Machine learning for flood forecasting : a case study of Yom River sub-basin, Thailand | |
| Author | Habib, Wahaj |
| Call Number | AIT Thesis no.RS-18-10 |
| Subject(s) | Machine learning--Statistical methods Flood forecasting--Thailand--Yom River Watershed |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Remote Sensing and Geographic Information System, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Abstract | Floods, much like other disasters have an influence on human society from time immemorial. Flood affects communities and consequently causes large economic losses. Owing to this, the significance of the flood forecasting system for disaster preparedness cannot be denied. Machine learning, with its application in many fields currently, is also being used to increase the effectiveness of these systems. However, most of the machine learning techniques predict the runoff consequences based on rainfall data and hence, predict the flood. The main aim of this study, therefore, is to develop a machine learning based model for flood forecasting, using remote sensing based flood observation and hydro-meteorological parameters (for the upstream area). A novel approach is made by combining hydrometeorological parameters; daily rainfall and discharge data (collected from ground stations), with flood extent derived from satellite remote sensing (MODIS Near Real Time Flood product) observations (from 2005 - 2011) for training different machine learning models (Artificial Neural Network, Support Vector Machine and Decision Tree). The effectiveness of each model is assessed using a different combination of parameters. Between these parameters, the remote sensing observation proved vital for this study as it is used as a flood indicator providing the exact date and geographical extent for flood on a daily basis. Since machine learning based models are dependent on a high number of samples, a thorough flood observation record is required which otherwise is not available normally for all the locations. This gap is filled using the remote sensing flood observation data in this study. Among the three machine, learning based models ANN (Artificial Neural Network) showed satisfactory results, with the ability to forecast the flood for the same day with an accuracy of 76 %. While the simulations that were based on SVM (Linear) were able to forecast the flood four days in advance with an accuracy of 80 %. With this kind of modeling, it is easy to pre-evaluate the flood scenarios promptly. As proven by testing the model withhold out data from the year (2016 and 2017). After the model was developed, it was provided with only the rainfall and discharge data to evaluate the accuracy of the model to predict the future flood scenario. The result showed that if the model can predict the flood, if only provided with rainfall and discharge data, with the accuracy of 95 %. Furthermore, the model can predict the flood four days ahead if only provided with the rainfall data, with the accuracy of 84 %. Altogether, the precautions and guidelines suggested in this study for developing a machine learning based model that is primarily dependent on satellite-based flood observations provide a good framework for risk assessment and flood forecasting for future studies. Keywords: Machine learning, Rainfall, Discharge, MODIS NRT Flood Product, Flood forecasting |
| Year | 2018 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing (RS) |
| Chairperson(s) | Sarawut Ninsawat |
| Examination Committee(s) | Shrestha, Sangam ; Vilas Nitivattananon ; Shipin, Oleg V. |
| Scholarship Donor(s) | Thailand (HM King) |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2018 |