A summary of all data used in this paper, along with the period of record, data type, source, impact class, and resolution is available in Table .

We obtained streamflow data by blending two main sources: (i) the databases of Italy's 21 Regional Administrations and Autonomous Provinces, made accessible to CIMA Research Foundation via the Italian Civil Protection system, and (ii) a comprehensive hydrological dataset covering the Alpine region for the period September 2004 to August 2023 (https://edp-portal.eurac.edu/geonetwork/srv/api/records/9e195271-02ae-40be-b3a7-525f57f53c80, last access 17 April 2025). Based on data availability at the time of the study, we focused here on the period of record 1 September 2010 through 31 August 2023.

Across more than 350 streamflow gauges that were available, we selected a subset of study catchments based on the following criteria: first, the time series of daily mean streamflow contained less than 20 % missing data for the period 2011–2023; second, the same time series did not exhibit spikes or suspicious values based on a visual inspection of the daily data (with suspicious values qualitatively defined as persistent periods of unseasonably or implausibly high or low streamflow based on expert knowledge); and third, the ratio between mean annual, basin-wide SWE estimated by the IT-SNOW reanalysis and cumulative annual streamflow during 2011–2023 was at least 5 %. The first two criteria ensured a high level of data completeness and quality while the third aimed to identify mountainous headwater catchments where snow significantly contributes to the water balance. In cases where more than one gauging station along the same river fulfilled all of the above requirements, we selected the one most upstream in order to focus our study specifically on mountain headwaters and to avoid the presence of nested catchments in our sample.

The final sample was composed of 38 catchments across the whole of the Italian latitudinal range. While most catchments were located in the Alps, additional coverage included the Apennines, a considerably less studied region of the Italian mountain landscape (see Fig. a). Median catchment size was 430 km2, with first and third quartiles equal to 200 and 870 km2, respectively. In terms of elevation, the median value across all catchments was 1830 ma.s.l., with first and third quartiles equal to 1270 and 1956 ma.s.l., respectively. Thus, the investigated catchments represent small-to-medium headwater river systems: for reference, the catchment areas at the sea outlet of the Po, Adige, and Tiber rivers, three of the largest rivers in Italy, are 71 000, 12 200, and 17 375 km2, respectively.

Spatially distributed SWE for the same 2011–2023 period was taken from the IT-SNOW reanalysis, an open-source, quasi-real-time, operational reanalysis of snow-cover patterns across Italy at 500 m and daily resolution . IT-SNOW has been extensively validated using both in-situ and remote-sensing data, typically returning root mean square errors (RMSE) on the order of 30–60 cm and 90–300 mm for in situ, measured snow depth and SWE, respectively. While these accuracies may appear large compared to the application of a snow model at the local scale see for example, they are in line with other similar large-scale reanalyses across the world . The reanalysis is currently available for the period 1 September 2010 through 31 August 2025 (10.5281/zenodo.6861722, ) and is constantly updated at the end of each water year (here defined as the period between 1 September and the following 31 August, labeled using the calendar year when it ends).

Monthly cumulative precipitation and average air temperatures were taken from the BIGBANG dataset, which provides quality-controlled estimates of the most essential water-balance variables across Italy at 1 km for the period of record 1951 to the near present. Both variables are derived from the interpolation of 2000+, quality-checked, in-situ stations across Italy . Precipitation maps are the result of a Natural-Neighbor algorithm as implemented in the Spatial Analyst of ESRI ArcGIS 10.3; a double interpolation is performed to filter out precipitation values below 1 mm. Air temperature maps are derived using a Double-Kriging approach that includes elevation and latitude as additional predictors .

For terrestrial-ecosystem impacts, we selected the remote-sensing-based GPP dataset integrated into the Penman–Monteith–Leuning Evapotranspiration product, PML_V2, with a resolution of 500 m and an 8 d interval. PML_V2 is estimated by a process-based water-carbon coupled model which uses vapor pressure deficit as a proxy of moisture stress . The PML_V2 products demonstrate strong performance when compared to observations at 95 flux sites worldwide, including 10 plant functional types and a wide range of climatic conditions worldwide. Validation was performed using a leave-one-out method, with Nash–Sutcliffe Efficiency (NSE)=0.76, R2=0.77, and RMSE=1.99 gCm-2d-1 for 8 d GPP. Seven of the calibration and validation sites were in Italy, where the PML_V2 GPP exhibited better performances than other global models, with NSE=0.40, R2=0.68, and RMSE=2.49 gCm-2d-1 on average. This performance is comparable to, or outperforms, major state-of-the-art ET and GPP products that are extensively utilized by the water and ecology research communities .

