East Asian regions in the North Pacific have recently experienced severe riverine flood disasters. State-of-the-art neural networks are currently utilized as a quick-response flood model. Neural networks typically require ample time in the training process because of the use of numerous datasets. To reduce the computational costs, we introduced a transfer-learning approach to a neural-network-based flood model. For a concept of transfer leaning, once the model is pretrained in a source domain with large datasets, it can be reused in other target domains. After retraining parts of the model with the target domain datasets, the training time can be reduced due to reuse. A convolutional neural network (CNN) was employed because the CNN with transfer learning has numerous successful applications in two-dimensional image classification. However, our flood model predicts time-series variables (e.g., water level). The CNN with transfer learning requires a conversion tool from time-series datasets to image datasets in preprocessing. First, the CNN time-series classification was verified in the source domain with less than 10% errors for the variation in water level. Second, the CNN with transfer learning in the target domain efficiently reduced the training time by 1/5 of and a mean error difference by 15% of those obtained by the CNN without transfer learning, respectively. Our method can provide another novel flood model in addition to physical-based models.
15 September 2008 in Limnology and Oceanography: Methods
An automated river-estuary discharge imaging system (AREDIS) is developed to measure the velocity of a wide tidal estuary. The system contains near- and far-field cameras that capture the entire 370-m-wide channel from an oblique angle with high image resolution. A new rotational scheme is developed to calibrate the cameras. Wakes generated by flow past bridge piers are found to be natural tracers for large-scale particle image velocimetry (LSPIV) to reliably and accurately measure the surface velocity, confirmed by buoy tracking velocimetry (BTV). The success of AREDIS is demonstrated in three field measurements on the Danshui River, the largest estuary in Taiwan. First, AREDIS is used to measure discharge over the entire tidal cycle under normal flow, and a 10% difference is found between discharges measured with AREDIS and a boat-mounted acoustic Doppler profiler. Second, AREDIS is employed to measure discharge under a typhoon event, resulting in discharges 45% greater than those at normal flow...
29 November 2008 in Environmental Monitoring and Assessment
We studied how typhoon strength affects the daily dynamics of ecosystem metabolism of a subtropical alpine lake in Taiwan. We identified proximal agents of typhoon disturbance and assessed the resistance (the extent of change induced by a disturbance) and resilience (the rate of recovery after a disturbance) of lake metabolism to them. Gross primary production (GPP), ecosystem respiration (ER), and net ecosystem production were estimated from high-frequency dissolved oxygen data provided by an instrumented buoy. Typhoons resulted in significantly lower GPP (3%–81% decrease), and higher ER (7%–828% increase) compared with immediately before the events, and thus the lake became more heterotrophic (28%–852% increase in heterotrophy). The resistance and resilience of lake metabolism depended on the intensity of the typhoon. Smaller typhoons (with average daily accumulated precipitation (ADAP) < 200 mm·day–1) had greater effects on lake metabolism than medium (ADAP = 200–350 mm·day–1) and large (ADAP > 350 mm·day–1) typhoons. However, metabolism also recovered more quickly after smaller typhoons than after medium or larger typhoons. Typhoon effects on ecosystem metabolism is likely mediated by the magnitude and duration of typhoon-induced changes in lake mixing, the quantity and quality of dissolved organic carbon, and the biomass of primary producers.