| Land cover type indicates different ecosystem and community composition.Land cover change will also lead to changes in the ecological environment.Land cover change will lead to changes in the structure and function of the ecosystem,thus affecting the ecological environment.In order to study land cover change better and extract more scientific and reliable land cover information,land cover change research has become the forefront of global change research.High precision land cover products can objectively and truly describe global and regional land surface ecosystems.The research status at home and abroad shows that time-series NDVI method and terrain factors method can effectively improve the accuracy of land cover classification.In this paper,these two methods will be further discussed and the classification accuracy of land cover products will be further improved.In this paper,the global land cover product MCD12Q1 was selected as the product to be improved,the Beijing-Tianjin-Hebei region was selected as the application demonstration area,and the 1:100000 land use vector data was selected as the reference data to improve the MCD12Q1 by the time-series NDVI method and terrain factors method.Through the regression model analysis of the terrain factors of the whole area and the terrain factors of the sub area in the Beijing-Tianjin-Hebei area,the model was used to improve the products to be improved.Time-series NDVI atlas library was built for spectrum recognition,which could extract land cover information,and improve the classification accuracy of products.Both terrain factors method and time-series NDVI method verify that they can effectively improve the accuracy of land cover classification.Combining the two methods,the classification accuracy was improved according to the two schemes,and high-precision land cover products were obtained.The main research work and conclusions of this paper were as follows:(1)The introduction of terrain factors can effectively improve the classification accuracy of land cover products,and the improvement of terrain factors zoning modeling was more significant than the improvement of overall area modeling.The MCD12Q1,a global land cover product,has a spatial resolution of 500 m.Its classification accuracy was low,with an overall accuracy of 56.50% and a kappa coefficient of 0.38.Through the introduction of terrain factors,the regression model was established,and the product MCD12Q1 was to be improved,and the accuracy of the improved results were evaluated.The overall accuracy of the improved product of terrain factors overall area modeling has increased by 6.73%,Kappa coefficient has increased by 0.09,and the overall accuracy of the improved products of zoning modeling has increased by 10.96%,Kappa coefficient has increased by 0.12,indicating that the introduction of terrain factors can improve the classification accuracy of land cover products.From the comparative analysis of accuracy evaluation,it can be seen that the improvement effect of terrain factors zoning modeling was more significant than that of overall area modeling.The overall accuracy of terrain factors zoning modeling improvement product was 4.23% higher than that of terrain factors overall area modeling improvement products,indicating that the accuracy of refining terrain factors zoning and constructing regression model could be improve.(2)The improvement effect of the introduction of terrain factors on the non plain part was greater than that on the plain part.The products to be improved and the improved products of terrain factors were divided into the plain part and the non plain part for accuracy analysis.The accuracy of the improved products of terrain factors overall area modeling was improved by 0.61% in the plain part,and the accuracy of the improved products of zoning modeling was improved by 0.22% in the plain part;The accuracy of the improved product of terrain factors overall area modeling in the non plain part was improved by 3.86%,and the zoning modeling was improved by22.15%.The analysis showed that the introduction of terrain factors was more effective for the improvement of the non plain part than the plain part.(3)The method of constructing atlas database based on time-series NDVI can effectively improve the accuracy of land cover classification,and the method based on time-series NDVI was more effective than the terrain factors modeling method in improving the accuracy of land cover product classification.The overall accuracy of time-series NDVI improved product was 74.97%,and that of MCD12Q1 land cover product was 56.50%.The overall accuracy of time-series NDVI improved product has increased by 18.47% and the kappa coefficient has increased by 0.26 based on the original data MCD12Q1.The comparison of the overall accuracy and Kappa coefficient between the improved products of time-series NDVI and the improved products of terrain factors showed that the overall accuracy and Kappa coefficient of the improved products of time-series NDVI were higher than those of the improved products of terrain factors,which showed that the method of constructing atlas library based on time-series NDVI was more effective and improved than the modeling method of terrain factors in improving the classification accuracy of land cover products.(4)The integration of terrain factors and time-series NDVI can further improve the classification accuracy of land cover products.In the research on the accuracy of land cover classification,the combination of terrain factors and time-series NDVI was used to improve the accuracy of land cover classification for the first time.According to the characteristics of the two methods and the accuracy analysis of the improved results,the combination of terrain factors and time-series NDVI was carried out according to two schemes.In the first scheme,because the improvement effect of terrain factors in the non plain part was greater than that in the plain part,the products to be improved were divided into plain part and the non plain part.The non plain part was improved by the method of terrain factors,and the plain part was improved by the time-series NDVI method.The overall accuracy of the confusion matrix of the improved results for the first scheme was 68.34%,and the Kappa coefficient was 0.52.In the second scheme,the two methods were combined and improved directly.After the time-series NDVI method was used to improve inconsistent area,the improved result was improved by the terrain factors method for the second time.The overall accuracy of the improved results was 82.78% and the Kappa coefficient was 0.74.It was proved that the fusion improvement of terrain factors and time-series NDVI can further improve the classification accuracy of land cover products and obtain the improved high-precision land cover products.To sum up,taking the Beijing-Tianjin-Hebei region as the research object,in this paper,the accuracy of overall area modeling and zoning modeling of terrain factors method were discussed,the improvement of terrain factors method on the plain part and the non plain part,the comparative analysis of terrain factors and time-series NDVI method,and the integration of terrain factors and time-series NDVI.Based on the accuracy analysis of improved products,it was proved that the integration of terrain factors and time-series NDVI can further improve the classification accuracy of land cover products.The research results and method conclusions of this paper not only provide a certain reference for the future research on the accuracy of land cover classification,but also provide guidance and guarantee for the sustainable development of ecological environment and land use. |
PDF Full Text Request