Identifying major drivers of daily streamflow from large-scale atmospheric circulation with machine learning DOI Creative Commons
Jenny Sjåstad Hagen, Étienne Leblois,

Deborah Lawrence

et al.

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 596, P. 126086 - 126086

Published: Feb. 23, 2021

Previous studies linking large-scale atmospheric circulation and river flow with traditional machine learning techniques have predominantly explored monthly, seasonal or annual streamflow modelling for applications in direct downscaling hydrological climate-impact studies. This paper identifies major drivers of daily from using two reanalysis datasets six catchments Norway representing various Köppen-Geiger climate types flood-generating processes. A nested loop roughly pruned random forests is used feature extraction, demonstrating the potential automated retrieval physically consistent interpretable input variables. Random forest (RF), support vector (SVM) regression multilayer perceptron (MLP) neural networks are compared to multiple-linear assess role model complexity utilizing identified reconstruct streamflow. The models were trained on 31 years aggregated data distinct moving windows each catchment, reflecting catchment-specific forcing-response relationships between atmosphere rivers. results show that accuracy improves some extent complexity. In all but smallest, rainfall-driven most complex model, MLP, gives a Nash-Sutcliffe Efficiency (NSE) ranging 0.71 0.81 testing spanning five years. poorer performance by smallest catchment discussed relation characteristics, sub-grid topography local variability. intra-model differences also viewed consistency automatically retrieved selections datasets. study provides benchmark future development deep variables Norway.

Language: Английский

COVID-19 Pandemic Prediction for Hungary; a Hybrid Machine Learning Approach DOI Creative Commons
Gergő Pintér, Imre Felde,

Amir Mosavi

et al.

Research Square (Research Square), Journal Year: 2020, Volume and Issue: unknown

Published: May 6, 2020

Abstract Several epidemiological models are being used around the world to project number of infected individuals and mortality rates COVID-19 outbreak. Advancing accurate prediction is utmost importance take proper actions. Due a high level uncertainty or even lack essential data, standard have been challenged regarding delivery higher accuracy for long-term prediction. As an alternative susceptible-infected-resistant (SIR)-based models, this study proposes hybrid machine learning approach predict we exemplify its potential using data from Hungary. The methods adaptive network-based fuzzy inference system (ANFIS) multi-layered perceptron-imperialist competitive algorithm (MLP-ICA) time series rate. that by late May, outbreak total morality will drop substantially. validation performed nine days with promising results, which confirms model accuracy. It expected maintains as long no significant interruption occurs. Based on results reported here, due complex nature variation in behavior nation-to-nation, suggests effective tool This paper provides initial benchmarking demonstrate future research.

Language: Английский

Citations

64
Separating the Impacts of Climate Change and Human Activities on Runoff: A Review of Method and Application DOI Open Access

F. J. Zeng,

Mingguo Ma, Dongrui Di

et al.

Water, Journal Year: 2020, Volume and Issue: 12(8), P. 2201 - 2201

Published: Aug. 5, 2020

Separating the impact of climate change and human activities on runoff is an important topic in hydrology, a large number methods theories have been widely used. In this paper, we review current papers separating impacts runoff, summarize progress relevant research applications recent years, discuss future needs directions.

Language: Английский

Citations

56
A Comprehensive Review of Deep Learning Applications in Hydrology and Water Resources DOI Open Access
Muhammed Sit, Bekir Zahit Demiray, Zhongrun Xiang

et al.

EarthArXiv (California Digital Library), Journal Year: 2020, Volume and Issue: unknown

Published: June 17, 2020

The global volume of digital data is expected to reach 175 zettabytes by 2025. volume, variety, and velocity water-related are increasing due large-scale sensor networks increased attention topics such as disaster response, water resources management, climate change. Combined with the growing availability computational popularity deep learning, these transformed into actionable practical knowledge, revolutionizing industry. In this article, a systematic review literature conducted identify existing research which incorporates learning methods in sector, regard monitoring, governance communication resources. study provides comprehensive state-of-the-art approaches used industry for generation, prediction, enhancement, classification tasks, serves guide how utilize available future challenges. Key issues challenges application techniques domain discussed, including ethics technologies decision-making management governance. Finally, we provide recommendations directions models hydrology

Language: Английский

Citations

55
Prediction of Combine Harvester Performance Using Hybrid Machine Learning Modeling and Response Surface Methodology DOI
Tarahom Mesri Gundoshmian, Sina Ardabili,

Amir Mosavi

et al.

Lecture notes in networks and systems, Journal Year: 2020, Volume and Issue: unknown, P. 345 - 360

Published: Jan. 1, 2020

Language: Английский

Citations

54
Identifying major drivers of daily streamflow from large-scale atmospheric circulation with machine learning DOI Creative Commons
Jenny Sjåstad Hagen, Étienne Leblois,

Deborah Lawrence

et al.

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 596, P. 126086 - 126086

Published: Feb. 23, 2021

Previous studies linking large-scale atmospheric circulation and river flow with traditional machine learning techniques have predominantly explored monthly, seasonal or annual streamflow modelling for applications in direct downscaling hydrological climate-impact studies. This paper identifies major drivers of daily from using two reanalysis datasets six catchments Norway representing various Köppen-Geiger climate types flood-generating processes. A nested loop roughly pruned random forests is used feature extraction, demonstrating the potential automated retrieval physically consistent interpretable input variables. Random forest (RF), support vector (SVM) regression multilayer perceptron (MLP) neural networks are compared to multiple-linear assess role model complexity utilizing identified reconstruct streamflow. The models were trained on 31 years aggregated data distinct moving windows each catchment, reflecting catchment-specific forcing-response relationships between atmosphere rivers. results show that accuracy improves some extent complexity. In all but smallest, rainfall-driven most complex model, MLP, gives a Nash-Sutcliffe Efficiency (NSE) ranging 0.71 0.81 testing spanning five years. poorer performance by smallest catchment discussed relation characteristics, sub-grid topography local variability. intra-model differences also viewed consistency automatically retrieved selections datasets. study provides benchmark future development deep variables Norway.

Language: Английский

Citations

52