Spatial distribution of the uncertainty in land cover maps obtained from remote sensing [Distribución espacial de la incertidumbre en mapas de cubiertas obtenidos mediante teledetección]

Xavier P., Sevillano E., More G., Serra P., Cornford D., Ninyerola M. (2014) Spatial distribution of the uncertainty in land cover maps obtained from remote sensing [Distribución espacial de la incertidumbre en mapas de cubiertas obtenidos mediante teledetección]. Revista de Teledeteccion. : 1-10.
Enllaç
Doi: 10.4995/raet.2014.3059

Resum:

When combining remote sensing imagery with statistical classifiers to obtain categorical thematic maps it is not usual to provide data about the spatial distribution of the error and uncertainty of the resulting maps. This paper describes, in the context of GeoViQua FP7 project, feasible approaches for methods based on several steps such as hybrid classifiers. Both for “per pixel” and “per polygon” strategies, the proposal is based on the use of the available ground truth, which is used to properly model the spatial distribution of the errors. Results allow mapping the classification success with a very high level of reliability (R2>0,94), providing users a sound knowledge of the accuracy at every area of the map.

Llegeix més

Effects of topography on the radiometry of CHRIS/PROBA images of successional stages within tropical dry forests

Garcia Millan V.E., Sanchez Azofeifa G.A., Malvarez G.C., More G., Pons X., Yamanaka-Ocampo M. (2013) Effects of topography on the radiometry of CHRIS/PROBA images of successional stages within tropical dry forests. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 6: 1584-1595.
Enllaç
Doi: 10.1109/JSTARS.2013.2259471

Resum:

In the present paper, the effect of shadows in the classification of three successional stages of a tropical dry forest (TDF) in Mexico, using hyperspectral and multi-angular CHRIS/PROBA images, is evaluated. An algorithm based on the cosine of the angle of solar incidence on the terrain is applied to correct the effect of topography on CHRIS/PROBA reflectances. Previous to the removal of shadows caused by topography, CHRIS/PROBA images were atmospherically corrected in BEAM software. Vegetation maps of the study site were generated using non-parametric decision trees, defining four main classes: late, intermediate and early stages of forest succession within a tropical dry forest, and riparian forests. By comparing the vegetation maps before and after shadow removal in CHRIS/PROBA spectral data, it was observed that the late stage of succession and riparian forests are overestimated for the non-corrected images while intermediate and early stages of succession are underestimated. Errors in classification are more important for the large CHRIS/PROBA viewing angles. Therefore, the removal of shadows caused by topography is necessary for an accurate classification of successional stages in tropical dry forests. © 2013 IEEE.

Llegeix més

Enhanced land use/cover classification of heterogeneous tropical landscapes using support vector machines and textural homogeneity

Paneque-Gálvez J., Mas J.-F., Moré G., Cristóbal J., Orta-Martínez M., Luz A.C., Guèze M., Macía M.J., Reyes-García V. (2013) Enhanced land use/cover classification of heterogeneous tropical landscapes using support vector machines and textural homogeneity. International Journal of Applied Earth Observation and Geoinformation. 23: 372-383.
Enllaç
Doi: 10.1016/j.jag.2012.10.007

Resum:

Land use/cover classification is a key research field in remote sensing and land change science as thematic maps derived from remotely sensed data have become the basis for analyzing many socio-ecological issues. However, land use/cover classification remains a difficult task and it is especially challenging in heterogeneous tropical landscapes where nonetheless such maps are of great importance. The present study aims at establishing an efficient classification approach to accurately map all broad land use/cover classes in a large, heterogeneous tropical area, as a basis for further studies (e.g., land use/cover change, deforestation and forest degradation). Specifically, we first compare the performance of parametric (maximum likelihood), non-parametric (k-nearest neighbor and four different support vector machines - SVM), and hybrid (unsupervised-supervised) classifiers, using hard and soft (fuzzy) accuracy assessments. We then assess, using the maximum likelihood algorithm, what textural indices from the gray-level co-occurrence matrix lead to greater classification improvements at the spatial resolution of Landsat imagery (30 m), and rank them accordingly. Finally, we use the textural index that provides the most accurate classification results to evaluate whether its usefulness varies significantly with the classifier used. We classified imagery corresponding to dry and wet seasons and found that SVM classifiers outperformed all the rest. We also found that the use of some textural indices, but particularly homogeneity and entropy, can significantly improve classifications. We focused on the use of the homogeneity index, which has so far been neglected in land use/cover classification efforts, and found that this index along with reflectance bands significantly increased the overall accuracy of all the classifiers, but particularly of SVM. We observed that improvements in producer's and user's accuracies through the inclusion of homogeneity were different depending on land use/cover classes. Early-growth/degraded forests, pastures, grasslands and savanna were the classes most improved, especially with the SVM radial basis function and SVM sigmoid classifiers, though with both classifiers all land use/cover classes were mapped with producer's and user's accuracies of ~90%. Our classification approach seems very well suited to accurately map land use/cover of heterogeneous landscapes, thus having great potential to contribute to climate change mitigation schemes, conservation initiatives, and the design of management plans and rural development policies. © 2012 Elsevier B.V.

