Atmospheric rivers (ARs) are narrow filaments of concentrated water-vapour transport that deliver a large share of mid-latitude extreme rainfall and flooding. This study couples a physically-based description of AR structure with machine learning to estimate where and when ARs translate into flood risk at global scale, rather than treating the ML model as a pure black box over raw fields.大气河（AR）是集中输送水汽的狭长气流带，贡献了中纬度相当大比例的极端降雨与洪水。本文把描述大气河结构的物理量与机器学习结合起来，在全球尺度上估计大气河何时、何地转化为洪水风险，而不是让模型在原始气象场上做纯黑箱拟合。
Key findings关键发现
Atmospheric rivers coincide with roughly one-fifth of large floods globally, rising to about 30–35% across the mid-latitudes — quantifying how much of the world's flood burden co-occurs with ARs.大气河与全球约五分之一的大洪水同时发生，在中纬度地区升至约 30–35% —— 量化了全球洪水中有多大比例与大气河相伴出现。
The physically-guided gradient-boosted model (LightGBM) reaches ROC-AUC ≈ 0.94; modelled flood risk climbs sharply when persistent AR activity coincides with high antecedent soil moisture.物理引导的梯度提升模型（LightGBM）ROC-AUC 达到约 0.94；当持续的大气河活动与前期高土壤含水量叠加时，模型给出的洪水风险急剧上升。
Framing the problem physically improves interpretability, making the model's risk attributions auditable against known AR-flood mechanisms.以物理框架切入提升了可解释性，模型给出的风险归因可对照已知的「大气河—洪水」机制进行核验。
Why it's worth reading为什么值得读
A clean example of hybrid physical–AI modelling done for a planetary-scale hazard: it shows how injecting domain structure buys both generalisation and interpretability at once. The author list overlaps with the team behind Google's operational flood-forecasting work, so the framing is grounded in operational reality.这是把「混合物理—AI 建模」用于行星尺度灾害的一个干净范例：它展示了注入领域结构如何同时换来泛化能力与可解释性。作者名单与谷歌业务化洪水预报团队有重叠，因此框架有很强的落地背景。
Relevance to your work与你研究的关联
Directly speaks to planetary-scale hydrology, hazard monitoring and the hybrid-modelling thread running through HyDROS / Hydrometeorology-AI: a template for coupling physical descriptors with learning rather than replacing physics wholesale.直接对应全球/行星尺度水文、灾害监测，以及贯穿 HyDROS / Hydrometeorology-AI 的混合建模主线：给出了「用物理描述子引导学习、而非整体替换物理」的一个可借鉴模板。
Reading path推荐阅读路径
Start with the global maps of AR–flood co-occurrence, then the feature importances; the soil-moisture × AR-persistence interaction is the core evidence. ~15 min.先看大气河—洪水同现的全球地图，再看特征重要性；「土壤含水量 × 大气河持续性」的交互作用是核心证据。约 15 分钟。
Limitations风险 / 局限
Global evaluation smooths over regional data quality; the gain from physical guidance depends on how faithfully AR variables are estimated, and "flood risk" here is a probabilistic index, not a calibrated inundation forecast.全球评估会掩盖区域数据质量差异；物理引导带来的增益取决于大气河变量估计的准确度，且此处的「洪水风险」是概率指数，并非经过率定的淹没预报。
Most ML streamflow models predict at a single outlet and bolt on an explicit routing module to move water downstream. This paper proposes a fully distributed, purely data-driven model that predicts spatially consistent streamflow across a large basin without a hand-built routing scheme, letting the network learn the spatial transfer of water directly.大多数机器学习径流模型只预测单个出口断面，再外挂一个显式汇流模块把水往下游输送。本文提出一个全分布式、纯数据驱动的模型，无需人工设计汇流方案即可在大流域内给出空间一致的径流预测，让网络直接学习水的空间传递。
Key findings关键发现
