Eléonore Wolff

Assessment of Very High Spatial Resolution Satellite Image Segmentations

Since 1999, very high spatial resolution satellite data represent the surface of the Earth with more detail. However, information extraction by per pixel multispectral classification techniques proves to be very complex owing to the internal variability increase in land-cover units and to the weakness of spectral resolution. Image segmentation before classification was proposed as an alternative approach, but a large variety of segmentation algorithms were developed during the last 20 years, and a comparison of their implementation on very high spatial resolution images is necessary. In this study, four algorithms from the two main groups of segmentation algorithms (boundarybased and region-based) were evaluated and compared. In order to compare the algorithms, an evaluation of each algorithm was carried out with empirical discrepancy evaluation methods. This evaluation is carried out with a visual segmentation of Ikonos panchromatic images. The results show that the choice of parameters is very important and has a great influence on the segmentation results. The selected boundary-based algorithms are sensitive to the noise or texture. Better results are obtained with regionbased algorithms, but a problem with the transition zones between the contrasted objects can be present.

Exploitation of Very High Resolution Satellite Data for Tree Species Identification

With the emergence of very high spatial resolution satellite images, the spatial resolution gap which existed between satellite images and aerial photographs has decreased. A study of the potential of these images for tree species in “monoculture stands” identification was conducted. Two Ikonos images were acquired, one in June 2000 and the other in October 2000, for an 11- by 11-km area covering the Sonian Forest in the southeastern part of the Brussels-Capital region (Belgium). The two images were orthorectified using a digital elevation model and 1256 geodetic control points. The identification of the tree species was carried out utilizing a supervised maximum-likelihood classification on a pixel-by-pixel basis. Classifications were performed on the orthorectified data, NDVI transformed data, and principal components imagery. In order to decrease the intraclass variance, a mean filter was applied to all the spectral bands and neo-channels used in the classification process. Training and validation areas were selected and digitized using detailed geographical databases of the tree species. The selection of the relevant bands and neo-channels was carried out by successive addition of information in order to improve the classification results. Seven different tree species of one to two different age classes were identified with an overall accuracy of 86 percent. The seven identified tree species or species groups are Oaks (Quercus sp.), Beech (Fagus sylvatica L.), Purple Beech (Fagus sylvatica purpurea), Douglas Fir (Pseudotsuga menziesii (Mirb.) Franco), Scots Pine (Pinus sylvestris L.), Corsican Pine (Pinus nigra Arn. subsp. laricio (Poir.) Maire var. corsican), and Larch (Larix decidua Mill.).

Urban land cover multi-level region-based classification of VHR data by selecting relevant features

The limited spatial resolution of satellite images used to be a problem for the adequate definition of the urban environment. This problem was expected to be solved with the availability of very high spatial resolution satellite images (IKONOS, QuickBird, OrbView‐3). However, these space‐borne sensors are limited to four multi‐spectral bands and may have specific limitations as far as detailed urban area mapping is concerned. It is therefore essential to combine spectral information with other information, such as the features used in visual interpretation (e.g. the degree and kind of texture and the shape) transposed to digital analysis. In this study, a feature selection method is used to show which features are useful for particular land‐cover classes. These features are used to improve the land‐cover classification of very high spatial resolution satellite images of urban areas. The useful features are compared with a visual feature selection. The features are calculated after segmentation into regions that become analysis units and ease the feature calculation.

Monitoring grass nutrients and biomass as indicators of rangeland quality and quantity using random forest modelling and WorldView-2 data

Land use and climate change could have huge impacts on food security and the health of various ecosystems. Leaf nitrogen (N) and above-ground biomass are some of the key factors limiting agricultural production and ecosystem functioning. Leaf N and biomass can be used as indicators of rangeland quality and quantity. Conventional methods for assessing these vegetation parameters at landscape scale level are time consuming and tedious. Remote sensing provides a bird-eye view of the landscape, which creates an opportunity to assess these vegetation parameters over wider rangeland areas. Estimation of leaf N has been successful during peak productivity or high biomass and limited studies estimated leaf N in dry season. The estimation of above-ground biomass has been hindered by the signal saturation problems using conventional vegetation indices. The objective of this study is to monitor leaf N and above-ground biomass as an indicator of rangeland quality and quantity using WorldView-2 satellite images and random forest technique in the north-eastern part of South Africa. Series of field work to collect samples for leaf N and biomass were undertaken in March 2013, April or May 2012 (end of wet season) and July 2012 (dry season). Several conventional and red edge based vegetation indices were computed. Overall results indicate that random forest and vegetation indices explained over 89% of leaf N concentrations for grass and trees, and less than 89% for all the years of assessment. The red edge based vegetation indices were among the important variables for predicting leaf N. For the biomass, random forest model explained over 84% of biomass variation in all years, and visible bands including red edge based vegetation indices were found to be important. The study demonstrated that leaf N could be monitored using high spatial resolution with the red edge band capability, and is important for rangeland assessment and monitoring.

