Hydrology × AI Weekly Reading — Issue #1水文 × AI 每周精读 — 第 1 期TEST ISSUE测试版
Curated, ranked reading on hydrology, remote sensing, natural hazards & AI.
围绕水文、遥感、自然灾害与人工智能的精选排序阅读。
Issue期号#1
Coverage window覆盖时间2026-07-09 → 2026-07-23
Pushed推送时间2026-07-23 18:19 CDT
Next issue下期预计2026-07-27
Frequency更新频率Every Mon 07:00每周一 07:00
All全部
Hydrology水文
Remote sensing遥感
Hazards灾害
Flood monitoring洪水监测
Scientific ML科学机器学习
Foundation model基础模型
◆
Key Papers重要文献
4 reads, ranked by quality + relevance4 篇，按质量+相关性排序
★
Top Priority重点优先
#1
Overall rank · by quality + relevance综合排名 · 按质量+相关性
Inunda: A GPU-Native, Differentiable Solver for High-Resolution Flood Inundation ModelingInunda：面向高分辨率洪水淹没模拟的 GPU 原生可微分求解器
Preprint · Not peer-reviewed预印本 · 未同行评审
Open access开放获取
Type: Methods / preprint类型：方法/预印本
~20 min read约 20 分钟
Keywordsdifferentiable solver2-D shallow-water equationslocal-inertial schemeGPU-native autogradmass-conservative flood inundation关键词可微分求解器二维浅水方程局地惯性格式GPU原生自动微分质量守恒淹没模拟
Author: Zhi Li · Source: arXiv:2607.09614 · Published: 2026-07-10
作者：Zhi Li · 来源：arXiv:2607.09614 · 发布：2026-07-10
Relevance 9.5/10相关性 9.5/10
Summary简要摘要
A single-GPU flood inundation model that solves the 2-D shallow-water equations with a mass-conservative local-inertial scheme — but built entirely from autograd-compatible operators, so it is fully differentiable. Parameters (e.g., roughness, hydraulic conductivity) are estimated by gradient descent through reverse-mode automatic differentiation, and the solver is driven by convection-allowing precipitation ensembles to produce probabilistic forecasts.
一个可在单块 GPU 上运行的洪水淹没模型：用质量守恒的局部惯性格式求解二维浅水方程，但整套算子都与自动微分兼容，因此完全可微分。糙率、导水率等参数可通过反向模式自动微分做梯度下降率定；模型由对流可分辨的降水集合驱动，输出概率性预报。
Key findings关键发现
- On Hurricane Harvey (2017): 0.67 m mean absolute error vs. surveyed high-water marks and a Nash–Sutcliffe efficiency of +0.72, versus +0.31 for the operational National Water Model v3.0.
在 2017 年 Harvey 飓风案例中：相对实测高水位标记的平均绝对误差 0.67 m，Nash–Sutcliffe 效率 +0.72，而业务化的 National Water Model v3.0 仅为 +0.31。
- For the July 2025 Central Texas flash flood, probabilistic forecasts showed improving skill as the crest approached.
对 2025 年 7 月中德州山洪，概率预报在临近洪峰时技巧逐步提升。
- In the post-fire Rio Ruidoso basin, gradient-based inversion recovered spatially-explicit hydraulic-conductivity fields across multiple years.
在过火后的 Rio Ruidoso 流域，基于梯度的反演还原出多年、空间显式的导水率场。
- Differentiability enables end-to-end calibration and ML coupling that traditional black-box solvers cannot offer.
可微分性带来传统黑箱求解器无法实现的端到端率定与机器学习耦合能力。
Why it's worth reading为什么值得读
It is a clean, concrete demonstration of the "differentiable modeling" thesis in operational-grade flood hydraulics: physics stays intact, yet gradients let you calibrate, assimilate, and couple to neural components — all at single-GPU cost. A rare case where a physics solver beats a large operational system on real events.
它把"可微分建模"的理念落到了接近业务级的洪水水力学上：物理过程保持不变，却能借梯度进行率定、同化并与神经网络耦合，而且只需单 GPU 成本。物理求解器在真实事件上胜过大型业务系统，实属难得。
Relevance to your research与你研究的关联
Directly on-target for HyDROS Lab / AQUAH and CREST-EF5-style flash-flood forecasting: a differentiable core could sit under an AI agent that calibrates and runs flood scenarios automatically. The July 2025 Central Texas case overlaps your flash-flood datasets, and the gradient-based parameter recovery is a template for hybrid physical–AI hydrology.
