Predicting Tree Species Diversity In Mountain Forests Based On Sentinel-2 Remote Sensing Data

Remote sensing of biodiversity is an emerging field in ecology,but the mismatch between scale of remote sensing data and the scale of ecological research limits its application in complex mountain forests where a variety of species exists.Satellites with medium spatial resolution and multispectral resolution,such as Sentinel-2,open new opportunities for predicting plant diversity at large scales.There have been some studies predicting plant diversity based on medium resolution satellites so far,but little consideration has been given to the effect of plant phenological seasonality,species abundance,tree size,and vegetation type on prediction of plant diversity.These factors may have a greater effect on mountain ecosystems with high habitat heterogeneity.This study collected the ground truth data of 233 plots from 11 mountain forests in tropical and subtropical regions of China and extracted Sentinel-2 satellite image time series using Google Earth Engine cloud platform.The relationship between multispectral data and tree species diversity was quantified to reveal the effect of plant phenological seasonality,species abundance and tree size on prediction of tree species diversity in mountain forests.Furthermore,this study evaluated the effect of vegetation type on remote sensing predicting tree species diversity by comparing the relationship between multispectral data and species richness of different vegetation types.The main conclusions are as follows:(1)Phenological seasonality,species abundance,and tree size significantly affect the prediction of tree species diversity in mountain forests:phenological periods corresponding to the optimal prediction of tree diversity in 11 mountain forests are different and most of them appear at the alternation of seasons;the spectral information correlates to species richness better than diversity indices based on abundance and the prediction accuracy of richness is the highest at three tree size classes in most mountain forests;there is a large difference between the tree size classes corresponding to the optimal prediction in each mountain forest.Some forests get better prediction for large trees(DBH≥10 cm),but some get better prediction for all trees(DBH≥1 cm).(2)Significant differences exist in the prediction of tree diversity across different vegetation types:when vegetation type is not accounted for,the prediction accuracy of richness is low(R~2=0.25);however,considering vegetation type significantly improves the prediction accuracy of richness,i.e.,evergreen broadleaf deciduous mixed forest(R~2=0.60),evergreen coniferous forest(R~2=0.50),coniferous and broadleaf mixed forest(R~2=0.36),evergreen broadleaf forest(R~2=0.32),deciduous broadleaf forest(R~2=0.21).These differences are primarily due to different vegetation composition and plant phenological characteristics.The rank of spectral bands based on importance measure varies among vegetation types,with blue,red and short-wave infrared bands playing crucial roles.In summary,this study demonstrates the great potential of Sentinel-2 satellite imagery for predicting species diversity in mountain forests at a regional scale.It reveals that phenological seasonality,species abundance,tree size,and vegetation type substantially influence the prediction of tree species diversity in highly heterogeneous mountainous areas.Accounting for these factors can effectively enhance the prediction of tree diversity using satellite remote sensing and further enable accurate biodiversity monitoring and conservation efforts at large scales.