Environmental Modelling & Software最新文献_第7页

RegreSSM: A novel software tool for downscaling the SMAP L3 soil moisture operational product utilizing the Ts/VI feature space and Sentinel-3 data 回归：一种利用Ts/VI特征空间和Sentinel-3数据对SMAP L3土壤湿度操作产品进行降尺度处理的新型软件工具

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-17 DOI: 10.1016/j.envsoft.2025.106836

George P. Petropoulos , Spyridon E. Detsikas , Vasileios Anagnostopoulos , Christina Lekka

Herein we present RegreSSM, a software tool that enables the downscaling of SMAP L3 Surface Soil Moisture (SSM) operational product from 36 km to 1 km by the fusion of optical and thermal data retrieved from Sentinel-3 platform. The downscaling method is based on the well-established properties of the Ts/VI feature space. Most of the existing soil moisture downscaling methods are computationally complex, require advanced expertise, and lack standalone tools suitable for operational or non-expert use. To address these limitations, this study proposes a simple and accessible framework for generating high-resolution SSM maps using only land surface temperature and vegetation cover as inputs. The tool has been developed in python programming language as a stand-alone application and can be executed in any operational system. The application offers automated and reproducible workflows for spatiotemporal matching and processing of SMAP L3 SSM products and Sentinel-3 dataset. The software tool's practical application is demonstrated over the Iberian Peninsula, where validation of the SMAP L3 product performed for all calendar year 2022 using in-situ observations from the REMEDHUS operational network stations. Results showed a satisfactory retrieval of SSM with a small average bias of 0.01 m³/m³, a MAD of 0.06 m³/m³, a RMSD of 0.07 m³/m³, and a satisfactory R² of 0.63, confirming the ability of the proposed downscaling framework and RegreSSM to retrieve SSM at the 1 km spatial resolution. Results obtained herein were also compared to the validation metrics reported for operational RS-based SSM products, with typically reported uncertainty of 0.04 m³/m³. The availability of RegreSSM to the SSM users' community consists an important step towards the standardization of downscaling procedures as well as bridging the spatial gap of existing operational SM products to the requirements of the fine-scale applications. It also contributes towards advancing the deployment of geo-processing tools utilizing the synergies between state-of-the-art methods and RS data available today from the most sophisticated satellites in orbit.

本文提出了一种软件工具RegreSSM，通过融合Sentinel-3平台检索的光学和热数据，可以将SMAP L3表层土壤湿度（SSM）业务产品从36公里降尺度到1公里。该降尺度方法是基于已建立的Ts/VI特征空间的特性。大多数现有的土壤湿度降尺度方法计算复杂，需要高级专业知识，并且缺乏适合操作或非专业使用的独立工具。为了解决这些限制，本研究提出了一个简单易用的框架，用于仅使用地表温度和植被覆盖作为输入来生成高分辨率SSM地图。该工具是用python编程语言开发的独立应用程序，可以在任何操作系统中执行。该应用程序为SMAP L3 SSM产品和Sentinel-3数据集的时空匹配和处理提供了自动化和可重复的工作流程。该软件工具的实际应用在伊比利亚半岛进行了演示，在那里，通过REMEDHUS操作网络站的现场观测，对2022年全年的SMAP L3产品进行了验证。结果表明，反演结果令人满意，平均偏差为0.01 m3/m3， MAD为0.06 m3/m3， RMSD为0.07 m3/m3， R2为0.63，证实了所提出的降尺度框架和回归模型在1 km空间分辨率下反演SSM的能力。本文获得的结果也与运行RS-based SSM产品报告的验证指标进行了比较，通常报告的不确定性为0.04 m3/m3。回归模型对SSM用户社区的可用性是实现降尺度过程标准化的重要一步，同时也弥补了现有可操作SM产品与精细尺度应用需求之间的空间差距。它还有助于利用最先进的方法与目前最先进的在轨卫星提供的遥感数据之间的协同作用，推进地理处理工具的部署。