与 HyDROS Lab / AQUAH 及 CREST-EF5 类山洪预报高度契合：可微分内核可作为 AI agent 自动率定并运行洪水情景的底座。2025 年 7 月中德州案例与你的山洪数据集相交，梯度式参数反演也是"物理–AI 混合水文"的可借鉴范式。
Suggested reading path推荐阅读路径
Start with the Harvey inundation figure and the NSE comparison table (the headline result). Then read the differentiable local-inertial scheme in the methods. You can skip the low-level CUDA/kernel implementation details on a first pass. ~20 min.
先看 Harvey 淹没图与 NSE 对比表（核心结论），再读方法中的可微分局部惯性格式。首轮可跳过底层 CUDA/核函数实现细节。约 20 分钟。
⚠ Risks / limitations风险 / 局限:
Preprint, not yet peer-reviewed; single author. Case studies are a handful of U.S. events — cross-region/climate generalization is untested. NSE gains should be checked for hydrologic significance, not just numbers, and near-crest probabilistic skill may be event-specific.
预印本、尚未评审、单一作者。案例仅为少数美国事件，跨区域/跨气候的泛化未验证。NSE 提升需检验其水文意义而非只看数字，临近洪峰的概率技巧也可能是个案。
Read original ↗阅读原文 ↗
·
Download PDF ↓下载 PDF ↓
#2
Overall rank · by quality + relevance综合排名 · 按质量+相关性
Quantification of Data and Model Uncertainty for Deep Learning-based Streamflow Prediction深度学习径流预报中的数据与模型不确定性量化
Peer-reviewed同行评审
Open access开放获取
Type: Research article类型：研究论文
~18 min read约 18 分钟
Keywordsdata vs model uncertaintydeep-learning streamflow predictionaleatoric-epistemic decompositionuncertainty quantificationrainfall input error关键词数据与模型不确定性深度学习径流预报偶然-认知不确定性分解不确定性量化降水输入误差
Authors: Kaidi Peng, Daniel B. Wright, Lei Yan, G. Aaron Alexander, Yagmur Derin · Source: Hydrology and Earth System Sciences (HESS) · Published: 2026-07-22
作者：Kaidi Peng、Daniel B. Wright、Lei Yan、G. Aaron Alexander、Yagmur Derin · 来源：Hydrology and Earth System Sciences (HESS) · 发布：2026-07-22
Relevance 9.0/10相关性 9.0/10
Summary简要摘要
An uncertainty-aware deep-learning streamflow prediction system that explicitly accounts for the three main error sources — uncertain satellite rainfall forcing, uncertain streamflow observations, and the model itself — without requiring perfect rainfall input. Rather than a single point forecast, it produces predictions with characterized uncertainty.
一套"不确定性感知"的深度学习径流预报系统：显式考虑三大误差来源——不确定的卫星降水驱动、不确定的径流观测，以及模型本身——且无需完美的降水输入。它给出的不是单点预报，而是带有不确定性刻画的预测。
Key findings关键发现
- Jointly propagating data and model uncertainty yields more reliable, better-calibrated streamflow predictions than a deterministic model.
同时传播数据与模型不确定性，比确定性模型给出更可靠、校准更好的径流预测。
- The framework works with imperfect satellite rainfall — the realistic setting for data-scarce and ungauged basins.
框架可在不完美的卫星降水下工作——这正是数据稀缺与无资料流域的真实情形。
- Separating data-driven vs. model-driven uncertainty clarifies whether better forcing or a better model would most improve a given forecast.
区分"数据来源"与"模型来源"的不确定性，能说明对某次预报而言，是改进驱动数据还是改进模型更有效。
Why it's worth reading为什么值得读
Operational flood/streamflow warning lives or dies on trustworthy uncertainty, not just a mean prediction. This paper treats satellite-rainfall error as a first-class citizen — exactly the regime you operate in — and offers a decomposition you can act on.
业务化洪水/径流预警的成败取决于可信的不确定性，而不仅是一个均值。本文把卫星降水误差当作"一等公民"处理——正是你所处的场景——并给出可据以行动的分解。
Relevance to your research与你研究的关联
Core to AQUAH, Hydrometeorology AI and satellite-precipitation-driven hydrology: a natural component for an AI hydrology agent that must report calibrated confidence, and directly useful for flash-flood datasets built on remotely-sensed rainfall.
