| Fully understanding of forestry vegetation is the basic requirement for rational management of forestry.Scientific protection and utilization of forestry vegetation resources is the major foundation for sustainable development both in national economic and healthy of the society.The nation wouldn't achieve the aim of forestry's sustainable development without the sustainable development of the forestry coverage.A comprehensive,system and accurate understanding for the forestry vegetation coverage can accelerate the realization of the goal of forestry sustainable development,also can provide scientific support to the monitoring of the forestry resources.In this study,Linzhi County,Tibet Autonomous Region was selected as the research field.To have a comprehensive understanding of the forestry vegetation coverage types and the growing stock volume in Linzhi County,the research chosen Landsat OLI data to extract the forestry vegetation coverage types of Linzhi.This research also discussed the impact of the size that extracted the texture information of Landsat OLI imagery that can affected the accuracy of the inversion model of the coniferous and broad-leaves forestry in Linzhi.The main conclusions are drawn as follows:(1)Classification of the forestry vegetation coverage types in LinzhiWhile taken the coverage information of the study area and resolution impact of Landsat OLI into consideration,the research firstly set up the classification system of the forestry vegetation coverage in Linzhi County,they were coniferous forest,broad-leaves forest,shrubbery,non-forestry land and water.After the definition of the classification system,3 different classifiers were used to classify the remote sensing imagery and extracted the forestry coverage information,the classifiers contained Maximum likelihood classifier,the mahalanobis distance classifier and artificial neural network,their overall accuracy were 71.52%,67.23%,81.49%,respectively.The results showed the ANNC classifier got the best results in accuracy.When took the single land feature into consideration,the research found no classifier can achieve the best classification results in every single land feature.Under such condition,the research utilize the D-S evidence theory to combine the single classifier to examine whether the combination of classifier will improve the classification accuracy.Based on ENVI+IDL and programming under the support of D-S evidence theory to combine the classifiers,the algorithm fused the different characteristics and complementary the advantages of three classifiers,and can realize the high accuracy extraction of the forestry vegetation coverage in the study area.After combined the 3 different single classifiers according to the D-S evidence theory,the research got 4 different combined classifiers,they were MLC-MaDS,MLC-ANNC,ANNC-MaDC,MLC-MaDC-ANNC,respectively.The classification accuracy got from the 4 combined classifiers were 72.12%,81.66%,80.98%,81.074%,respectively.The average accuracy of MLC,MaDC,ANNC were 78.53%,74.63%,76.81%,respectively;The average accuracy of MLC-MaDC,MLC-ANNC,ANNC-MaDC,MLC-ANNC-MaDC were 77.45%,77.23%,78.23%,78.75%,respectively.The comparison indicated when introudced a new classifier into the classification system,the classifer which got a low classification accuracy would improved and the classification accuracy would raise up,the MLC-ANNC-MaDC got the best average accuracy.Took every combined classifier into observation,it was not hard to find out that after introduced the ANNC classifier into the other two classifier,their classification accuracy have risen more or less,this indicates the ANNC classifier have more significant contribution to the combined classifiers than the other two classifier.which are MLC classifier and MaDC classifier.Took both the overall accuration and the single land feature's accuracy into consideration and statistical comparison,it turned out the combination of 3 different classifiers got the best results of all the different combined classifiers,though the MLC-ANNC classifier has the highest accuracy,but after compared every land feature's classification accuracy,the study have the best result obtained from the classifier which was combined by the 3 single classifiers.The research results indicates,the ANNC classifier can obtain the highest accuracy when compared with the single classifiers,and both MLC and MaDC classifier can improve their classification results after the introduce of ANNC classifier,which means,if the evidence source have a relatively high accuracy,then the combined classifiers would obtained a higher accuracy.The investigated Coniferous forest areas are 2728034 km2,the inverstigated broad-leaves forest area are 464242 km2,the coniferous forest areas and the broad-leaves forest areas extracted by MLC-ANNC-Ma DC classifier were 3319302 km2,418815 km2,respectively.(2)Inversion study of the growing stock volume in Linzhi CountyIn this part,the texture information were extracted with 3 different window sizes,based on the texture information,the improved texture information and the improved vegetable indices were calculated.The indices mentioned above would be developed to describe the relationship between the indices and the growing stock volume of both coniferous and broad-leaves forest.The study results showed,different window sizes could affect the inversion accuracy of the forest growing stock volume.The best growing stock volume inversion model of coniferous forest got the adjusted R2 was 0.553,the standard estimation error was 6.3049,the RMSE of the inversion model was 2.7484.After the statistically analyzed the T-test between the estimated value and field measured data,the comparison result showed,the best growing stock volume inversion model of coniferous forest had a significant p-value was 0.891,this value is greater than 0.05.While the best growing stock volume inversion model of broad-leaves forest got the adjusted R2 was 0.202,the standard estimation error was 2.0023,the RMSE of the inversion model was1.8757.After the statistically analyzed the T-test between the estimated value and field measured data,the comparison result showed,the best growing stock volume inversion model of coniferous forest had a significant p-value was 0.06,this value is little bit greater than 0.05.Both the test results of coniferous and broad-leaves forest growing stock volume inversion model indicated the estimation value had no significant difference with the measured data,the inversion model established with the texture information and the improved texture information was reliable to determine the forest growing stock volume,and it would be of great significance in monitoring and management of forest resources.Overall,the study have three innovations.(1)Based on the D-S evidence theory,three different classifiers were combined to extract the coverage information in Linzhi County,Tibet Autonomous Region,the classification results' overall accuracy were relatively better than single classifiers.(2)Putting forward a parameter C which is a definition of the ability to distinguish different land features for the classifiers,it is a quantitative indicator.(3)Putting forward different improved indices based on texture information,they were Average_Mean,Average_Variance,Average_Contrast,Average_Entropy,Average_Correlation,Average_Homogeneity,Average_Dissimilarity,Average_ASM,AverageBlue,AverageGreen,AverageRed,AverageNIR,Average_texture_NDVI,Average_texture_DVI,Average_texture_SRVI,respectively.Inverse the forest growing stock volume in Linzhi County with these improved indices had improved the ability of traditional forest growing stock volume estimation study which was based on spectral information,it has a significant meaning and feasibility for the forest growing stock volume estimation in Linzhi County. |
PDF Full Text Request