{"title":"RegreSSM: A novel software tool for downscaling the SMAP L3 soil moisture operational product utilizing the Ts/VI feature space and Sentinel-3 data","authors":"George P. Petropoulos , Spyridon E. Detsikas , Vasileios Anagnostopoulos , Christina Lekka","doi":"10.1016/j.envsoft.2025.106836","DOIUrl":"10.1016/j.envsoft.2025.106836","url":null,"abstract":"<div><div>Herein we present RegreSSM, a software tool that enables the downscaling of SMAP L3 Surface Soil Moisture (SSM) operational product from 36 km to 1 km by the fusion of optical and thermal data retrieved from Sentinel-3 platform. The downscaling method is based on the well-established properties of the Ts/VI feature space. Most of the existing soil moisture downscaling methods are computationally complex, require advanced expertise, and lack standalone tools suitable for operational or non-expert use. To address these limitations, this study proposes a simple and accessible framework for generating high-resolution SSM maps using only land surface temperature and vegetation cover as inputs. The tool has been developed in python programming language as a stand-alone application and can be executed in any operational system. The application offers automated and reproducible workflows for spatiotemporal matching and processing of SMAP L3 SSM products and Sentinel-3 dataset. The software tool's practical application is demonstrated over the Iberian Peninsula, where validation of the SMAP L3 product performed for all calendar year 2022 using in-situ observations from the REMEDHUS operational network stations. Results showed a satisfactory retrieval of SSM with a small average bias of 0.01 m<sup>3</sup>/m<sup>3</sup>, a MAD of 0.06 m<sup>3</sup>/m<sup>3</sup>, a RMSD of 0.07 m<sup>3</sup>/m<sup>3</sup>, and a satisfactory R<sup>2</sup> of 0.63, confirming the ability of the proposed downscaling framework and RegreSSM to retrieve SSM at the 1 km spatial resolution. Results obtained herein were also compared to the validation metrics reported for operational RS-based SSM products, with typically reported uncertainty of 0.04 m<sup>3</sup>/m<sup>3</sup>. The availability of RegreSSM to the SSM users' community consists an important step towards the standardization of downscaling procedures as well as bridging the spatial gap of existing operational SM products to the requirements of the fine-scale applications. It also contributes towards advancing the deployment of geo-processing tools utilizing the synergies between state-of-the-art methods and RS data available today from the most sophisticated satellites in orbit.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106836"},"PeriodicalIF":4.6,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785085","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Meteorological observation research based on an improved EfficientNetV2 model 基于改进的EfficientNetV2模型的气象观测研究

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-16 DOI: 10.1016/j.envsoft.2025.106835

Haozheng Yin , Yu Cao , Linlin Liu , Dan Chen , Qiong Zhang

Meteorological observation plays a critical role in ensuring safety, promoting agricultural development, optimizing energy management, and achieving sustainable development. Although image recognition methods based on deep learning have made notable progress, existing models still face challenges in complex weather scenarios, such as insufficient feature extraction, inadequate utilization of scale information, and poor robustness to interference. To address these issues, this study proposes a novel deep learning model based on EfficientNetV2-CBAM-PANet. By leveraging transfer learning, the training efficiency and accuracy of the EfficientNetV2 pretrained model are enhanced. The integration of the CBAM attention mechanism improves the perception of meteorological features, while the PANet structure enables multi-level feature fusion. Experimental results demonstrate that the proposed model achieves a recognition accuracy of 97.6% on a self-constructed dataset, indicating strong classification capability across various weather conditions and providing useful insights for future research in weather image classification and forecasting.

气象观测在保障安全、促进农业发展、优化能源管理、实现可持续发展等方面发挥着重要作用。尽管基于深度学习的图像识别方法取得了显著进展，但现有模型在复杂天气场景下仍然面临着特征提取不足、尺度信息利用不足、抗干扰性差等挑战。为了解决这些问题，本研究提出了一种基于EfficientNetV2-CBAM-PANet的新型深度学习模型。利用迁移学习，提高了EfficientNetV2预训练模型的训练效率和准确性。CBAM注意机制的集成提高了对气象特征的感知，而PANet结构实现了多层次特征融合。实验结果表明，该模型在自构建数据集上的识别准确率达到97.6%，表明该模型具有较强的天气分类能力，为未来天气图像分类预报的研究提供了有益的见解。

{"title":"Meteorological observation research based on an improved EfficientNetV2 model","authors":"Haozheng Yin , Yu Cao , Linlin Liu , Dan Chen , Qiong Zhang","doi":"10.1016/j.envsoft.2025.106835","DOIUrl":"10.1016/j.envsoft.2025.106835","url":null,"abstract":"<div><div>Meteorological observation plays a critical role in ensuring safety, promoting agricultural development, optimizing energy management, and achieving sustainable development. Although image recognition methods based on deep learning have made notable progress, existing models still face challenges in complex weather scenarios, such as insufficient feature extraction, inadequate utilization of scale information, and poor robustness to interference. To address these issues, this study proposes a novel deep learning model based on EfficientNetV2-CBAM-PANet. By leveraging transfer learning, the training efficiency and accuracy of the EfficientNetV2 pretrained model are enhanced. The integration of the CBAM attention mechanism improves the perception of meteorological features, while the PANet structure enables multi-level feature fusion. Experimental results demonstrate that the proposed model achieves a recognition accuracy of 97.6% on a self-constructed dataset, indicating strong classification capability across various weather conditions and providing useful insights for future research in weather image classification and forecasting.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106835"},"PeriodicalIF":4.6,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of a River Dynamical Core for E3SM to simulate compound flooding on Exascale-class heterogeneous supercomputers E3SM河流动力核在百亿亿级异构超级计算机上模拟复合驱的开发

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-16 DOI: 10.1016/j.envsoft.2025.106804

Gautam Bisht , Donghui Xu , Jeffrey Johnson , Jed Brown , Matthew Knepley , Mark Adams , Dongyu Feng , Dalei Hao , Darren Engwirda , Mukesh Kumar , Zeli Tan

