Land degradation and desertification in arid and semi-arid areas is of great concern. Accurately mapping percentage vegetation cover (PVC) of the areas is critical but challenging because the areas are often remote, sparsely vegetated, and rarely populated, and it is difficult to collect field observations of PVC. Traditional methods such as regression modeling cannot provide accurate predictions of PVC in the areas. Nonparametric constant k-nearest neighbors (Cons_kNN) has been widely used in estimation of forest parameters and is a good alternative because of its flexibility. However, using a globally constant k value in Cons_kNN limits its ability of increasing prediction accuracy because the spatial variability of PVC in the areas leads to spatially variable k values. In this study, a novel method that spatially optimizes determining the spatially variable k values of Cons_kNN, denoted with Opt_kNN, was proposed to map the PVC in both Duolun and Kangbao County located in Inner Mongolia and Hebei Province of China, respectively, using Landsat 8 images and sample plot data. The Opt_kNN was compared with Cons_kNN, a linear stepwise regression (LSR), a geographically weighted regression (GWR), and random forests (RF) to improve the mapping for the study areas. The results showed that (1) most of the red and near infrared band relevant vegetation indices derived from the Landsat 8 images had significant contributions to improving the mapping accuracy; (2) compared with LSR, GWR, RF and Cons_kNN, Opt_kNN resulted in consistently higher prediction accuracies of PVC and decreased relative root mean square errors by 5%, 11%, 5%, and 3%, respectively, for Duolun, and 12%, 1%, 23%, and 9%, respectively, for Kangbao. The Opt_kNN also led to spatially variable and locally optimal k values, which made it possible to automatically and locally optimize k values; and (3) the RF that has become very popular in recent years did not perform the predictions better than the Opt_kNN for the both areas. Thus, the proposed method is very promising to improve mapping the PVC in the arid and semi-arid areas.

Accurately mapping urban vegetation carbon density is challenging because of complex landscapes and mixed pixels. In this study, a novel methodology was proposed that combines a linear spectral unmixing analysis (LSUA) with a linear stepwise regression (LSR), a logistic model-based stepwise regression (LMSR) and k-Nearest Neighbors (kNN), to map the forest carbon density of Shenzhen City of China, using Landsat 8 imagery and sample plot data collected in 2014. The independent variables that contributed to statistically significantly improving the fit of a model to data and reducing the sum of squared errors were first selected from a total of 284 spectral variables derived from the image bands. The vegetation fraction from LSUA was then added as an independent variable. The results obtained using cross-validation showed that: (1) Compared to the methods without the vegetation information, adding the vegetation fraction increased the accuracy of mapping carbon density by 1%–9.3%; (2) As the observed values increased, the LSR and kNN residuals showed overestimates and underestimates for the smaller and larger observations, respectively, while LMSR improved the systematical over and underestimations; (3) LSR resulted in illogically negative and unreasonably large estimates, while KNN produced the greatest values of root mean square error (RMSE). The results indicate that combining the spatial modeling method LMSR and the spectral unmixing analysis LUSA, coupled with Landsat imagery, is most promising for increasing the accuracy of urban forest carbon density maps. In addition, this method has considerable potential for accurate, rapid and nondestructive prediction of urban and peri-urban forest carbon stocks with an acceptable level of error and low cost.

Forest stand structure (the characteristics and interrelationships of live trees) and site conditions (the physical and environmental characteristics of a specific location) have been linked to forest regeneration, nutrient cycling, wildlife habitat, and climate regulation. While the effects of stand structure (i.e., spatial and non-spatial) and site conditions on the single function of Cunninghamia lanceolata and Phoebe bournei (CLPB) mixed forest have been studied in previous studies, the relative importance of stand structure and site conditions in terms of productivity, species diversity, and carbon sequestration remains unresolved. In this study, a structural equation model (SEM) was adopted to analyze the relative importance of stand structure and site conditions for the forest productivity, species diversity, and carbon sequestration of CLPB mixed forest in Jindong Forestry in Hunan Province. Our research demonstrates that site conditions have a greater influence on forest functions than stand structure, and that non-spatial structures have a greater overall impact on forest functions than spatial structures. Specifically, the intensity of the influence of site conditions and non-spatial structure on functions is greatest for productivity, followed by carbon sequestration and species diversity. In contrast, the intensity of the influence of spatial structure on functions is greatest for carbon sequestration, followed by species diversity and productivity. These findings provide valuable insights for the management of CLPB mixed forest in Jindong Forestry and have significant reference value for the close-to-natural forest management (CTNFM) of pure Cunninghamia lanceolata forests.

Remote sensing is a powerful tool for precision forestry, providing the forestry industry with spatial information on environment impacts, growth and yield, site variables and damage assessment. Nonetheless the extraction of information from remotely sensed imagery is presently labor intensive requiring highly qualified remote sensing experts, making this information source expensive and slow. With the improvement of spatial resolution, very high resolution remote sensing image are now a competitive alternative to aerial photography and field visits in forest resource survey. In recent years, numerous classification methods were described in the literature and they can be classified into two large classes: traditional pixel-based classification and object-oriented image analysis method. Traditional pixel-based classification techniques either supervised methods or unsupervised method all based on spectral analysis of individual pixels and significant progress has been achieved in recent years. However, these approaches have their limitations since the problem of mixed pixels is indeed reduced, but the internal variability and the noise within land cover classes are increased the improved spatial resolution. In order to improve the classification accuracy, object-oriented image analysis concept has been proposed. This paper explores the use of object oriented image analysis approaches in mapping forest resource and introduces a fast and robust segmentation algorithm--mean shift. The study is based on SPOT-5 image covering the national forest park of Tian'eshan, Zixing City, Hunan, China. Image processing included geometric and atmospheric correction and image segmentation and classification using spectral and spatial information to separate 5 classes. 86.5342% overall accuracy was achieved with this approach. In additional, object oriented image analysis method is compared with traditional pixel based method. The results show the importance, capabilities and challenges of object oriented approaches in providing detailed and accurate information about the physical structure of forest areas.

Image classification is an important technology in the application of remote sensing. Traditional methods of image classification are based on low or medium-resolution images, and the accuracy of classification is always very low. In recent years, high-resolution remote sensing images have significant improvements, but there is still no good method of classification. Studies showed that the accuracy of classified high-resolution images is even lower than that of low or medium -resolution images by traditional classification methods. This turns out that traditional classification technologies appeared to have serious error when using high-resolution images. In this paper, a method of multi-feature classification was introduced to high-resolution remote sensing image, thus avoiding the method of single-feature and pixel-based classification. In this method, pixel-based high-resolution images are changed into object-based images by segmentation. Models of area, perimeter, length, width, symmetry, ratio of length and width, rectangular fit and compactness were established to measure features of segmented objects. More over, the new method of using spectral and texture features to classify high-resolution images was completed. The result showed that the accuracy of image classification can be up to 91.6% by the multi-featured classification, which proved to have improved high-resolution remote sensing image classification.