[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3reTdwXwJ6pMKVrGbVddQOuv4i3vvKaIkjMHeh30Ie0":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1249,"南科大环境学院郑一团队提出生成式人工智能预报洪水的新方法","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">近日，南方科技大学环境科学与工程学院教授郑一团队与中国科学院大气物理研究所等多家单位合作，在地球科学领域旗舰期刊\t\u003Cem>Geophysical Research Letters\u003C\u002Fem>\t发表题为“Probabilistic Diffusion Models Advance Extreme Flood Forecasting”的论文，首次将生成式人工智能（GAI）的前沿技术——扩散模型（diffusion model）——成功用于洪水预报。这项研究不仅为洪水预报技术带来了重大变革，更为水文学乃至整个地球系统科学领域的AI应用探索了新路径。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F3e387d9e03dc4097941c4282ef72de4d\u002F151755266579760975.jpg\" width=\"724\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">气候变化加剧导致全球极端洪水事件频发，严重威胁人类社会发展。联合国减灾署最新数据显示，本世纪以来全球灾难性洪水灾害发生频率激增134%，造成逾10万人罹难，直接经济损失超过6510亿美元。然而，传统降雨径流模型易低估峰值流量，难以预报最危险、最具破坏性的大洪峰，也无法直接给出基于概率的风险评估。该研究提出了基于扩散模型的DRUM（diffusion-based runoff model）方法，利用深度神经网络训练径流数据的噪声模型，再使用该模型进行多步去噪操作，生成径流的集合预报数据，径流生成过程如图1所示。DRUM无需预定义径流的概率分布形式，直接从数据中学习概率分布；能完成多尺度任务分解，将复杂的洪水预报任务分解为一系列相对简单的子问题；以及具有灵活的条件生成机制，可有效利用条件信息（如洪水形成的气象条件）。这些特点使DRUM能有效处理洪水预报中的非线性、多尺度和高不确定性特征。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F653a1597e35e4996b2e40824d9102cb8\u002F151755232294465148.png\" width=\"711\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">图1 洪水预报模型DRUM的分布重建（预测结果采样）过程\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">研究团队基于CAMELS数据集，在美国531个代表性流域上对DRUM的性能进行了检验，并与现有的深度学习标杆模型进行对比。短临预报（0天预见期）实验结果表明，DRUM提升预报准确性的幅度随洪水量级的增加而增大（图2a）。在72.3%的研究流域中，DRUM对前千分之一流量（即最极端洪水）的短临预报能力超越了标杆模型（图2b）。此外，DRUM在洪水概率预报方面的优势在8个超出历史数据极大值的极端洪水事件中进一步凸显（图2c–j）。研究团队进一步使用欧洲中期天气预报中心综合预报系统（ECMWF-IFS）的降水预报数据驱动模型，将短临预报扩展到业务预报（7天预见期）。结果显示，DRUM在各种洪水量级和不同预见期上始终优于标杆方法（图3a）。DRUM还表现出更优的洪水提前预警能力（图3b），特别是对极端事件（20年和50年一遇洪水），成功将平均预警提前期从约0.2天延长至约1.2天，实现了近一整天的预警时间提升。研究成果充分展示了利用生成式人工智能进行业务化洪水预报的光明前景，对于全球洪水风险评估、预警及应急响应具有重要意义。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F73422a437e3d422e86da771d9c7ff843\u002F151755232375782406.png\" width=\"768\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">图2 DRUM在极端洪水事件短临预报中的性能表现\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002Fcb45da42619543d7a5019a06062537dc\u002F151755232397230483.png\" width=\"766\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">图3 DRUM在业务洪水预报中的性能评估\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">南科大环境学院2024级博士生欧志刚和中科院大气所奈聪毅为论文共同第一作者，南科大郑一教授和中国科学院大气物理研究所潘宝祥副研究员为论文共同通讯作者，南方科技大学为论文第一单位。论文合作者还包括宾夕法尼亚州立大学教授申朝鹏、太平洋西北国家实验室助理教授蒋佩诗、中科院地理所副研究员刘星才、中科院地理所研究员汤秋鸿、中国水科院博士生李雯晴和加利福尼亚大学圣迭戈分校教授潘铭等。该研究得到了国家自然科学基金委杰出青年科学基金项目、高水平专项资金等经费支持。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">论文链接：\u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fdoi.org\u002F10.1029\u002F2025GL115705\" rel=\"noopener noreferrer\" target=\"_blank\" class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fdoi.org\u002F10.1029\u002F2025GL115705\u003C\u002Fa>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">【新闻来源】南方科技大学新闻网 \u003C\u002Fspan>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187); background-color: rgb(56, 56, 56);\">供稿单位：环境科学与工程学院 通讯员：周亦潆 主图：丘妍 编辑：周易霖 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fnewshub.sustech.edu.cn\u002Fhtml\u002F202508\u002F46739.html\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fnewshub.sustech.edu.cn\u002Fhtml\u002F202508\u002F46739.html\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F20111a330026429695030468d8cacd42\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002Fthumbs\u002F20111a330026429695030468d8cacd42\u002FAI领域.jpg",0,1,207,"2025-08-20 19:22",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A59718b2b-a23e-489a-86e5-c53fe34d080f%3A0.wav?Expires=1755695619&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=UT2q4aPnjwijXLwKINesuYWtvNI%3D",7504330,"59718b2b-a23e-489a-86e5-c53fe34d080f","2025-08-20 19:13","Nankai University's School of Environment, Professor Zheng Yi's team proposes a new method for flood forecasting using generative artificial