Physically-consistent quantification of flood risks in global models requires kilometer scale flood simulations using 2D physics schemes, both of which are unavailable in the current generation Earth System Models. In this work, we have developed the River Dynamical Core (RDycore), which is an open-source, 2D shallow water equation library for the Energy Exascale Earth System Model (E3SM). RDycore uses PETSc and libCEED libraries to run efficiently on CPUs and GPUs. RDycore was validated for analytical, manufactured, and a well-studied dam break problem. For a problem with 471 million grid cells, RDycore achieves a speedup of 6.6x and 7.6x on GPUs compared to CPUs on Perlmutter and Frontier supercomputers, respectively. The one-way E3SM-RDycore coupling is demonstrated by performing multiple 5-day flooding simulations during Hurricane Harvey driven by five precipitation datasets. The work presented here is the foundational step in providing hardware and algorithmic portability for simulating kilometer-scale river dynamics within E3SM.

全球模型中洪水风险的物理一致性量化需要使用二维物理方案进行千米尺度的洪水模拟，而这两种方案在当前的地球系统模型中都是不可用的。在这项工作中，我们开发了河流动力核心（RDycore），这是一个开源的二维浅水方程库，用于能量百亿亿次地球系统模型（E3SM）。RDycore使用PETSc和libCEED库在cpu和gpu上高效运行。RDycore经过了分析、制造和充分研究的溃坝问题的验证。对于一个有4.71亿个网格单元的问题，RDycore在gpu上的速度比Perlmutter和Frontier超级计算机上的cpu分别提高了6.6倍和7.6倍。在飓风哈维期间，5个降水数据集驱动的多个5天洪水模拟证明了单向E3SM-RDycore耦合。本文提出的工作是在E3SM中为模拟千米尺度的河流动态提供硬件和算法可移植性的基础步骤。

{"title":"Development of a River Dynamical Core for E3SM to simulate compound flooding on Exascale-class heterogeneous supercomputers","authors":"Gautam Bisht , Donghui Xu , Jeffrey Johnson , Jed Brown , Matthew Knepley , Mark Adams , Dongyu Feng , Dalei Hao , Darren Engwirda , Mukesh Kumar , Zeli Tan","doi":"10.1016/j.envsoft.2025.106804","DOIUrl":"10.1016/j.envsoft.2025.106804","url":null,"abstract":"<div><div>Physically-consistent quantification of flood risks in global models requires kilometer scale flood simulations using 2D physics schemes, both of which are unavailable in the current generation Earth System Models. In this work, we have developed the River Dynamical Core (RDycore), which is an open-source, 2D shallow water equation library for the Energy Exascale Earth System Model (E3SM). RDycore uses PETSc and libCEED libraries to run efficiently on CPUs and GPUs. RDycore was validated for analytical, manufactured, and a well-studied dam break problem. For a problem with 471 million grid cells, RDycore achieves a speedup of 6.6x and 7.6x on GPUs compared to CPUs on Perlmutter and Frontier supercomputers, respectively. The one-way E3SM-RDycore coupling is demonstrated by performing multiple 5-day flooding simulations during Hurricane Harvey driven by five precipitation datasets. The work presented here is the foundational step in providing hardware and algorithmic portability for simulating kilometer-scale river dynamics within E3SM.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106804"},"PeriodicalIF":4.6,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785089","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Soft computing techniques for atmospheric pollution and traffic emission prediction 大气污染与交通排放预测的软计算技术

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-16 DOI: 10.1016/j.envsoft.2025.106828

Vivek Mathur , Divya Srivastava , Jaya Pandey , Vivek Mishra

Air pollution, especially from traffic emissions, poses a significant threat to public health and environmental sustainability. Traditional monitoring systems are resource-intensive, prompting a shift toward computational forecasting techniques. This mini-review evaluates the application of soft computing methods—such as Artificial Neural Networks (ANN), Fuzzy Logic, Genetic Algorithms, and hybrid models—for atmospheric pollution and traffic emission prediction. These models offer flexibility, adaptability, and high accuracy in handling nonlinear and uncertain data. The review compares model architectures, input features, and performance metrics, emphasizing the superior predictive ability of hybrid and deep learning models. Additionally, the potential integration of these models with IoT and smart city frameworks is discussed. Key limitations, including lack of model generalizability and uncertainty handling, are highlighted alongside suggestions for future improvement. This work provides a concise overview of emerging data-driven strategies for air quality forecasting, offering direction for researchers and policymakers in sustainable urban planning.