In addition to GPP, we also analyzed the greening date, as day of the calendar year, which was derived from the “Greenup” layer of the MODIS product MCD12Q2v061 . This refers to the first date when the Enhanced Vegetation Index (EVI2) exceeded 15 % of the EVI2 amplitude within the greenup segment . GPP and phenology timeseries for the target catchments were obtained from Google Earth Engine (collections “CAS/IGSNRR/PML/V2_v018” and “MODIS/061/MCD12Q2” respectively).

We complemented these remote-sensing data on terrestrial-ecosystem impacts with two ground-based case studies of GPP measured with flux towers: the first located in a European larch forest (Larix decidua Mill.), IT-TrF 45°49^′25.641^′′ N, 7°33^′39.131^′′ N, and the second in a subalpine grassland dominated by matgrass (Nardus stricta), IT-Tor 45°50^′40^′′ N, 7°34^′41^′′ N,. Both sites are part of the Integrated Carbon Observation System (ICOS) and are located in Aosta Valley (western Italian Alps) at 2160 m. In the last 20 years, no land management interventions or major disturbance have occurred at either site. As described in , the forest site is dominated by European larch (Larix decidua Mill., 92 % by relative abundance as a percentage of cover), with sporadic (8 %) Norway spruce (Picea abies (L.) H.Karst) individuals. Larch trees have no needles during dormancy, from nearly November to April/May. The forest understory is mainly composed of shrubs (Juniperus communis Willd., Rhododendron ferrugineum L., Vaccinium myrtillus L.). The grassland site is an abandoned mountain pasture, with dominant vegetation composed of Nardus stricta L., Festuca nigrescens All., Arnica montana L., Carex sempervirens Vill., and other minor species. Measurements of H2O and CO2 fluxes have been carried out since 2008 at the grassland site and since 2012 at the forest site by means of the eddy covariance technique . More details on measurement devices are available in Sect. S3 in the Supplement. The period of record for these flux-tower data went from water years 2012–2024 based on data availability.

Collectively, these datasets contribute to the terrestrial-ecosystem category of the classification by the European Drought Impact Report Inventory EDII, see: 10.5 “Reduced plant growth” and 10.6 “(Mid-/long-term) deterioration of habitats”.

Societal drought impacts were characterized using a dataset of emergency water-use restrictions issued by mayors. This dataset was built by blending two sources. The first is the inventory developed by by manually surveying online catalogues of such restrictions as maintained by regional administrations; peripheral water-resources-management agencies (e.g. Ambiti Territoriali Ottimali, the public bodies in charge of local-to-regional freshwater-supply management in Italy); and municipal repositories (in Italian, “Albi Comunali”). This survey generated an initial list of 886 water-use restrictions for the Po River basin in 2022, the most severe drought in this region since the 1800s .

This first dataset was then blended with a list of 76 658 news articles collected by automated scraping of Google News during 2022 and 2023 . The implementation of the web scraper is described in the Supporting Information (Sect. S1) and yielded 76 658 articles that were then filtered according to whether they contained the word “decree” (i.e. “ordinanza”) in their title or main body as well as based on the names of those regions or provinces they applied to, eliminating those that did not belong to the Alps or the Appennines (and would thus not be related to snow droughts). This process led to a set of 2054 articles. This web-scraped dataset was finally manually checked by the authors following a set of defined rules for extracting the municipality names for which a decree of restriction was explicitly issued (see again Sect. S1). The dataset resulting from the manual annotation was comprised of 919 news articles.

Blending the two datasets (the first resulting from the manual survey and the second from the web scraper) was performed by removing duplicate news items, such as those pointing to the same municipality or those published very close in time to each other (i.e. within less than 30 d). The resulting dataset identified 1291 municipalities that issued a local decree for water restriction during 2022 and 2023 across Italy. The majority of these decrees covered areas outside our 38 study catchments, which was intentional for two reasons: first, to gain a broader regional perspective and second, to capture the full elevation gradient of emergency water-use restrictions, including areas at lower altitudes than the 38 study catchments.

In order to monitor societal impacts across high-elevation areas that were not monitored by our water-restriction inventory (say, above 1500 m), we also performed an ad-hoc impact survey for mountain huts, that is,e accommodation facilities for hikers providing freshwater supply, refreshment, and overnight-stays. In the European Alps, mountain huts are not only the most iconic component of tourism at high elevations, but also a key driver of high-elevation mountain economy and, often, the only source of fresh-water supply for humans at those elevations . Survey questions were organized in five sections: (i) hut name, coordinates, and elevation; (ii) observed recent changes in water-supply patterns as well as other long-term changes to flora and fauna; (iii) information on seasonal opening and closing dates, both on average and specifically for 2022 and 2023, the two most recent snow droughts in Italy (see Sect. ); and (iv) information on water-supply methods and storage options. Section (ii) of the survey included a question on the number of years of management of the respondent as a way to quantify their “period of record”. The survey was distributed via email to all guarded huts belonging to the Italian Alpine Club (Club Alpino Italiano, https://www.cai.it/, last access 18 May 2025) and is summarized in the Supporting Information (Sect. S2). We received 113 answers, both from the Alps (where the vast majority of mountain huts are located) and from the Apennines (about 20 % of response rate).

