Evidence of human influence on Northern Hemisphere snow loss

Seasonal snow is often considered a sentinel of climate change because warmer winters shift precipitation from snow to rain and enhance snowmelt, altering water storage and hydrologic risks. Yet despite robust warming across the Northern Hemisphere since 1981, observational snow products do not agree on the sign and magnitude of snowpack trends at hemispheric, continental or basin scales. This discordance hampers clear detection and attribution of anthropogenic impacts on snow and limits preparedness for water security risks. The study asks where, when and by how much human-caused climate change has altered Northern Hemisphere March snow water equivalent (SWE), and whether a robust anthropogenic fingerprint can be detected at hemispheric and river-basin scales. It further examines how forced changes in temperature and precipitation have influenced snowpack and characterizes the nonlinear sensitivity of snowpack to warming to inform projections of water resource impacts.

The IPCC AR6 assessed with high confidence that Northern Hemisphere spring SWE has generally declined since 1981, but uncertainties in SWE products complicate precise attribution. Prior work highlights large observational uncertainties in SWE estimates and substantial internal climate variability (e.g., Pacific Decadal Oscillation, Atlantic Multidecadal Variability) that modulates snowpack. Earlier attribution studies in specific regions have used limited models or realizations and sometimes normalized SWE by cold-season precipitation to isolate temperature effects, but such approaches may miss the full influence of climate change and conflate internal variability with model structural differences. Climate models show large spread in snow-related processes and regional precipitation responses. Studies document that warming can increase cold-season precipitation and snowfall extremes in some cold regions, potentially offsetting warming-driven SWE losses. Consequently, a robust, multi-dataset, multi-model approach is needed to detect and attribute human-forced snow changes while accounting for internal variability, observational uncertainty, and the potentially compensating effects of precipitation changes.

The study combines observations, empirical modeling, and climate model simulations to detect and attribute anthropogenic influences on Northern Hemisphere March SWE (1981–2020) at hemispheric and river-basin (169 major basins) scales.

1. Detection and attribution of spatial patterns: Using CMIP6 multi-model ensembles, the authors compare HIST (historical anthropogenic + natural forcing) and HIST-NAT (natural-only forcing) experiments. They compute spatial pattern correlations between observed 40-year SWE trends and the multimodel mean trends from HIST and HIST-NAT. A null distribution is constructed by correlating HIST trends against all possible 40-year trends from pre-industrial control (PIC) simulations to quantify the probability that observed similarity could arise from internal variability alone. Observational datasets include in situ stations and multiple gridded SWE products (ERA5-Land, JRA-55, MERRA-2, Snow-CCI, TerraClimate).

2. Basin-scale empirical reconstructions: The authors build an ensemble of observation-based statistical reconstructions of March SWE at the basin scale using a random forest machine-learning approach. Inputs include combinations of gridded snowpack, temperature, and precipitation datasets; models are trained to reproduce trends and variability in SWE at basin scale. Skill is assessed via spatial pattern correlations (0.9–0.97 against gridded products) and median RMSE (<8% across products/basins; ~22% against ~3,000 in situ sites). This ensemble samples observational, empirical model, and climate model uncertainties.

3. Forced vs. counterfactual decomposition: The forced responses of temperature and precipitation are estimated by differencing CMIP6 HIST and HIST-NAT simulations. These forced components are removed from observed temperature and precipitation time series to produce counterfactual, no-anthropogenic-climate-change reconstructions of SWE for each basin. Comparing historical reconstructions to counterfactual ones isolates the effects of anthropogenic forcing on SWE trends and allows separate attribution to forced temperature and forced precipitation changes.

4. Nonlinear sensitivity analysis: The marginal sensitivity of March SWE to 1 °C warming is quantified as a function of climatological winter (Nov–Mar) temperature using in situ, gridded products, climate models, and basin-scale reconstructions. A change-point analysis identifies the temperature at which SWE sensitivity becomes nonlinear.

5. Projections and exposure: Using CMIP6 projections under SSP2-4.5, the study assesses percentage changes in March SWE and April–June runoff (SWE-driven) for 2070–2099 relative to 1981–2020 and relates these to basin populations to quantify exposure. Population is binned by climatological temperature relative to the identified change point.

6. Uncertainty partitioning: Contributions to uncertainty from internal variability, climate model structural differences, and observational product differences are quantified across basins to identify dominant sources limiting attribution precision.