空气污染，特别是交通排放造成的空气污染，对公众健康和环境可持续性构成重大威胁。传统的监测系统是资源密集型的，促使向计算预测技术的转变。这篇小型综述评估了软计算方法的应用，如人工神经网络（ANN）、模糊逻辑、遗传算法和混合模型，用于大气污染和交通排放预测。这些模型在处理非线性和不确定数据方面具有灵活性、适应性和高精度。该综述比较了模型架构、输入特征和性能指标，强调了混合和深度学习模型的卓越预测能力。此外，还讨论了这些模型与物联网和智慧城市框架的潜在集成。关键的限制，包括缺乏模型通用性和不确定性处理，强调了未来的改进建议。这项工作提供了新兴的数据驱动的空气质量预测策略的简要概述，为可持续城市规划的研究人员和政策制定者提供了方向。

{"title":"Soft computing techniques for atmospheric pollution and traffic emission prediction","authors":"Vivek Mathur , Divya Srivastava , Jaya Pandey , Vivek Mishra","doi":"10.1016/j.envsoft.2025.106828","DOIUrl":"10.1016/j.envsoft.2025.106828","url":null,"abstract":"<div><div>Air pollution, especially from traffic emissions, poses a significant threat to public health and environmental sustainability. Traditional monitoring systems are resource-intensive, prompting a shift toward computational forecasting techniques. This mini-review evaluates the application of soft computing methods—such as Artificial Neural Networks (ANN), Fuzzy Logic, Genetic Algorithms, and hybrid models—for atmospheric pollution and traffic emission prediction. These models offer flexibility, adaptability, and high accuracy in handling nonlinear and uncertain data. The review compares model architectures, input features, and performance metrics, emphasizing the superior predictive ability of hybrid and deep learning models. Additionally, the potential integration of these models with IoT and smart city frameworks is discussed. Key limitations, including lack of model generalizability and uncertainty handling, are highlighted alongside suggestions for future improvement. This work provides a concise overview of emerging data-driven strategies for air quality forecasting, offering direction for researchers and policymakers in sustainable urban planning.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106828"},"PeriodicalIF":4.6,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

GeoClimate intelligence platform: A web-based framework for environmental data analysis 地理气候情报平台：一个基于网络的环境数据分析框架

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-13 DOI: 10.1016/j.envsoft.2025.106826

Saurav Bhattarai , Nawa Raj Pradhan , Rocky Talchabhadel

Environmental science education faces a critical barrier: programming requirements prevent students, novice researchers, and domain experts from accessing planetary-scale datasets. This study presents the GeoClimate Intelligence Platform, a web-based framework powered by Google Earth Engine (GEE) that eliminates programming barriers while maintaining research-grade analytical capabilities. The platform comprises five integrated modules: GeoData Explorer for climate dataset access, Climate Analytics implementing 20+ ETCCDI-compliant climate indices, Hydrology Analyzer for precipitation analysis and return periods, Product Selector for dataset validation, and Data Visualizer for interactive analysis. This modular design supports integrated workflows while maintaining analytical independence across specialized functions. Development was motivated by workshops where students found programming barriers insurmountable despite strong motivation. Educational validation through university coursework demonstrated effectiveness. Performance evaluation shows robust scalability from educational to research-scale applications. The platform requires only a GEE account and operates through web browsers, eliminating software installation. This accessibility transformation enables broader participation in data-driven environmental problem-solving with scientific rigor, democratizing sophisticated environmental analysis for educational and research communities.

环境科学教育面临着一个关键的障碍：编程要求阻碍了学生、研究新手和领域专家访问行星尺度的数据集。本研究提出了地球气候情报平台，这是一个基于网络的框架，由谷歌地球引擎（GEE）提供支持，在保持研究级分析能力的同时消除了编程障碍。该平台包括五个集成模块：用于气候数据集访问的GeoData Explorer，实现20多个符合etccdi的气候指数的climate Analytics，用于降水分析和回归期的Hydrology Analyzer，用于数据集验证的Product Selector，以及用于交互式分析的Data Visualizer。这种模块化设计支持集成工作流，同时保持跨专门功能的分析独立性。开发是由研讨会推动的，学生们发现编程障碍难以克服，尽管有很强的动机。通过大学课程证明教育有效性。性能评估显示了从教育到研究规模应用的强大可扩展性。该平台只需要一个GEE账户，并通过网络浏览器操作，无需安装软件。这种可访问性转变使更广泛的参与到数据驱动的环境问题解决中，并具有科学的严谨性，使教育和研究社区的复杂环境分析民主化。

{"title":"GeoClimate intelligence platform: A web-based framework for environmental data analysis","authors":"Saurav Bhattarai , Nawa Raj Pradhan , Rocky Talchabhadel","doi":"10.1016/j.envsoft.2025.106826","DOIUrl":"10.1016/j.envsoft.2025.106826","url":null,"abstract":"<div><div>Environmental science education faces a critical barrier: programming requirements prevent students, novice researchers, and domain experts from accessing planetary-scale datasets. This study presents the GeoClimate Intelligence Platform, a web-based framework powered by Google Earth Engine (GEE) that eliminates programming barriers while maintaining research-grade analytical capabilities. The platform comprises five integrated modules: GeoData Explorer for climate dataset access, Climate Analytics implementing 20+ ETCCDI-compliant climate indices, Hydrology Analyzer for precipitation analysis and return periods, Product Selector for dataset validation, and Data Visualizer for interactive analysis. This modular design supports integrated workflows while maintaining analytical independence across specialized functions. Development was motivated by workshops where students found programming barriers insurmountable despite strong motivation. Educational validation through university coursework demonstrated effectiveness. Performance evaluation shows robust scalability from educational to research-scale applications. The platform requires only a GEE account and operates through web browsers, eliminating software installation. This accessibility transformation enables broader participation in data-driven environmental problem-solving with scientific rigor, democratizing sophisticated environmental analysis for educational and research communities.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106826"},"PeriodicalIF":4.6,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732496","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