Comparison of very high spatial resolution satellite image segmentations

Since 1999, very high spatial resolution data represent the surface of the earth with more details. However, information extraction by computer-assisted classification techniques proves to be very complex owing to the internal variability increase in land-cover units and to the weakness of spectral resolution. The increase in variability decreases the statistical separability of land-cover classes in the spectral space. Per pixel multispectral classification techniques are then insufficient for an extraction of complex categories and spectrally heterogeneous land-cover, like urban areas. Per region classification was proposed as an alternative approach. The first step of this approach is the segmentation. A large variety of segmentation algorithms were developed these last 20 years and a comparison of their implementation on very high spatial resolution images is necessary. For this study, four algorithms from the two main groups of segmentation algorithms (boundary-based and region-based algorithms) were selected. In order to compare the algorithms, an evaluation of each algorithm was carried out with empirical discrepancy evaluation methods. This evaluation is carried out with a visual segmentation of IKONOS panchromatic images.

Effects of Historical Urbanization in the Brussels Capital Region on Surface Air Temperature Time Series: A Model Study

Abstract The authors examine the local impact of change in impervious surfaces in the Brussels capital region (BCR), Belgium, on trends in maximum, minimum, and mean temperatures between 1960 and 1999. Specifically, data are combined from remote sensing imagery and a land surface model including state-of-the-art urban parameterization—the Town Energy Balance scheme. To (i) isolate effects of urban growth on near-surface temperature independent of atmospheric circulations and (ii) be able to run the model over a very long period without any computational cost restrictions, the land surface model is run in a stand-alone mode coupled to downscaled 40-yr ECMWF reanalysis data. BCR was considered a lumped urban volume and the rate of urbanization was assessed by estimating the percentage of impervious surfaces from Landsat images acquired for various years. Model simulations show that (i) the annual mean urban bias (AMUB) on minimum temperature is rising at a higher rate (almost 3 times more) than on maximum t...

An Open-Source Semi-Automated Processing Chain for Urban Object-Based Classification

This study presents the development of a semi-automated processing chain for urban object-based land-cover and land-use classification. The processing chain is implemented in Python and relies on existing open-source software GRASS GIS and R. The complete tool chain is available in open access and is adaptable to specific user needs. For automation purposes, we developed two GRASS GIS add-ons enabling users (1) to optimize segmentation parameters in an unsupervised manner and (2) to classify remote sensing data using several individual machine learning classifiers or their prediction combinations through voting-schemes. We tested the performance of the processing chain using sub-metric multispectral and height data on two very different urban environments: Ouagadougou, Burkina Faso in sub-Saharan Africa and Liege, Belgium in Western Europe. Using a hierarchical classification scheme, the overall accuracy reached 93% at the first level (5 classes) and about 80% at the second level (11 and 9 classes, respectively).

Seasonal separation of African savanna components using WorldView-2 imagery : a comparison of pixel- and object-based approaches and selected classification algorithms

Separation of savanna land cover components is challenging due to the high heterogeneity of this landscape and spectral similarity of compositionally different vegetation types. In this study, we tested the usability of very high spatial and spectral resolution WorldView-2 (WV-2) imagery to classify land cover components of African savanna in wet and dry season. We compared the performance of Object-Based Image Analysis (OBIA) and pixel-based approach with several algorithms: k-nearest neighbor (k-NN), maximum likelihood (ML), random forests (RF), classification and regression trees (CART) and support vector machines (SVM). Results showed that classifications of WV-2 imagery produce high accuracy results (>77%) regardless of the applied classification approach. However, OBIA had a significantly higher accuracy for almost every classifier with the highest overall accuracy score of 93%. Amongst tested classifiers, SVM and RF provided highest accuracies. Overall classifications of the wet season image provided better results with 93% for RF. However, considering woody leaf-off conditions, the dry season classification also performed well with overall accuracy of 83% (SVM) and high producer accuracy for the tree cover (91%). Our findings demonstrate the potential of imagery like WorldView-2 with OBIA and advanced supervised machine-learning algorithms in seasonal fine-scale land cover classification of African savanna.

Less is more: Optimizing classification performance through feature selection in a very-high-resolution remote sensing object-based urban application

This study evaluates the impact of four feature selection (FS) algorithms in an object-based image analysis framework for very-high-resolution land use-land cover classification. The selected FS algorithms, correlation-based feature selection, mean decrease in accuracy, random forest (RF) based recursive feature elimination, and variable selection using random forest, were tested on the extreme gradient boosting, support vector machine, K-nearest neighbor, RF, and recursive partitioningclassifiers, respectively. The results demonstrate that the selection of an appropriate FS method can be crucial to the performance of a machine learning classifier in terms of accuracy but also parsimony. In this scope, we propose a new metric to perform model selection named classification optimization score (COS) that rewards model simplicity and indirectly penalizes for increased computational time and processing requirements using the number of features for a given classification model as a surrogate. Our findings suggest that applying rigorous FS along with utilizing the COS metric may significantly reduce the processing time and the storage space while at the same time producing higher classification accuracy than using the initial dataset.

Fuzzy Multi-temporal Land-use Analysis and Mine Clearance Application

This paper puts forward a pixel-based and a region-based method for detecting land-use changes using a multitemporal pair of satellite KVR panchromatic and aerial Daedalus multispectral images. These methods, based on fuzzy logic, provide a confidence level of their post-classification change detection results. They were applied to support an early stage of mine clearance in Croatia, where mines laid during the latest conflict (1991 to 1995) cause disruption in the land-use evolution. They were used for locating the agricultural land that was abandoned during the war and is therefore considered as a mine contamination indicator. The mere location of the abandoned agricultural land is of limited use for supporting mine clearance. On the other hand, the integration of the location and spatial confidence level can help the mine experts in their decision making by evaluating both unlikelihood and likelihood of a plot of land to be abandoned.