Forest Types Precise Classification And Forest Resources Change Monitoring Based On Medium And High Spatial Resolution Remote Sensing Images

The study of forest types precise classification and forest resources change monitoring based on medium and high spatial resolution remote sensing images has been recently a key research hotspot with the rapid development of remote sensing technology and the further application in forestry. Although the remote sensing image classification technology has made great progress, some research have shown that there are also some outstanding problems. In forest types information products, the precision is not preferable, and the level of detail should be further improved, and the reliability is low. The process of forest resources dynamic information extracting and change monitoring is a rather difficult process, which has the following problems such as long information acquisition cycle, difficult change information extraction, less practical application of new methods, low degree of automation and poor quality and accuracy of results. In order to meet industry application needs of the national forest resources monitoring survey, this paper focuses on the methods of forest types precise classification and the technology of forest resources change monitoring, which could provide support for abtaining timely and accurate information about forest resources status and dynamic changes. And then it could provide important technical support for reasonable spatial configuration, optimization and adjustment and assistant decision making of forest resources. The main contents of this paper are as follows.(1) This paper presents a new forest types precise classification method based on hierarchical classification strategy using high spatial resolution remote sensing images and multi-source auxiliary data in complex mountainous area, which locates in BaiHua forest farm, Xiaolongshan Forestry Experimental Bureau, GanSu Province. In the test area, SPOT5 HRG2 and two scenes of GF-1 PMS2 images, ancillary information such as forest resources inventory results, forest form map and forest distribution map, and field measurement sample points of the land cover types and fine forest types were applied in classification process. The multivariate characteristics based on high resolution remote sensing images, including image spectral features, vegetation indices, textural features and image time-phase changed features, and topography characteristics, were used as the significant indicators to develop multi-level information extraction methods and forest types fine identification methods, which were particularly suitable for the complex mountainous terrain area of typical natural secondary forest region in warm temperate zone. Independent test samples of the seven forest land or forest types based on stratified random sampling were selected and the confusion matrix and Kappa coefficient are examined to evaluate the accuracy of classification results. In order to further evaluate the validity of classification method proposed and reliability of classification results on the whole, the performance of the classification method was discussed by comparing the statistic area about five main forest types based on classification results with the statistic result of forest resources inventory and image visual interpretation.The results showed that the method proposed in this paper had a good performance in forest types information extraction and overall accuracy of the seven forest land or forest types, including closed forest land, other types of forest land, nursery and fine forest types, reached 92.28% and overall Kappa coefficient 0.8996. Average relative accuracy of area statistic results of the five main fine forest types, including Chinese Pine, Pinus armandi, Japanese larch, Oak species dominant broad-leaved deciduous forest and Other(hardwood species dominant) deciduous broad-leaved mixed forest, reached 92.4%. The results of this study suggest that the hierarchical information extraction method based on multi-source data support is an effective approach of precise classification of forest land types, fine identification of forest types and accurate monitoring of forest resources, especially in mountainous area of complicated topography conditions. The proposed method have advantages in fine identification of forest types with high accuracy and reliability, and the detail degree of fine identification reaches dominant tree species, which could fully meet the needs of forestry applications such as forest resources investigation, forest land change monitoring and thematic map digital update.(2)This paper presents two new forest resources change monitoring technology based on based on post-classification comparison of nonparametric classifiers, that is Random Forest(RF) and Parameter Optimized Support Vector Machine(POSVM) classifier. The forest resources monitoring technology have obtained good application effect in five phase summer and winter Landsat TM/OLI remote sensing images, auxiliary data and field sample points in Tianshui, Gansu Province from 1990 to 2015. According to the comprehensive analysis of the spectral features, index features and time phase features, four indices are used as the characteristic variables in the classification process, such as NDVI, NDWI, NDI and MTVI.The research results showed that the introduction of multiple features and the application of robust and optimal nonparametric classifiers could significantly improve the classification accuracy and the credibility of the classification results. Two kinds of classification methods have achieved good classification results with high spatial consistency and the sequence classification results and the analysis of phase change could accurately and objectively reflect the spatial and temporal dynamic changes of forest resources in the region during the past nearly thirty years. The RF classification method is obviously better than the POSVM classification method, especially in the classification accuracy, efficiency, computation and