与 AQUAH、Hydrometeorology AI 及卫星降水驱动的水文密切相关：可作为"必须报告校准置信度"的水文 AI agent 的天然组件，也直接服务于基于遥感降水构建的山洪数据集。
Suggested reading path推荐阅读路径
Read the introduction's framing of the three uncertainty sources, then jump to the results that decompose data vs. model uncertainty, then the conclusions. ~18 min.
先读引言对三类不确定性的界定，再看分解"数据 vs 模型"不确定性的结果，最后读结论。约 18 分钟。
⚠ Risks / limitations 风险 / 局限:
Verify the exact basins/regions and how the uncertainty estimates are validated (coverage/reliability diagrams). Decomposition quality depends on the assumed error models; generalization beyond the tested catchments is not guaranteed.
需核实具体流域/区域，以及不确定性估计如何验证（覆盖率/可靠性图）。分解质量依赖所假设的误差模型；对测试流域之外的泛化并无保证。
Read original ↗ (HESS / ESS Open Archive)阅读原文 ↗（HESS / ESS Open Archive）
＋
Strong Selections优选推荐
#3
Overall rank综合排名
An Unprecedented View of Ocean Currents from Geostationary Satellites (GOFLOW)GOFLOW：从地球静止卫星获得前所未有的海洋流场视角
Peer-reviewed同行评审
Subscription (not OA)订阅（非开放获取）
Type: Research article类型：研究论文
~15 min read约 15 分钟
Keywordsgeostationary SST imagerysurface current reconstructionocean-simulation-trained AIinfrared image sequenceshigh-resolution current fields关键词静止卫星海表温度影像表层流场重建海洋模拟训练AI红外图像序列高分辨率流场
Authors: Roy Barkan et al. (Tel Aviv Univ.; Scripps, UCLA, Univ. Rhode Island) · Source: Nature Geoscience · Published: 2026-07-20
作者：Roy Barkan 等（特拉维夫大学；Scripps、UCLA、罗德岛大学）· 来源：Nature Geoscience · 发布：2026-07-20
Relevance 7.0/10相关性 7.0/10
Summary简要摘要
GOFLOW is an AI system trained on ocean simulations that reads sequences of infrared (sea-surface-temperature) satellite images and reconstructs high-resolution surface current fields — resolving dynamics below ~30 km and delivering the first direct, satellite-based estimates of horizontal ocean divergence (a proxy for vertical water movement).
GOFLOW 是一套在海洋模拟上训练的 AI 系统：读取红外（海表温度）卫星影像序列，重建高分辨率表层流场——可分辨约 30 km 以下的动力过程，并首次由卫星直接估计海洋水平散度（垂向运动的代理量）。
Key findings关键发现
- Reconstructs fine-scale surface currents (<30 km) purely from thermal image sequences — no new sensor required.
仅凭热红外影像序列即可重建 30 km 以下的精细表层流场，无需新增传感器。
- Provides the first direct satellite measurement of horizontal divergence, revealing vertical motion in features like the Gulf Stream.
首次由卫星直接测得水平散度，揭示如湾流等结构中的垂向运动。
- These fine-scale processes govern mixing, heat and gas exchange, storm intensity and CO₂ uptake.
这些精细过程支配着混合、热与气体交换、风暴强度及 CO₂ 吸收。
Why it's worth reading为什么值得读
A flagship example of learning an inverse operator: AI turns an existing observable (thermal radiance) into a new geophysical variable (currents, divergence). The "simulation-trained, applied to real imagery" recipe transfers cleanly to hydrology.
这是学习反演算子的标杆案例：AI 把已有的可观测量（热辐射）转化为新的地球物理量（流场、散度）。"以模拟训练、应用于真实影像"的思路可平滑迁移到水文。
Relevance to your research与你研究的关联
The radiance-to-state inversion paradigm mirrors satellite-hydrology and foundation-model efforts (e.g., mapping radiance to precipitation or water storage) central to planetary/global hydrology and scientific foundation models. The storm-intensity link also touches hazards.
