Mario Caetano

A method to incorporate uncertainty in the classification of remote sensing images

The aim of this paper is to investigate if the incorporation of the uncertainty associated with the classification of surface elements into the classification of landscape units (LUs) increases the results accuracy. To this end, a hybrid classification method is developed, including uncertainty information in the classification of very high spatial resolution multi-spectral satellite images, to obtain a map of LUs. The developed classification methodology includes the following steps: (1) a pixel-based hard classification with a probabilistic Bayesian classifier; (2) computation of the posterior probabilities and quantification of the classification uncertainty using an uncertainty measure; (3) image segmentation and (4) object classification based on decision rules. The classification of the resulting objects into LUs is performed considering a set of decision rules that incorporate the pixel-based classification uncertainty. The proposed methodology was tested on the classification of an IKONOS satellite image. The accuracy of the classification was computed using an error matrix. The comparison between the results obtained with the proposed approach and those obtained without considering the classification uncertainty revealed a 12% increase in the overall accuracy. This shows that the information about uncertainty can be valuable when making decisions and can actually increase the accuracy of the classification results.

Combining per-pixel and object-based classifications for mapping land cover over large areas

A plethora of national and regional applications need land-cover information covering large areas. Manual classification based on visual interpretation and digital per-pixel classification are the two most commonly applied methods for land-cover mapping over large areas using remote-sensing images, but both present several drawbacks. This paper tests a method with moderate spatial resolution images for deriving a product with a predefined minimum mapping unit (MMU) unconstrained by spatial resolution. The approach consists of a traditional supervised per-pixel classification followed by a post-classification processing that includes image segmentation and semantic map generalization. The approach was tested with AWiFS data collected over a region in Portugal to map 15 land-cover classes with 10 ha MMU. The map presents a thematic accuracy of 72.6 ± 3.7% at the 95% confidence level, which is approximately 10% higher than the per-pixel classification accuracy. The results show that segmentation of moderate-spatial resolution images and semantic map generalization can be used in an operational context to automatically produce land-cover maps with a predefined MMU over large areas.

Multitemporal MERIS images for land-cover mapping at a national scale: a case study of Portugal

Medium-spatial-resolution satellite images have already proved to be successful in automatic production of global land-cover maps. However, their operational use for land-cover mapping at a national scale has not yet been well established. We find that the reasons for this are not data-source dependent, but are due to the land-cover nomenclatures properties adopted, regional landscape specificities and the methodological approaches used. The aim of this paper is to evaluate the suitability for national applications of land-cover maps derived from automatic classification of medium-spatial-resolution satellite images. To tackle this issue, we produce a land-cover map of Continental Portugal from multitemporal MEdium Resolution Imaging Spectrometer (MERIS) full-resolution satellite images of 2005 and evaluate its accuracy. For the accuracy assessment of the final map, we compute unbiased estimates of overall, user and producer accuracies using an independent testing sample collected through a stratified random sampling design. The overall accuracy of the final map is 80%, with an absolute precision of 2% at the 95% confidence level. High independent accuracy assessment results demonstrate that medium-spatial-resolution satellite images can be used on an operational basis for annual production of land-cover maps suitable for national applications.

Trapped between antiquity and urbanism – a multi-criteria assessment model of the greater Cairo Metropolitan area

You can tell whether a man is clever by his answers. You can tell whether a man is wise by his questions. Naguib Mahfouz This article attempts to provide systematic policy information regarding land use/land cover change in the vicinity of the Giza Pyramids in Egypt. As a result of the rapid urban growth Cairo has experienced in the past couple of decades, a surrounding enclave of urban development seems to be forming around the Pyramids and the highly valued historical legacy of the area, designated in 1979 as a World Heritage site. Hence, assessing land use changes and future urban sprawl prediction is of major importance for strategic planning and avoiding further endangerment. The data used in this study are derived from remote sensing imagery, taken by Landsat MSS satellite on 31 August 1972 and on 20 September 1984 by Landsat TM as well as Landsat ETM + imagery from 11 November 2000. The use of the different bands of that imagery allowed the classification of the land cover classes: urban, vegetation, desert and water. A temporal comparison of the different types of land cover indicates which land use changes that have occurred over the years considered are associated with the potential for endangering the Giza complex in the study area.

Incorporating reference classification uncertainty into the analysis of land cover accuracy

To accommodate the difficulty of identifying a single ‘true’ or ‘reference’ class, the reference data protocol of an accuracy assessment may include identifying both a primary and alternate reference land cover label along with a rating of the interpreters confidence in the reference classification obtained for each sample location. This additional reference information is used to construct a nominal variable (called CONF) in which the categories represent the ‘confidence’ in the correctness of the map land cover classification at a given location. An accuracy measure that incorporates uncertainty in the reference classification is then derived by assigning partial credit weights to each CONF class. Further, the accuracy reporting format can be organized by CONF classes to provide additional understanding of the relationship between accuracy and uncertainty in the reference classification. The analysis is illustrated using an accuracy assessment of a land cover map of Portugal. These analyses incorporating uncertainty in the reference classification are intended to supplement traditional analyses to further enhance understanding of the accuracy of land cover maps.

