Dr. Zuyuan Wang

Accurate and comprehensive tree cover mapping plays a vital role in forest management and ecosystem assessment. However, the lack of spectral information and the low quality of available black and white (B&W) imagery makes it difficult to retroactively predict an historical tree cover map using common methods. Here, a network structure called B&WTreeNet, based on semantic segmentation, is proposed to make full use of the limited labeled training remote sensing data to obtain tree cover information from historical datasets. The proposed B&WTreeNet is capable of cross-temporal semantic segmentation, including various data augmentation methods, to address inconsistent tree cover features caused by variations in image quality. The luminance enhancer and other modules in B&WTreeNet can extract the characteristics of tree cover effectively, successfully compensating for the limited spectral information in B&W image datasets. The countrywide historical tree cover map in Switzerland generated for the 1980s using a limited training dataset from 2018 and 2019 agrees well with manual interpretation results.

Foliage, wood (i.e., trunk, branch), ground, and lower objects (i.e., grass, shrubs), are key semantic components of forests that play different roles in the forest ecosystem. For understanding forest ecosystem structure and function, Light Detection And Ranging (LiDAR) point cloud is a valuable form of remote sensing observation. The understanding intensively relies on precisely performing semantic segmentation task to segment semantic components from forest point cloud data. However, the semantic segmentation of massive point cloud data from forest scenes remains a significant challenge. The forest environment is highly heterogeneous and complex due to tree species and terrain conditions. Different climate zones lead to varying canopy characteristics and the diversity of LiDAR platforms delivers inconsistent point cloud properties. Heuristic approaches and conventional machine learning approaches inevitably suffer from poor generalization. Additionally, most deep learning methods lack a dedicated network design to address the characteristics of forests. This paper introduces Sen-net, a point cloud understanding network specifically constructed for semantic component segmentation of forest scene point cloud. Sen-net implements three modules tailored for forest characteristics. First, a spatial context enhancement module (SCEM) is designed for providing both global and dataset-level perspectives to mine geometric information and robust features hidden in heterogeneous forest. Second, a semantic-driven detail enrichment module (SDEM) is incorporated to preserve rich geometric details and semantic information thereby enhancing the learning of complex structures in the forests. Finally, an adaptive guidance flow (AGF) is added to seamlessly fuse the semantic and detailed features. Comprehensive experiments were conducted on both self-built Lin3D dataset and public datasets. Sen-net achieved an OA of 97.6 % and 85.1 % MIoU on the Lin3D dataset, and an OA of 94.5 % and 78.2 % MIoU on the public dataset FOR-instance. Results show that Sen-net outperformed the representative forest scene point cloud semantic segmentation approaches and state-of-the-art deep learning networks, and it has the potential to generalize to point cloud data collected by LiDAR from other platforms. It is concluded that Sen-net is a powerful and robust framework with substantial potential for being widely and deeply explored in forest ecosystem studies.

Forest edges represent the transition zone (ecotone) between the forest interior and the surrounding open land. Due to their great ecological importance, the value of assessing the structure of forest edges has been recognized. In Switzerland, for example, forest edge structure is assessed during field surveys of the Swiss National Forest Inventory (NFI). However, these assessments are time consuming and limited to sample plots. Publicly available countrywide airborne laser scanning (ALS) data, in contrast, offers possibilities to retrieve forest edge structure information over large spatial extents. In this study, we derived five metrics from ALS point clouds, namely the canopy height variability; ratios of the areas of the three edge components, i.e. shrub belt, shelterbelt and forest layer; the sky-view fraction; the shelterbelt slope; and the front density of the forest edge. These metrics describe the three-dimensional edge structure and therefore could enhance existing NFI edge metrics, which focus on two-dimensional structure characteristics. An expert assessment of the ALS edge metrics demonstrated the ability of the defined set of edge metrics to capture ecologically relevant and indicative characteristics of forest edges. Understanding this relationship between edge metrics and ecological functions is a prerequisite if ALS metrics are to be integrated into NFIs. We subsequently used the ALS edge metrics to group 284 forest edge into three classes with respect to their structural complexity using k-means clustering. The results indicated that the structual complexity was low for 173, medium for 46 and high for 65 forest edges, respectively. Applied to countrywide ALS data sets, our approach allows to retrieve area-wide, spatially continuous information on the forest edge conditions, and, if multitemporal ALS data is available, to monitor the development of the forest edges.

