University of Twente Student Theses
Spatiotemporal analysis for fire forecasting using deep learning techniques in google earth engine : a case study for the Indian state of Uttarakhand
Kanagasabapathi, Dhanasekaran (2023) Spatiotemporal analysis for fire forecasting using deep learning techniques in google earth engine : a case study for the Indian state of Uttarakhand.
|
PDF
7MB |
| Abstract: | Fire forecasting models are built with different fire-influencing factors such as terrain, vegetation, meteorological, anthropogenic, and day of year. Two different approaches were used for fire forecasting. One is to forecast with the data of fire-influencing variables of a single day, and the other with the data of fire-influencing variables of the past five days. A variant of U-Net architecture and Convolutional LSTM are used for the non-temporal and temporal datasets, respectively. Datasets are obtained from Google Earth Engine and processed for the fire modelling process. In order to reduce the imbalance in the proportion of fire and non-fire classes, data is selected from two different patches that represent all the state's climatic, land use land cover, and terrain conditions. In this way, the imbalance in proportion has been reduced to a certain extent. In the approach which includes the fire-influencing variables of a single day, three different scenarios are performed to analyse whether the fire-influencing variable of a single can be used to forecast the fire of the next day, the next two days, and the next five days. This has been performed by aggregating the fire labels of the next two days and the next five days to find the relationship. In the approach that utilized the temporal data of fire influencing variables, the last five days of fire influencing variables data are used to model the next day’s fire forecast. The model built to forecast the next two days of fire had high prediction accuracy for both fire and non-fire pixels. The model is selected, and a web application is hosted on a local server using the Python library ‘greppo’. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 38 earth sciences, 43 environmental science, 54 computer science, 74 (human) geography, cartography, town and country planning, demography |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/97211 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page