Llegeix més

Influence of the nature and number of ground control points to the quality of remote sensing geometric corrections

More G., Pons X. (2012) Influence of the nature and number of ground control points to the quality of remote sensing geometric corrections. International Geoscience and Remote Sensing Symposium (IGARSS). : 2356-2359.
Enllaç
Doi: 10.1109/IGARSS.2012.6351021

Resum:

Georeferencing satellite images is an essential procedure to carry out most remote sensing applications. The quality of this process will affect all the ulterior procedures and products. Independent test ground control points (GCPs) are required to assess the quality of the correction. However, a representative number is hardly obtained when they are manually located. This work studies the effect of the number of GCPs in the geometric correction quality when they are manually located. The methodology has been applied to Landsat TM images in a region with complex relief (heights ranging from 0 to 3000+ m). The work presents a spatial representation of the error and discusses its role in the visualisation of the quality. Moreover, we critically discuss the usage of indicators as the RMS error without considering the number of GCPs or the method used in their placement in the realistic assessment of the geometric quality of the imagery. Indeed, it is shown that, for the studied scenes, a minimum of 25 GCPs is needed to achieve a test RMS smaller than a pixel and that not using independent GCPs leads to unrealistic quality indicators. Moreover, manual placement of GCPs gives clearly worst results than automatic procedures. © 2012 IEEE.

Llegeix més

Multitemporal flooding dynamics of rice fields by means of discriminant analysis of radiometrically corrected remote sensing imagery

Moré G., Serra P., Pons X. (2011) Multitemporal flooding dynamics of rice fields by means of discriminant analysis of radiometrically corrected remote sensing imagery. International Journal of Remote Sensing. 32: 1983-2011.
Enllaç
Doi: 10.1080/01431161003645816

Resum:

An automatic classifier based on a discriminant analysis (DA) was used to classify eight classes in relation to different stages of rice fields during the flooding season. This methodology is characterized by the fact that, once the training phase has been carried out, training areas are not required to perform new classifications. If the images have been radiometrically corrected in a consistent way, the classifier can be used in a retrospective mode using past images. For this study, the training phase was conducted with data taken in October 2006 and January 2007 while the automatic classifier was applied to a total of 10 Landsat-5 Thematic Mapper (TM) images from the 2004-05 and 2006-07 seasons. An average level of accuracy of 93.4% (range 89.7-98.7%) demonstrates the capability of the method to obtain high-quality and quasi-instantaneous classifications and to carry out retrospective studies even when training areas are not available for past dates. Two examples of how the method can be used are included in this article: (i) a study of the temporal evolution of flooding covers by period and (ii) the use of vector enrichment as a thematic updating tool for the cadastre. An additional objective of the study was to analyse the importance of the different bands to ascertain the suitability of alternative sensors with spectral configurations other than those provided by Landsat. This analysis demonstrates that the absence of shortwave infrared (SWIR) bands results in a decrease of almost nine percentage points in the accuracy levels of the classification while the blue band can be excluded with minimal impact on the results. © 2011 Taylor & Francis.

Llegeix més

Classification of hyperspectral images compressed through 3D-JPEG2000

Blanes I., Zabala A., Moré G., Pons X., Serra-Sagristà J. (2008) Classification of hyperspectral images compressed through 3D-JPEG2000. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). 5179 LNAI: 416-423.
Enllaç
Doi: 10.1007/978-3-540-85567-5-52

Resum:

Classification of hyperspectral images is paramount to an increasing number of user applications. With the advent of more powerful technology, sensed images demand for larger requirements in computational and memory capabilities, which has led to devise compression techniques to alleviate the transmission and storage necessities. Classification of compressed images is addressed in this paper. Compression takes into account the spectral correlation of hyperspectral images together with more simple approaches. Experiments have been performed on a large hyperspectral CASI image with 72 bands. Both coding and classification results indicate that the performance of 3d-DWT is superior to the other two lossy coding approaches, providing consistent improvements of more than 10 dB for the coding process, and maintaining both the global accuracy and the percentage of classified area for the classification process. © 2008 Springer-Verlag Berlin Heidelberg.

