Study On Extraction Of Stand Age Information Of Typical Coniferous Forests In Northeast China And Its Impact On Tree Species Classification

As one of the most important ecosystems on the global land,the forest ecosystem quickly and accurately obtains the vegetation type and its forest age composition,which not only affects the formulation of regional forest and woodland management measures,but also affects the forest organisms of different forest land dominant tree species Quantity,diversity and ecological function.Stand age structure information is one of the most important factors that characterize the health status of the forest.However,due to the almost uniform similarity of the spectral information of different dominant tree species in the forest,and the difference in growth information of different dominant tree species,how to accurately extract from the remote sensing image The age information of different forest trees has become an urgent problem to be solved in current research.In the past 30 years,the rapid development of remote sensing technology has accumulated a large amount of remote sensing data.At the same time,with the continuous improvement of remote sensing quantitative technology,it is possible to extract forest age information of different dominant tree species from higher and better resolution remote sensing band.In this paper,Wangyedian Forest Farm in Chifeng City,Inner Mongolia is taken as the main research area.The main coniferous forest tree species are Pinus tabulaeformis and Larix gmelinii.Through field investigation and sorting and analysis of various data and information,Sentinel-2 satellite is used to combine different phenological periods.The characteristics of the spectral information use 5 types of machine learning algorithms(multiple linear regression model(MLR),feedforward back propagation neural network model(BP),support vector machine regression model(SVR),random forest model(RF)and multiple adaptive Regression spline model(MARS))to realize the inversion study of larch forest age information.By studying the growth characteristics of Pinus tabulaeformis and using Landsat series of long-time NDVI time series data in winter,the inversion of plantation age information of Pinus tabulaeformis was completed.On this basis,the influence of different types of forest age sample distribution on tree species classification was analyzed.The main contents and achievements of this study are as follows:1.The leaf spreading period is the best remote sensing inversion phenology period of Larix gmelinii forest.The age of Larix gmelinii forest is basically negatively correlated with various indicators,of which the correlation with the canopy water content(CWC)is the highest,and the pearson correlation coefficient reaches-0.74(p<0.01).The inversion results of different models show that the random forest model(RF)is the best estimation model of Larix gmelinii stand age,and its average determination coefficient R~2 and average root mean square error(RMSE)are 0.89 and2.91 yr,respectively;the multiple linear regression model(MLR)The forest age estimation results are the worst.The average determination coefficient R~2 and the average root mean square error(RMSE)are only 0.57 and 5.69 yr.The nonlinear model can better explain the relationship between forest age and modeling variables.2.Data standardization can effectively reduce the difference in NDVI calculated by different sensors,and Savitzky-Golay(S-G)filtering can better smooth long-term NDVI time series data.The judgment criterion of the Pinus tabulaeformis felling node is the point where NDVI starts to change from a high value to a low value and presents a downward trend.The decision condition of planting node is the minimum value in time series,and NDVI will gradually increase until NDVI is stable 0.4 Above,the planting node can be defined as the starting node of Pinus tabulaeformis plantation age,where the minimum value of the planting node is between 0.22 and0.33.Through this method,the age distribution of Pinus tabulaeformis Plantation in the whole study area was determined.The final validation result shows that the total classification accuracy is 72.22%(kappa coefficient is 0.62),which can better extract the age of Pinus tabulaeformis plantation.3.The selection of different time node and different algorithms can affect the accuracy of the classification results of tree species.Among them,the random forest(RF)algorithm and April perform best in this study.Including all types of forest age samples can improve the classification accuracy of tree species(total classification accuracy is 69.51%,kappa coefficient is 0.62),which is higher than the results of tree species classification only including near-mature forest and mature forest type forest age samples.2.44%and 3.76%,the kappa coefficient is increased by 0.03 and 0.05respectively,so considering different types of samples before classifying tree species is one of the ways to improve the classification accuracy.