Inflow Prediction for Agricultural Reservoirs Using Disaster Prevention Measurement Data: A Comparison of TANK Model and Machine Learning
This study compared the inflow prediction performance of the TANK model and Ridge regression for agricultural reservoirs equipped with disaster prevention measurement systems. Two models were developed using rainfall, water level, and inflow data collected from monitoring equipment installed at Baek...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Korean Society of Environmental Engineers
2025-05-01
|
| Series: | 대한환경공학회지 |
| Subjects: | |
| Online Access: | http://www.jksee.or.kr/upload/pdf/KSEE-2025-47-5-303.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study compared the inflow prediction performance of the TANK model and Ridge regression for agricultural reservoirs equipped with disaster prevention measurement systems. Two models were developed using rainfall, water level, and inflow data collected from monitoring equipment installed at Baekrok Reservoir in Boeun-gun, Chungcheongbuk-do and Nangye Reservoir in Sunchang-gun, Jeollabuk-do. Through analysis of rainfall-inflow relationships, 4-hour moving average rainfall for Baekrok Reservoir and 8-hour moving average rainfall for Nangye Reservoir were selected as optimal input data for the RidgeCV model. For the calibration period from January to August 2024, the TANK model showed NSE values of 0.893 for Baekrok Reservoir and 0.502 for Nangye Reservoir. In contrast, the RidgeCV model demonstrated superior performance with NSE values of 0.989 for Baekrok Reservoir and 0.983 for Nangye Reservoir. During the validation period (September-October 2024), the TANK model's performance significantly deteriorated to NSE values of 0.141 for Baekrok Reservoir and 0.547 for Nangye Reservoir, while the RidgeCV model maintained stable performance with NSE values of 0.978 for Baekrok Reservoir and 0.984 for Nangye Reservoir. This superior performance of the RidgeCV model can be attributed to its effective learning of the relationship between inflow data and optimal moving average rainfall, as well as the prevention of overfitting through regularization. The results of this study demonstrate the potential of machine learning techniques for inflow prediction in agricultural reservoirs and suggest the need for further research on model improvement using various algorithms and input variables. |
|---|---|
| ISSN: | 1225-5025 2383-7810 |