Estimation Of Volume Of Cunninghamia Lanceolata Based On GF-2 Images

Combing forest inventory sample plot data and remotely sensed images has been rarely applied to single tree volume inversion by spatial interpolation methods such as regression and spatial simulation.However,these studies only focus on a larger scale,and there are few studies on single tree volume inversion.With the increase of the resolution of remote sensing image,the inversion of single tree volume is possible.However,various uncertainty factors seriously affect the inversion accuracy,and the position error is the most deadly uncertainty factor in the single-tree volume inversion.In this study,909 Cunninghamia lanceolata measured in 15 sample plots of middle-aged planted in the experimental area of Huangfengqiao Forest Farm in Hunan Province were selected as the research object.Multivariate Stepwise Regression,Partial Least-Squares Regression and BP Neural Network were used to establish the estimation model of Cunninghamia lanceolata volume based on GF-2 images.The position error is simulated through the location of moving tree.Two experimental schemes of fixed-direction and random-direction moving were designed.In the fixed direction movement experiment,the moving direction was determined to be north by random array,and the moving distance is lm,2m,until 10m in sequence.In the random direction moving experiment,Using a random array,each tree can be randomly divided into 8 directions(east,west,south,north,northeast,northwest,southwest,and southeast),and the moving distance is 1m,2m,until 10m in sequence.The inversion and accuracy evaluation of Cunninghamia lanceolata volume with three established models for each movement.Map of inversion accuracy changes with positional deviation were obtained.The main findings were as follows:(1)The BP neural network model of the three models had the best fitting effect and prediction effect,the prediction accuracy was 81.60%,the partial least squares regression model was the second,the prediction accuracy was 80.98%,and the multivariate stepwise regression model has the lowest accuracy of 79.22%.The prediction accuracy of the three models was high,indicating that it is feasible to perform inversion of Chinese fir volume.(2)Significance test of residual difference between the three models showed that there is no significant difference between multiple stepwise regression model and PLS regression mode and there is significant difference between BP neural network model and other two models.It showed that the prediction results of BP neural network model are not only better than those of the other two models in accuracy,but also essentially different from those of the other two models.(3)The experimental results of fixed-direction and random-direction movement showed that the inversion accuracy of the three models is the highest when the position is not moved and the inversion accuracy decreases slowly with the offset distance within 2m.When the offset exceeds the 2m,the inversion accuracy drops sharply,and Since then,with the increase of moving distance,the estimation accuracy is lower.(4)The effect of position error on the inversion accuracy of single tree volume is not simply"The greater the error,the lower the accuracy".When the position deviation is within the range of the crown width,the inversion accuracy decreases slowly with the offset distance.When the offset exceeds the crown width,the inversion accuracy drops sharply,and then subtle fluctuations occur,maintaining a steady state with low accuracy.It was found that there is a spatial autocorrelation between the blades,which can effectively alleviate the effect of the position error on the inversion accuracy in the crown.But when the position error exceeds the crown,the inversion of single tree volume is completely meaningless.