On generalization, language, interpretability and the future of geo-scientific machine learning 论地球科学机器学习的泛化、语言、可解释性和未来

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-12 DOI: 10.1016/j.envsoft.2025.106834

Hoshin V. Gupta

There are several types of generalization ability that we may wish our models to be capable of. All but the most basic of these require the representation to be suitably interpretable so that it can provide meaningful support for scenario analysis, scientific reasoning, and decision making under system non-stationarity and model transfer. However interpretability of a model can only be meaningfully understood in the context of the ‘language’ used for its construction. In this regard it is important to recognize that, while machine-learning-based (MLB) models tend to prioritize accuracy and precision (in service of predictive performance) and physics-based (PB) models tend to emphasize physical/geo-scientific interpretability (in service of understanding), their learned representations are actually based in related but somewhat different languages, levels of linguistic abstraction, and grammatical rules.

Importantly, these differences are not fundamentally necessary. It is my opinion that the future of geo-scientific ML need not compromise accuracy and precision to achieve improved understanding. Instead, we must develop “telescopic” hierarchical representations that prioritize “learning from data” at their fundamental levels, while simultaneously enabling “geo-scientific abstraction” so that higher-level interpretable and understandable representations can be extracted by directed compression. Ultimately, the geosciences will benefit from a specific kind of “interpretable generative modeling” that can learn how to construct causal and/or understandable representations of the underlying physical data generating processes from data, and that can facilitate the kind of hierarchical, multi-level abstraction processes alluded to above.

有几种类型的泛化能力，我们可能希望我们的模型能够。除了最基本的之外，所有这些都要求表示具有适当的可解释性，以便它可以为系统非平稳性和模型转移下的场景分析、科学推理和决策提供有意义的支持。然而，模型的可解释性只能在用于其构建的“语言”上下文中被有意义地理解。在这方面，重要的是要认识到，虽然基于机器学习（MLB）的模型倾向于优先考虑准确性和精度（为预测性能服务），而基于物理的（PB）模型倾向于强调物理/地球科学的可解释性（为理解服务），但它们的学习表示实际上是基于相关但有些不同的语言，语言抽象水平和语法规则。重要的是，这些差异并不是根本必要的。我认为，地球科学机器学习的未来不需要牺牲准确性和精度来提高理解能力。相反，我们必须开发“可伸缩的”分层表示，优先考虑“从数据中学习”，同时实现“地球科学抽象”，以便通过定向压缩提取更高级别的可解释和可理解的表示。最终，地球科学将受益于一种特定类型的“可解释的生成建模”，它可以学习如何构建因果关系和/或可理解的基础物理数据生成过程的表示，并且可以促进上述提到的分层、多层次抽象过程。

{"title":"On generalization, language, interpretability and the future of geo-scientific machine learning","authors":"Hoshin V. Gupta","doi":"10.1016/j.envsoft.2025.106834","DOIUrl":"10.1016/j.envsoft.2025.106834","url":null,"abstract":"<div><div>There are several types of generalization ability that we may wish our models to be capable of. All but the most basic of these require the representation to be suitably interpretable so that it can provide meaningful support for scenario analysis, scientific reasoning, and decision making under system non-stationarity and model transfer. However interpretability of a model can only be meaningfully understood in the context of the ‘<em>language’</em> used for its construction. In this regard it is important to recognize that, while machine-learning-based (MLB) models tend to prioritize accuracy and precision (in service of predictive performance) and physics-based (PB) models tend to emphasize physical/geo-scientific interpretability (in service of understanding), their learned representations are actually based in related but somewhat different languages, levels of linguistic abstraction, and grammatical rules.</div><div>Importantly, these differences are not fundamentally necessary. It is my opinion that the future of geo-scientific ML need not compromise accuracy and precision to achieve improved understanding. Instead, we must develop “<em>telescopic</em>” hierarchical representations that prioritize “<em>learning from data</em>” at their fundamental levels, while simultaneously enabling “<em>geo-scientific abstraction</em>” so that higher-level interpretable and understandable representations can be extracted by directed compression. Ultimately, the geosciences will benefit from a specific kind of “<em>interpretable generative modeling</em>” that can learn how to construct causal and/or understandable representations of the underlying physical data generating processes from data, and that can facilitate the kind of hierarchical, multi-level abstraction processes alluded to above.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106834"},"PeriodicalIF":4.6,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732499","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Towards democratized flood risk management: An advanced AI assistant enabled by GPT-4 for enhanced interpretability and public engagement 走向民主化的洪水风险管理：由GPT-4支持的高级人工智能助手，以增强可解释性和公众参与