intelligence","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Recently, Professor Zheng Yi's team from the School of Environment Science and Engineering at Southern University of Science and Technology collaborated with multiple institutions including the Institute of Atmospheric Physics, Chinese Academy of Sciences, to publish a paper titled \"Probabilistic Diffusion Models Advance Extreme Flood Forecasting\" in the flagship journal of Earth sciences \u003Cem>Geophysical Research Letters\u003C\u002Fem>. This study is the first to successfully apply the cutting-edge technology of generative artificial intelligence (GAI), specifically diffusion models, to flood forecasting. This research not only brings significant changes to flood forecasting technology but also explores new paths for AI applications in hydrology and the entire Earth system science field.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F3e387d9e03dc4097941c4282ef72de4d\u002F151755266579760975.jpg\" width=\"724\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Climate change intensification has led to frequent global extreme flood events, posing serious threats to human social development. According to the latest data from the United Nations Office for Disaster Risk Reduction, the frequency of catastrophic flood disasters worldwide has increased by 134% since this century, resulting in over 100,000 deaths and direct economic losses exceeding $651 billion. However, traditional rainfall-runoff models tend to underestimate peak flow and are unable to forecast the most dangerous and destructive large floods, nor can they directly provide probabilistic risk assessments. The study proposed the DRUM (diffusion-based runoff model) method based on diffusion models. It uses deep neural networks to train noise models of runoff data and then performs multi-step denoising operations using the model to generate ensemble runoff forecasts. The runoff generation process is shown in Figure 1. DRUM does not require predefined probability distribution forms for runoff and directly learns the probability distribution from the data. It can complete multi-scale task decomposition, breaking complex flood forecasting tasks into a series of relatively simple sub-problems. It also has a flexible conditional generation mechanism that can effectively utilize condition information (such as meteorological conditions for flood formation). These characteristics enable DRUM to effectively handle nonlinear, multi-scale, and high-uncertainty features in flood forecasting.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F653a1597e35e4996b2e40824d9102cb8\u002F151755232294465148.png\" width=\"711\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">Figure 1 The distribution reconstruction (prediction result sampling) process of the flood forecasting model DRUM\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The research team tested the performance of DRUM on 531 representative basins in the United States using the CAMELS dataset and compared it with existing deep learning benchmark models. Short-term forecast (0-day lead time) results show that the improvement in forecast accuracy by DRUM increases with the magnitude of the flood (Figure 2a). In 72.3% of the studied basins, DRUM outperformed the benchmark models in short-term forecasting of the top 0.1% flow (i.e., the most extreme floods) (Figure 2b). Additionally, the advantages of DRUM in flood probability forecasting were further highlighted in eight extreme flood events exceeding historical data maximums (Figures 2c–j). The research team further used precipitation forecast data from the European Centre for Medium-Range Weather Forecasts (ECMWF-IFS) to drive the model, extending the short-term forecast to operational forecasting (7-day lead time). The results showed that DRUM consistently outperformed the benchmark methods across various flood magnitudes and different lead times (Figure 3a). DRUM also demonstrated better flood early warning capabilities (Figure 3b), especially for extreme events (20-year and 50-year floods), successfully extending the average lead time from about 0.2 days to about 1.2 days, achieving nearly a full day of early warning time improvement. The research results fully demonstrate the bright prospects of using generative artificial intelligence for operational flood forecasting, which is of great significance for global flood risk assessment, early warning, and emergency response.