Chinese fir (Cunninghamia lanceolata (Lamb.) Hook.) is the most widely distributed conifer species in the subtropical region of southern China. Due to its ecological and economic significance, accurate prediction of growth of Chinese fir plantations is crucial for forest management and industrial timber production. In this study, we employed a physiologically process-based model (3-PG) to simulate the growth dynamic of Chinese fir plantations under various initial planting densities in southern China. Calibration and validation results indicated that the model outputs have strong correlations with the observed data (R² > 0.79, p < 0.01), except foliage biomass and root biomass. Self-thinning will occur earlier at higher initial planting densities, and stand DBH and height will increase dramatically afterwards. Sensitivity analysis further revealed that the FR (soil fertility rating), alphaCx and gammaN1 parameters were highly sensitive in the 3-PG model (p < 0.01), while their sensitivity was influenced by stand age and initial density variations. This study confirmed that the parameter-specific optimized 3-PG model could accurately predict the growth process of Chinese fir plantations under different densities. It will provide a scientific reference for regional-scale management of Chinese fir plantations.

Remote sensing scene classification (RSSC) is a hotspot and play very important role in the field of remote sensing image interpretation in recent years. With the recent development of the convolutional neural networks, a significant breakthrough has been made in the classification of remote sensing scenes. Many objects form complex and diverse scenes through spatial combination and association, which makes it difficult to classify remote sensing image scenes. The problem of insufficient differentiation of feature representations extracted by Convolutional Neural Networks (CNNs) still exists, which is mainly due to the characteristics of similarity for inter-class images and diversity for intra-class images. In this paper, we propose a remote sensing image scene classification method via Multi-Branch Local Attention Network (MBLANet), where Convolutional Local Attention Module (CLAM) is embedded into all down-sampling blocks and residual blocks of ResNet backbone. CLAM contains two submodules, Convolutional Channel Attention Module (CCAM) and Local Spatial Attention Module (LSAM). The two submodules are placed in parallel to obtain both channel and spatial attentions, which helps to emphasize the main target in the complex background and improve the ability of feature representation. Extensive experiments on three benchmark datasets show that our method is better than state-of-the-art methods.

Historical black-and-white (B&W) aerial images have been recognized as an important source of information for assessing vegetation dynamics. However, the use of these images is limited by the lack of multispectral information, as well as by their varying quality. It is therefore important to study and develop methods that are capable of automatic and accurate classification of these B&W images while reducing the need for tedious manual work. The goal of this study was to assess changes over 30 years in woody vegetation cover along alpine treeline ecotones using B&W aerial images from two time points. A convolutional neural networks model was firstly set up based on three structure classes calculated from Airborne Laser Scanning data using the B&W aerial images from 2010. Then, the model was improved by active addition of training samples of those that were wrongly predicted from historical B&W aerial images from 1980. A comparison with visual image interpretation revealed generally high agreement for the class "dense forest" and lower agreement for the class "group of trees". The study illustrates that vegetation changes at the treeline ecotone can be detected in order to assess areawide long-term vegetation dynamics at a fine spatial resolution.

The spectral clustering method has notable advantages in segmentation. But the high computational complexity and time consuming limit its application in large-scale and dense airborne Light Detection and Ranging (LiDAR) point cloud data. We proposed the Nyström-based spectral clustering (NSC) algorithm to decrease the computational burden. This novel NSC method showed accurate and rapid in individual tree segmentation using point cloud data. The K-nearest neighbour-based sampling (KNNS) was proposed for the Nyström approximation of voxels to improve the efficiency. The NSC algorithm showed good performance for 32 plots in China and Europe. The overall matching rate and extraction rate of proposed algorithm reached 69% and 103%. For all trees located by Global Navigation Satellite System (GNSS) calibrated tape-measures, the tree height regression of the matching results showed an value of 0.88 and a relative root mean square error (RMSE) of 5.97%. For all trees located by GNSS calibrated total-station measures, the values were 0.89 and 4.49%. The method also showed good performance in a benchmark dataset with an improvement of 7% for the average matching rate. The results demonstrate that the proposed NSC algorithm provides an accurate individual tree segmentation and parameter estimation using airborne LiDAR point cloud data.

