{"title":"Exploring the influence of training sampling strategies on time-series deep learning model in hydrology","authors":"Sunghyun Yoon , Kuk-Hyun Ahn","doi":"10.1016/j.jhydrol.2025.132774","DOIUrl":null,"url":null,"abstract":"<div><div>Numerous deep-learning models have been introduced to achieve reliable predictions in hydrology. In particular, recent works have shown that leveraging abundant training samples significantly improves the generalization performance due to the increased opportunity to learn fundamental processes from the samples. However, these studies often overlook the diverse levels of importance uncertainty in sample datasets. In our exploration of the rainfall-runoff relationship through modeling, we show that applying heterogeneous weights to training samples could yield even more enhanced predictions. We systematically examine the usefulness of the curriculum learning (CL)-based approach, which can dynamically assign weights to each training sample based on the concept of easy and hard samples. We particularly develop five distinct training strategies that prioritize different subsets of the training samples. Results demonstrate that employing the “easy-preferred mode” strategy—gradually increasing the complexity of data samples from easier ones—yields improved predictions compared to a random sampling strategy, similar to those prevalent in many studies. Specifically, the “easy-hard mode” strategy, allowing a shift in priority mode from “easy-first” to “hard-first” during the training process, exhibits the most remarkable performance. Lastly, our experiment also highlights that utilizing heterogeneous weights, known as a soft weighting scheme, for each sample proves more effective in enhancing the model’s performance compared to employing binary weights, referred to as a hard weighting scheme, for distinguishing between easy and hard samples in the proposed sampling strategy. Overall, our findings support the considerable benefits of employing the training sampling strategy for deep-learning models in hydrological tasks, ultimately enhancing prediction accuracy and contributing to improved water resource management and related policy-making.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"653 ","pages":"Article 132774"},"PeriodicalIF":5.9000,"publicationDate":"2025-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Hydrology","FirstCategoryId":"89","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S002216942500112X","RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENGINEERING, CIVIL","Score":null,"Total":0}
引用次数: 0
Abstract
Numerous deep-learning models have been introduced to achieve reliable predictions in hydrology. In particular, recent works have shown that leveraging abundant training samples significantly improves the generalization performance due to the increased opportunity to learn fundamental processes from the samples. However, these studies often overlook the diverse levels of importance uncertainty in sample datasets. In our exploration of the rainfall-runoff relationship through modeling, we show that applying heterogeneous weights to training samples could yield even more enhanced predictions. We systematically examine the usefulness of the curriculum learning (CL)-based approach, which can dynamically assign weights to each training sample based on the concept of easy and hard samples. We particularly develop five distinct training strategies that prioritize different subsets of the training samples. Results demonstrate that employing the “easy-preferred mode” strategy—gradually increasing the complexity of data samples from easier ones—yields improved predictions compared to a random sampling strategy, similar to those prevalent in many studies. Specifically, the “easy-hard mode” strategy, allowing a shift in priority mode from “easy-first” to “hard-first” during the training process, exhibits the most remarkable performance. Lastly, our experiment also highlights that utilizing heterogeneous weights, known as a soft weighting scheme, for each sample proves more effective in enhancing the model’s performance compared to employing binary weights, referred to as a hard weighting scheme, for distinguishing between easy and hard samples in the proposed sampling strategy. Overall, our findings support the considerable benefits of employing the training sampling strategy for deep-learning models in hydrological tasks, ultimately enhancing prediction accuracy and contributing to improved water resource management and related policy-making.
期刊介绍:
The Journal of Hydrology publishes original research papers and comprehensive reviews in all the subfields of the hydrological sciences including water based management and policy issues that impact on economics and society. These comprise, but are not limited to the physical, chemical, biogeochemical, stochastic and systems aspects of surface and groundwater hydrology, hydrometeorology and hydrogeology. Relevant topics incorporating the insights and methodologies of disciplines such as climatology, water resource systems, hydraulics, agrohydrology, geomorphology, soil science, instrumentation and remote sensing, civil and environmental engineering are included. Social science perspectives on hydrological problems such as resource and ecological economics, environmental sociology, psychology and behavioural science, management and policy analysis are also invited. Multi-and interdisciplinary analyses of hydrological problems are within scope. The science published in the Journal of Hydrology is relevant to catchment scales rather than exclusively to a local scale or site.