2021 Fiscal Year Research-status Report
Improving flood and drought prediction using downscaled GRACE terrestrial water storage
| Project/Area Number |
21K20443
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
尹 高虹 東京大学, 生産技術研究所, 特任研究員 (00906282)
|
|---|
| Project Period (FY) |
2021-08-30 – 2023-03-31
|
| Keywords | GRACE / TWS / Downscaling / Deep Learning / Flood / Drought |
|---|
| Outline of Annual Research Achievements |
(1) All required satellite-based and model-based data has been prepared. (2) A synthetic experiment has been set up. Synthetic TWS was generated by upscaling model-based TWS to GRACE mascon scale, afterward, a observational error was added to the model-based TWS. (3) Synthetic experiment has been tested using Long short-term memory (LSTM) deep learning method. (4) Synthetic results has been evaluated against synthetic truth, and results demonstrated the feasibility of the proposed method across the globe.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The progress of the project is more smoothly than initially planned. As I am familiar with the used satellite-based and model-based data sets, data preparation process was faster than expected. An synthetic experiment was conducted based on plan. The learning process of deep learning algorithms goes well, and the computational efficiency of the selected method, i.e., LSTM, was higher than expected. Therefore, the synthetic experiment was almost finished at a global scale.
|
| Strategy for Future Research Activity |
For the future work, following work is expected to be accomplished:
(1) More investigation on the synthetic models in order to select the best combination of predictors, the best deep learning model, as well as the modeling strategy.
(2) Conduct downscaling experiment using real GRACE and GRACE-FO data based on the selected method from step 1.
(3) Assimilating the downscaled TWS into land surface models for flood and drought monitoring.
|
| Causes of Carryover |
For the next year, article costs including official goods, computational spent will be needed. Most important results will come out in the next fiscal year. Therefore, publication fee, conference attendee and traveling fee will be required in the next year.
|