The demand for precise mapping and monitoring of forest resources, such as above ground biomass (AGB), has increased rapidly. National accounting and monitoring of AGB requires regularly updated information based on consistent methods. While remote sensing technologies such as airborne laser scanning (ALS) and digital aerial photogrammetry (DAP) have been shown to deliver the necessary 3D spatial data for AGB mapping, the capacity of repeat acquisition, remotely sensed, vegetation structure data for AGB monitoring has received less attention. Here, we use vegetation height models (VHMs) derived from repeat acquisition DAP data (with ALS terrain correction) to map and monitor woody AGB dynamics across Switzerland over 35 years (1983-2017 inclusive), using a linear least-squares regression approach. We demonstrate a consistent relationship between canopy height derived from DAP and field-based NFI measures of woody AGB across four inventory periods. Over the environmentally heterogeneous area of Switzerland, our models have a comparable predictive performance (R² = 0.54) to previous work predicting AGB based on ALS metrics. Pearson correlation coefficients between measured and predicted changes in woody AGB over time increased with shorter time gaps (< 2 years) between image capture and field-based measurements, ranging between 0.76 and 0.34. A close temporal match between field surveys and remote sensing data acquisition is thus key to reliable mapping and monitoring of AGB dynamics, especially in areas where forest management and natural disturbances trigger relatively fast canopy dynamics. We show that VHMs derived from repeat DAP capture constitute a cost effective and reliable approach to map and monitor changes in woody AGB at a national extent and can provide an important information source for national carbon accounting and monitoring of ecosystem service provisioning.

Due to the strong impact of the composition and structure of the forest edge on forest ecosystems, the value of investigating changes at the forest edge has been recognized for a few decades. Automatic quantification and detection of forest edge changes has advantages over fieldwork with respect to the time and cost required. Based on the structural features extracted from Airborne Laser Scanning (ALS) data, in this study an automatic method using Kernel Density Estimation (KDE) was developed to describe ALS point data related to the forest edge in terms of a finite number of interpretable indicators. The method was applied to identify patterns in data sets at the forest edge between two given dates (2001 and 2014) in European temperate mixed forests. The study area has an extent of 1403 km² and the total length of forest boundary is approx. 5000 km.
The calculated structural features explained the vegetation’s height distribution and the structural variance at the forest edge. The kernel density curves represented the general structural diversity of the study area. Three extracted indicators described the shape of the kernel density curves and reflected the dynamics of the studied forest edge area. Overall, results from this study show that the proposed approach is a promising tool for using remote sensing data to analyze whether or not the vertical structure of the forest edge for the entire study area has changed and thus to assess automatically the progress of forest management at forest edge.

Methods and workflows for creating area-wide products using remote sensing methods have been developed for various reasons. For the sample-based estimates in the Swiss National Forest Inventory (NFI), area-wide data sets are used for the two-phase estimations, which have substantially lower estimation errors than one-phase inventories. Using area-wide data sets, it is possible to skip dense manual interpretation of the stereo-images and thus save resources. Further, many functions of the forest, such as biodiversity and protection against natural hazards, can be described and quantified better with area-wide spatial data than with field plot data.