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F73422a437e3d422e86da771d9c7ff843\u002F151755232375782406.png\" width=\"768\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">Figure 2 Performance of DRUM in short-term forecasting of extreme flood events\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002Fcb45da42619543d7a5019a06062537dc\u002F151755232397230483.png\" width=\"766\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">Figure 3 Performance evaluation of DRUM in operational flood forecasting\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Doctoral student Ou Zhigang from the 2024 cohort of the School of Environment at Nankai University and Nai Congyi from the Institute of Atmospheric Physics, Chinese Academy of Sciences, are the co-first authors of the paper. Professor Zheng Yi from Nankai University and Associate Researcher Pan Baoxiang from the Institute of Atmospheric Physics, Chinese Academy of Sciences, are the co-corresponding authors of the paper. Nankai University is the first author's institution. The paper also includes collaborators such as Professor Shen Chaopeng from Pennsylvania State University, Assistant Professor Jiang Peishi from the Pacific Northwest National Laboratory, Associate Researcher Liu Xingcai from the Institute of Geography, Chinese Academy of Sciences, Researcher Tang Qiuhong from the Institute of Geography, Chinese Academy of Sciences, Doctoral student Li Wenqing from the China Institute of Water Resources and Hydropower Research, and Professor Pan Ming from the University of California, San Diego. This research was supported by the National Natural Science Foundation of China's Excellent Young Scientists Fund Project and other funding sources.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">Paper link:\u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fdoi.org\u002F10.1029\u002F2025GL115705\" rel=\"noopener noreferrer\" target=\"_blank\" class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fdoi.org\u002F10.1029\u002F2025GL115705\u003C\u002Fa>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\t\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">[News Source] Southern University of Science and Technology News Network \u003C\u002Fspan>\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187); background-color: rgb(56, 56, 56);\">Contributing unit: School of Environment Science and Engineering, Correspondent: Zhou Yiying, Main image: Qiu Yan, Editor: Zhou Yilin \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fnewshub.sustech.edu.cn\u002Fhtml\u002F202508\u002F46739.html\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fnewshub.sustech.edu.cn\u002Fhtml\u002F202508\u002F46739.html\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\"> (This article is reprinted by this website to provide readers with more information. The content involved does not constitute investment or consumption advice. If there are any questions about the facts of the article, please verify with the relevant parties. The views expressed in the article are not the views of this website and are for reference only.)\u003C\u002Fspan>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A0fdaaefa-acc1-4a27-91f2-2873b88868d8%3A0.wav?Expires=1774838492&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=BvWZZeali6vekCSFEB3Ddb8rw7Q%3D","0fdaaefa-acc1-4a27-91f2-2873b88868d8",10941930]