Removing the explicit routing module does not cost accuracy; spatial consistency across nested sub-basins is learned rather than imposed.去掉显式汇流模块并未损失精度；嵌套子流域之间的空间一致性是「学」出来的，而非强加的。
Predictions remain coherent at ungauged interior locations, addressing the spatial heterogeneity that defeats lumped models in large basins.在流域内部无观测点位上预测仍保持连贯，缓解了集总模型在大流域面对空间异质性时的失效。
The architecture points toward gauge-sparse, continental-scale streamflow mapping without per-basin routing calibration.该架构指向一种「站点稀疏、大陆尺度」的径流制图路线，无需逐流域率定汇流参数。
Why it's worth reading为什么值得读
It challenges the default that hydrological ML needs an explicit physical routing scaffold, and it does so in GRL with a spatially-consistent formulation — a useful counterpoint if you build distributed models.它挑战了「水文机器学习必须依赖显式物理汇流骨架」这一默认假设，并在 GRL 上以空间一致的形式给出证据 —— 对做分布式建模的人是一个有价值的对照。
Relevance to your work与你研究的关联
Central to streamflow prediction, ungauged-interior inference and distributed hydrologic modelling (CREST/EF5-style problems): a data-driven take on the routing question you meet when scaling models across space.直接切中径流预测、内部无资料点位推断与分布式水文建模（CREST/EF5 类问题）：为你在跨空间扩展模型时遇到的汇流问题提供了一个数据驱动的视角。
Reading path推荐阅读路径
Read the architecture figure and the spatial-consistency metric definition first, then the maps comparing predicted vs observed across sub-basins. Skip the hyperparameter appendix. ~14 min.先看架构图与「空间一致性」指标的定义，再看各子流域预测与观测对比图；超参数附录可略过。约 14 分钟。
Limitations风险 / 局限
"Pure data-driven" trades physical guarantees (mass conservation, hydraulics) for flexibility; cross-region transfer beyond the study basins is asserted more than demonstrated here.「纯数据驱动」以物理保证（质量守恒、水力学）换取灵活性；对研究流域之外的跨区域迁移，本文更多是主张而非充分论证。
CISRMamba: Cross-Modal Interaction and Scan-Routing Mamba for Multi-Sensor Flood Inundation Mapping
CISRMamba：面向多传感器洪水淹没制图的跨模态交互与扫描路由 Mamba
Haoran Feng, Chenyang Xiao, Ruiyang Lin, Yuxuan Chen, Linxing Liang, Bin Lin · Sensors · online 2026-08-18Haoran Feng、Chenyang Xiao、Ruiyang Lin、Yuxuan Chen、Linxing Liang、Bin Lin · Sensors · 在线发表 2026-08-18
#3Peer-reviewed同行评审Open access开放获取Q 4.2 · R 4.7~13 min约 13 分钟
KeywordsSAR–optical fusionflood inundation mappingstate-space model (Mamba)modality collapsemulti-sensor关键词SAR—光学融合洪水淹没制图状态空间模型（Mamba）模态坍缩多传感器
Summary摘要
Flood mapping that fuses optical and SAR imagery breaks down when clouds, speckle and terrain make the two sensors disagree — models "collapse" onto whichever modality looks cleaner. CISRMamba uses a state-space (Mamba) backbone with cross-modal interaction and adaptive scan-routing so evidence from SAR and optical is combined even when one is degraded.融合光学与 SAR 影像的洪水制图，在云、斑点噪声与地形导致两种传感器相互矛盾时会失效 —— 模型往往「坍缩」到看起来更干净的那一种模态。CISRMamba 采用状态空间（Mamba）主干，配合跨模态交互与自适应扫描路由，即使一种模态质量较差也能融合 SAR 与光学的证据。
Key findings关键发现
Adaptive scan-routing beats fixed scan patterns for irregular inundation boundaries, where a single raster-scan order loses spatial context.对形状不规则的淹没边界，自适应扫描路由优于固定扫描模式 —— 单一的栅格扫描顺序会丢失空间上下文。
Explicit cross-modal interaction mitigates "modality collapse", keeping SAR's all-weather signal useful when optical is cloud-blocked.显式的跨模态交互缓解了「模态坍缩」，在光学被云遮挡时仍能保留 SAR 的全天候信号价值。
