Julien Radoux

Thematic accuracy assessment of geographic object-based image classification

Geographic object-based image analysis is an image-processing method where groups of spatially adjacent pixels are classified as if they were behaving as a whole unit. This approach raises concerns about the way subsequent validation studies must be conducted. Indeed, classical point-based sampling strategies based on the spatial distribution of sample points (using systematic, probabilistic or stratified probabilistic sampling) do not rely on the same concept of objects and may prove to be less appropriate than the methods explicitly built on the concept of objects used for the classification step. In this study, an original object-based sampling strategy is compared with other approaches used in the literature for the thematic accuracy assessment of object-based classifications. The new sampling scheme and sample analysis are founded on a sound theoretical framework based on few working hypotheses. The performance of the sampling strategies is quantified using simulated object-based classifications results of a Quickbird imagery. The bias and the variance of the overall accuracy estimates were used as indicators of the methods benefits. The main advantage of the object-based predictor of the overall accuracy is its performance: for a given confidence interval, it requires fewer sampling units than the other methods. In many cases, this can help to noticeably reduce the sampling effort. Beyond the efficiency, more conceptual differences between point-based and object-based samplings are discussed. First, geolocation errors do not influence the object-based thematic accuracy as they do for point-based accuracy. These errors need to be addressed independently to provide the geolocation precision. Second, the response design is more complex in object-based accuracy assessment. This is interesting for complex classes but might be an issue in case of large segmentation errors. Finally, there is a larger likelihood to reach the minimum sample size for each class with an object-based sampling than in a point-based sampling. Further work is necessary to reach the same suitability than point-based sampling for pixel-based classification, but this pioneer study shows that object-based sampling could be implemented within a statistically sound framework.

Sentinel-2’s Potential for Sub-Pixel Landscape Feature Detection

Land cover and land use maps derived from satellite remote sensing imagery are critical to support biodiversity and conservation, especially over large areas. With its 10 m to 20 m spatial resolution, Sentinel-2 is a promising sensor for the detection of a variety of landscape features of ecological relevance. However, many components of the ecological network are still smaller than the 10 m pixel, i.e., they are sub-pixel targets that stretch the sensor’s resolution to its limit. This paper proposes a framework to empirically estimate the minimum object size for an accurate detection of a set of structuring landscape foreground/background pairs. The developed method combines a spectral separability analysis and an empirical point spread function estimation for Sentinel-2. The same approach was also applied to Landsat-8 and SPOT-5 (Take 5), which can be considered as similar in terms of spectral definition and spatial resolution, respectively. Results show that Sentinel-2 performs consistently on both aspects. A large number of indices have been tested along with the individual spectral bands and target discrimination was possible in all but one case. Overall, results for Sentinel-2 highlight the critical importance of a good compromise between the spatial and spectral resolution. For instance, the Sentinel-2 roads detection limit was of 3 m and small water bodies are separable with a diameter larger than 11 m. In addition, the analysis of spectral mixtures draws attention to the uneven sensitivity of a variety of spectral indices. The proposed framework could be implemented to assess the fitness for purpose of future sensors within a large range of applications.

Automated Training Sample Extraction for Global Land Cover Mapping

Land cover is one of the essential climate variables of the ESA Climate Change Initiative (CCI). In this context, the Land Cover CCI (LC CCI) project aims at building global land cover maps suitable for climate modeling based on Earth observation by satellite sensors. The challenge is to generate a set of successive maps that are both accurate and consistent over time. To do so, operational methods for the automated classification of optical images are investigated. The proposed approach consists of a locally trained classification using an automated selection of training samples from existing, but outdated land cover information. Combinations of local extraction (based on spatial criteria) and self-cleaning of training samples (based on spectral criteria) are quantitatively assessed. Two large study areas, one in Eurasia and the other in South America, are considered. The proposed morphological cleaning of the training samples leads to higher accuracies than the statistical outlier removal in the spectral domain. An optimal neighborhood has been identified for the local sample extraction. The results are coherent for the two test areas, showing an improvement of the overall accuracy compared with the original reference datasets and a significant reduction of macroscopic errors. More importantly, the proposed method partly controls the reliability of existing land cover maps as sources of training samples for supervised classification.

Quality assessment of segmentation results devoted to object-based classification

Object-based image analysis often uses image segmentation as a preliminary step to enhance classification. Object-based classification therefore relies on the quality of the segmentation output. This study evaluates the relevance of quantitative segmentation quality indices to object-based classification. Image segmentation is expected to improve the thematic accuracy of classification but the counterpart is an increased chance of boundary artefacts. Goodness indices were used to assess the former while discrepancy indices evaluated boundary quality. Inter-class Bhattacharyya distance was used to test the relevance of the goodness indices. The results showed that the use of global goodness indices, which did not require a priori information about the study area, was relevant in the case of object-based classification. In this context, the goodness index based on intra-class standard deviation was more useful than the one based on mean object size. On the other hand, it was shown that object size improved class discrimination but this could deteriorate the boundary quality. The use of complementary discrepancy indices is therefore required in the case of frequent under-segmentation.

