Preprints
https://doi.org/10.5194/hess-2024-261
https://doi.org/10.5194/hess-2024-261
07 Oct 2024
 | 07 Oct 2024
Status: this preprint is currently under review for the journal HESS.

Probabilistic Hierarchical Interpolation and Interpretable Configuration for Flood Prediction

Mostafa Saberian, Vidya Samadi, and Ioana Popescu

Abstract. The last few years have witnessed the rise of Neural Networks (NNs) applications for hydrological time series modeling. By virtue of their capabilities, NN models can achieve unprecedented levels of performance when learn how to solve increasingly complex rainfall-runoff processes via data, making them pivotal for the development of computational hydrologic tasks such as flood predictions. The NN models should, in order to be considered practical, provide a probabilistic understanding of the model mechanisms and predictions and hints on what could perturb the model. In this paper, we developed two probabilistic NN models, i.e., Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS) and Network-Based Expansion Analysis for Interpretable Time Series Forecasting (N-BEATS), and benchmarked them with long short-term memory (LSTM) for flood prediction across two headwater streams in Georgia and North Carolina, USA. To generate a probabilistic prediction, a Multi-Quantile Loss was used to assess the 95th percentile prediction uncertainty (95PPU) of multiple flooding events. We conducted extensive flood prediction experiments demonstrating the advantages of hierarchical interpolation and interpretable architecture, where both N-HiTS and N-BEATS provided an average accuracy improvement of almost 5 % (NSE) over the LSTM benchmarking model. On a variety of flooding events with different timing and magnitudes, both N-HiTS and N-BEATS demonstrated significant performance improvements over the LSTM benchmark and showcased their probabilistic predictions by specifying a likelihood parameter.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Mostafa Saberian, Vidya Samadi, and Ioana Popescu

Status: open (until 05 Dec 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • CC1: 'Comment on hess-2024-261', Nima Zafarmomen, 18 Oct 2024 reply
  • RC1: 'Comment on hess-2024-261', Anonymous Referee #1, 04 Nov 2024 reply
  • RC2: 'Comment on hess-2024-261', Anonymous Referee #2, 06 Nov 2024 reply
Mostafa Saberian, Vidya Samadi, and Ioana Popescu
Mostafa Saberian, Vidya Samadi, and Ioana Popescu

Viewed

Total article views: 316 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
206 71 39 316 4 6
  • HTML: 206
  • PDF: 71
  • XML: 39
  • Total: 316
  • BibTeX: 4
  • EndNote: 6
Views and downloads (calculated since 07 Oct 2024)
Cumulative views and downloads (calculated since 07 Oct 2024)

Viewed (geographical distribution)

Total article views: 308 (including HTML, PDF, and XML) Thereof 308 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 23 Nov 2024
Download
Short summary
Recent progress in neural network accelerated improvements in the performance of catchment modeling systems. Yet flood modeling remains a very difficult task. Focusing on two headwater streams, this paper developed N-HiTS and N-BEATS models and benchmarked them with LSTM to predict flooding events. Analysis suggested that both N-HiTS and N-BEATS outperformed LSTM for short-term (1-hour) flood predictions.