Llegeix més

Improvements on classification by tolerating NoData values application to a hybrid classifier to discriminate mediterranean vegetation with a detailed legend using multitemporal series of images

Moré G., Pons X., Serra P. (2006) Improvements on classification by tolerating NoData values application to a hybrid classifier to discriminate mediterranean vegetation with a detailed legend using multitemporal series of images. International Geoscience and Remote Sensing Symposium (IGARSS). : 192-195.
Enllaç
Doi: 10.1109/IGARSS.2006.54

Resum:

Natural and crop vegetation phenologic data become indispensable when creating thematically and geographically detailed maps through satellite images classification. Several date acquisition is necessary to achieve this cartography. However, the presence of clouds, shadows, snow, etc, is usual when many different dates are used and that fact implies an important loss in classifiable surface. This work presents a hybrid classifier designed to deal with the common problems appeared in the classification of Mediterranean vegetation. Specifically, IsoMM, the first phase of the hybrid methodology, is an unsupervised classifier that allows a better use of temporal series thanks to a particular treatment of NoData values (or missing values) in the images. This methodology has been applied to a Mediterranean forestry zone with a legend of eleven categories and has been compared to a Maximum Likelihood classifier. The presented improvements allow classifying more surface than a common NoData treatment strategy (wheter unsupervised, Maximum Likelihood classification or the extraction of a problematic date) and achieving high accuracy level.

Llegeix més

Reassessing global change research priorities in mediterranean terrestrial ecosystems: How far have we come and where do we go from here?

Doblas-Miranda E., Martinez-Vilalta J., Lloret F., Alvarez A., Avila A., Bonet F.J., Brotons L., Castro J., Curiel Yuste J., Diaz M., Ferrandis P., Garcia-Hurtado E., Iriondo J.M., Keenan T.F., Latron J., Llusia J., Loepfe L., Mayol M., More G., Moya D., Penuelas J., Pons X., Poyatos R., Sardans J., Sus O., Vallejo V.R., Vayreda J., Retana J. (0) Reassessing global change research priorities in mediterranean terrestrial ecosystems: How far have we come and where do we go from here?. Global Ecology and Biogeography. 24: 25-43.
Enllaç
Doi: 10.1111/geb.12224

Resum:

Aim: Mediterranean terrestrial ecosystems serve as reference laboratories for the investigation of global change because of their transitional climate, the high spatiotemporal variability of their environmental conditions, a rich and unique biodiversity and a wide range of socio-economic conditions. As scientific development and environmental pressures increase, it is increasingly necessary to evaluate recent progress and to challenge research priorities in the face of global change. Location: Mediterranean terrestrial ecosystems. Methods: This article revisits the research priorities proposed in a 1998 assessment. Results: A new set of research priorities is proposed: (1) to establish the role of the landscape mosaic on fire-spread; (2) to further research the combined effect of different drivers on pest expansion; (3) to address the interaction between drivers of global change and recent forest management practices; (4) to obtain more realistic information on the impacts of global change and ecosystem services; (5) to assess forest mortality events associated with climatic extremes; (6) to focus global change research on identifying and managing vulnerable areas; (7) to use the functional traits concept to study resilience after disturbance; (8) to study the relationship between genotypic and phenotypic diversity as a source of forest resilience; (9) to understand the balance between C storage and water resources; (10) to analyse the interplay between landscape-scale processes and biodiversity conservation; (11) to refine models by including interactions between drivers and socio-economic contexts; (12) to understand forest-atmosphere feedbacks; (13) to represent key mechanisms linking plant hydraulics with landscape hydrology. Main conclusions: (1) The interactive nature of different global change drivers remains poorly understood. (2) There is a critical need for the rapid development of regional- and global-scale models that are more tightly connected with large-scale experiments, data networks and management practice. (3) More attention should be directed to drought-related forest decline and the current relevance of historical land use.

Llegeix més