Linear-time state-space attention keeps the model lighter than transformer fusion at comparable accuracy — relevant for near-real-time use.线性时间复杂度的状态空间注意力，在相近精度下比 Transformer 融合更轻量 —— 对近实时应用有意义。
Why it's worth reading为什么值得读
A concrete answer to a reader request — SAR-based flood mapping that is engineering-minded about the failure modes (clouds, speckle, sensor disagreement) that separate a method paper from an operational one.这是对读者请求的一个具体回应 —— SAR 洪水制图，且以工程视角认真处理了那些区分「方法论文」与「业务系统」的失效模式（云、斑点、传感器不一致）。
Relevance to your work与你研究的关联
Direct fit for flood monitoring, satellite-based inundation mapping and remote-sensing precipitation/water pipelines; the modality-collapse framing is transferable to any multi-sensor product.与洪水监测、卫星淹没制图、遥感降水/水体处理链高度契合；「模态坍缩」这一视角可迁移到任何多传感器产品。
Reading path推荐阅读路径
Look at the failure-case figure (where fixed-scan fusion breaks) then the scan-routing diagram; the cloud-cover ablation is the payoff. ~13 min.先看失效案例图（固定扫描融合在哪失败），再看扫描路由示意图；云遮挡消融实验是重点收获。约 13 分钟。
Limitations风险 / 局限
A benchmark-dataset study, not a deployed service; real near-real-time latency, orbit revisit gaps and label quality for true operational use are out of scope.这是基准数据集研究，并非已部署的业务系统；真实近实时时延、轨道重访间隙与业务级标注质量都不在讨论范围内。
A Hierarchical Deep Learning Framework for Runoff Prediction Using Raster-Based Spatial Representations
基于栅格空间表征的分层深度学习径流预测框架
De-Hui Ouyang, E Deng, Yi-Qing Ni · Water Resources Research · online 2026-08-18De-Hui Ouyang、E Deng、Yi-Qing Ni · Water Resources Research · 在线发表 2026-08-18
#4Peer-reviewed同行评审Open access开放获取Q 4.3 · R 4.5~15 min约 15 分钟
Large-sample deep-learning runoff models usually feed the network catchment attributes aggregated to a single number per basin — losing the spatial pattern inside the watershed. Using 531 CONUS watersheds, this framework represents static properties as rasters and learns them hierarchically, preserving where the soil, slope and land-cover heterogeneity actually sits.大样本深度学习径流模型通常把流域属性聚合成「每个流域一个数」喂给网络，从而丢失流域内部的空间格局。本文在 531 个 CONUS 流域上，将静态属性表征为栅格并进行分层学习，保留了土壤、坡度与土地覆盖异质性「实际分布在哪里」的信息。
Key findings关键发现
Raster (spatially-resolved) attributes beat aggregated scalars, confirming that within-basin spatial structure carries predictive signal usually discarded.栅格化（保留空间分辨）的属性优于聚合标量，证实流域内部空间结构携带了通常被丢弃的预测信号。
The hierarchy lets the model share representation across basins while still adapting to each watershed's internal layout — useful for transfer to sparsely-gauged catchments.分层结构让模型在跨流域共享表征的同时，仍能适应每个流域的内部布局 —— 有利于向观测稀疏流域迁移。
Gains are largest in heterogeneous basins, where single-value attributes are the crudest approximation.在异质性强的流域增益最大 —— 这些地方「单一数值」属性恰恰是最粗糙的近似。
Why it's worth reading为什么值得读
Responds to the reader ask on ungauged-basin runoff and transfer learning: it isolates one lever — spatial representation of static attributes — and shows it matters, a cleaner ablation than most large-sample papers.回应了读者关于无资料流域径流与迁移学习的请求：它单独抓住「静态属性的空间表征」这一杠杆并证明其重要性，消融实验比多数大样本论文更干净。
Relevance to your work与你研究的关联
Bears on prediction in ungauged basins, large-sample/foundation-style hydrology and the flash-flood dataset work — a practical recipe for feeding spatial structure into learned runoff models.关联无资料流域预测、大样本/基础模型式水文，以及山洪数据集工作 —— 给出把空间结构喂进学习型径流模型的可操作配方。