Automated image-to-map discrepancy detection using iterative trimming.

Keeping existing vector databases Lip to date is a real challenge for GIS data providers. This study directly compares a a map with a more recent image in order to detect the discrepancies between them. An automatic workflow was designed to process the image based on existing information extracted from the vector database. First, geographic object-based image anal is provided automatically labeled image segments after matching, the vector database to the image. Then, discrepancies were detected using a statistical iterative trimming, where outliers were excluded based on a likelihood threshold. Applied on forest map updating, the proposed workflow, was able to detect about 75 percent of the forest regeneration, and 100 percent of the clear cuts with less than 10 percent of commission errors. This discrepancy detection approach assumes that discrepancy corresponds to small proportion of the map area and is very promising in diverse applications thanks to its flexibility.

Bayesian Data Fusion: Spatial and Temporal Applications

Because the characteristics of remotely sensed data vary greatly with the sensors, spectral and spatial resolutions are practically unique for each sensor. Therefore, there is a real need for a theoretical framework that aims at merging information from two or more different sources. In this paper, a new Bayesian data fusion (BDF) framework is used in order to tackle several classical remote sensing issues. This BDF framework is dedicated to spatial prediction, which draws new avenues for applications in remote sensing. An existing BDF method proposed for the pansharpening of IKONOS image is adapted in the case of SPOT 5 image. The BDF approach is then tested for the enhancement of the spatial resolution of coarse images with high-resolution images. In order to illustrate these methods, SPOT 5 and SPOT VEGETATION images were purchased at two different dates in die province of Ninh Thuan (Vietnam). Finally, prospective considerations are addressed for updating past high-resolution images with recent coarse images.

Multimodal accessibility modeling from coarse transportation networks in Africa

Accessibility is a key driving factor for economic development, social welfare, resources management, and land use planning. In many studies, modeling accessibility relies on proxy variables such as estimated travel time to selected destinations. In developing countries, estimating the travel time is hindered by scarce information about the transportation network, making it necessary to take into account off-network travel coupled with considerations of multimodal options available within the existing network. This research proposes such a hybrid approach that computes the travel time to selected destinations by optimizing together a fully modeled multimodal network and off-network travel. The model was applied in a region around Kisangani located in northeastern Democratic Republic of the Congo. Travel times to Kisangani from the hybrid approach were found to be in close agreement with field-based information (R 2 = 0.98). The developed approach also proved to better support real-world transportation constraints (such as transfer points between travel modes or barriers) than cost-distance-based travel-time modeling. Demonstration results from the hybrid approach highlight the potential for impact assessment of road construction or rehabilitation, development of secondary towns or markets, and for land use planning in general.

Good Practices for Object-Based Accuracy Assessment

Thematic accuracy assessment of a map is a necessary condition for the comparison of research results and the appropriate use of geographic data analysis. Good practices of accuracy assessment already exist, but Geographic Object-Based Image Analysis (GEOBIA) is based on a partition of the spatial area of interest into polygons, which leads to specific issues. In this study, additional guidelines for the validation of object-based maps are provided. These guidelines include recommendations about sampling design, response design and analysis, as well as the evaluation of structural and positional quality. Different types of GEOBIA applications are considered with their specific issues. In particular, accuracy assessment could either focus on the count of spatial entities or on the area of the map that is correctly classified. Two practical examples are given at the end of the manuscript.

Assessment of airborne and spaceborne thermal infrared remote sensing for detecting and characterizing landfills

This work deals with the application of thermal infrared remote sensing for landfills management. It aims to address two research questions: (i) Can we detect all past and present landfills using spaceborne data archive? and (ii) Is airborne remote sensing an efficient tool for landfill sites characterization? The results extracted from the spaceborne data analysis show that the detection of landfills is possible during their activity phase. The importance of the context of acquisition is highlighted by the application of the developed detection methods on distinct geographical contexts. High spatial resolution airborne data acquired at dawn proves to be a relevant information source for the characterization of thermal anomalies present on multiple landfills.

Saigas on the brink: Multidisciplinary analysis of the factors influencing mass mortality events

An opportunistic bacterial infection preceded by weather of unusually high humidity and temperature caused mass death of saigas. In 2015, more than 200,000 saiga antelopes died in 3 weeks in central Kazakhstan. The proximate cause of death is confirmed as hemorrhagic septicemia caused by the bacterium Pasteurella multocida type B, based on multiple strands of evidence. Statistical modeling suggests that there was unusually high relative humidity and temperature in the days leading up to the mortality event; temperature and humidity anomalies were also observed in two previous similar events in the same region. The modeled influence of environmental covariates is consistent with known drivers of hemorrhagic septicemia. Given the saiga population’s vulnerability to mass mortality and the likely exacerbation of climate-related and environmental stressors in the future, management of risks to population viability such as poaching and viral livestock disease is urgently needed, as well as robust ongoing veterinary surveillance. A multidisciplinary approach is needed to research mass mortality events under rapid environmental change.