"辐射→状态量"的反演范式，与卫星水文及基础模型（如把辐射映射为降水或水储量）一脉相承，正是行星/全球尺度水文与科学基础模型的核心；其与风暴强度的联系亦触及灾害主题。
Suggested reading path推荐阅读路径
Look first at the Gulf Stream divergence maps, then read how the model is trained from simulations and applied to real imagery (the sim-to-real design). Skip deep oceanographic derivations if your interest is the method. ~15 min.
先看湾流散度图，再读模型如何"以模拟训练、用于真实影像"（sim-to-real 设计）。若只关心方法，可跳过深入的海洋学推导。约 15 分钟。
⚠ Risks / limitations风险 / 局限:
Ocean, not land hydrology — relevance is by analogy. Simulation-trained models carry a sim-to-real gap, and satellite-derived divergence is hard to validate directly. Paywalled.
属海洋而非陆地水文，相关性以类比为主。以模拟训练的模型存在 sim-to-real 差距，卫星反演的散度也难以直接验证。付费墙。
Read original ↗阅读原文 ↗
#4
Overall rank综合排名
Causal AI for U.S. Water Risk: Semiconductor Fabs vs. Agricultural Irrigation用因果 AI 评估美国水资源风险：半导体工厂与农业灌溉的权衡
Peer-reviewed同行评审
Subscription (news open)订阅（新闻可读）
Type: Research article类型：研究论文
~12 min read约 12 分钟
Keywordscausal inference frameworksemiconductor fab water demandagricultural irrigation trade-offmulti-sector data fusionwater availability attribution关键词因果推断框架半导体工厂用水农业灌溉权衡多部门数据融合水资源可用性归因
Authors: Pincus et al. (lead: Feras Batarseh, Virginia Tech) · Source: J. Water Resources Planning & Management (ASCE) · Published: 2026-07-10
作者：Pincus 等（负责人：Feras Batarseh，弗吉尼亚理工）· 来源：Journal of Water Resources Planning & Management (ASCE) · 发布：2026-07-10
Relevance 7.5/10相关性 7.5/10
Summary简要摘要
A causal AI framework fuses agriculture, hydrology, climate and industrial data to quantify cause-and-effect between water availability, crop needs, and semiconductor-fab expansion across all 50 U.S. states and their basins — and to simulate policy and infrastructure scenarios.
一套因果 AI 框架，融合农业、水文、气候与工业数据，量化"水资源可得性、作物需水与半导体扩张"之间的因果关系，覆盖全美 50 州及其流域，并可模拟政策与基础设施情景。
Key findings关键发现
- Decisions in one region ripple nationally — e.g., a new fab in Arizona can affect irrigation capacity in neighboring states.
一地的决策会向全国传导——例如亚利桑那新建晶圆厂会影响邻州的灌溉能力。
- Agriculture accounts for ~70% of national freshwater withdrawals.
农业约占全美淡水取用量的 70%。
- Improving irrigation efficiency by 10–20% could free capacity for industrial growth without raising water stress.
将灌溉效率提高 10–20%，可在不加剧水压力的前提下为工业增长腾出容量。
- Designed to support decisions from state managers up to federal agencies.
面向从州级管理者到联邦机构的多层级决策。
Why it's worth reading为什么值得读
Most water-AI work is correlational and physical; this one foregrounds causal reasoning and scenario simulation over coupled human–water systems — a useful complement when the question is "what should we do," not just "what will the river do."
多数水-AI 工作偏相关与物理；这篇突出因果推理与耦合"人-水"系统的情景模拟——当问题是"我们该怎么做"而不仅是"河流会怎样"时，是有益的补充。
Relevance to your research与你研究的关联
Points to a decision/agent layer on top of physical hydrology — the kind of reasoning an AI hydrology agent (AQUAH-style) would need for drought and allocation questions, and a bridge between hydrologic simulation and water-policy analysis.
指向物理水文之上的决策/agent 层——正是水文 AI agent（AQUAH 式）处理干旱与配水问题所需的推理，也是水文模拟与水政策分析之间的桥梁。
Suggested reading path推荐阅读路径
The phys.org write-up is the fastest way in (~6 min). For the method, skim the framework diagram and the Arizona-fab scenario, then the policy implications. Skip the detailed economic tables. ~12 min total.
phys.org 的报道是最快入口（约 6 分钟）。看方法时略读框架图与"亚利桑那晶圆厂"情景，再看政策含义，跳过详细经济表格。合计约 12 分钟。
⚠ Risks / limitations风险 / 局限:
Causal claims from largely observational data need scrutiny (identifiability of the causal graph). National aggregation can hide local nuance, and this is a decision model — not a physically-based hydrologic model.