Reading path推荐阅读路径
Read the raster-vs-scalar schematic and the per-basin skill map; the transfer experiment to held-out basins is the part worth dwelling on. ~15 min.先看「栅格 vs 标量」示意图与逐流域精度图；对留出流域的迁移实验值得多花时间。约 15 分钟。
Limitations风险 / 局限
CONUS-only with the usual well-gauged-country bias; "transfer" is within-CONUS held-out basins, not a true cross-continent ungauged test, and raster attributes raise the data-preparation cost.仅限 CONUS，存在「观测良好国家」的常见偏差；此处「迁移」是 CONUS 内部留出流域，并非真正的跨大陆无资料测试，且栅格属性抬高了数据准备成本。
SeaScope: A Transparent and Reproducible LLM-Assisted Framework for Maritime Earth Observation Analysis
SeaScope：面向海洋地球观测分析的透明、可复现 LLM 辅助框架
C. Sekas, L. Mavrofidopoulou, I. Agathangelidis, C. Cartalis, K. Philippopoulos, et al. · Remote Sensing · online 2026-08-22C. Sekas、L. Mavrofidopoulou、I. Agathangelidis、C. Cartalis、K. Philippopoulos 等 · Remote Sensing · 在线发表 2026-08-22
#5Peer-reviewed同行评审Open access开放获取Q 4.0 · R 4.4~12 min约 12 分钟
Building Earth-observation workflows demands expertise in data selection, geospatial coding and cloud processing. SeaScope wraps a large language model as an assistant that turns natural-language requests into transparent, reproducible EO analysis pipelines for maritime monitoring — with the provenance of each step exposed rather than hidden inside a chat.构建地球观测工作流需要在数据选择、地理空间编程与云处理上的专业能力。SeaScope 把大语言模型包装成助手，将自然语言请求转化为透明、可复现的海洋监测 EO 分析流水线 —— 每一步的来源可追溯，而非藏在一段对话里。
Key findings关键发现
Emphasises transparency and reproducibility as first-class goals — the LLM emits inspectable, re-runnable steps rather than one-off answers.把透明性与可复现性作为一等目标 —— LLM 输出可检查、可重跑的步骤，而非一次性答案。
Lowers the expertise barrier to standard EO tasks (data retrieval, geospatial processing) for non-programmer domain users.为非编程背景的领域用户降低了完成标准 EO 任务（数据获取、地理空间处理）的门槛。
A concrete instance of the "scientific agent" pattern applied to remote sensing, transferable beyond the maritime case.「科学智能体」范式在遥感中的一个具体实例，可迁移到海洋之外的场景。
Why it's worth reading为什么值得读
This issue's closest in-window match to the repeated reader request for LLM agents in the Earth sciences. Its transparency/reproducibility stance is exactly what the AQUAH / hydrologic-agent line needs, and the design choices carry over to hydrology.本期最接近读者反复请求的「地球科学 LLM 智能体」的在窗口内论文。它对透明性/可复现性的坚持，正是 AQUAH / 水文智能体这条线所需要的，其设计取舍可迁移到水文领域。
Relevance to your work与你研究的关联
Direct bearing on AQUAH and scientific-workflow automation: a reproducibility-first template for wrapping LLMs around geoscience tooling. See also the Frontier Concept on scientific agents below.与 AQUAH 及科学工作流自动化直接相关：一个「可复现优先」、把 LLM 包裹在地球科学工具外围的模板。另见下方「前沿概念」中的科学智能体条目。
Reading path推荐阅读路径
Read the architecture/provenance diagram and one worked maritime example end-to-end; the reproducibility discussion is the transferable lesson. ~12 min.看架构/来源追溯图，并完整跟一个海洋案例走一遍；关于可复现性的讨论是可迁移的要点。约 12 分钟。
Limitations风险 / 局限
Maritime EO, not hydrology; the LLM still depends on curated tools and can err on ambiguous requests. Read it for the pattern, not as a drop-in hydrologic agent.面向海洋 EO 而非水文；LLM 仍依赖经过整理的工具，对含糊请求可能出错。应把它当作「范式」来读，而非可直接套用的水文智能体。