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-12 DOI: 10.1016/j.envsoft.2025.106821

Rafaela Martelo , Kimia Ahmadiyehyazdi , Ruo-Qian Wang

Traditional flood risk communication fails to bridge the gap between complex technical data and the needs of the public, hindering effective response. This research addresses this gap by developing and validating a novel AI-powered assistant that uses GPT-4 to democratize flood risk information. Our core methodology includes a Retrieval-Augmented Generation (RAG) framework that synthesizes real-time flood warnings, geospatial data, and social vulnerability indices into clear, conversational responses. To validate its effectiveness, we conducted a mixed-methods evaluation, including a comparison across different GPT models. Key quantitative findings reveal that the assistant achieved high performance scores in general flood knowledge (5/5) and handling flash flood alerts (4.3/5). Response times averaged a rapid 12 s for non-function-calling queries, though more complex data retrieval tasks averaged 36 s, highlighting areas for optimization. Our comparison identified GPT-4o as the optimal model for balancing accuracy with response time. The broader implications of this work demonstrate that large language models can serve as powerful tools to translate complex environmental data for non-experts, paving the way for more equitable, engaging, and effective public participation in disaster risk management.

传统的洪水风险沟通无法弥合复杂的技术数据与公众需求之间的差距，阻碍了有效的应对。本研究通过开发和验证一种新型人工智能助手来解决这一差距，该助手使用GPT-4来实现洪水风险信息的民主化。我们的核心方法包括检索-增强生成（RAG）框架，该框架将实时洪水预警、地理空间数据和社会脆弱性指数综合为清晰的对话响应。为了验证其有效性，我们进行了混合方法评估，包括不同GPT模型的比较。关键的定量研究结果显示，该助手在一般洪水知识（5/5）和处理山洪警报（4.3/5）方面取得了很高的成绩。非函数调用查询的平均响应时间为12秒，而更复杂的数据检索任务的平均响应时间为36秒，这突出了需要优化的领域。我们的比较确定gpt - 40是平衡精度和响应时间的最佳模型。这项工作的更广泛意义表明，大型语言模型可以作为强大的工具，为非专家翻译复杂的环境数据，为更公平、更有吸引力和更有效的公众参与灾害风险管理铺平道路。

{"title":"Towards democratized flood risk management: An advanced AI assistant enabled by GPT-4 for enhanced interpretability and public engagement","authors":"Rafaela Martelo , Kimia Ahmadiyehyazdi , Ruo-Qian Wang","doi":"10.1016/j.envsoft.2025.106821","DOIUrl":"10.1016/j.envsoft.2025.106821","url":null,"abstract":"<div><div>Traditional flood risk communication fails to bridge the gap between complex technical data and the needs of the public, hindering effective response. This research addresses this gap by developing and validating a novel AI-powered assistant that uses GPT-4 to democratize flood risk information. Our core methodology includes a Retrieval-Augmented Generation (RAG) framework that synthesizes real-time flood warnings, geospatial data, and social vulnerability indices into clear, conversational responses. To validate its effectiveness, we conducted a mixed-methods evaluation, including a comparison across different GPT models. Key quantitative findings reveal that the assistant achieved high performance scores in general flood knowledge (5/5) and handling flash flood alerts (4.3/5). Response times averaged a rapid 12 s for non-function-calling queries, though more complex data retrieval tasks averaged 36 s, highlighting areas for optimization. Our comparison identified GPT-4o as the optimal model for balancing accuracy with response time. The broader implications of this work demonstrate that large language models can serve as powerful tools to translate complex environmental data for non-experts, paving the way for more equitable, engaging, and effective public participation in disaster risk management.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106821"},"PeriodicalIF":4.6,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732498","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Integrating field surveys and visual interpretation to enhance CSLE model of soil erosion response to LUCC in Southwest China 结合野外调查和目视解译改进西南地区土地利用变化对土壤侵蚀响应的CSLE模型

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-11 DOI: 10.1016/j.envsoft.2025.106831

Rui Tan , Geng Guo , Kaiwen Huang , Zicheng Liu , Chaorui Wang , Jie Lin , Yizhong Huang

Absence of high-resolution spatial data on Soil and water conservation measures (SWCM) hampers the accuracy of erosion modeling, particularly in regions with complex terrain and frequent land use/cover changes (LUCC). This study integrated multi-source remote sensing (RS), field surveys, and visual interpretation to map SWCM distribution and estimate soil erosion. It further quantified the response of erosion to LUCC. Soil erosion conditions have improved, with an average annual decrease in erosion modulus of 0.51 % and a total reduction of approximately 9.5 × 10⁵ t. LUCC was characterized by cropland reduction, expansion of garden, and increasing landscape fragmentation. Garden development enhances economic returns but may exacerbate erosion when vegetation cover is insufficient. Nonetheless, under similar conservation intensity, slope, and elevation, conversion of cropland or bare land to woodland or garden effectively reduces erosion. The findings provide a new perspective for evaluating soil erosion in fragmented mountainous landscapes with complex management measures.

