Suchergebnisse

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques. ; ISSN:2072-4292

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

International audience ; Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

The Epidemiological Surveillance System for Malaria (SIVEP-Malaria) is the Brazilian governmental program that registers all information about compulsory reporting of detected cases of malaria by all medical units and medical practitioners. The objective of this study is to point out the main sources of errors in the SIVEP-Malaria database by applying a data cleaning method to assist researchers about the best way to use it and to report the problems to authorities. The aim of this study was to assess the quality of the data collected by the surveillance system and its accuracy. The SIVEP-Malaria data base used was for the state of Amazonas, Brazil, with data collected from 2003 to 2014. A data cleaning method was applied to the database to detect and remove erroneous records. It was observed that the collecting procedure of the database is not homogeneous among the municipalities and over the years. Some of the variables had different data collection periods, missing data, outliers and inconsistencies. Variables depending on the health agents showed a good quality but those that rely on patients were often inaccurate. We showed that a punctilious preprocessing is needed to produce statistically correct data from the SIVEP-Malaria data base. Fine spatial scale and multi-temporal analysis are of particular concern due to the local concentration of uncertainties and the data collecting seasonality observed. This assessment should help to enhance the quality of studies and the monitoring of the use of the SIVEP database. ; O Sistema de Vigilância Epidemiológica de Malária (SIVEP-Malária) é um programa governamental brasileiro que arquiva automaticamente todas as informações sobre casos de malária registrados em todas as unidades de saúde e consultórios medicos. O objetivo deste estudo foi avaliar a qualidade dos dados coletados pelo sistema de vigilância e sua precisão. Foram utilizados os dados do SIVEP-Malária para o estado do Amazonas, Brasil, de 2003 a 2014. Um método de limpeza de dados foi aplicado para detectar e remover registros errôneos. Observamos que a coleta de dados não é homogênea entre os municipios e ao longo dos anos. Algumas variaveis tinham diferentes padrões de coleta, falta de dados, dados discrepantes e inconsistências. Dados que dependem do agente de saúde possuem boa qualidade mas aqueles que dependem dos pacientes são frequentemente imprecisos. Mostramos que um pre-processamento meticuloso é necessário para produzir dados estatisticamente corretos a partir do SIVEP-Malária. Analises em escala espacial detalhada ou multi-temporais são particularmente afetadas devido à concentração local de incertezas e a sazonalidade observada na coleta de dados. Esta avaliação deve auxiliar a melhorar os estudos e monitoramentos que fazem uso dos dados do SIVEP.

Le moustique Aedes (Stegomyia) albopictus (Skuse) (Diptera : Culicidae) est une espèce particulièrement invasive, qui démontre de remarquables capacités d'adaptation à des conditions climatiques multiples. Le suivi de son aire de diffusion répond à une préoccupation importante de santé publique, puisque cette espèce présente, en plus d'une forte capacité de nuisance diurne, la capacité de transmettre les virus de la dengue, du chikungunya, et du Zika. Sa forte adaptation aux milieux anthropisés, et son caractère hautement invasif justifient une politique de surveillance appropriée. Dans cette étude publiée dans la revue International Journal of Environnemental Research and Public Health sous le titre " A rainfall- and temperature-driven abundance model for Aedes albopictus populations ", la dynamique de population du moustique " tigre " Aedes Albopictus a pu être modélisée pour la première fois en climat tempéré au travers d'une approche mécaniste. La dynamique du modèle est pilotée par les variables météorologiques de température et de précipitation. Cette approche a été validée par comparaison avec des données entomologiques relevées sur quatre années dans la région de Nice (coefficient de pearson 0,73-0,93). La validation satisfaisante du modèle en climat tempéré s'explique principalement par la prise en compte de la diapause, une période défavorable au développement du moustique, et pendant laquelle seuls les oeufs des moustiques survivent. Dans un second temps, ce modèle a été spatialisé et implémenté à La Réunion en 2015 dans le cadre du projet ALBORUN, avec pour objectif le développement d'un outil opérationnel à destination du service de lutte anti-vectorielle (LAV) de l'Agence de Santé Océan Indien (ARS OI). Les prédictions du modèle ont montré un bon accord avec les observations de terrain, ce qui a conduit à la construction d'un outil opérationnel de cartographie des densités de moustique intégrant les données de stations météorologiques in situ. Cet outil est actuellement utilisé en routine par les services de LAV à La Réunion. La généricité de ce modèle mécaniste permet aussi d'envisager son application sur l'ensemble des aires du territoire métropolitain où l'espèce Aedes albopictus est installée. C'est l'objet du projet ARBOCARTO, financé par la Direction Générale de la Santé (DGS) et les Agences Régionales de Santé (ARS Auvergne-Rhône-Alpes, Occitanie et Nouvelle- Aquitaine). Dans cette nouvelle étape, l'outil est à nouveau paramétré pour servir de démonstrateur sur les trois sites pilotes de Montpellier, Grenoble, et Bordeaux. L'intérêt des sorties cartographiques des densités vectorielles d'Aedes albopictus est en cours d'évaluation, pour répondre à terme à une double intégration : (i) comme support décisionnel par les ARS mentionnées, et ii) comme support sur le terrain par les Ententes Interdépartementales de Démoustication (EID). Les possibilités d'adaptation du modèle aux différentes aires de distribution reposent en grande partie sur l'étude de sensibilité réalisée dans l'étude publiée. Celle-ci a permis d'identifier en particulier un paramètre clef dans sa contextualisation géographique : la variable dite de " capacité de charge de l'environnement ". Cette variable reflète la capacité de l'environnement à offrir des conditions favorables à la présence de gîtes de ponte pour les moustiques. Dans les projets ALBORUN et ARBOCARTO, le paramétrage du modèle a pu être affiné en établissant une valeur de capacité de charge à partir de données issues de la littérature et des échanges réalisés avec les acteurs de terrain de la démoustication. L'estimation de cette valeur de " capacité de charge de l'environnement " bénéficie de l'intégration de données géographiques complémentaires pour caractériser l'environnement en termes d'occupation (p.ex. densité de bâti) et d'usage du sol (p.ex. hotspot type cimetière). Une meilleure estimation de la capacité de charge est également attendue avec l'intégration d'informations issues des données de télédétection, en particulier avec l'estimation de la densité de végétation dans la maille élémentaire du modèle. Des développements méthodologiques portant sur une meilleure intégration des données de télédétection, typologie urbaine en particulier, sont également en cours dans le cadre du projet APUREZA (" Analyses des relations entre paysages urbains dengue et Zika ", financement TOSCA CNES 2017-2020). Enfin, une nouvelle perspective d'application est envisagée en Asie du Sud-Est avec le projet ECOMORE II (ECOnomic development, ECOsystem MOdifications, and emerging infectious diseases Risk Evaluation), financé par l'Agence Française de Développement.