A Diffusion-Based Framework for High-Resolution Precipitation Forecasting over CONUS
面向 CONUS 的高分辨率降水预报扩散模型框架
M. Vicens-Miquel, A. McGovern, A. J. Hill, E. Foufoula-Georgiou, C. Guilloteau, S. S. P. Shen · Artificial Intelligence for the Earth Systems · print 2026-08-17M. Vicens-Miquel、A. McGovern、A. J. Hill、E. Foufoula-Georgiou、C. Guilloteau、S. S. P. Shen · Artificial Intelligence for the Earth Systems · 印刷 2026-08-17
#6Peer-reviewed同行评审Open access开放获取Q 4.4 · R 4.3~16 min约 16 分钟
Keywordsdiffusion model1 km precipitationHRRR correctionMRMSuncertainty quantification关键词扩散模型1 公里降水HRRR 订正MRMS不确定性量化
Summary摘要
The team builds a diffusion deep-learning framework for 1 km, 1–12 h precipitation forecasts over CONUS and systematically compares three residual strategies: data-driven (MRMS only), corrective (HRRR only), and hybrid (both). Autoregressive rollouts extend the lead time, with calibrated uncertainty attached.作者构建了一个扩散深度学习框架，用于 CONUS 上 1 公里、1–12 小时的降水预报，并系统比较三种残差策略：纯数据驱动（仅 MRMS）、订正型（仅 HRRR）、混合型（两者兼用）。通过自回归滚动延长预见期，并附带经过校准的不确定性。
Key findings关键发现
The diffusion framework consistently beats the HRRR baseline on pixel-wise and spatio-statistical metrics across all lead times.在所有预见期上，扩散框架在逐像元与时空统计指标上都稳定优于 HRRR 基线。
The hybrid model wins at short lead times; the HRRR-corrective model takes over at longer leads through 12 h — a clean division of labour.混合模型在短预见期占优；HRRR 订正模型在更长预见期（直到 12 小时）接手 —— 分工清晰。
Calibrated uncertainty tailored to the residual setup makes the forecasts usable for emergency preparedness, not just point estimates.针对残差设置定制的校准不确定性，使预报可用于应急准备，而不只是点估计。
Why it's worth reading为什么值得读
A well-controlled study (McGovern, Foufoula-Georgiou) that separates data-driven, corrective and hybrid roles instead of shipping one monolithic model — a useful design map for anyone building high-resolution precipitation nowcasts for flash-flood warning.一项控制良好的研究（McGovern、Foufoula-Georgiou），把数据驱动、订正与混合三种角色分开，而非交付一个「一体化」模型 —— 对做山洪预警高分辨率降水临近预报的人是一张有用的设计地图。
Relevance to your work与你研究的关联
Squarely in Hydrometeorology-AI and flash-flood territory: high-resolution QPF/QPE with uncertainty is the input layer for the flood models you build.正处于 Hydrometeorology-AI 与山洪的核心地带：带不确定性的高分辨率定量降水预报/估计，正是你所构建洪水模型的输入层。
Reading path推荐阅读路径
Read the three-strategy comparison table first, then the lead-time skill curves; the uncertainty-calibration section is the part to adapt. ~16 min. Preprint PDF on arXiv (2512.09059).先看三种策略对比表，再看随预见期变化的技能曲线；不确定性校准一节最值得借鉴。约 16 分钟。预印本 PDF 见 arXiv（2512.09059）。
Limitations风险 / 局限
CONUS with dense radar (MRMS/HRRR) — the approach leans on data infrastructure that gauge-sparse regions lack; diffusion rollouts are also compute-heavy for real-time operations.CONUS 拥有密集雷达（MRMS/HRRR）—— 该方法依赖观测稀疏地区所缺乏的数据基础设施；扩散滚动预报在实时业务中计算开销也较大。
LiveEO funded for optical+SAR vision foundation modelLiveEO 获资助研发「光学+SAR」视觉基础模型
The German Space Agency (DLR) backed an 18-month feasibility study for a single vision foundation model that detects change across combined optical and radar imagery — another step toward multimodal EO foundation models.德国航天局（DLR）资助一项 18 个月的可行性研究，目标是用单一视觉基础模型在「光学+雷达」联合影像上检测变化 —— 又一步迈向多模态 EO 基础模型。