主要基于观测数据的因果结论需谨慎审视（因果图的可辨识性）。全国尺度聚合可能掩盖局部差异；它是决策模型，并非基于物理的水文模型。
Read summary ↗ (phys.org)阅读综述 ↗（phys.org）
·
Journal (DOI) ↗期刊原文 (DOI) ↗
≡
Briefs关键资讯
6 short items — scan, don't study6 条短讯 —— 供快速扫读
AI
Moonshot AI releases Kimi K3, the largest open-weight LLM yet月之暗面发布 Kimi K3，迄今最大的开源权重大模型
A ~2.8T-parameter MoE with a 1M-token context and native multimodality — a strong open base Earth/geo teams can fine-tune, narrowing much of the gap with frontier models.
约 2.8 万亿参数的 MoE，具备 100 万 token 上下文与原生多模态 —— 为地球/地学团队提供可微调的强开源底座，大幅缩小与前沿模型的差距。
AI
Alibaba previews Qwen3.8-Max, a 2.4T multimodal model阿里预览 Qwen3.8-Max，2.4 万亿参数多模态模型
Alibaba's 2.4T sparse-MoE Qwen handles text, image, video and documents with a 1M-token context — days after Kimi K3, a sign of how fast the open-model frontier is moving for imagery-reasoning pipelines.
阿里 2.4 万亿参数稀疏 MoE 的 Qwen，可处理文本、图像、视频与文档，具 100 万 token 上下文 —— 紧随 Kimi K3，显示影像推理管线所依赖的开源前沿推进之快。
Earth observation对地观测
NASA issues the Landsat 10 spacecraft Request for ProposalNASA 发布 Landsat 10 卫星招标（RFP）
The RFP (launch readiness by 2031) continues the ~50-year Landsat land-imaging record — vital for long-term continuity of the water-quality, snow and land data streams the community relies on.
该招标（2031 年具备发射条件）延续约 50 年的 Landsat 陆地成像记录 —— 对水质、积雪与陆地数据流的长期连续性至关重要。
Policy · funding政策 · 经费
WRDA 2026 advances with drought-resilience & flood-risk programsWRDA 2026 推进，新增抗旱与洪水风险管理项目
The House committee advanced the Water Resources Development Act of 2026 with two new continuing-authority programs for drought resilience and flood-risk management, plus 133 new feasibility studies — a major US water-infrastructure policy signal.
众议院委员会推进《2026 水资源开发法案》，新增两项针对抗旱与洪水风险管理的常设授权项目，并新增 133 项可行性研究 —— 美国水基础设施领域的重大政策信号。
Report报告
ISciences Worldwide Water Watch List, July 2026ISciences 全球水情观察清单（2026 年 7 月）
The monthly global outlook flags a possibly record El Niño, a historic European water emergency, and a drought-to-flood reversal across Central Asia and the Middle East — concise scanning of global hydrological anomalies.
该月度全球展望提示可能创纪录的厄尔尼诺、欧洲历史性缺水危机，以及中亚/中东"旱转涝"反转 —— 全球水文异常的简明扫读来源。
Report · attribution报告 · 归因
World Weather Attribution: warming is driving Europe's drought世界天气归因：变暖正在加剧欧洲干旱
A rapid study found 2026 Europe's agricultural soil droughts ~5× (west) to 11× (east) more likely due to climate change, with the evaporative extremes up to ~80× more probable — timely and quotable.
一项快速归因研究发现，2026 年欧洲农业土壤干旱因气候变化而出现的可能性提高约 5 倍（西）至 11 倍（东），高蒸发极端事件概率最高提升约 80 倍 —— 及时且可引用。
✦
Frontier Concepts前沿概念
3 concepts recurring across this issue's picks本期选题中反复出现的 3 个概念
1 · Differentiable modeling1 · 可微分建模
Definition: Building physical models (e.g., PDE solvers) entirely from differentiable operators, so gradients flow through the whole simulation — unifying process-based physics with ML calibration and coupling.
