| Forest park's land type is complex,and has many tree species.Human interference is serious.Using remote sensing to estimate the main tree species may provide a scientific basis for the formulation of sustainable management of landscape forest.Zijin Mountain Forest Park in Nanjing was selected as the study area,and WorldView-2 data in December 2011 was chosen as the main information sources.Based on the comparison of bands combination,we chose the preferred for the following study.Try using decision tree classifier,neural networks and support vector machine classification method respectively,to distinguish the land use and the main species in the study area.The results showed that:The optimal band combination in the study area was 368.The overall classification accuracy and Kappa coefficients of decision tree classification were 87.10% and 0.85,while the indexes of other methods that based on neural networks and support vector machine classification were lower than it,which were 73.85%,0.70 and76.91%,0.73,respectively.In the results of decision tree classification,classification accuracy of the majority species was high,while the accuracy of foreign pine and cypress was low.Vegetation coverage is an important factor to measure vegetation cover,and is an important parameter of vegetation community and ecosystem.This study was based on the NDVI*of dimidiate pixel model to estimate the vegetation coverage of the study area.The results showed that:The terrain conditions in the middle,southwest and southeast of the Zijin mountain restricted the large-scale reclamation and utilization of the land,and the vegetation coverage was high;In addition to the water,the areas with low vegetation coverage were mostly distributed in low-lying flat areas.In these areas,human activities were more frequent and there were lots of buildings and traffic arteries which had a big effect on the vegetation coverage.Other than that,there was also a small area located on the northern part of the mountain ridge.Because of the steep slope,low density and small tree species,which lead to the shadow dark area.The spectral characteristics of these areas were different from the common vegetation,and the vegetation coverage could not be estimated correctly.Therefore,we did not analyze these areas in detail.Instead,we directly classified it into the low vegetation coverage area.Forest biomass is closely related to the level of forest production and is the basic data which reflects forest ecological system function.In this paper,the data of 90 plots in September 2011 were also used to analyze the forest biomass in the study area.Four remote sensing based models namely multiple linear regression(MLR),k-Nearest Neighbor(KNN),bagging(Bagging)and random forest(RF)were established using 5 vegetation index,7 texture features,2 spectral characteristic factors,2 artificial factors and three terrain variables.Fiveindicators of correlation coefficient(COR),mean absolute error(MAE),root mean squared error(RMSE),relative absolute error(%RAE),root relative squared error(%RMSE)were figured out to evaluate the performance of the four models using ten-fold cross validation method.Then the model with the best performance was applied to analyze the forest biomass of study area.Results showed that: Among the four models,the performance of random forest was the highest,followed by multiple linear regression,while the performance of Bagging was the lowest;Among the 19 independent variables,the spectral characteristic factors(B7),the texture features(MEAN),the terrain factors(elevation)and the difference vegetation index(DVI)were important environmental variables affecting the aboveground biomass of the Forest Park;The spatial hot/cold spots analysis showed that,the cold spots with most gentle change of forest aboveground biomass during research period were mainly distributed in the southern area with low elevation and gentle slope,while the hot spots with most dramatic change were located in the middle and northern area with high elevation,steep slope and less human disturbance. |
PDF Full Text Request