stability. The RF classification method has fairly good adaptability to complex terrain and landform area and typical vegetation(forest, shrub and grassland) transitional zone, which can be used for vegetation and forest mapping and dynamic change monitoring in a large area, complex terrain and transition region.The forest change monitoring results showed that from 1990 to 1996, the ratio of forest land to non-forest land was 3.764% and the ratio of non-forest land to forest land was 3.024%, which meaned that forest land area decreased by 0.74%. From 1996 to 2002, the ratio of forest land to non-forest land was 5.648% and the ratio of non-forest land to forest land was 2.914%, which meaned that forest land area decreased by 2.734% and the decrease of forest land showed an increasing trend. From 2002 to 2008, the ratio of forest land to non-forest land was 5.574% and the ratio of non-forest land to forest land was 6.631%, which meaned that forest land area increased by 1.057%. From 2008 to 2015, the ratio of forest land to non-forest land was 6.563% and the ratio of non-forest land to forest land was 15.446%, which meaned that forest land area increased by 8.883%. The overall trend of forest resources change in the past 30 years in the region is that the area of forest land decreases at first and then increases, and especially after 2002, the area of forest land increases significantly. The main reason for the increase of forest land area is that the other types are transformed into closed forest. After 1999, with the implementation of the National Key Forestry Programs, such as the Natural Forest Protection Project and the Returning Farmland to Forest Project, the regional forest coverage rate obviously increases and the forest land area significantly enlarges. Obvious reduction of forest land area is mainly distributed in Wushan County and Zhangjiachuan Hui Autonomous County, especially in the transition zone of forest land and other types and the forest edge region after 2002. The possible reasons for the decrease of forest land area are natural factors and the expansion of human activities, so that the original forest and shrub vegetation in the area have been partially damaged and the forest land have changed into cultivated land, grassland and construction land. It is worth noting that the local ecological environment in this region has a further deterioration of risk.(3)On the basis of the above classification results and dynamic changes analysis of land types and forest and non-forest, dynamic conversion of the three typical forest types, including evergreen coniferous forest, broadleaved deciduous forest and other forest types, were further analyzed. In the four time periods, the obvious conversion areas, including the ’increased area’ and the ’decreased area’, the spatial distribution and change process of the three typical forest types were analyzed.The research results showed as follows. From 1990 to 1996, the dynamic conversions(’increased’ and ’decreased’) of evergreen coniferous forest were quite obvious. The ’increased’ area of evergreen coniferous forest was distributed in all forest farms of Xiaolongshan forest region in the south of Qinling Mountains. And the ’decreased’ area of evergreen coniferous forest was distributed in the north of Qinling Mountains, along the strip region of the Weihe River and both sides of the river and the valley in lower altitude area of the upper reaches of the Jialing River. From 1996 to 2002 and from 2002 to 2008, the conversions(’increased’ and ’decreased’) of evergreen coniferous forest were not obvious as a whole. From 2008 to 2015, the ’increased’ of evergreen coniferous forest was quite significant. The ’increased’ area of evergreen coniferous forest was distributed in Xiaolongshan forest region, West Qinling forest region and all forest farms of Guanshan forest region. And the ’decreased’ area of evergreen coniferous forest was only distributed in the central region of Longmen forest farm. From 1990 to 1996, the dynamic conversions(’increased’ and ’decreased’) of broadleaved deciduous forest were quite obvious. The ’increased’ area of broadleaved deciduous forest was distributed in Tailu, Liyuan, Dongcha forest farms and the southern region of Dangchuan, Guanyin and Longmen forest farm. And the ’decreased’ area of broadleaved deciduous forest was distributed in all forest farms of Xiaolongshan forest region in the south of Qinling Mountains, especially in the northern margin and the central region of forest region. From 1996 to 2002 and from 2002 to 2008, the ’increased’ area of broadleaved deciduous forest was mainly distributed in Maiji, Dongcha and Liyuan forest farm of Xiaolongshan forest region. And the ’decreased’ area of broadleaved deciduous forest was mainly distributed in Dongcha, Liyuan and the northern margin region of Xiaolongshan forest region. From 2008 to 2015, the ’increased’ area of broadleaved deciduous forest was mainly distributed in the north of Qinling Mountains and the south of the Weihe River, Maiji, Wenquan and Jianshan forest farm. And the ’decreased’ area of broadleaved deciduous forest was mainly distributed in Maiji and Tange forest farm. From 1990 to 2015, the dynamic conversions(’increased’ and ’decreased’) of other forest types were not obvious and the conversion areas were mainly distributed in the edge of the forest region.The proposed change monitoring technology based on nonparametric classifiers and post-classification comparison methods in this paper have revealed the spatial distribution pattern, time changing trend and influencing factors of forest resources in the typical transition region of the Loess Plateau and the West Qinling Mountains in the past nearly thirty years, aiming to provide certain reference to quantitative analysis and comprehensive evaluation of forest dynamic change, spatial allocation and optimization adjustment of forest resources, management and assistant decision making, progress monitoring of forestry engineering, assessment of the ecological environment and development of the forest protection measures.