A reference sample database for the accuracy assessment of medium spatial resolution land cover products in Portugal

This paper introduces a reference sample database that is being developed by the Remote Sensing Unit of the Portuguese Geographic Institute for the accuracy assessment of medium scale land cover products in Portugal. The goal is to provide the worldwide remote sensing community with sufficient data for the accurate estimation of overall and per class proportions of correctly classified area in regional and global land cover products at this part of the globe. This is a massive database that encloses various descriptive attributes for each sample observation, namely primary and alternate reference land cover labels, nominally scored interpretation ratings and location confidence ratings. We present in detail the attached land cover nomenclature and the database design, i.e. the sampling design used to collect sample observations, as well as the process used to identify the most pertinent reference land cover label for each observation. In addition, we briefly describe some statistics about attribute information that was recorded for each observation and that can be used as auxiliary information for the accuracy assessment of land cover products.

Specific Land Cover Class Mapping by Semi-Supervised Weighted Support Vector Machines

In many remote sensing projects on land cover mapping, the interest is often in a sub-set of classes presented in the study area. Conventional multi-class classification may lead to a considerable training effort and to the underestimation of the classes of interest. On the other hand, one-class classifiers require much less training, but may overestimate the real extension of the class of interest. This paper illustrates the combined use of cost-sensitive and semi-supervised learning to overcome these difficulties. This method utilises a manually-collected set of pixels of the class of interest and a random sample of pixels, keeping the training effort low. Each data point is then weighted according to its distance to its near positive data point to inform the learning algorithm. The proposed approach was compared with a conventional multi-class classifier, a one-class classifier, and a semi-supervised classifier in the discrimination of high-mangrove in Saloum estuary, Senegal, from Landsat imagery. The derived classification accuracies were high: 93.90% for the multi-class supervised classifier, 90.75% for the semi-supervised classifier, 88.75% for the one-class classifier, and 93.75% for the proposed method. The results show that accuracy achieved with the proposed method is statistically non-inferior to that achieved with standard binary classification, requiring however much less training effort.

Information Extraction for Forest Fires Management

Forest fires are a recurrent environmental and economic emergency worldwide. In Southern Europe the year 2003 is considered one of the worst ever with a total burnt area of 740,000 hectares [22]. Even if in some countries according to the national burnt area statistics the problem seems to be under control and of steady or decreasing magnitude, in regions like the western Iberian Peninsula or large parts of North America the trend is worryingly opposite [23, 29]. The causes of such catastrophic events depend on a number of factors, ranging from meteorology to societal changes to vegetation type and emergency response readiness and efficiency.

Fire risk mapping using satellite imagery and ancillary data: towards operationality

Forest fires affect significant areas of the Portuguese forest annually, depending upon seasonal moisture and temperature conditions. Although a large percentage of those fires are not originated by natural causes, there is still a need to develop an effective and timely warning system for fire-prevention. The combination of Earth observation (EO) information with ancillary data of natural parameters for daily monitoring of fire risk, which is allowed by Geographic Information Systems (GIS), offers an appropriate response to that need. The PREMFIRE project was funded by the European Spatial Agency (ESA) for the selection and implementation of the most adequate method for production of fire risk maps for Portugal. The project aims at building a wireless system, enabling its use by fire prevention services in the field with real-time or near real-time exchange of data. We present a methodology to produce fire risk maps using satellite imagery and ancillary data. The approach combines a detailed land cover map and other spatial data sets with vegetation greenness maps and meteorological information to produce a fire potential index map. The vegetation greenness is characterized using NDVI 10-day composites derived from NOOA AVHRR imagery. This methodology is being tested in central Portugal yielding encouraging results.

Improving specific class mapping from remotely sensed data by cost-sensitive learning

ABSTRACT In many remote-sensing projects, one is usually interested in a small number of land-cover classes present in a study area and not in all the land-cover classes that make-up the landscape. Previous studies in supervised classification of satellite images have tackled specific class mapping problem by isolating the classes of interest and combining all other classes into one large class, usually called others, and by developing a binary classifier to discriminate the class of interest from the others. Here, this approach is called focused approach. The strength of the focused approach is to decompose the original multi-class supervised classification problem into a binary classification problem, focusing the process on the discrimination of the class of interest. Previous studies have shown that this method is able to discriminate more accurately the classes of interest when compared with the standard multi-class supervised approach. However, it may be susceptible to data imbalance problems present in the training data set, since the classes of interest are often a small part of the training set. A result the classification may be biased towards the largest classes and, thus, be sub-optimal for the discrimination of the classes of interest. This study presents a way to minimize the effects of data imbalance problems in specific class mapping using cost-sensitive learning. In this approach errors committed in the minority class are treated as being costlier than errors committed in the majority class. Cost-sensitive approaches are typically implemented by weighting training data points accordingly to their importance to the analysis. By changing the weight of individual data points, it is possible to shift the weight from the larger classes to the smaller ones, balancing the data set. To illustrate the use of the cost-sensitive approach to map specific classes of interest, a series of experiments with weighted support vector machines classifier and Landsat Thematic Mapper data were conducted to discriminate two types of mangrove forest (high-mangrove and low-mangrove) in Saloum estuary, Senegal, a United Nations Educational, Scientific and Cultural Organisation World Heritage site. Results suggest an increase in overall classification accuracy with the use of cost-sensitive method (97.3%) over the standard multi-class (94.3%) and the focused approach (91.0%). In particular, cost-sensitive method yielded higher sensitivity and specificity values on the discrimination of the classes of interest when compared with the standard multi-class and focused approaches.