New papers, each cut down to one plain-language page. Read the finding in two minutes, then follow the link to the original.
In one climate model, the Southern Ocean's heat uptake under 4×CO2 is mostly set by the old circulation, with upwelling fixing its size and along-isopycnal diffusion fixing its slow decline.
In a model, hydrogen-powered aircraft make contrail cirrus with much less warming than jet fuel aircraft in most cases, but lubrication oil droplets could erase the benefit.
AICON, a graph neural network trained on high-resolution ICON data without multi-step rollout, is now operational at DWD and competes with ICON near the surface, but it smooths fine scales and underestimates cyclone intensity.
A differentiable weather model lets a simple long-window 4D-Var, with no background-error term, fit surface-pressure observations and give analyses closer to ERA5 than 20CRv3 over three months.
A nine-year satellite map shows slush is the largest mapped surface meltwater feature on the Greenland Ice Sheet, growing and reaching higher in high-melt years.
The fourth annual indicator update finds human-induced warming at 1.37 °C in 2025, rising at 0.27 °C per decade, with emissions at a record high but no longer growing rapidly.
Fossil emissions hit a new high in 2024 and are projected to rise again in 2025, while 2024's record atmospheric CO2 growth left the carbon budget about 1.7 GtC out of balance.
ScenarioMIP for CMIP7 sets out seven emission scenarios, from High to Very Low, to be run mostly with CO2 emissions driving the models, with some extended to 2500.
In an industrial coal region of Shanxi, using the observed three-peak particle size and internal mixing, instead of simple standard assumptions, changes estimated absorption and radiative forcing by amounts comparable to local greenhouse gas forcing.
With eight GPUs, the RIM2D flood model can simulate a 48-hour rainstorm over all of Berlin fast enough for early warning, at 10 m, 5 m and even 2 m resolution.
In the 2022 drought, extra glacier melt offset water deficits in catchments with around 15 % glacier cover, yet shrinking glacier area meant total meltwater was already below the 2003 level in most catchments.
Swiss rain and river TFA levels have climbed several-fold since 1996/1997, and simulated breakdown of known fluorinated gases explains only about 63 % of the TFA deposited on the Swiss Plateau.
The authors used a neural network to turn Argo float temperature, salinity and sometimes oxygen into 1°×1° maps of Southern Ocean carbon chemistry, including anthropogenic carbon, with uncertainties.
A 30 m annual map of China's forest biomass from 1985 to 2023 shows aboveground carbon stock rising from 5.50 to 13.97 PgC, mostly through growth of existing forest.
GFEI v3 maps nationally reported 2020 fossil fuel methane emissions onto a 0.1° grid using new infrastructure and coal mine data, and it stays below other inventories and satellite-based estimates.
Two radionuclide tracers show that in 2021 Pacific-influenced surface water met Atlantic-origin water at the Lomonosov Ridge, that this front moved from the Makarov Basin toward the ridge between 2011/2012 and 2015, and that Atlantic layer ages rose from 2015 to 2021 in several regions.
The authors released a labeled set of radar image patches with 3225 oil objects and many look-alike scenes, so oil spill detectors can be tested on a common benchmark.
Satellite data show that flash droughts starting from water-stressed land have stronger surface heating, and in parts of Africa low vegetation water content 1–2 months earlier goes with hotter peaks.
Extreme summer sea temperatures in the Arabian Gulf come from weaker local winds and trapped moist air, and ENSO and the NAO together explain over 50 % of the year-to-year variation.
In a benchmark of 11 Transformer-type models against an LSTM, the LSTM was best at regression and short forecasts, while attention-based models did better at long-horizon autoregression and zero-shot forecasting.
Using only PlanetScope satellite images plus existing footprint data, the authors built the first complete global set of building outlines, 3 m height maps and simple 3D models, with height errors of 1.5 to 8.9 m by continent.
Adding a two-moment ice scheme and a ten-member ensemble to the ICON model raises the share of ice-supersaturated regions detected from 0.4 to 0.6 for one run, and to 0.8 for a three-member vote.
