Desert Vegetation Classification And Aboveground Biomass Inversion Based On UAV Remote Sensing Image

Using UAV remote sensing technology to monitor vegetation growth and distribution is a research hotspot in the field of remote sensing.Desert vegetation is one of the important components of desert ecosystem.It is of great significance to obtain the growth and distribution of desert vegetation in time and explore the impact of climate on desert vegetation for the protection of desert ecosystem.In this paper,the Gurbantunggut Desert is selected as the study area.According to different temperature and precipitation gradients,the sample plots with less interference from human activities are arranged.UAV observation and field investigation are carried out in August 2018（shrub growing season）and April 2019（herb growing season）.Three machine learning algorithms,k-nearest neighbor（KNN）,support vector machines（SVM）and random forest（RF）,were used to classify desert vegetation;Based on the accurate classification of desert vegetation,combined with the field measured data,the above ground biomass inversion models of Haloxylon ammodendron shrubs and herbs were constructed by using linear regression method and RF;The effects of climate factors on desert vegetation coverage and aboveground biomass were analyzed based on temperature and precipitation indexes.The results show that:（1）Based on the spectral information enhancement of visible light image,the object-oriented RF algorithm can realize the high-precision classification of desert vegetation.The correlation between R-band and G-band decreased from 0.97 to 0.52（p<0.01）after decorrelation stretching.At the same time,the shadow area in the image became brighter,which made the vegetation contour clearer and eliminated the shadow interference,which was more conducive to the classification of desert vegetation.The results show that the overall classification accuracy（OA）of three machine learning algorithms for UAV remote sensing images in August 2018 and April 2019 is RF（OA=89.26%in August 2018 and 88.06%in April 2019）>SVM（OA=84.43%in August 2018 and 83.57%in April 2019）>KNN（OA=80.52%in August 2018 and78.95%in April 2019）.The object-oriented RF has the highest classification accuracy for desert vegetation.（2）Based on the UAV remote sensing image and the field measured data,the aboveground biomass of Haloxylon ammodendron and herbage can be retrieved more accurately.The model H=0.763HI+0.620（R²=0.70,RMSE=0.56 m,MAE=0.45 m）and the model C=1.32CI+0.212（R²=0.77,RMSE=1.57m²,MAE=1.06 m²）were compared with the model AGB=0.3628×CH^0.9605.The aboveground biomass of Haloxylon ammodendron（H and C represent the measured values of plant height and crown area,respectively,and HI and CI represent the predicted values of plant height and crown area,respectively）was retrieved by remote sensing.The fitting analysis of all measured values and all predicted values showed that the model could accurately retrieve the aboveground biomass of Haloxylon ammodendron.Nine visible light vegetation indices were selected to construct three regression models to retrieve the biomass of grass.The results show that:the inversion precision of three regression models are RF（R²=0.89,RMSE=5.50g/m²,MAE=3.06 g/m²）>multiple stepwise regression model（R²=0.63,RMSE=9.05 g/m²,MAE=6.62g/m²）>Ex G univariate linear regression model（R²=0.41,RMSE=11.01 g/m²,MAE=6.70 g/m²）.（3）Temperature and precipitation had significant effects on vegetation coverage and aboveground biomass（p<0.05）.The results of one-way ANOVA showed that:with the increase of temperature gradient,shrub coverage,herb coverage and the sum of shrub and herb coverage showed a downward trend on the whole,with the decrease rates of 74.09%,74.65%and 78.25%respectively;With the increase of temperature gradient,the aboveground biomass of shrub and grass showed a downward trend on the whole,with the decrease rates of 17.32%and 31.95%respectively;With the increase of precipitation gradient,shrub coverage,herb coverage and the sum of shrub and herb coverage showed a significant growth trend,with the increase rates of 1033.41%,337.9%and 614.9%respectively.With the increase of precipitation gradient,the aboveground biomass of shrub and herb showed a significant growth trend,with an increase of 86.63%and 104.20%respectively.