Method to apply and visualize physical models associated to a land cover performed by CNN: A case study of vegetation and water cooling effect in Bangkok Thailand - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Remote Sensing Applications: Society and Environment Année : 2022

Method to apply and visualize physical models associated to a land cover performed by CNN: A case study of vegetation and water cooling effect in Bangkok Thailand

Résumé

Convolutional Neural Networks (CNNs) are useful tools to perform land cover analysis, in particular when working on large areas. The information extracted from these land cover analyses is useful in many practical applications and can be used as input data for physical models capable of simulating any type of phenomenon of interest. Models results can be used to support policy making and can be visualized on the associated remote sensing image to aid interpretation. A method covering these aspects and illustrated on a practical use case is proposed in this paper. The use case focuses on the cooling effect of vegetation and water in Bangkok. For this use case, a dedicated dataset of approximately 680,000 64 × 64 pixels tiles with a resolution of 0.15 m/pixel is created. Three CNN models are built and optimized to classify each of the following classes: city, vegetation and water. The validation of the models shows that, with the exception of water where the accuracy is only 84%, the other two classes have an accuracy of over 90%. Post-processing is performed on each prediction before aggregating these results to obtain the land cover. Vegetation and water cooling models, given in the literature, are successfully applied using the land cover and their effects are plotted on the associated satellite images. Results show that small areas of trees close to water have little influence on the cooling effect and that it is not efficient to plant isolated trees near a larger vegetation site. Through this study, the proposed tool has demonstrated its usefulness.
Fichier non déposé

Dates et versions

hal-04103260 , version 1 (23-05-2023)

Identifiants

Citer

M. Mentet, N. Hongkarnjanakul, C. Schwob, L. Mezeix. Method to apply and visualize physical models associated to a land cover performed by CNN: A case study of vegetation and water cooling effect in Bangkok Thailand. Remote Sensing Applications: Society and Environment, 2022, 28, pp.100856. ⟨10.1016/j.rsase.2022.100856⟩. ⟨hal-04103260⟩
21 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More