ML for flood-related emergency (112) calls用机器学习预测洪水相关报警（112）
A new NHESS study predicts flood-related emergency-call volume, pushing impact-based early warning (EW4All) from hazard forecasts toward forecasting societal impact directly.NHESS 一项新研究预测洪水相关报警电话量，把「面向影响的预警」（EW4All）从预报灾害本身推进到直接预报社会影响。
Global Landsat forest-fire-patch dataset (1984–2022)全球 Landsat 森林火灾斑块数据集（1984–2022）
A new ESSD release maps four decades of global forest-fire patches from Landsat — a long, consistent record for hazard, carbon and land-cover-change studies.ESSD 新发布的数据集用 Landsat 绘制了近四十年全球森林火灾斑块 —— 为灾害、碳循环与土地覆盖变化研究提供了长期一致的记录。
Global forest-disturbance regimes from satellite biomass基于卫星生物量的全球森林扰动机制数据集
Another ESSD dataset derives global forest-disturbance regimes from satellite biomass observations, quantifying how the number, size and severity of disturbances threaten forest carbon sinks.ESSD 另一数据集从卫星生物量观测中反演全球森林扰动机制，量化扰动的数量、规模与强度如何威胁森林碳汇。
AI weather on a stretched cubed sphere「拉伸立方球」上的 AI 天气预报
A GRL paper runs global-plus-regional AI weather forecasting on a stretched cubed-sphere grid, giving local hazard detail without external boundary forcing — a step past uniform lat-lon AI forecasts.GRL 一篇论文在「拉伸立方球」网格上实现「全球+区域」AI 天气预报，无需外部边界强迫即可给出局地灾害细节 —— 超越了均匀经纬网的 AI 预报。
Review: remote sensing of clouds & precipitation综述：云与降水的遥感
A Remote Sensing review charts the shift from descriptive cloud/precip observation to process-oriented diagnosis across multi-source platforms — a good map of where satellite precipitation is heading.《Remote Sensing》一篇综述梳理了多源平台上云/降水遥感从「描述性观测」转向「过程诊断」的趋势 —— 是了解卫星降水走向的一张好地图。
Definition. An LLM-driven system that plans and executes a scientific workflow — retrieving data, configuring and running a model, then reading results and writing a report — via tools, with human oversight at chosen checkpoints.定义。由 LLM 驱动、通过工具规划并执行一整套科学工作流的系统 —— 获取数据、配置并运行模型、再解读结果并撰写报告 —— 并在设定的检查点接受人工监督。
Why it matters for hydrology. It lowers the operational barrier to running distributed hydrologic models (data prep, calibration, reporting), turning days of manual setup into a natural-language request — and, done right, keeps every step auditable. This is exactly the reader-requested topic.为何对水文重要。它降低了运行分布式水文模型的操作门槛（数据准备、率定、报告撰写），把数天的手工搭建变成一句自然语言请求 —— 做得好还能让每一步可核验。这正是读者点名想学的主题。
Foundation models for Earth observation面向地球观测的基础模型
Definition. Large models pre-trained self-supervised on massive multi-sensor satellite archives, producing general-purpose embeddings that fine-tune to many downstream tasks (land cover, flood, biomass) with little labelled data.定义。在海量多传感器卫星档案上以自监督方式预训练的大模型，产出通用嵌入表示，只需少量标注即可微调到诸多下游任务（土地覆盖、洪水、生物量）。
Why it matters. They promise transfer to data-sparse regions and tasks — the same generalisation problem as prediction in ungauged basins — and are converging on optical+SAR multimodality (see this week's LiveEO brief).为何重要。它们承诺向数据稀疏地区与任务迁移 —— 这与无资料流域预测是同一个泛化问题 —— 并正朝「光学+SAR」多模态收敛（见本周 LiveEO 短讯）。
Read first — OlmoEarth (Ai2): an accessible overview of a current open EO foundation-model family.先读 —— OlmoEarth（Ai2）：对当前开放 EO 基础模型家族的通俗介绍。
Then — Prithvi-EO-2.0 (arXiv 2412.02732): a concrete multi-temporal architecture and its benchmarks.再读 —— Prithvi-EO-2.0（arXiv 2412.02732）：一个具体的多时相架构及其基准测试。
