Christian Berger

Multi-Modal and Multi-Temporal Data Fusion: Outcome of the 2012 GRSS Data Fusion Contest

The 2012 Data Fusion Contest organized by the Data Fusion Technical Committee (DFTC) of the IEEE Geoscience and Remote Sensing Society (GRSS) aimed at investigating the potential use of very high spatial resolution (VHR) multi-modal/multi-temporal image fusion. Three different types of data sets, including spaceborne multi-spectral, spaceborne synthetic aperture radar (SAR), and airborne light detection and ranging (LiDAR) data collected over the downtown San Francisco area were distributed during the Contest. This paper highlights the three awarded research contributions which investigate (i) a new metric to assess urban density (UD) from multi-spectral and LiDAR data, (ii) simulation-based techniques to jointly use SAR and LiDAR data for image interpretation and change detection, and (iii) radiosity methods to improve surface reflectance retrievals of optical data in complex illumination environments. In particular, they demonstrate the usefulness of LiDAR data when fused with optical or SAR data. We believe these interesting investigations will stimulate further research in the related areas.

Robust Extraction of Urban Land Cover Information From HSR Multi-Spectral and LiDAR Data

This paper focuses on the description and demonstration of a simple, but effective object-based image analysis (OBIA) approach to extract urban land cover information from high spatial resolution (HSR) multi-spectral and light detection and ranging (LiDAR) data. Particular emphasis is put on the evaluation of the proposed method with regard to its generalization capabilities across varying situations. For this purpose, the experimental setup of this work includes three urban study areas featuring different physical structures, four sets of HSR optical and LiDAR input data, as well as statistical measures to enable the assessment of classification accuracies and methodological transferability. The results of this study highlight the great potential of the developed approach for accurate, robust and large-area mapping of urban environments. Users and producers accuracies observed for all maps are almost consistently above 80%, in many cases even above 90%. Only few larger class-specific errors occur mainly due to the simple assumptions on which the method is based. The presented feature extraction workflow can therefore be used as a template or starting point in the framework of future urban land cover mapping efforts.

From land cover-graphs to urban structure types

Urban structure types (UST) are an initial interest and basic instrument for monitoring, controlling and modeling tasks of urban planners and decision makers during ongoing urbanization processes. This study focuses on a method to classify UST from land cover (LC) objects, which were derived from high resolution satellite images. The topology of urban LC objects is analyzed by implementing neighborhood LC-graphs. Various graph measures are examined by their potential to distinguish between different UST, using the machine learning classifier random forest. Additionally the influence of different parameter settings of the random forest model, the reduction of training samples, and the graph measure importance is analyzed. An independent test set is classified and validated, achieving an overall accuracy of 87%. It was found that the height of the building with the highest node degree has a strong impact on the classification result.

Stratified aboveground forest biomass estimation by remote sensing data

Abstract Remote sensing-assisted estimates of aboveground forest biomass are essential for modeling carbon budgets. It has been suggested that estimates can be improved by building species- or strata-specific biomass models. However, few studies have attempted a systematic analysis of the benefits of such stratification, especially in combination with other factors such as sensor type, statistical prediction method and sampling design of the reference inventory data. We addressed this topic by analyzing the impact of stratifying forest data into three classes (broadleaved, coniferous and mixed forest). We compare predictive accuracy (a) between the strata (b) to a case without stratification for a set of pre-selected predictors from airborne LiDAR and hyperspectral data obtained in a managed mixed forest site in southwestern Germany. We used 5 commonly applied algorithms for biomass predictions on bootstrapped subsamples of the data to obtain cross validated RMSE and r2 diagnostics. Those values were analyzed in a factorial design by an analysis of variance (ANOVA) to rank the relative importance of each factor. Selected models were used for wall-to-wall mapping of biomass estimates and their associated uncertainty. The results revealed marginal advantages for the strata-specific prediction models over the unstratified ones, which were more obvious on the wall-to-wall mapped area-based predictions. Yet further tests are necessary to establish the generality of these results. Input data type and statistical prediction method are concluded to remain the two most crucial factors for the quality of remote sensing-assisted biomass models.

Removal of Optically Thick Clouds from Multi-Spectral Satellite Images Using Multi-Frequency SAR Data

This study presents a method for the reconstruction of pixels contaminated by optical thick clouds in multi-spectral Landsat images using multi-frequency SAR data. A number of reconstruction techniques have already been proposed in the scientific literature. However, all of the existing techniques have certain limitations. In order to overcome these limitations, we expose the Closest Spectral Fit (CSF) method proposed by Meng et al. to a new, synergistic approach using optical and SAR data. Therefore, the term Closest Feature Vector (CFV) is introduced. The technique facilitates an elegant way to avoid radiometric distortions in the course of image reconstruction. Furthermore the cloud cover removal is independent from underlying land cover types and assumptions on seasonality, etc. The methodology is applied to mono-temporal, multi-frequency SAR data from TerraSAR-X (X-Band), ERS (C-Band) and ALOS Palsar (L-Band). This represents a way of thinking about Radar data not as foreign, but as additional data source in multi-spectral remote sensing. For the assessment of the image restoration performance, an experimental framework is established and a statistical evaluation protocol is designed. The results show the potential of a synergistic usage of multi-spectral and SAR data to overcome the loss of data due to cloud cover.