水土保持措施（SWCM）高分辨率空间数据的缺乏影响了侵蚀模型的准确性，特别是在地形复杂、土地利用/覆盖变化频繁的地区。本研究将多源遥感、野外调查和目视解译相结合，绘制SWCM分布和估算土壤侵蚀。进一步量化了侵蚀对土地利用变化的响应。土壤侵蚀条件得到改善，侵蚀模数年均减少0.51%，总减少量约9.5 × 105 t。土地利用变化呈现耕地减少、园林扩大、景观破碎化加剧的特征。园林的发展提高了经济效益，但当植被覆盖不足时，可能会加剧侵蚀。然而，在相似的保护强度、坡度和高程下，将耕地或裸地转化为林地或花园可以有效地减少侵蚀。研究结果为复杂管理条件下破碎化山地景观土壤侵蚀评价提供了新的视角。

{"title":"Integrating field surveys and visual interpretation to enhance CSLE model of soil erosion response to LUCC in Southwest China","authors":"Rui Tan , Geng Guo , Kaiwen Huang , Zicheng Liu , Chaorui Wang , Jie Lin , Yizhong Huang","doi":"10.1016/j.envsoft.2025.106831","DOIUrl":"10.1016/j.envsoft.2025.106831","url":null,"abstract":"<div><div>Absence of high-resolution spatial data on Soil and water conservation measures (SWCM) hampers the accuracy of erosion modeling, particularly in regions with complex terrain and frequent land use/cover changes (LUCC). This study integrated multi-source remote sensing (RS), field surveys, and visual interpretation to map SWCM distribution and estimate soil erosion. It further quantified the response of erosion to LUCC. Soil erosion conditions have improved, with an average annual decrease in erosion modulus of 0.51 % and a total reduction of approximately 9.5 × 10<sup>5</sup> t. LUCC was characterized by cropland reduction, expansion of garden, and increasing landscape fragmentation. Garden development enhances economic returns but may exacerbate erosion when vegetation cover is insufficient. Nonetheless, under similar conservation intensity, slope, and elevation, conversion of cropland or bare land to woodland or garden effectively reduces erosion. The findings provide a new perspective for evaluating soil erosion in fragmented mountainous landscapes with complex management measures.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106831"},"PeriodicalIF":4.6,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732500","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

FloodTransformer: Efficient real-time high-resolution flood forecasting 洪水变压器：高效的实时高分辨率洪水预报

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-11 DOI: 10.1016/j.envsoft.2025.106832

Zhanzhong Gu , Jiachen Kang , Wenzheng Jin , Feifei Tong , Y. Jay Guo , Wenjing Jia

Flood forecasting is crucial for disaster planning and risk management, yet conventional hydrodynamic-based approaches are often slow in response and computationally intensive. We present a hybrid framework leveraging traditional hydrodynamic modelling with a novel AI model to enable accurate, real-time, and high-resolution flood prediction. To address the computational challenges of large-scale, dense flood prediction, we develop an efficient flood prediction model, FloodTransformer, which possesses three key novelties: variable-size cell embedding, tokenised time-sequence encoding, and physics-informed multi-task optimisation. These components effectively capture complex spatiotemporal dependencies, allowing accurate sequential predictions in a single run. Comprehensive evaluations on both simulated and historical flood events demonstrate FloodTransformer’s excellent accuracy and efficiency: NSE 0.9445, KGE 0.9759 for water-depth prediction, and IoU 0.8180, F1 0.8997 for inundation classification, outperforming all comparative models. With 3s inference enabling multiple horizons in one pass, FloodTransformer offers a robust and practical solution for operational flood risk management.

洪水预报对灾害规划和风险管理至关重要，然而传统的基于水动力学的方法往往反应缓慢，计算量大。我们提出了一个混合框架，利用传统的水动力学建模和一种新的人工智能模型，实现准确、实时和高分辨率的洪水预测。为了解决大规模、密集洪水预测的计算挑战，我们开发了一种高效的洪水预测模型，FloodTransformer，它具有三个关键的新颖之处：可变大小的单元嵌入、标记化时间序列编码和物理信息多任务优化。这些组件有效地捕获复杂的时空依赖关系，允许在一次运行中进行准确的顺序预测。对模拟和历史洪水事件的综合评价表明，FloodTransformer具有良好的精度和效率：深度预测的NSE为0.9445，KGE为0.9759，洪水分类的IoU为0.8180，F1为0.8997，优于所有比较模型。通过3秒推理，可以一次通过多个视界，FloodTransformer为操作洪水风险管理提供了强大而实用的解决方案。

