| Forest ecosystems are main sources of livelihood for human beings and play a crucial role in the social development.With the continuous development of afforestation,and the implementation of“Grain-for-Green Project”and“Closure of Mountains for Forestry Project”,the area of plantation forests has been increasing in China.However,plantations still face with problems such as low forest productivity,unreasonable structure,simple management measures and low quality.Especially on the Loess Plateau with poor habitat conditions,the health condition,quality and function of plantations face enormous challenges.In order to improve the health condition of plantations on the Loess Plateau and achieve accurate improvement of plantation quality,it is necessary to have a preliminary assessment of the health condition of the plantations.So as to take management measures to improve the health condition of the plantations,thereby improving their ecological and service functions.However,relying on data from field investigation is not only time-consuming and labor-intensive,but also limits in small areas.The remote sensing technology has made it possible to evaluate the forests health condition on a large scale,both low cost and high efficiency.In this study,the black locust(Robinia pseudoacacia)plantations,which is the typical plantation type in the southern part of the Loess Plateau was studied.To find the best regression model between Quick Bird image and field inventory data,the performance of different variable types with different regression models were compared in forest parameters estimations.The estimated forest parameters were then used to evaluate the black locust plantations health condition.Referring to the best combination of Quick Bird image variable type and regression model,the best regression model between Landsat images and field inventory data was found out.The dynamic changes of black locust plantations health condition from 1985 to 2015 in this region were analyzed.And analyze the reason for the changes.This method not only solves the problem that the current forest health assessment only relies on the field inventory data,but also can monitor the forest health in real time and provide a basis for sustainable forest management.In addition,long-term observations of forest health are carried out using multi-temporal remote sensing images.Moreover,reasonable management measures are taken in combination with the changing trend of forest health during the growth process.Thereby cultivating a forest ecosystem with complete functions,good quality,healthy and stable condition.The main results of this study are as follows:(1)Quick Bird images variables and models selection:Select stepwise multiple linear regression(SML),classification and regression tree(CART),support vector machine(SVM),artificial neural network(ANN)and random forest(RF)regression model estimate the forest parameters(stand age,canopy cover(CC),stand density,aboveground biomass(AGB),the average crown diameter(CD),the average diameter at breast height(DBH),the average tree height(H),and leaf area index(LAI)according to the three type of variables extracted from Quick Bird image(spectral variable,textural variables,combination of spectral and textural variables).Adjusting the model parameters of SVM,ANN and RF.Finally,the RF regression model with the combination of spectral and textural variables is selected to obtain the highest accuracy in the estimation of all forest parameters,R~2are range from 0.66 to 0.85,r RMSE are between 0.07 and 0.34.Among them,the DBH obtained the highest accuracy and AGB was the lowest.The optimal parameters of the RF regression model for each forest parameters are as follows:for stand age,the ntree equals300 and mtry equals 6;for AGB ntree equals 350,mtry equals 6;for CC ntree equals 150,mtry equals 6;for CD ntree equals 100 and mtry equals 7;for DBH the ntree equals 300and mtry equals 8;stand density ntree equals 100 and mtry equals 3;for H the ntree equals300 and mtry equals 5;for LAI ntree equals 350 and mtry equals 2.(2)Landsat images variables selection and models adjustment:Referring to the best model and variable used of Quick Bird image,selecting the RF regression model,and the spectral and texture variables of Landsat image.After adjusting the RF model parameters to find the the optimal estimation models for forest parameters.As a result,R~2are between0.60-0.76,and r RMSE are 0.09-0.39.Similar to the results of Quick Bird images,the highest precision is the DBH,and the lowest is AGB.The optimal parameters of the RF regression model for each forest parameters are as follows:for stand age,the ntree equals300 and mtry equals 4;for AGB ntree equals 300,mtry equals 6;for CC ntree equals 400,mtry equals 2;for CD ntree equals 1100 and mtry equals 4;for DBH the ntree equals 150and mtry equals 4;for stand density ntree equals 2000 and mtry equals 6;for H the ntree equals 400 and mtry equals 3;for LAI ntree equals 300 and mtry equals 6.(3)Forest health assessment:Using the optimal model of Quick Bird images to estimate the forest parameters.Combine the forest parameter maps with the topographic factors(aspect,slope and slope position)extracted from the digital elevation model(DEM),using the analytic hierarchy process assess the forest health condition of the study area in2012.The sub-healthy forests accounted for 47.95%of the total,and the healthy accounted for 37.06%.Among the four stand age groups,the proportion of unhealthy forests was lowest and that for healthy forest was the highest in the middle-aged forests.While in the overmature forests,the situation was opposite,with unhealthy forests accounting for the highest proportion,and healthy forests accounting for the lowest proportion.In young forests,the proportion of healthy and sub-healthy forests is roughly equal,which account for 44.16%and 44.97%,respectively.In the middle-aged forest stage,more than 50%of the forests are in healthy condition,and unhealthy forests account for only 6.58%.When growing to the mature forests,the sub-healthy forests occupy the main position,the healthy forests are gradually reduced,and the unhealthy is gradually increasing.When it reached the stage of overmature forests,most of them were still in sub-health condition,healthy forests decreased to 21.02%,and unhealthy forests increased to 23.12%.(4)Dynamic monitoring of forest health condition:Estimate the forest parameters in1985,1995,2005 and 2015 using the optimal regression model based on Landsat images.Combine the forest parameter maps and the topographic factors extracted in DEM,using the analytic hierarchy process assess the health condition in the study area between 1985and 2015.Comparing the forest health conditions in 1985,1995,2005 and 2015,it can be concluded that the proportion of healthy forests was the highest in 2015,followed by 1985,2005 and 1995,while the unhealthy forests do the opposite trend.In the four decades,sub-healthy forests have remained at around 50%.In addition,in 1985,healthy forests accounted for 35.42%of the total,and unhealthy accounted for 15.68%.Unhealthy forests increased significantly in 1995,accounting for 34.40%of the total,and only 17.58%of healthy forests were left.In 2005,healthy forests gradually increased,accounting for28.33%of the total,and unhealthy forests accounted for 21.62%.In 2015,healthy forests increased to 36.09%,and unhealthy decreased to 15.64%.As time goes on,the proportion of healthy forests in the different age groups were decreased first and then increased.While the sub-healthy forests have different trends in different age groups.In young and mid-aged forests,the proportion of sub-healthy forests increased first and then decreased,while the mature and overmature forests showed an opposite trend.The changes of unhealthy forests were consistent in different age groups.The proportion of unhealthy forests rose to the highest in 1995,and then gradually decreased. |
PDF Full Text Request