Both sources of impact data regarding society (emergency water-use restrictions and impacts on huts) are restricted to the period 2022–2023, thus covering a shorter period compared to all other data sources in this paper (see Table ). For the former, this limitation was related to the massive amount of information to be manually processed, which restricted this research to the most recent (and locally most intense) drought. Regarding the latter, we chose to limit our survey to hut managers to the most recent drought episode at the time of our survey and so those years for which recollection of impacts would be strongest.

Collectively, these datasets contribute to the public-water-supply (7.1 “Local water supply shortage / problems”, 7.3 “Bans on domestic and public water use”, 7.4 “Limitations in water supply to households in rural areas”, and 7.5 “Limitations in water supply to households in urban areas”) and the tourism-recreation categories of the classification by the European Drought Impact Report Inventory EDII, see.

We performed a snow-drought inventory for our 38 catchments across the 2011–2023 water years by first computing basin-wide average SWE between 1 December–30 June for each water year and each catchment this period was chosen according to the typical climatology of snow seasons across elevation bands in Italy as discussed in. A snow-drought water year for a given catchment was then defined as one in which this mean SWE fell below the 30th percentile, following .

In parallel, we also computed basin-wide cumulative precipitation and average temperature during the typical accumulation period (December–April) for each catchment and water year using the BIGBANG dataset. These variables were used to classify snow-drought water years by catchment (hereafter, simply snow droughts) into three categories: (i) cold-dry snow droughts, when both cumulative accumulation-period precipitation and mean accumulation-period temperature were below the median; (ii) warm-dry snow droughts, when accumulation-period mean temperature was above the median but precipitation remained below the median; and (iii) warm-wet snow droughts, when both variables were above the median. Additionally, we identified a small fraction of cold-wet snow droughts, a rare condition where SWE was below the 30th percentile despite mean accumulation-period temperature being below the median and precipitation above the median.

In the literature, multiple methods have been proposed to classify snow-drought years . Some approaches use SWE on a specific date and compare it to climatological averages , whereas others rely on mean or daily SWE falling below a given percentile . Thresholds based on SWE are frequently combined with analogous thresholds on precipitation or temperature to further distinguish snow-drought types, as done in this study . The selection of both the snow-drought metric and the SWE variable (e.g. mean SWE, peak SWE, or 1 April SWE) inevitably affects whether a given year is classified as experiencing a snow drought . Although there is still no scientific consensus on a universal definition of snow drought, we opted for average SWE instead of peak SWE because marginal, Mediterranean snowpacks tend to be short-lived , often resulting in multiple “peak” values within a single season. We further evaluated whether using peak SWE instead of average SWE would substantially change our results and found 82 % agreement between the two metrics, largely due to their strong correlation (mean correlation coefficient of 0.94 across all catchments). Finally, as we elaborate in Sect. , our definition and the resulting set of snow-drought years correspond closely with our understanding of years characterized by low snow accumulation in the Italian mountains, providing indirect support for our methodological choices.

In this study, each water year was treated as independent from all others, and snow droughts were defined on an annual basis. Consequently, multi-year snow drought events are not explicitly analyzed. Furthermore, only a single snow drought event was identified per water year, and the entire water year was classified as a snow-drought year according to this criterion.

We characterized the phenomenology of snow droughts in terms of snow-water resources using three metrics: average snow-season SWE (December to June), snow-season duration, and number of seasonal melt-out episodes. For average snow-season SWE, we computed mean SWE between 1 December–30 June, including no-snow days. Snow-season duration was estimated as the maximum number of consecutive days with SWE above 5 mm, a threshold chosen to reflect the expected accuracy of snow data assimilated in IT-SNOW . Finally, the number of seasonal melt-out episodes was computed as the number of instances in which a pre-existing SWE value dropped below 5 mm between 1 December–30 June. This last statistic is meant as a measure of the tendency of a given snowpack towards ephemerality , that is, intra-seasonal melt-out events (a behavior that is likely increasing in a warming climate and that might have compounding effects with snow droughts). Each of these metrics was first computed on a per-pixel basis and then averaged across 500-m elevation bands within each basin, from 500 to 3500 ma.s.l., which represents the typical range of mountain elevations across Italy (excluding peaks, which are challenging to characterize using a 500-m snow reanalysis).

