| Titre : | Deep learning approaches for water stress forecasting in arboriculture using remote sensing images: Comparative study between ConvLSTM and CNN-LSTM models |
| Auteurs : | BOUNOUA,Ismail;SAIDI,Youssef, Auteur ; YAAGOUBI Reda, Auteur ; BOUZIANI Mourad, Auteur |
| Type de document : | texte imprimé |
| Editeur : | Rabat : IAV Hassan II, 2023 |
| Format : | 145 |
| Langues: | Français |
| Mots-clés: | water stress ; evapotranspiration ; arboriculture ; CWSI ; time series ; forecasting ; stochastic models ; machine learning ; deep learning ; remote sensing ; ConvLSTM ; CNN-LSTM |
| Résumé : |
Amid the context of global water stress, a pressing concern impacting agricultural productivity, plant health, and environmental sustainability, the need for accurate forecasting has gained paramount importance. Remote sensing has emerged as an invaluable tool for assessing water stress, offering critical insights into crop conditions. However, traditional machine learning methods and stochastic approaches, while successful in various contexts, encounter inherent limitations when dealing with raster data and complex spatiotemporal patterns. To address this limitation, deep learning techniques have gained prominence, owing to their capacity to handle extensive data volumes and capture intricate relationships. Therefore, for spatiotemporal forecasting harnessing remote sensing data, deep learning methods stand out as the most fitting choice. In our research, we focus on comparing two deep learning models, ConvLSTM and CNN-LSTM, for forecasting water stress using remote sensing data. An arboriculture farm in Morocco has been chosen as a case study. The foundation of our methodology involves meticulous preparation of time series data, where we calculate the Crop Water Stress Index (CWSI) utilizing Landsat 8 satellite imagery through Google Earth Engine. Subsequently, we employ the diverse functionalities available to fine-tune hyperparameters and mitigate overfitting or underfitting risks during the training of ConvLSTM and CNN-LSTM models. Our analysis of the results reveals that the CNN-LSTM model excels over the ConvLSTM model for sequences comprising nine images. The CNN-LSTM model achieves a root mean square error (RMSE) of 0.119, outperforming ConvLSTM model, which achieves an RMSE of 0.123. |
| En ligne : | http://10.2.0.27//cda/ebooks/BOUNOUA_SAIDI_2023.pdf |
Exemplaires (1)
| Code-barres | Cote | Support | Localisation | Section | Disponibilité |
|---|---|---|---|---|---|
| 200041162 | 9987 | Support papier | Salle des thèses/PFE (RDC) | Ingénieur topographe | Consultation sur place Exclu du prêt |
