectral and Stereo Data for Modeling Aboveground Biomass of Larch Plantations in North China

“…In order to solve this problem, an alternative is to divide the sample plots into k folds (here k is the number of sample plots) and use cross-validation for the model evaluation, that is, k-1 folds are used for model calibration and the remaining one fold is used for model validation, and this process is iterated for k times. In this research, we used the leave-one-out cross-validation for calculating correlation coefficient (r), root mean squared error (RMSE), relative RMSE (RMSEr), and mean absolute error (MAE) to assess the models' prediction performance [16]. The higher r value and lower RMSE, RMSEr, and MAE values indicate better modeling performance.…”

“…Previous studies also show that Landsat data are especially valuable for the AGB estimation of successional forests in the Amazon basin, but are not suitable for primary forests because spectral signatures cannot effectively reflect the small difference of forest stand structures, although their AGB may be considerably different [12,15]. Similar conclusions were also obtained in subtropical regions [7,16]. This situation is often called data saturation.…”

“…Although the data saturation problem has long been recognized as an important factor that results in poor forest AGB estimation using optical sensor data [7,12,16], this problem cannot be solved without incorporation of height-relevant variables in the AGB estimation models. With the advantages of lidar in providing tree height features, lidar-based AGB estimation does not have the data saturation problem [25,26]; however, lidar data are not available for most of the study areas, especially in a large area.…”

Estimating the Growing Stem Volume of Chinese Pine and Larch Plantations based on Fused Optical Data Using an Improved Variable Screening Method and Stacking Algorithm

“…Selecting suitable feature variables from optical images is a key step in developing forest GSV estimation models [45]. Besides raw spectral data, many variables such as various band ratios, vegetation indices, and texture measures can be derived by band calculations and transformations [13,32], which offers the possibility of obtaining a large number of feature variables. However, this also leads to difficulty in selecting the variables that can significantly increase the GSV estimation accuracy.…”