Paper Harvest Report

Date range: August 28, 2026

11 top-tier papers selected out of 42 total publications

Today’s Highlights

Today’s harvest is headlined by a landmark Nature paper demonstrating operational AI-based tropical cyclone forecasting that delivers an extra day of warning compared to conventional numerical models—a result with direct implications for flood early-warning and coastal risk management. Complementing this, a GRL attribution study shows that warming-driven precipitation phase transitions, not changes in precipitation amount, are the primary driver of declining global seasonal snowfall since 1983, with critical implications for future water resources. A cluster of AI-focused papers examines the reliability and biases of neural weather models for extreme tropical cyclone prediction, including the sobering finding that current AI models struggle to forecast “gray swan” cyclones—record-breaking events too rare to have appeared in their training data.

Today’s Highlights
Top-Tier Journal Papers
AI for Science
1. How AI is changing research
2. Cross-discipline sparks
Statistics
1. Papers by journal
Filtering Criteria

Top-Tier Journal Papers

Operational Tropical Cyclone Forecasting with AI

Authors: Ferran Alet, Tom R. Andersson, Ilan Price, Stratis Markou, Andrew El-Kadi, Dominic Masters et al.

Journal: Nature · DOI: 10.1038/s41586-026-10953-2

Matched topics: tropical cyclone, flood, coastal, AI forecasting

Abstract not available.

Global Improvement of Tropical Cyclone Representation Enhances Hindcasts of Extreme Sea Levels Over 68 Years

Authors: Pengcheng Wang, Natacha B. Bernier, Sean Xie

Journal: Geophysical Research Letters · DOI: 10.1029/2026gl124965

Matched topics: flood, coastal

The underrepresentation of tropical cyclones (TCs) in state‐of‐the‐art atmospheric reanalyses such as ERA5 is well known, yet their impact on global hindcasts of extreme sea levels remains largely unaddressed. We show that weak TC signals in ERA5 cause severe underestimation, and in some cases near omission, of TC‐induced storm surges. To address this limitation, we leverage recent advances in TC observations and parametric modeling to improve TC representation in ERA5 surface winds and pressure fields, producing a 68‐year global sea level hindcast. The hindcast shows consistent improvements in TC‐induced surges of ∼25% globally, with improvements also extending to areas north of 44°N. The largest gains (∼40%) occur around Australia, where individual surge underestimation is reduced by up to 1.6 m. These improvements translate into more reliable return period estimates at tide gauges and substantial changes in estimated return periods across TC‐prone regions, with important implications for coastal flood risk assessment.

Thermodynamic Control of Global Seasonal Snow Decline

Authors: Yang Song, Dawei Han, Guojun Gu, Jie Song, Ali Behrangi

Journal: Geophysical Research Letters · DOI: 10.1029/2026gl122806

Matched topics: seasonal

Seasonal snowfall is declining globally, yet the relative roles of precipitation variability versus warming‐induced phase transitions remain poorly quantified. We develop a physically based attribution framework that decomposes snowfall trends (1983–2023) into hydrological (precipitation‐driven) and thermodynamic (phase‐driven) components, constrained by a wet‐bulb‐temperature‐based phase relationship. Across most seasonal snow regions, snowfall decline is primarily driven by thermodynamic phase shifts rather than changes in precipitation amount. Regions near the rain–snow transition exhibit the strongest sensitivity to warming, consistent with systematic snowline retreat. Results are robust across independent data sets, including ERA5 reanalysis and the satellite‐based GPCP v3.3 products. These findings indicate that precipitation‐phase transitions exert pivotal control on contemporary snowfall loss, suggesting that projected precipitation increases are unlikely to offset warming‐driven reductions in snowfall. This has critical implications for future hydrological regimes and cryosphere‐climate feedbacks.

Dynamic Responses of Groundwater Flow and Salt Transport to Storm Surge Inundation in Unconfined Coastal Aquifers

Authors: Canhao Cai, Chengji Shen, Shichang Li, Chunhui Lu, Ling Li

Journal: Geophysical Research Letters · DOI: 10.1029/2026gl124091

Matched topics: coastal

Storm surges severely salinize coastal aquifers, yet post‐storm subsurface hydrodynamics remain poorly understood. We investigated surge inundation in unconfined tidal aquifers, revealing that subsurface flow recovers much faster than salinity distributions. Immediately post‐storm, submarine groundwater discharge rapidly increases to greatly exceed the pre‐storm level under extreme surge scenarios. Concurrently, downward surge infiltration creates a transient hydraulic barrier, forcing the saltwater wedge to temporarily retreat seaward, depending on surge height. Furthermore, we reveal the upper saline plume overshoot effect, as infiltrated storm water continues to migrate downward and seaward. This overshoot is exacerbated in micro‐tidal, low‐permeability aquifers. Understanding these asynchronous responses is critical for predicting groundwater quality and managing coastal resources under intensifying storm events.

