Designing Artificial Intelligence-based Probabilistic Methods for Early Flood Warning System for Community Risk Management in Fiji Islands.
06 FEB 2023
9.30 AM - 11.00 AM
9.30 AM - 11.00 AM
Fiji is one of the Pacific Island Countries severely affected by flooding. Apart from coastal floods, Fiji often experiences fluvial and pluvial floods. The flood-related asset losses for Fiji are anticipated to exceed 5 percent of the GDP by 2050. Consequently, timely and accurate forecasting of flood events is anticipated to provide governments, businesses, and individuals more time to make well-informed decisions to be better prepared for flood events. In the first phase of this study, the novel hourly flood index (SWRI24-hr-S) will be formulated, which includes the development of the relevant software codes to calculate the index in a normalized way for the existing 24-hourly water resources index (WRI24-hr-S). The feasibility of SWRI24-hr-S in determining a flood situation and its characteristics (i.e., duration, severity, and intensity) will be evaluated for various sites in Fiji from 2014-2018. The advanced statistical copula-based techniques will also be used to model the joint distributions amongst the extreme flood characteristics to extract their joint exceedance probability, which is vital information for managing flood risks in Fiji. In the second phase of this study, the deep learning-based model will be developed to produce hourly forecasts of SWRI24-hr-S values and test its capability to forecast future flood events at hourly intervals at the selected study sites in Fiji. More specifically, a novel framework based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and Long Short-term Memory (LSTM), i.e., the ICEEMDAN-LSTM model, is proposed. Using the real-time hourly precipitation dataset from 2014 to 2018 obtained from Fiji Meteorological Services, the newly developed SWRI24-hr-S was applied at seven sites in Fiji. The SWRI24-hr-S was used to identify all the flood situations between 2014 to 2018 and compute their characteristics. The flood events between the 3rd and 5th of April 2016 were quantified. Based on the results, the flood that occurred during this period was only severe in the western part of Viti Levu, Fiji. Moreover, the analysis of the five severest floods in Nadi indicated that the most significant flood reached a total severity of 157.28 and lasted 48 hours. This flood started on 29th January 2014 at 8 a.m. and reached a peak danger of 6.80. This was also the most severe flood event with the most prolonged duration for the 5-year period among all the study sites. The results also showed that between 2014 and 2018, Tavua recorded 61 flood situations, followed by Nadi with 51, while Nasinu recorded only 41 flood events. Thus, these primary results demonstrate the usefulness of SWRI24-hr-S in identifying flood situations and determining the severity, duration, and intensity of flood events on an hourly scale.
For more information, please email the Graduate Research School or phone 07 4631 1088.