Construction And Application Research Of MODIS-NDVI Time-series Tupu

The coverage level of earth's surface can be reflected by land cover information,which will be greatly affected by human activities.Human utilization of land resulted in the change of land cover,and furtherly affected the biological,physical and chemical recycling and energy conversion of the earth,as well as the structure and function of earth biogeochemical layers.Recently,with the increasingly serious global environmental problems and development of the research of global change,people have gradually realized the importance and urgency of the study of land cover change.In order to carry out the research of land cover change,we must obtain the land cover data in a timely,efficient and accurate way firstly.In order to obtain precise land cover data,currently the relevant research mainly centers on the land cover classification,classification feature,classification system and other related methods,like the method for reducing noise pollution of remote sensing image.Through these methods,we can conclude that the accuracy of the land cover data can be improved by ameliorating several procedures in the process of the generation of land cover data.The accuracy of the existing land cover products often can not meet the practical demand,especially the land cover products at a large scale.But we cannot deny that the accuracies of these products in some areas are actually pretty high,while in other areas relatively low.That is to say,mybe we can improve the accuracy of the existing land cover products by ameliorating the precision of some areas where the accuracy are relatively low.Concerning this purpose,this paper took Henan Province as an example.The following research work has been carried out and the corresponding results were obtained,using MODIS data.?1?Time-series NDVI curves of each land cover type were obtained,and the Tupu library was created.In this paper,national 1:25 million land cover data was used as reference data,through combing the relevant research and data,38 sample regions were finally selected within the whole country,which contains 17 types of land cover.And the time-series MODIS NDVI were reconstructed using several filtering methods.Then time-series NDVI curves of each land cover type were extracted using NDVI time-series data of the sample area.Finally,the Tupu library of time-series NDVI curves was created basing on this.?2?In order to investigate the feasibility of using NDVI to improve the accuracy of land cover classification,firstly the land cover information in Henan province was extracted based on the Tupu library,then the relationship between NDVI and accuracy of land cover calssification was explored.The land cover information of Henan Province in 2005 and 2013 were extracted using Tupu library of time-series NDVI curves.Then the accuracies of these results were evaluated by comparative evaluation method and sample evaluation method respectively.And the result of comparative evaluation showed that overall accuracy of land cover information extraction results in Henan Province was 71.08% and kappa coefficient reached 0.46 which was evaluated in the first category of land cover,meanwhile the overall accuracy was 48.26%,and kappa coefficient reached 0.33 which was evaluated in the second category of land cover.The result of sample evaluation showed that the overall accuracy of land cover information extraction results in Henan Province reached 79.57%.Therefore,NDVI is an important feature of land cover classification,and to a certain extent,the use of NDVI can ensure the accuracy of land cover classification.So,we can safely conculude that the accuracy of land cover classification can be improved by using NDVI.?3?Improving the accuracy of existing land cover data based on the Tupu library of time-series NDVI curves.New land cover data AMCD₂005 and AMCD₂013 were produced combining MCD12Q1 and the extraction data in Henan Province based on the Tupu library.Then the accuracy of AMCD₂005 and MCD12Q1 were evaluated where national 1:25 million land cover data was used as reference data.The evaluation result showed that the overall accuracy and kappa coefficient increased 11.81% and 0.27 respectively from MCD12Q1 to AMCD₂005.Meanwhile,the accuracy of AMCD₂013 and MCD12Q1 were evaluated where national 1:10 million land use data was used as reference data,and the evaluation result showed that the overall accuracy and kappa coefficient increased 10.29% and 0.25 respectively from MCD12Q1 to AMCD₂013.In addition,the absolute accuracy of AMCD₂013 and MCD12Q1 were obtained using the exploration data which carried out in 2015.And the result showed that the absolute accuracy increased 17.20% from MCD12Q1 to AMCD₂013.That is,the accuracy of the existing land cover data can be improved using the Tupu library of time-series NDVI curves.?4?Discussion about the advantages and disadvantages of time-series NDVI curves similarity matching model.Land cover information in Henan Province were extracted using Minimum Distance method and Spectral Angle Mapper method respectively,and four sets of land cover products were obtained such as SAM₂005,MD₂005,SAM₂013 and MD₂013.By comparing the relative overall accuracy and kappa coefficient of these four sets of products,we found that the overall accuracy and kappa coefficient increased by 1.81% and 0.05 respectively from SAM₂005 to MD₂005,and improved 8.51%,0.13 correspondingly from SAM₂013 to MD₂013.At the same time,compared with SAM₂013 and MD₂013,the results showed that the absolute accuracy of MD₂013 was 12.90% higher than SAM₂013.Thus,in this study,Minimum Distance method was better than Spectral Angle Mapper method.?5?Attempts to find the major land cover categories in which we can improve the accuracy of existing land cover products in Henan Province.By comparing AMCD₂013 which was generated in our research and MCD12Q1 representing the existing land cover products in Henan Province,the following results were found: the user's accuracy of woodland decreased from 88.53% to 83.85%,while producer's accuracy increased by 35.36% from MCD12Q1 to AMCD₂013;the user's accuracy of meadow increased by 48.99%,and producer's accuracy improved 29.51%;the user's accuracy and producer's accuracy of farmland improved 9.6% and 2.98% respectively;the UA and PA of artificial surface improved 39.49% and 6.96% respectively;while the UA and PA of water improved 5.98% and 6.11% correspondingly.It can be seen that the accuracy of each land cover category has been improved,but the accuracy improvement of MCD12Q1 products was mainly attributed to woodland,meadow and artificial surface.Specific to the sub categories of woodland,the UA and PA of evergreen needleleaf forest increased by 21.01% and 39.9% respectively;evergreen broadleaf forest increased by 25.61% and 46.67% respectively;deciduous needleleaf forest improved 1.39% and 2.66% respectively;mixed forest increased by 38.09% and 0.19%;shrubland increased by 54.62% and 0.01%;although the UA of deciduous broadleaf forest was slightly reduced?11.87%?,the PA was significantly improved?57.32%?.On the whole,the accuracy of each sub categories of woodland has been improved,and the evergreen needleleaf forest and evergreen broadleaf forest was more significant.In summary,the author established a Tupu library of time-series NDVI curves,through selecting samples of land cover types and using MODIS time-series data,and the land cover information in Henan province was extracted based on this Tupu library.The results showed that NDVI is an important feature of land cover classification,and the Tupu library can be used to improve the accuracy of the existing land cover data.Therefore,the research ideas,methods and final results from this paper will have a certain guidance and reference value on research of land cover in the future.