 

The need for up-to-date and accurate information on forest resources has rapidly increased in recent years. Forest mapping is an important source of information for the assessment of woodland resources and a key issue for any National Forest Inventory (NFI). Nowadays, new perspectives for automated forest mapping are emerging through the latest developments in remote sensing data and techniques. n this chapter, an overview of current remote sensing data and techniques for mapping woodland and forests, the challenges and requirements for optimization and automation, and the need for validating final products are presented. Special attention is paid to land use - a crucial criterion for forest mapping which, in contrast to land cover, cannot be easily derived from remotely sensed data. Three different approaches for extracting woodland areas (i.e., patches of trees and shrubs) are presented, all of which involve a high degree of automation. Two additional approaches, which are based on NFI forest definitions, are presented. These require the subdivision of woodlands into the classes “used for forestry” and “other use” and implement the criteria “height,” “minimum crown coverage,” “minimum area,” “minimum width,” and “land use”. Special attention is paid to connecting patches using distance criteria from national forest definitions. The main points of this chapter are as follows: (1) forest needs an exact definition which may differ depending on the country, (2) mapping woodland can be highly automated and is indispensable prior to mapping forests, and (3) forest mapping is now feasible using remote sensing data and techniques; however, it is less automated due to the implementation of a forest definition.

Assessing forest cover change is a key issue for any national forest inventory. This was tested in two study areas in Switzerland on the basis of stereo airborne digital sensor (ADS) images and advanced digital surface model (DSM) generation techniques based on image point clouds. In the present study, an adaptive multi-scale approach to detect forest cover change with high spatial and temporal resolution was applied to two study areas in Switzerland. The challenge of this approach is to minimize DSM height uncertainties that may affect the accuracy of the forest cover change results. The approach consisted of two steps. In the first step, a 'change index' parameter indicated the overall change status at a coarser scale. The tendency towards change was indicated by derivative analysis of the normalized histograms of the difference between the two canopy height models (DCHMs) in different years. In the second step, detection of forest cover change at a refined scale was based on an automatic threshold and a moving window technique. Promising results were obtained and reveal that real forest cover changes can be distinguished from non-changes with a high degree of accuracy in managed mixed forests. Results had a lower accuracy for forests located on steep alpine terrain. A major benefit of the proposed method is that the magnitude of forest cover change of any specific region can be made available within a short time as often required by forest managers or policy-makers, especially after unexpected natural disturbances.

Forest mapping is an important source of information for assessing woodland resources and a key issue for any National Forest Inventory (NFI). In the present study, a detailed wall-to-wall forest cover map was generated for all of Switzerland, which meets the requirement of the Swiss NFI forest definition. The workflow is highly automated and based on digital surface models from image-based point clouds of airborne digital sensor data. It fully takes into account the four key criteria of minimum tree height, crown coverage, width, and land use. The forest cover map was validated using almost 10,000 terrestrial and stereo-interpreted NFI plots, which verified 97% agreement overall. This validation implies different categories such as five production regions, altitude, tree type, and distance to the forest border. Overall accuracy was lower at forest borders but increased with increasing distance from the forest border. Commission errors remained stable at around 10%, but increased to 17.6% at the upper tree line. Omission errors were low at 1%-10%, but also increased with altitude and mainly occurred at the upper tree line (19.7%). The main reasons for this are the lower image quality and the NFI height definition for forest which apparently excludes shrub forest from the mask. The presented forest mapping approach is superior to existing products due to its national coverage, high level of detail, regular updating, and implementation of the land use criteria.

Clearly delineated forest boundaries are important for sustainable forest management. Interpreting aerial images for forest boundary delineation is a necessary, but labour-intensive process. The results are partly subjective and may include inconsistencies. In this article, we present an automatic approach to delineating forest boundaries in natural colour aerial images from the Swiss National Forest Inventory (NFI). This method is based on JSEG (J-measure-based Segmentation) image segmentation and can be used to obtain forest delineations using the green vegetation index (GVI) feature, Gabor wavelet texture features as well as curvature features from airborne laser scanning (ALS). This approach is compared with the commonly used method of supervised classification, nearest neighbour (NN), which requires local training, and is found to work well in several different landscape types after establishing thresholds. Although we tested the method in a variety of different landscape types, further development and testing are required to cope with complex terrain topography, for example, in regions at the upper tree line in the Alps.