The cascading effect of snow droughts on streamflow was studied using three metrics widely adopted in previous studies of snow-drought propagation into streamflow : summer cumulative runoff, annual low-flow days, and the annual runoff coefficient. Summer runoff, normalized by catchment area and expressed as a depth, was calculated by accumulating observed streamflow at the outlet of each study catchment over the period 1 May–31 August, corresponding to the typical timing of snowmelt-driven streamflow in our study region (recall that only cumulative values based on samples with less than 20 % missing data were retained). We defined a low-flow day as one with streamflow below the 20th percentile according to the entire streamflow time-series at that outlet; we then computed the total number of these days for each water year, again retaining only those records with less than 20 % missing data. The annual runoff coefficient was computed as the ratio of water-year cumulative streamflow to cumulative precipitation. The use of an annual, rather than seasonal, time scale for low-flow days and for the runoff coefficient was meant to capture the asynchronicity between winter precipitation (and so snow accumulation) and spring-summer runoff . In order to provide additional context to this analysis, we computed summer cumulative precipitation and summer mean air temperature for each catchment and water year according to the BIGBANG dataset.

The role of snow drought in dictating SWE and streamflow phenomenology was investigated by performing a Kolmogorov–Smirnov two-sample test between the distribution of each of the metrics during snow-drought and non-snow-drought water years to test if the difference between these distributions was statistically significant . We considered this difference to be significant with a significance level α=0.05. We did not calculate these statistics using the raw values. Instead, we first derived deviations from the mean values within each basin (hereafter referred to simply as anomalies), and then applied the statistical tests to these anomalies. This approach removes the influence of climatological differences among catchments, allowing us to focus on the contrast between snow drought and non-snow drought years.

Snow-drought impacts on terrestrial ecosystems were characterized by computing weekly, pixel-by-pixel statistics of daily cumulative GPP, which were then averaged according to the same elevation bands as SWE (see Sect. ). We then explored statistical differences in GPP between snow-drought years and non-snow-drought years by calculating the respective weekly quartiles across all 38 study catchments. These statistics are thus average values across all land-cover classes within each elevation band. We also computed differences between mean values of seasonal GPP for each year, catchment, and elevation band. Similarly, we also computed differences in average greening date for all elevation bands and across all catchments, by again differentiating between snow-drought and non-snow-drought years. Statistical differences between snow drought and non-snow drought years were evaluated using a Student's t test .

The above analysis on GPP and greening date was performed using moderate-resolution satellite products. In addition, we averaged responses from various vegetation-cover types, while it is well known the this factor plays a key role in dictating how droughts impacts the bioshpere . To complement the analysis of snow drought effects on GPP and phenology at the watershed scale, we also looked at ground-based flux-tower data at one forest (IT-TrF) and one grassland (IT-Tor) site in Aosta Valley. The key added value of this analysis was to to gain further insights into the potentially different impacts of snow droughts on forests and grasslands at the plot scale. Daily cumulative precipitation, average air temperature, snow depth, soil water content at 30 cm, ET, and GPP were collected at both IT-TrF and IT-Tor. We computed quartiles of each of these variables between 2012–2024 and then evaluated deviations between these quartiles and the observed time-series for the exemplary 2022, a recent snow-drought year in this region that had profound impacts on ecosystems see Sect.  andfor a context on 2022. This was done separately for the grassland and the forest sites, to elucidate whether the response of GPP and ET was different between these two land-cover classes and how this response related to weather (air temperature and precipitation) and surface hydrology (snow depth and soil water content).

Emergency water-use restrictions were binned into four elevation bands based on the typical distribution of municipalities in Italy: below 1000 ma.s.l.; between 1000–1500 ma.s.l.; between 1500–2000 ma.s.l.; and above 2000 ma.s.l. Restrictions were then clustered by month, separately for 2022 and for 2023, in order to evaluate the potential trend in elevation as the snowmelt deficit propagates downstream. Note that this computation was performed using all available water-use restrictions, including those outside the 38 study catchments, again because our study catchments covered only the headwater portions of rivers while water-supply impacts propagate from these headwaters to adjacent lowlands.

The assessment of the impact on mountain huts was conducted through an analysis of the most pertinent variables for our research objectives: (i) duration of management (in years); (ii) hut elevation; (iii) details on impact on water supply; (iv) impact on seasonal opening and closing periods; (v) water-supply source; and (vi) available water storage options.