Graph-Based Mapping of Urban Structure Types From High-Resolution Satellite Image Objects—Case Study of the German Cities Rostock and Erfurt

Ongoing urbanization processes have increased the demand for monitoring, controlling, and modeling services, with urban structure types as an initial interest. While urban land cover (LC) can be derived directly from high-resolution satellite images, urban land use (LU) is achieved through analyzing a combination of structural, functional, spatial, morphological, and topological attributes of the various LC classes. The objective of this letter is to distinguish urban LU classes on the basis of distances between buildings incorporated into a graph-based concept. The method was developed using cadastral data (ALK) for the German city of Rostock, then applied to the LC building objects derived from Quickbird data. Building distribution was examined and distances between buildings were used as an attribute for graph generation. Two graph measures (beta index, clustering coefficient) were analyzed resulting in two groups of LU categories. Transferability to a different urban area was tested without adaptions. Similar building distribution and LC extraction quality were found to be crucial for transferability tests. Distances between buildings are an important property for deriving LU classes, but should be accompanied by additional LC attributes to improve LU separability.

Evaluation of red-edge spectral information for biotope mapping using RapidEye

Mapping of Landscape Protection Areas with regard to user requirements for detailed land cover and biotope classes has been limited by the spatial and temporal resolution of Earth observation data. With the new spatial high resolution RapidEye data providing an additional channel in the red-edge region potentially new possibilities for vegetation mapping should be investigated. The presented work is part of the ENVILAND-2 project, which focuses on the complementary use of RapidEye and TerraSAR-X data to derive land cover and biotope classes as needed by the environmental agencies. The goal is to semi-automatically update the corresponding maps by utilising more Earth observation data and less field work derived information. The red-edge spectral region located between the red and near infrared (NIR) wavelengths, has proven to held valuable information on vegetation type, age and condition. In this study the goal is to evaluate the red-edge spectral information compared to the shorter and longer wavelength of the RapidEye sensor. This is done with regard to the classification capability of different land cover classes. Four RapidEye images were used covering two study sites: 1. Rostocker Heide, Mecklenburg-Vorpommern and 2. Elsteraue, Saxony. The spectral bands were analysed for redundant information by using regression and hypothesis testing. For the rededge band and for every class combination present in the study area different separability measurements like divergence or Bhattacharyya distance were computed. As result there are for every class a separability values. The separability values are provided for all spectral bands. A comparison of the values showed the applicability of the red-edge for the classification. Results have shown that additional red-edge information leads to similar class separability for vegetation classes as using red and NIR spectral information. Some specific classes can be classified with a higher accuracy by additional using the red-edge information.

Expanding an urban structure type mapping approach from a subarea to the entire city of Berlin

Each city exhibits recurring patterns consisting of similar building types, vegetation structures, and open spaces, enabling environmental and socio-economic investigations of the urban fabric. In this study, urban structure types (UST) of the city of Berlin are mapped on the basis of a prior land cover classification utilizing a synergistic approach of knowledge based classification and Random Forests. The results are then compared to the outcomes of a previous analysis regarding a subarea of the utilized high spatial resolution airborne data. Results show that UST classification based on a combination of prototype objects and Random Forests is suitable to generate accurate UST maps for these areas with only minor adaptations. Future analyses will focus on transferring the processes to different German cities and data of several sensors.

An object-based multisensoral approach for the derivation of urban land use structures in the city of Rostock, Germany

The present work is part of the Enviland-2 research project, which investigates the synergism between radar- and optical satellite data for ENVIronment and LAND use applications. The urban work package of Enviland aims at the combined analysis of RapidEye and TerraSAR-X data for the parameterization of different urban land use structures. This study focuses on the development of a transferable, object-based rule set for the derivation of urban land use structures at block level. The data base consists of RapidEye and TerraSAR-X imagery, as well as height information of a LiDAR nDSM (normalized Digital Surface Model) and object boundaries of ATKIS (Official Topographic Cartographic Information System) vector data for a study area in the city of Rostock, Germany. The classification of various land cover units forms the basis of the analysis. Therefore, an object-based land cover classification is implemented that uses feature level fusion to combine the information of all available input data. Besides spectral values also shape and context features are employed to characterize and extract specific land cover objects as indicators for the prevalent land use. The different land use structures are then determined by typical combinations and constellations of the extracted land use indicators and land cover proportions. Accuracy assessment is done by utilizing the available ATKIS information. From this analysis the land use structure classes residential, industrial/commercial, other built-up, allotments, sports facility, forest, grassland, other green spaces, squares/parking areas and water are distinguished with an overall accuracy of 63.2 %.

An Image Transform Based on Temporal Decomposition

Today, very dense synthetic aperture radar (SAR) time series are available through the framework of the European Copernicus Programme. These time series require innovative processing and preprocessing approaches including novel speckle suppression algorithms. Here we propose an image transform for hypertemporal SAR image time stacks. This proposed image transform relies on the temporal patterns only, and therefore fully preserves the spatial resolution. Specifically, we explore the potential of empirical mode decomposition (EMD), a data-driven approach to decompose the temporal signal into components of different frequencies. Based on the assumption that the high-frequency components are corresponding to speckle, these effects can be isolated and removed. We assessed the speckle filtering performance of the transform using hypertemporal Sentinel-1 data acquired over central Germany comprising 53 scenes. We investigated speckle suppression, ratio images, and edge preservation. For the latter, a novel approach was developed. Our findings suggest that EMD features speckle suppression capabilities similar to that of the Quegan filter while preserving the original image resolution.