Re‐Evaluating Inherited SIF Anomaly Detection Methods for the 2024 Amazon Drought: A 3D Convolutional Autoencoder Approach

Authors: Carmen Oliver Huidobro, Gerbrand Koren

Journal: Geophysical Research Letters · DOI: 10.1029/2026gl122340

Matched topics: drought

In the past decade, the Amazon has experienced multiple severe droughts, raising critical questions about how vegetation stress can be detected using remote sensing indicators such as Solar‐Induced Chlorophyll Fluorescence (SIF). In this study, we compare two anomaly detection approaches applied to TROPOSIF data. First a classical pixel wise z‐score method, and second a 3D Convolutional Autoencoder (CAE) that learns spatiotemporal SIF patterns. We evaluate both methods against two drought products, the global Standardised Precipitation‐Evapotranspiration Index (SPEI) metric and Brazil’s Monitor de Secas (MSB). Our results show that the CAE identifies broader, more persistent anomalies than the z‐score method and aligns more closely with drought patterns detected by SPEI and MSB, particularly during the peak dry period in September 2024 and the months of delayed recovery that followed. These findings highlight how methodological choices shape the interpretation of vegetation stress and why anomaly detection methodologies should be carefully selected.

Emergent Regimes of River Meandering and Permafrost Extent in Arctic Floodplains

Authors: Michael P. Lamb, Emily C. Geyman, Madison D. Douglas, Ajay B. Limaye

Journal: Geophysical Research Letters · DOI: 10.1029/2026gl124384

Matched topics: river, flood

Permafrost in Arctic floodplains impacts carbon stores, ecosystems, and infrastructure, yet the processes controlling its distribution remain poorly constrained. Here we present a numerical model for the coupled evolution of river meandering and floodplain permafrost that captures the competition between bank erosion and permafrost growth. We show that permafrost extent reaches a dynamic equilibrium in which removal by channel migration is balanced by epigenetic permafrost generation on newly deposited land. This behavior collapses onto a regime space defined by two dimensionless ratios—the rate of bank thaw relative to sediment entrainment and the timescale of permafrost growth relative to meander‐bend cutoff—which organizes three emergent regimes: permafrost‐bounded, confined, and alluvial meanders. Rivers in the confined regime are most sensitive to warming, in which modest increases in bank thaw rates or the permafrost‐growth timescale produce large increases in migration rates and substantial permafrost loss.

Probabilistic Storylines: Characterizing the Likelihood of Impactful Events in an Uncertain World

Authors: Andrew D. Jones, Julia M. Longmate, Smitha Buddhavarapu, William Gutowski, Antonia Hadjimichael, Kripa Jagannathan et al.

Journal: Bulletin of the American Meteorological Society · DOI: 10.1175/bams-d-24-0301.1

Matched topics: flood, drought

Storylines are a useful approach for understanding and preparing for impactful events. For instance, resource managers often use “design droughts” or “floods of record” to stress-test system vulnerabilities and consider alternative responses. While plausible storylines are sufficient for many applications, there is a growing interest in rigorously evaluating storyline probabilities to inform risk-based adaptation investments, especially in cases where preparing for storylines is costly or where storyline probabilities are changing over time. However, assigning actionable probabilities to storylines presents conceptual, methodological, computational, and data challenges. One key challenge is how to reconcile deeply uncertain aspects of the future arising from different scenarios and model structural biases with those aspects that are more amenable to statistical characterization, such as stochastic variability. Another challenge is translating the idiosyncratic features of specific storylines into decision-relevant metrics that can be understood statistically, ideally delivering probability distributions of salient events. Informed by our experience co-producing storyline simulations with resource managers and Earth system scientists, and by a critical literature review, we propose a conceptual framework for rigorously incorporating probabilities into storyline-based analysis and planning.

Global Forecasting of Tropical Cyclone Intensity Using Neural Weather Models

Authors: Milton Gomez, Louis Poulain‐Auzéau, Alexis Berne, Tom Beucler

Journal: Artificial Intelligence for the Earth Systems · DOI: 10.1175/AIES-D-25-0073.1

Matched topics: tropical cyclone, machine learning, flood, coastal

Numerical Weather Prediction (NWP) models that integrate coupled physical equations forward in time are the traditional tools for simulating atmospheric processes and forecasting weather. With recent advancements in deep learning, AI-based Weather Prediction models that rely on neural network architectures–Neural Weather Models (NeWMs)–have emerged as competent medium-range NWP emulators, with performances that compare favorably to state-of-the-art NWP models. However, they are commonly trained on reanalyses with limited spatial resolution (e.g., 0.25° horizontal grid spacing), which smooths out key features of weather systems. For example, tropical cyclones (TCs)—among the most impactful weather events due to their devastating effects on human activities—are challenging to forecast, as extrema are smoothed in deterministic forecasts at 0.25° resolution. To address this, we use our best observational estimates of wind gusts and minimum sea level pressure to train a hierarchy of post-processing models on NeWM outputs. Applied to Pangu-Weather and FourCastNet v2, the post-processing models produce accurate and reliable forecasts of TC intensity up to five days ahead.

