Khazaeiathar, Mahshid ; Schmalz, Britta
Hrsg.: Park, Jeehyun ; ETH Zürich (2024)
Enhancing Short-Term Discharge Predictions: An innovative ARIMA-iGARCH Model for Improved Flood Forecasting and Disaster Resilience.
5th International Conference on Resilient Systems (ICRS 2024). Singapore, Singapore (28.08.2024-30.08.2024)
doi: 10.3929/ethz-b-000696625
Konferenzveröffentlichung, Bibliographie

Kurzbeschreibung (Abstract)

Choosing the most suitable model for discharge simulation is challenging, especially with short-term data. While artificial neural networks excel at detecting river flow patterns, they require substantial data for training, making them less effective with limited datasets. As an alternative, Autoregressive Integrated Moving Average (ARIMA) models can be utilized for short-term data. However, severe volatilities and inherent non-stationarity in hydrological time series can introduce significant errors. This study introduces a new adaptive hybrid model, ARIMA-iGARCH (Integrated Generalized AutoRegressive Conditional Heteroscedasticity), designed to account for volatility and non-stationarity, thus minimizing errors in short-term time series modeling. The ARIMA-iGARCH model specifically addresses the inconsistency of variance and non-stationary behavior in discharge time series. We applied the ARIMA-iGARCH model to four hourly discharge time series of the Schwarzbach River upstream of the gauge Nauheim in Hesse, Germany. In this process, the iGARCH model was used for prediction, and hybrid model parameters were obtained by combining ARIMA and GARCH models, assuming a normal distribution for residuals. The results demonstrate that the new adaptive hybrid model, based on this special parameter estimation method, offers less complexity, greater accuracy, and more reliable predictions. By capturing fluctuations in time series variance, the ARIMA-iGARCH model significantly improves the modeling of long-memory, non-linear, non-stationary, and particularly short-term datasets. This improvement is crucial for disaster resilience, as accurate discharge predictions enhance flood forecasting and management. Effective flood forecasting leads to better preparedness and response strategies, mitigating the impacts of hydrological disasters. In conclusion, the ARIMA-iGARCH model represents a significant advancement for hydrological time series modeling, particularly with short-term data, contributing to disaster resilience by enabling more accurate and reliable flood predictions.

Frage zum Eintrag

Redaktionelle Details anzeigen