Spatial downscaling of satellite precipitation data using machine learning approach | |
| Author | Wangmo, Pema |
| Call Number | AIT Thesis no.AE-23-02 |
| Subject(s) | Machine learning Precipitation forecasting Satellite meteorology |
| Note | A Thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Agricultural Systems and Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | Precipitation is one of the crucial factors that govern the livelihood on earth. Accurate measurement of the precipitation distribution in real-time allows farmers to make farming decisions. It is challenging to provide fine-resolution precipitation data at the field scale due to the significant spatial-temporal variability of precipitation and the relatively uneven or sparse distribution of rain-gauges. Satellite precipitation estimates (SPEs) can be used to fix these issues. In areas with uneven or sparse distribution, SPEs can aid in overcoming the constraints of conventional rain gauge networks. However, the grid resolution provided by SPEs is unsuitable for water resource management and agricultural applications. Therefore, the study was conducted to assess four types of SPEs, namely, CHIRPS, IMERG, PERSIANN, and GSMaP, among which can be used as an alternative to ground observed data over Chao Phraya river basin in Thailand and Drangme Chhu river basin in Bhutan. The assessment was carried out using categorical and statistical analysis for their ability to detect and capture rainfall in these two basins. The most suitable SPEs was then re-gridded to the spatial resolution of 0.01 degree, and the ground observed precipitation was also interpolated to the same resolution of the selected SPEs for developing a support vector machine (SVM)-based machine learning algorithm for downscaling. The result obtained showed IMERG as the most suitable SPEs that can be used as an alternative to observed data over the Chao Phraya river basin and the Drangme Chhu river basin. The SVM algorithm developed showed a performance of more than 80% over both basins. As per the index of agreement (IOA), the agreement between the SPEs and observed precipitation was 0.90 and 0.82 for CPRB and DCRB, respectively, after validation. The NSE values of the model were 0.89 for CPRB and 0.82 for DCRB when validated at the daily time scale. The study’s findings indicate that IMERG SPEs can be an alternative to observed data over these two basin areas. Additionally, the developed SVM-based algorithm can be used to spatially downscale SPEs however, comparative evaluation with other available SPEs and machine learning algorithms is suggested to validate the present model. |
| Year | 2023 |
| Type | Thesis |
| School | School of Environment, Resources, and Development |
| Department | Department of Food, Agriculture and Natural Resources (Former title: Department of Food Agriculture, and BioResources (DFAB)) |
| Academic Program/FoS | Agricultural Systems and Engineering (ASE) |
| Chairperson(s) | Himanshu, Sushil Kumar; |
| Examination Committee(s) | Datta, Avishek;Zulfiqar, Farhad; |
| Scholarship Donor(s) | Asian Development Bank-Japan Scholarship Program (ADB-JSP); |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2023 |