Transformer neural networks for interpretable flood forecasting

Castangia, Marco; Grajales, Lina Maria Medina; Aliberti, Alessandro; Rossi, Claudio; Macii, Alberto; Macii, Enrico; Patti, Edoardo

doi:10.1016/j.envsoft.2022.105581

Cited by 51 publications

(24 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Interpretable AI. Interpretable AI has accelerated the development of reliable models in water resource management (Ding et al, 2020;Castangia et al, 2023). Among them, many applications have been conducted based on a TFT, which has both improved interpretability and notable predictive performances (Civitarese et al, 2021;Castangia et al, 2023).…”

Section: Related Workmentioning

confidence: 99%

“…Interpretable AI has accelerated the development of reliable models in water resource management (Ding et al, 2020;Castangia et al, 2023). Among them, many applications have been conducted based on a TFT, which has both improved interpretability and notable predictive performances (Civitarese et al, 2021;Castangia et al, 2023). The TFT is a transformer-based model that quantifies the importance of variables with a variable selection network and attention mechanism, thereby enhancing interpretability.…”

Section: Related Workmentioning

confidence: 99%

“…In developing a water level forecasting model, the need for the model interpretability has been raised in the same context (Aguilera et al, 2019;Ding et al, 2020;Castangia et al, 2023). The model interpretability is required, especially when the forecasting model provides a differing outcome from the expert's opinion or physics law.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Interpretable Water Level Forecaster with Spatiotemporal Causal Attention Mechanisms

Hong¹,

Choi²,

Jeon³

2023

Preprint

View full text Add to dashboard Cite

Forecasting the water level of the Han river is essential to control traffic and avoid natural disasters. There are many variables related to the Han river, and they are intricately connected. In this work, we propose a novel transformer that exploits the causal relationship based on the prior knowledge among the variables and forecasts the water level at the Jamsu bridge in the Han river. Our proposed model considers spatial and temporal causation by formalizing the causal structure as a multilayer network and using masking methods. Due to this approach, we can have interpretability that consistent with prior knowledge. In real data analysis, we use the Han river dataset from 2016 to 2021 and compare the proposed model with deep learning models.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Interpretable Water Level Forecaster with Spatiotemporal Causal Attention Mechanisms

Hong¹,

Choi²,

Jeon³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…As one of the most popular models, CNNs have the advantages of offering 1) powerful selflearning ability, 2) high processing efficiency for multipledimensional data, and 3) self-adaptability (Krizhevsky et al, 2012;Oquab et al, 2014;LeCun et al, 2015). These models can be potentially beneficial in geoscience studies and have been successfully used in object detection (Salberg, 2015;Liu et al, 2016;Long et al, 2017;Zhao et al, 2019;Santana et al, 2022), classification (Castelluccio et al, 2015;Luus et al, 2015;Chen et al, 2016;Masoumi, 2021), extreme weather prediction (Gorricha et al, 2013;Zhuang and Ding, 2016;Castangia et al, 2023), etc. In addition, these models are also used for predicting marine variables.…”

Section: Introductionmentioning

confidence: 99%

Monthly extended ocean predictions based on a convolutional neural network via the transfer learning method

Miao

Zhang

et al. 2023

Front. Mar. Sci.

View full text Add to dashboard Cite

Sea surface temperature anomalies (SSTAs) and sea surface height anomalies (SSHAs) are indispensable parts of scientific research, such as mesoscale eddy, current, ocean-atmosphere interaction and so on. Nowadays, extended-range predictions of ocean dynamics, especially in SSTA and SSHA, can provide daily prediction services in the range of 30 days, which bridges the gap between synoptic-scale weather forecasts and monthly average scale climate predictions. However, the forecast efficiency of extended range remains problematic. With the development of ocean reanalysis and satellite remote sensing products, large amounts datasets provide an unprecedented opportunity to use big data for the extended range prediction of ocean dynamics. In this study, a hybrid model, combing convolutional neural network (CNN) model with transfer learning (TL), was established to predict SSTA and SSHA at monthly scales, which makes full use of these data resources that arise from delayed gridding reanalysis products and real-time satellite remote sensing observations. The proposed model, where both ocean and atmosphere reanalysis datasets serve as the pretraining dataset and the satellite remote sensing observations are employed for fine-tuning based on the transfer learning (TL) method, can effectively capture the evolving spatial characteristics of SSTAs and SSHAs with low prediction errors over the 30 days range. When the forecast lead time is 30 days, the root means square errors for the SSTAs and SSHAs model results are 0.32°C and 0.027 m in the South China Sea, respectively, indicating that this model has not only satisfactory prediction performance but also offers great potential for practical operational applications in improving the skill of extended-range predictions.

show abstract

“…Research that has been conducted by [11]- [13] is not up to making a prediction table for flood events even though the results of training and testing the learning process for past flood events were quite good. Likewise, research conducted by [14]- [19] also still has the same shortcomings in the research he does, namely not carrying out a simulation of making a flood prediction table as an illustration for future conditions.…”

Section: Introductionmentioning

confidence: 99%

Prediction of flood events in the city of Bandar Lampung using the artificial neural network

2023

View full text Add to dashboard Cite

The city of Bandar Lampung is currently experiencing seasonal flooding which occurs almost every year, resulting in significant losses. Floods recorded by BNPB in the last 10 years there were 16 incidents of flooding in the Bandar Lampung area. More than 14,000 people suffered, more than 500 people had to be evacuated, more than 900 houses were damaged, and 4 public facilities were damaged. To study the pattern of flood events in the past, the Artificial Neural Network Backpropagation learning method will be used which will utilize its non-linear variable learning abilities. The configuration settings for the Artificial Neural Network were carried out experimentally without any basis for assigning values, especially for the parameters of the number of hidden layers, number of neurons, and epochs used in training and variable testing. The results obtained from this study are the results of training and testing of datasets that have been carried out by ANN backpropagation are able to properly study patterns of flood events and also non-flood events in the dataset, this is evidenced by the results of high model configuration accuracy and also the results of predictive tables that able to describe actual conditions, setting the configuration model experimentally is able to produce an accuracy value of 90-100%, an average training correlation value of 0.96 and an average test correlation value of 0.89, and an average error value of 0.0089 out of 20 model configuration, and the flood prediction table are made based on the 1 best configuration with a training and testing accuracy rate of 100% with an error value of 0.00134, namely configuration model 20, the prediction table uses an average air temperature of 27˚C with 80% humidity. The prediction table is able to produce excellent flood potential results which are able to represent flood events as well as non-flood events based on the results of the dataset learning.

show abstract

Transformer neural networks for interpretable flood forecasting

Cited by 51 publications

References 24 publications

Interpretable Water Level Forecaster with Spatiotemporal Causal Attention Mechanisms

Interpretable Water Level Forecaster with Spatiotemporal Causal Attention Mechanisms

Monthly extended ocean predictions based on a convolutional neural network via the transfer learning method

Prediction of flood events in the city of Bandar Lampung using the artificial neural network

Contact Info

Product

Resources

About