定义：用可微分算子完整地构建物理模型（如 PDE 求解器），使梯度能贯穿整个模拟——把过程物理与机器学习的率定/耦合统一起来。
Why it matters: It enables gradient-based calibration, learnable parameterizations, and physics–ML hybrids at scale — the engine behind this issue's #1 (Inunda).
为何重要：它支撑基于梯度的率定、可学习参数化与大规模的物理–ML 混合，正是本期 #1（Inunda）背后的引擎。
Shen et al. 2023, Nat. Rev. Earth Environ. — Differentiable modeling ↗
neuralhydrology (tutorials + code) ↗
JAX (autodiff framework) ↗
2 · Aleatoric vs. epistemic uncertainty2 · 偶然性 vs. 认知性不确定性
Definition: Aleatoric uncertainty is irreducible noise in the data/forcing; epistemic uncertainty is reducible model/parameter uncertainty. Separating them tells you whether more data or a better model will help most.
定义：偶然性不确定性是数据/驱动中不可约减的噪声；认知性不确定性来自模型/参数，可通过更多数据或更好模型减少。区分二者能告诉你"补数据还是改模型"更有效。
Why it matters: Trustworthy operational streamflow/flood forecasting needs calibrated uncertainty, not just a mean — the core of this issue's #2 (HESS).
为何重要：可信的业务化径流/洪水预报需要校准的不确定性而非仅有均值，正是本期 #2（HESS）的核心。
Kendall & Gal 2017 — Uncertainties in Bayesian deep learning ↗
Klotz et al. 2022, HESS — Uncertainty in DL streamflow ↗
3 · Satellite radiance-to-variable inversion with AI3 · 卫星"辐射→地球物理量"的 AI 反演
Definition: Using ML to learn the inverse operator that maps raw satellite radiances/imagery directly to geophysical state variables (currents, water storage, precipitation).
定义：用机器学习学习反演算子，把原始卫星辐射/影像直接映射为地球物理状态量（洋流、水储量、降水）。
Why it matters: It turns existing EO archives into new observations without new sensors — the idea behind this issue's #3 (GOFLOW) and radiance-based hydrology foundation models.
为何重要：它无需新传感器就能把已有对地观测存档变成新观测，是本期 #3（GOFLOW）及"辐射驱动"水文基础模型的共同思路。
Prithvi-EO-2.0 (NASA/IBM geospatial FM) ↗
Survey: Remote-sensing foundation models ↗
Appendix附录
Past issues: browse the full archive ↗往期内容：浏览全部往期 ↗
Coverage & method note覆盖范围与方法说明
Window: 2026-07-09 → 2026-07-23. This inaugural issue features 4 genuinely strong, in-window items — quality over quantity, with no padding by weakly-related work. Two are peer-reviewed journal papers, one is a peer-reviewed Nature Geoscience study, and one is a clearly-marked preprint. Summaries, analysis, relevance and reading tips all switch with the EN/中文 toggle.
窗口：2026-07-09 → 2026-07-23。本创刊号收录 4 篇真正高质量且落在窗口内的文章 —— 宁缺毋滥，不用弱相关内容凑数。其中两篇为同行评审期刊论文，一篇为同行评审的 Nature Geoscience 研究，一篇为明确标注的预印本。摘要、分析、关联与阅读建议均随中英切换。
Disclaimer. Content is curated and summarized by an automated agent — we summarize and interpret only; no article text is reproduced, and all links go to original sources whose authors and publishers hold all rights. Preprints are marked and are not peer-reviewed. Scores are editorial judgments, not the venues'. Please verify claims against the originals before citing.
免责声明。本刊内容由自动化 agent 整理与摘要 —— 我们只做摘要与解读，不复制文章正文；所有链接均指向原始来源，版权归原作者与出版方所有。预印本已标注且未经同行评审。评分为编辑判断，非期刊官方评分。引用前请对照原文核实。
Hydrology × AI Weekly Reading · Issue #1 · Coverage 2026-07-09 → 2026-07-23 · Pushed 2026-07-23 18:19 CDT · Next issue 2026-07-30 · Updated every Monday 07:00 America/Chicago.
水文 × AI 每周精读 · 第 1 期 · 覆盖 2026-07-09 → 2026-07-23 · 推送 2026-07-23 18:19 CDT · 下期 2026-07-30 · 每周一 07:00（America/Chicago）更新。