Definition. Approaches that inject physical structure — governing equations, conserved quantities, process descriptors — into learning, ranging from physics-guided features to fully differentiable process models trained with gradient descent.定义。把物理结构 —— 控制方程、守恒量、过程描述子 —— 注入学习过程的方法，从「物理引导的特征」到用梯度下降训练的完全可微过程模型都属此列。
Why it matters. It buys generalisation and interpretability at once, as this week's atmospheric-river flood paper and the neural-operator shallow-water work show — the middle path between pure black boxes and rigid physical models.为何重要。它能同时换来泛化能力与可解释性 —— 正如本周的大气河洪水论文与神经算子浅水方程工作所示 —— 是纯黑箱与僵硬物理模型之间的中间道路。
Issue 2026-07-23 · @hydros-ou asked for (1) ungauged-basin runoff via foundation / transfer learning and (2) operational SAR near-real-time flood mapping — plus shorter summaries and ordered Learning paths. This issue: the hierarchical raster-runoff paper (#4) and the distributed-streamflow paper (#2) target the ungauged/transfer ask; CISRMamba (#3) covers multi-sensor SAR flood mapping; every summary is trimmed to a short paragraph with findings kept as bullets; all three Frontier Concepts now carry a numbered "read first → then" path.2026-07-23 期 · @hydros-ou 请求：(1) 用基础模型/迁移学习做无资料流域径流；(2) 业务化 SAR 近实时洪水制图；并希望摘要更短、Learning 给出学习顺序。本期：分层栅格径流论文（#4）与分布式径流论文（#2）回应无资料/迁移诉求；CISRMamba（#3）覆盖多传感器 SAR 洪水制图；每篇摘要压缩为一小段、关键发现以要点保留；三个前沿概念均已给出带编号的「先读→再读」路径。
Issue 2026-07-23 · @hydros-ou asked to see LLM AI agents for hydrologic modeling and to learn more. This issue: SeaScope (#5) is the closest in-window peer-reviewed LLM-agent-for-EO paper, and the first Frontier Concept is a dedicated, ordered path through AQUAH → HydroAgent → a scientific-agent survey. Note: the peer-reviewed "AI Agent for Hydrologic Modeling" (GRL) was featured in a prior issue, so it is not repeated here.2026-07-23 期 · @hydros-ou 希望看到面向水文建模的 LLM 智能体并深入学习。本期：SeaScope（#5）是窗口内最接近的、经同行评审的「LLM 智能体×地球观测」论文；第一个前沿概念给出 AQUAH → HydroAgent → 科学智能体综述 的专门有序路径。说明：经同行评审的「AI Agent for Hydrologic Modeling」（GRL）已在往期收录，故此处不重复。
Coverage & method note覆盖与方法说明
Window used: 2026-08-10 → 2026-08-24 (14 days), not widened. A paper counts as in-window if EITHER its online-first OR print date falls inside. Candidates came from both web search and a Crossref/Unpaywall metadata sweep (keyword, journal and lab-author patrol). Some publishers (Wiley/AGU, Elsevier) block automated fetching; where authoritative metadata and abstracts were available, those papers were still assessed and cited by DOI.覆盖窗口：2026-08-10 → 2026-08-24（14 天），未扩窗。只要论文的在线首发或印刷日期任一落在窗口内即视为在窗口内。候选同时来自网络检索与 Crossref/Unpaywall 元数据巡查（关键词、期刊与实验室作者巡查）。部分出版商（Wiley/AGU、Elsevier）会拦截自动抓取；在有权威元数据与摘要时，这些论文仍被评估并以 DOI 引用。
Curated by an automated agent. Links point to originals; preprints are marked and not peer-reviewed. Quality (Q) and relevance (R) scores are editorial judgments on a 5-point scale, not bibliometrics. Selection reflects relevance to hydrology, remote sensing, hazards and AI — not endorsement.由自动化 agent 整理。链接指向原文；预印本已标注且未经同行评审。质量（Q）与相关性（R）评分为 5 分制的编辑判断，非文献计量指标。选题反映与水文、遥感、灾害与 AI 的相关性，不代表背书。