Can AI‐Based Weather Prediction Models Simulate the Butterfly Effect? The Role of Architecture and Implementation

Authors: T. Selz, G. C. Craig

Journal: Journal of Geophysical Research: Hydrology · DOI: 10.1029/2025JH001180

Matched topics: machine learning, climate change, earth system model

Simulations of numerical weather prediction models indicate that the atmosphere possesses an intrinsic limit of predictability. Initial perturbations of tiny amplitude grow quickly in areas of convection and latent heat release, then spread out and move upscale, eventually affecting even the largest planetary scales after about 2 weeks. In this study, we investigate the ability of several state‐of‐the‐art AI‐based weather prediction models to reproduce this phenomenon, which is sometimes referred to as the “butterfly effect.” The AI results are compared to those of a conventional, physics‐based, weather prediction model run at various resolutions. Evaluating six key characteristics of this butterfly effect, we find that the behavior of the AI models can be separated into two groups. The first group did not reproduce any of the key characteristics, while the second group did reproduce some, in particular fast initial uncertainty growth and indication of an intrinsic limit. However, the behavior was physically inconsistent and based on the production of numerical noise. It seems likely that the inability of AI models to simulate the butterfly effect results from limitations in the analysis data used for training, since their size, design and architecture turned out to be largely irrelevant.

Can AI Weather Models Predict Out‐of‐Distribution Gray Swan Tropical Cyclones?

Authors: Y. Qiang Sun, Pedram Hassanzadeh, Mohsen Zand, Ashesh Chattopadhyay, Jonathan Weare, Dorian S. Abbot

Journal: Proceedings of the National Academy of Sciences · DOI: 10.1073/pnas.2420914122

Matched topics: tropical cyclone, flood, machine learning

AI models produce skillful weather forecasts, including for some extreme events. However, forecasting the strongest events that are so rare they did not exist in the training set (the so-called gray swans) remains a major concern for these models’ operational use, especially as climate change introduces unprecedented conditions. Here, we train an AI weather model after removing Category 3–5 tropical cyclones from its training set and test it on Category 5 storms. The model could not accurately forecast these unseen cyclones. However, the model shows promise in learning from strong storms in one region and forecasting them in another region. Our work highlights the need for better understanding the limitations of AI weather models and innovations to improve them.

Identifying and Categorizing Bias in AI/ML for Earth Sciences

Authors: Amy McGovern, Ann Bostrom, Marie McGraw, Randy J. Chase, David John Gagne, Imme Ebert-Uphoff, Kate D. Musgrave, Andrea Schumacher

Journal: Bulletin of the American Meteorological Society · DOI: 10.1175/BAMS-D-23-0196.1

Matched topics: machine learning, earth system model

Artificial Intelligence (AI) can be used to improve performance across a wide range of Earth System prediction tasks. As with any application of AI, it is important for AI to be developed in an ethical and responsible manner to minimize bias and other effects. In this work, we extend our previous work demonstrating how AI can go wrong with weather and climate applications by presenting a categorization of bias for AI in the Earth Sciences. This categorization can assist AI developers to identify potential biases that can affect their model throughout the AI development life-cycle. We highlight examples from a variety of Earth System prediction tasks of each category of bias.

AI for Science

How AI is changing research

Tropical cyclones could be predicted with an extra day’s warning, thanks to an AI model (Nature, 2026-09-04) — An AI-based weather model now delivers tropical cyclone track forecasts with an additional 24-hour lead time compared to conventional numerical prediction — directly relevant to flood early-warning systems and hydrological impact modeling, where a single day’s extra warning can mean the difference between preparation and disaster.

Cross-discipline sparks

Embedding AI in biology — part 2 (Nature Methods, 2026-09-04) — Nature Methods’ ongoing series on integrating AI into biological research workflows — covering automated experimental design, hypothesis generation, and multimodal data fusion — maps directly onto unmet needs in hydrology: the same recipe of foundation-model pretraining on raw time-series (MODIS reflectance, river gauge records, reanalysis fields) followed by task-specific fine-tuning is something a small earth-science team could now replicate without a dedicated ML staff, dissolving a barrier that existed just two years ago.

Statistics

Metric	Count
Journals searched	11
Total papers fetched	42
Passed deterministic filter	11
Additionally from editorial citations	9
After LLM relevance filtering	11
Rejected (not relevant)	9
AI for Science items picked	2

Papers by journal

Journal	Papers
Geophysical Research Letters	5
Bulletin of the American Meteorological Society	2
Nature	1
Artificial Intelligence for the Earth Systems	1
Journal of Geophysical Research: Hydrology	1
Proceedings of the National Academy of Sciences	1

Filtering Criteria

Topics: hydrology, hydrologic model, river, runoff, streamflow, reservoir, water management, flood, drought, seasonal, land surface model, climate change, hydropower, surface water, irrigation, earth system model, estuary, coastal, freshwater discharge, river plume, ocean biogeochemistry, marine heatwave, paleohydrology, paleoclimate, Quaternary, Holocene, Pleistocene, fluvial geomorphology, river terrace, loess, drainage network, river capture, landscape evolution, luminescence dating

Fields: engineering, environmental science, computer science, geology, geography