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.

No abstract

No abstract

Drainage management in a complicated system in an agricultural lowland must operate pumps flexibly and quickly, based on the water level at the pumping station. A data-driven model without any physical-based information was implemented in a complicated drainage management system to predict the water level of a lagoon near a main drainage pumping station. We employed a long shortterm memory (LSTM) model as an advanced neural network model to utilize the field datasets obtained from water-related facilities and sensors over about eight years as model input data. We performed sensitivity tests for model accuracy with different types of data and locations of data using cross-validation with an error quantity between observed and predicted water levels at the main drainage pumping station. The results showed that the LSTM model with the input of all available datasets predicted better than the models using several parts of datasets or it was roughly equivalent to those for water levels over the entire observed period in 3-h and 6-h lead times. In addition, the LSTM with only inputs of the water level and rainfall observed by drainage pumping stations performed better for the observed subperiod, including the severest flood event.

No abstract