• Anthropogenic fingerprint in observed patterns: It is virtually certain (>99% probability) that human emissions contributed to the observed spatial pattern of Northern Hemisphere March SWE trends over 1981–2020 in in situ observations, the gridded ensemble mean, TerraClimate, and JRA-55. ERA5-Land and Snow-CCI show extremely likely probabilities (~97%). MERRA-2 shows no detectable anthropogenic influence (78%). Simulations without anthropogenic forcing (HIST-NAT) do not produce patterns distinguishable from internal variability.
• Basin-scale trends and attribution: Empirical reconstructions show consistent trend directions in 82/169 major basins, with many mid-latitude declines (~10% per decade in southwestern USA and Europe) and modest increases in cold high-latitudes. Anthropogenic forcing has altered March SWE trends in 31 basins. Forced changes reduce SWE in mid-latitudes (south of 60° N) by 4.1 ± 3.4% per decade and increase SWE in cold, high-latitude Arctic-draining basins by 2.5 ± 1.8% per decade.
• Masking by internal variability: In some basins (e.g., Columbia, Saint Lawrence), large forced SWE declines are detected despite small or positive observed trends, indicating internal variability has masked reductions. Conversely, basins like the Rio Grande show large observed declines but weak evidence of forced contributions, consistent with low-frequency variability (e.g., PDO) driving trends.
• Countervailing drivers: Forced temperature increases generally reduce SWE across most basins; forced precipitation increases offset a portion of these losses primarily in cold continental interiors. Outside the coldest regions, forced precipitation-driven increases are generally insignificant due to higher uncertainty and dominant internal hydroclimate variability.
• Nonlinear temperature sensitivity: A robust change point near −8 °C (climatological Nov–Mar temperature) marks where SWE sensitivity to 1 °C warming becomes highly nonlinear. Below ~−8 °C, SWE is relatively insensitive; above, marginal warming causes accelerating SWE losses. This change point is consistent across in situ, gridded, climate model, and reconstruction datasets.
• Distribution of snow mass vs. population: Over 80% of March snow mass lies in regions colder than −8 °C (insensitive to past warming), but about 80% of the Northern Hemisphere population lives in snow-dependent basins warmer than this threshold, implying rapidly emerging water security risks with further warming.
• Projected runoff declines under SSP2-4.5: Large spring (Apr–Jun) runoff declines are projected in populous mid-latitude basins due to nonlinear SWE losses, including upper Mississippi (−30.2%), Colorado (−42.2%), Columbia (−32.7%), and San Joaquin (−40.9%). In Europe: Danube (−41.0%), Volga (−39.5%), Rhine (−33.0%), and Po (−40.5%). Cold, sparsely populated Arctic-draining basins may see increases >10% in SWE-driven spring runoff.
• Uncertainty sources: Internal variability is the dominant contributor to forced SWE trend uncertainty in about 1 in 8 basins. Climate model structural differences dominate uncertainty in more than half of basins. Observational SWE product uncertainty is critical where in situ data are sparse.

The study demonstrates that human-caused warming has already altered the spatial pattern of Northern Hemisphere March SWE trends, addressing the long-standing question of detectability and attribution despite large observational uncertainties. By fusing multi-product observations, climate simulations, and empirical reconstructions, the authors isolate anthropogenic influences on basin-scale snowpack and reconcile why robust warming has not universally translated into detected snow loss: most snow mass resides in very cold regions that are insensitive to modest warming, while internal variability can mask forced declines elsewhere.

The identification of a consistent −8 °C change point in snow sensitivity provides a thermodynamic framework linking climatological temperature to SWE responsiveness. This framework explains observed spatial patterns—larger declines where winters are warmest—and implies rapid, widespread SWE losses as more basins warm beyond the threshold. Because most people live in basins to the warm side of this threshold, the results signal imminent water security challenges, including diminished spring runoff, shifts in seasonal streamflow timing, and increased competition for water resources.

The findings underscore that reliance on single observational products or small model ensembles can obscure forced signals; instead, multi-dataset, multi-model approaches can yield robust attribution at scales relevant for water management. The work also clarifies that natural variability will continue to modulate regional outcomes, necessitating risk-based planning that accounts for a range of plausible trajectories.

This study provides robust evidence that anthropogenic warming has contributed to Northern Hemisphere March snowpack declines since 1981, with clear attribution of forced changes in 31 major river basins. It reveals a generalizable, highly nonlinear sensitivity of snowpack to warming, with an inflection near −8 °C that explains muted historical detection across cold regions and forewarns of rapidly accelerating losses in populous, mid-latitude basins. Projections indicate substantial declines in SWE-driven spring runoff under SSP2-4.5 in many of the world’s most populated basins, while cold Arctic-draining basins may see increases.

Future research should prioritize: improving regional skill and process representation in climate models (especially cold-season precipitation and snow processes); enhancing SWE observing systems and data assimilation to reduce product uncertainty; better quantifying internal variability and its impacts on detection timelines; and integrating these improved constraints into water resource planning to mitigate emerging water security risks.

• Observational uncertainty: Large discrepancies among gridded SWE products and sparse in situ observations in key regions (e.g., High Mountain Asia, parts of Siberia) limit confidence in local trend magnitudes and validation.
• Model biases and uncertainty: CMIP6 models tend to overestimate historical winter warming in some regions and misrepresent multidecadal precipitation trends (e.g., drying in the southwestern USA, wetting over the Tibetan Plateau), though most biases fall within internal variability ranges. Structural differences among models are a dominant uncertainty source for forced SWE trends in many basins.
• Internal variability: Decadal to multidecadal variability (e.g., PDO, AMV) can mask or amplify forced signals over the 40-year period, complicating attribution at basin scales and making some results sensitive to sampling of variability.
• Product dependence: One dataset (MERRA-2) does not show a detectable anthropogenic influence on SWE pattern trends, illustrating sensitivity to observational product choice.
• Spatial heterogeneity of thresholds: The rain–snow and melt thresholds vary with topography and moisture sources, adding spread to estimated temperature sensitivities around the −8 °C change point.
• Precipitation uncertainty: Forced precipitation effects are less certain than temperature effects, limiting confidence in compensating precipitation-driven SWE changes outside cold interiors.