Phenology Examination,Classification And Aboveground Biomass Estimation Of Moso Bamboo Forests Using Time Series Remote Sensing Data

Moso bamboo(phyllostachys edulis)is the most widely distributed bamboo forest type in tropical and subtropical areas.It have have unique characteristics such as on/off-year phenomenon and different phenology periods due to the frequent disturbance by human activities.Plant phenology plays an important role in regulating carbon sequestration of the bamboo forest ecosystem.However,it is a challenge task to identify the on/off-year bamboo distribution and capture the phenological features of Moso bamboo forests in a regional scale due to frequent change of canopy structures and lack of high spatiotemporal remotely sensed data.In this study,we employs the Ven?s time series data to analyze the spectral characteristics of on-year/off-year Moso bamboo forests and other three evergreen vegetation types.Three red-edge based vegetation indices were compared,Red edge position index(REPI)time series curves were reconstructed using the Harmonic Analysis of Time Series(HANTS),and were selected to identify different phenological periods of Moso bamboo forests and other evergreen vegetation types.Then,based on Sentinel-2 and Landsat 8 time-series data,changes in the spectral characteristics during the growth cycle of bamboo forest were analyzed and the optimum time window for the classification of the bamboo forest was determined.The seasonal index of the bamboo forest based on spectral differences was the proposed approach,and on-year and off-year bamboo forest mapping was conducted seasonally using Landsat 8 and Sentinel-2 data.Then,On the consideration of phenology,the aboveground biomass(AGB)estimation function of a single Moso bamboo forest was reconstructed,aboveground biomass of Moso bamboo forests on single and sample scale were calculated.The temporal consistency of the plot and remote sensing data was unified,and the aboveground biomass estimation model of the bamboo forest was constructed.The main conclusions are as follows:(1)The spectral differences between bamboo forest and other forest types are mainly reflected in the re1-edge and near-infrared(NIR)bands(730–920 nm),especially red-edge band.Red-edge based vegetation index can more effectively identify the bamboo forests frequent changes of Moso bamboo forests.The REPI can more effectively identify the two-year growing cycle of the bamboo forests than other vegetation indices,especially the bamboo shoots period and bamboo rhizome growth period.The start of growing season(SOS)of off-year bamboo forest is approximately 50-60 days earlier than on-year bamboo forest.(2)On-year and off-year phenomena were found to be critical features influencing bamboo forest mapping.The best period to distinguish bamboo forest from other forest types is April–May,followed by December–February,and the best month for distinguishing between on-year and off-year bamboo forests is May.The multi-temporal bamboo forest index has a better overall classification accuracy(OA = 91.2)for distinguishing on-year and off-year bamboo forests than the mono-temporal index and can be applied to other regions.Sentinel-2 data have obvious advantages over Landsat 8 data in distinguishing the bamboo forest because of better spectral,temporal and spatial resolutions.It is recommended to use the multi-period remote sensing data to map on-year and off-year bamboo forests,then the May images and then the February images.(3)The aboveground biomass of Bamboo forests plot is in a dynamic cyclic system,with the characteristic like rapid growth,then selective cutting reduction,and stable growth.The AGB estimation function of single bamboo forest was reconstructed,which reduce the time series AGB error of on-year bamboo.Based on random forest,the result show that the AGB estimation based on Sentinel-2 and Ven?s data have better performance than Landsat 8 data,and the spectral variables are more important than texture variables in AGB modeling,the on-year and off-year stratification have better performance than non-stratification.Added time series variables could improve the estimation accuracy,however,the result still show the phenomenon of high value underestimation and low value over estimation.Spectral saturation is the main reason for estimation of aboveground biomass of Moso bamboo using optical remote sensing.