{"title":"FloodTransformer: Efficient real-time high-resolution flood forecasting","authors":"Zhanzhong Gu , Jiachen Kang , Wenzheng Jin , Feifei Tong , Y. Jay Guo , Wenjing Jia","doi":"10.1016/j.envsoft.2025.106832","DOIUrl":"10.1016/j.envsoft.2025.106832","url":null,"abstract":"<div><div>Flood forecasting is crucial for disaster planning and risk management, yet conventional hydrodynamic-based approaches are often slow in response and computationally intensive. We present a hybrid framework leveraging traditional hydrodynamic modelling with a novel AI model to enable accurate, real-time, and high-resolution flood prediction. To address the computational challenges of large-scale, dense flood prediction, we develop an efficient flood prediction model, <em>FloodTransformer</em>, which possesses three key novelties: variable-size cell embedding, tokenised time-sequence encoding, and physics-informed multi-task optimisation. These components effectively capture complex spatiotemporal dependencies, allowing accurate sequential predictions in a single run. Comprehensive evaluations on both simulated and historical flood events demonstrate FloodTransformer’s excellent accuracy and efficiency: NSE 0.9445, KGE 0.9759 for water-depth prediction, and IoU 0.8180, F1 0.8997 for inundation classification, outperforming all comparative models. With 3s inference enabling multiple horizons in one pass, FloodTransformer offers a robust and practical solution for operational flood risk management.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106832"},"PeriodicalIF":4.6,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732501","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of a self-supervised deep learning framework for chlorophyll-a retrieval in data-scarce inland waters 数据稀缺内陆水域叶绿素a检索的自监督深度学习框架开发

IF 4.6 2区环境科学与生态学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Environmental Modelling & Software

Pub Date : 2025-12-10 DOI: 10.1016/j.envsoft.2025.106817

Bongseok Jeong , Jihoon Shin , YoonKyung Cha

Deep learning and remote sensing-based chlorophyll-a (Chl-a) monitoring face challenges due to the optical complexity of inland waters and the scarcity of labeled data. To address these limitations, this study develops a self-supervised learning-based deep learning (SSL-DL) framework that leverages both labeled and unlabeled data. Three SSL-DL models are developed: a predictive SSL-DL model, which learns weak labels (incomplete labels); a generative SSL-DL model, which reconstructs input reflectance to capture underlying features; and an integrated SSL-DL model, which combines both. The models are applied to Sentinel-2 imagery of Daecheong and Paldang Lakes in South Korea. Results indicate that SSL-DL models outperform baseline models, with the integrated SSL-DL model achieving the highest test NSE (improvements of 0.1–0.36 over baselines in Daecheong Lake, improvements of 0.03–0.58 in Paldang Lake). The findings highlight the significance of SSL-DL in overcoming data limitations and enhancing scalability, demonstrating the potential for broader environmental remote sensing applications.

由于内陆水域的光学复杂性和标记数据的稀缺性，基于深度学习和遥感的叶绿素-a监测面临着挑战。为了解决这些限制，本研究开发了一个基于自我监督学习的深度学习（SSL-DL）框架，该框架利用了标记和未标记的数据。开发了三个SSL-DL模型：一个预测SSL-DL模型，它学习弱标签（不完整标签）；生成式SSL-DL模型，重建输入反射率以捕获底层特征；以及将两者结合起来的集成SSL-DL模型。该模型应用于韩国大清湖和八堂湖的Sentinel-2图像。结果表明，SSL-DL模型优于基线模型，其中综合SSL-DL模型的测试NSE最高（大清湖比基线提高0.1 ~ 0.36，八堂湖比基线提高0.03 ~ 0.58）。研究结果强调了SSL-DL在克服数据限制和增强可扩展性方面的重要性，展示了更广泛的环境遥感应用的潜力。

{"title":"Development of a self-supervised deep learning framework for chlorophyll-a retrieval in data-scarce inland waters","authors":"Bongseok Jeong , Jihoon Shin , YoonKyung Cha","doi":"10.1016/j.envsoft.2025.106817","DOIUrl":"10.1016/j.envsoft.2025.106817","url":null,"abstract":"<div><div>Deep learning and remote sensing-based chlorophyll-a (Chl-a) monitoring face challenges due to the optical complexity of inland waters and the scarcity of labeled data<strong>.</strong> To address these limitations, this study develops a self-supervised learning-based deep learning (SSL-DL) framework that leverages both labeled and unlabeled data. Three SSL-DL models are developed: a predictive SSL-DL model, which learns weak labels (incomplete labels); a generative SSL-DL model, which reconstructs input reflectance to capture underlying features; and an integrated SSL-DL model, which combines both. The models are applied to Sentinel-2 imagery of Daecheong and Paldang Lakes in South Korea. Results indicate that SSL-DL models outperform baseline models, with the integrated SSL-DL model achieving the highest test NSE (improvements of 0.1–0.36 over baselines in Daecheong Lake, improvements of 0.03–0.58 in Paldang Lake). The findings highlight the significance of SSL-DL in overcoming data limitations and enhancing scalability, demonstrating the potential for broader environmental remote sensing applications.</div></div>","PeriodicalId":310,"journal":{"name":"Environmental Modelling & Software","volume":"197 ","pages":"Article 106817"},"PeriodicalIF":4.6,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0