Using AGAGE station measurements and a simple 12-box atmosphere model, the authors publish yearly global emissions and abundances for 42 trace gases through 2023, and explain how they are derived and where they fall short.
In a climate model, removing CO2 back to its starting level does not bring back large diatoms in the Subpolar North Atlantic, because weaker ocean circulation keeps surface nutrients low, and this cuts carbon export.
Agricultural land expansion, not climate change, is the primary driver of intensifying water crises in arid regions.
Bright is the first open, global dataset combining pre-disaster optical and post-disaster radar images of buildings, and models trained on it work well within known events but struggle on new ones.
The EAWS Matrix is a decision-support tool developed by European avalanche forecasters that assigns danger levels by linking snowpack stability, how frequently it occurs, and expected avalanche size.
Over 2003–2023, Saharan dust reached an Italian mountain station on about 58 days a year, with no trend, and longer summer events brought the largest coarse-particle rises.
The 2022 drought showed that European water managers see rising drought risk but mostly respond with short-term, supply-side fixes, so the authors call for a binding European Drought Directive.
HTAP_v3.2 blends official regional emission inventories with the EDGAR global inventory into one monthly, gridded dataset for nine air pollutants over 2000–2020.
Using 16 years of satellite lidar data, the authors find that dust in the lowest atmospheric layer exceeds WHO annual guidelines for fine particles in 87.7 % of 81 large cities, though dust is mostly declining.
Combining a glider with SWOT altimetry, researchers saw a smaller, stronger eddy approach, co-rotate with and merge into the Lofoten Vortex in April 2023, spinning it up.
An XGBoost model predicted next-month hydrological drought category correctly 79.9 % of the time in the Huaihe River Basin, and the precipitation drought index (SPI) mattered most, though it was weak on severe drought.
Using TROPOMI satellite data, the authors estimate global methane emissions rose from 560 Tg a−1 in 2019 to 587–592 Tg a−1 in 2020–2021, with sub-Saharan African wetlands, especially in South Sudan, likely playing a major role.
When fed extreme design rainfall, a stand-alone LSTM runoff model hits a ceiling well below flows it saw in training, and its runoff share falls as rain rises, while a hybrid model scales more sensibly.
Many studies labelled multi-hazard simply layer separate single-hazard results, and few develop indicators that capture how hazards interact or how risks combine.
Swath radar shows long, streamlined bed ridges (mega-scale glacial lineations) beneath ice moving only about 60 m per year near the Northeast Greenland Ice Stream onset, which challenges the idea that they need fast flow.
In a snow model of Argentière Glacier, mineral dust lowered annual surface mass balance by 0.31 to 0.45 m w.e. in 2019–2021 and by 0.63 m w.e. in 2022, most strongly just above the previous summer's snowline.
Across three weather datasets and three methods, meteorology almost always raised China's surface ozone from 2013 to 2022, but the choice of method changes the size of that effect more than the choice of dataset.
Winter PM2.5 at a Pearl River Delta background site fell from 87.1 to 34.0 µg m−3 over 14 years, but nitrate dropped more slowly than sulfate, so secondary species and nitrate now matter more.
In a Tibetan Plateau lake basin, ground resistivity surveys and borehole temperatures showed that three small thermokarst lakes have thawed the permafrost below them all the way through.
In 15 isolated South China thunderstorms, graupel and supercooled rain content gave the best 6-minute lightning forecasts, while ZDR column volume looked better at 12 minutes.
In parts of the Northern Hemisphere, mostly deserts, montane shrublands and tundra, soil moisture can be negatively tied to precipitation, and temperature and vegetation productivity help explain it.
Giving an LSTM recent streamflow observations raises its accuracy most at the daily scale, while snow observations help only at the monthly scale, mainly in snow-dominated basins.
An integrated system forecasting farm profits and crop yields provides reliable drought warnings months ahead of rainfall-based indicators.
When nine CMIP6 models double dust emissions, they differ a lot in how much energy dust adds or removes, yet eight of nine show less precipitation.
Year-round lidar and radar data from near the North Pole suggest long-lived Arctic mixed-phase clouds persist because new droplets keep forming and ice-nucleating particles never run short.