Image segmentation is an important component in object based forest management with increasing high resolution remote sensing data such as ADS40 images. However, evaluation of segmentation algorithms has been largely subjective. To be able to evaluate image segmentation methods quantitatively and objectively, we applied a new error measurement which quantifies the performance of an image segmentation algorithm based on object shape and size as well. Experimental results show that the applied measurement is sensitive to over-segmentation and under-segmentation that can be served as a useful tool for further establishing an evaluation system.

Accurate shadow detection is very important for many remote sensing applications. In this paper, we applied an algorithm which takes advantages of combined information from RGB and near infrared images (NIR) to ADS40 data in Swiss national inventory ( NFI). Preliminary experimental results shows that shadows can be successfully detected by this simple methods.

Tree cover is one of the important forest parameters in SWISS National Forest Inventory (NFI). Therefore, automatic tree cover detection is of practical interest. However, it difficult to remover shadow areas because of some effects related to ADS40 data processing of Digital Surface Model (DSM) by NGATE. This paper proposes an automatic and flexible threshold selection algorithm for shadow removing based on the mean-shift segmentation results. Experimental results show that the shadows can be successfully removed by the proposed algorithm.

Zoige Wetland is one of the largest plateau wetlands in the world. This paper provides a dynamic analysis of spatial and temporal patterns of the wetland in Zoige, Eastern Qinghai-Tibetan Plateau, supported by ERDAS8.7 and ArcGIS9.0. It is the first comparative analysis of a system of rapidly changing wetland with landscape patterns in Zoige, using 3 classified landsat Thematic Mapper images of 1977, 1994 and 2001. The classified images were used to generate wetland distributing maps, and shape index (S), diversity index (H), dominance index (D), evenness index (E), fragmentation index (F) and fractal dimension (F_d ) were calculated and analyzed spatiotemporally across pure grazing area in Zoige for each landscape type and in different periods (before 1977, during 1977~1994 and 1994~2001), as well as the driving forces of natural and anthropogenic.
The study shows that for a comprehensive understanding of the shapes and trajectories of the shrinking and desertificated land expansion of the wetland, a spatiotemporal landscape metrics analysis in different periods is an improvement than only with landscape changing rates. This type of analysis can also be used to infer underlying social, economic, and political processes that drive the observed wetland forms. The results indicate that wetland patterns can be changed over relatively short periods of time. The total area of lake reduced by 164.86 km², grassland extended by 141.74 km², semi-marsh extended by 105.94 km², marsh reduced by 86.00 km²; the number of landscape patches reduced by 56, and their average area decreased by 2.68 km²; the successions within lake, marsh, semi-marsh and grassland were found obviously; S decreased stepwise; D and F increased but H decreased; The changing rate after 1994 was 2.3 to 2.9 times greater than that before. The change of the wetland landscape patterns resulted in the interaction between socio-economic and natural forces of positive and negative aspects; and natural factors affected as assistant aspect. Some important human activities in this period led to the change of the landscape patterns in this region directly. Some measurements made by government and NGO delayed the converting process partly.

Unsupervised segmentation methods are important to extract boundary features from large forest vegetation databases. Finding optimized segmentation algorithms for images with natural vegetation is crucial because of the computational load and the required reproducibility of results. In this paper, we present an approach how to automatically select optimized parameter values for JSEG segmentation. The parameter evaluation is based on a spatial comparison between segmented regions and manually acquired ground truth. City block distance will be used as error metric to define discrepancies between available ground truth and segmentation. Varying the parameter range of values systematically allows to compute corresponding error areas. The smallest error area represents the optimized parameter value.Dependent on the lightness distribution of the selected images and the chosen color quantization, the spatial comparison with the ground truth is limited to local optimization.

In this paper we incorporate different kinds of features, namely GVI, Gabor texture from high resolution aerial images and curvature feature from LIDAR for forest delineation in Swiss national forest inventory. The GVI is helpful for non vegetation areas removing while the curvature feature is useful for buildings removing in urban area. The texture feature constructed by Gabor wavelet successfully distinguishes forest from other landscape types. Experimental results show that the proposed method provides an automatic forest boundary delineation method in a semantic way.