Daily average streamflow measurements at the catchment outlet were provided by the FOEN via the Hydrological Service (https://www.hydrodaten.admin.ch, last access: 18 May 2026) for more than 200 stations. The data availability is station-specific and depends on the installation and FOEN-internal data quality checking. The HYD-RESPONSES dataset only provides a subset of 184 catchments and considers only stations for which an analysis of hydrological drought dynamics in response to cumulative water deficits was deemed to be meaningful in correspondence with the FOEN (Caroline Kan; see Fig. ). These are stations that provide reliable streamflow (Q) time series and are associated with a physical/natural catchment. Stations are therefore excluded if they (i) only provide water-level information (no Q, 3 stations), (ii) are not part of the main streamflow measurement network (e.g., stations from other networks such as the National Surface Water Monitoring Programme NAWA, 4 stations), (iii) secondary stations (11 stations), (iv) stations with potential return streamflow (= negative Q values, 2 stations), (v) Q measured at derivations (2 stations), (vi) stations without watershed delineation (i.e., subterranean; 1 station) and (vii) uncertainties in time series composition due to displacement and/or temporarily missing Q of contributing stations (4 stations). The complete list of stations included is provided in Tables , , , and in Appendix .

Hydro-meteorological variables used in this study were compiled from multiple complementary data sources, combining station-based spatial climate analyses, dedicated snow model products, and reanalysis data (Table ). This multi-source approach allows both comprehensive coverage of relevant variables and comparative analyses between different data products.

Meteorological variables derived from station-based observations were obtained from the high-resolution (1 × 1 km) spatial climate analyses provided by MeteoSwiss . These include average 2 m temperature (TabsD), daily minimum and maximum 2 m temperature (TminD, TmaxD), daily precipitation sums (RhiresD), and daily sunshine duration (SrelD) . Data availability is product-specific and covers the period 1961–2023 for RhiresD and TabsD and 1971–2023 for TminD, TmaxD, and SrelD. The spatial climate analyses generally cover the Swiss territory only, with the exception of RhiresD, which also includes catchments outside Switzerland that drain through Swiss territory. Note that RhiresD is not available for catchments covering regions in France and Italy before 1992 due to limited meteorological station availability and reduced data reliability . Catchments with substantial areas in these regions should therefore be treated with caution or excluded from analyses prior to 1992 (see Sect. ).

Snow water equivalent (SWE) data were compiled from two independent high-resolution (1 × 1 km) snow datasets. The primary source is the spatial Snow Climatology for Switzerland (SPASS) developed jointly by MeteoSwiss and SLF . This dataset provides modelled and bias-corrected daily SWE for September 1961–September 2022, derived using the SnowQM model based on TabsD and RhiresD. The SnowQM model is presented in detail in . The spatial coverage is restricted to Switzerland. A second snow dataset is derived from the Swiss Operational Snow-Hydrological model system (OSHD), provided by WSL (SLF), which supplies SWE and modelled snowmelt runoff for the period 1998–2022 .

All remaining hydro-meteorological variables, including evaporation, potential evaporation, soil moisture, and additional variables overlapping with the datasets described above, were extracted from the ERA5-Land reanalysis provided by ECMWF . Several variables are thus available from multiple sources and are retained in the HYD-RESPONSES dataset to enable inter-product comparisons. Variables covered by more than one data source include air temperature (TabsD, TminD and TmaxD from MeteoSwiss, t2m from ERA5-Land), precipitation (RhiresD from MeteoSwiss, precipitation from ERA5-Land), snow water equivalent (SWE; from SPASS, OSHD, and ERA5-Land) and modelled snowmelt (from OSHD and ERA5-Land). Additional ERA5-Land-specific variables include soil water volume (swvl) at four depths, total solar radiation (ssr), total runoff (ro), and surface runoff (sro). Detailed variable descriptions are provided in the dataset documentation on Zenodo . A glossary for the variable abbreviations is provided in Table .

ERA5-Land data are available at hourly resolution for the period 1950–2023 via the Copernicus Climate Data Store (CDS) (https://cds.climate.copernicus.eu/datasets/reanalysis-era5-land, last access: 18 May 2026). ERA5-Land consists of numerical model output from the ECMWF land surface model which itself is driven by downscaled and elevation-corrected meteorological forcing from ERA5 . The higher spatial resolution (0.1 × 0.1°, approximately 9 × 9 km) results in an enhanced soil moisture representation and river discharge estimations making ERA5-Land more suitable for analyses based on the hydrological cycle than ERA5 .

The procedures used to extract time series from all gridded datasets are detailed in Sect. .

Datasets used to compile an extensive set of catchment descriptors include station metadata and information on time series availability and homogeneity provided by the FOEN as well as spatial (polygon) data on hydro-terrestrial characteristics (e.g., soil characteristics, hydro-geology) provided by the FOEN, FOAG and Swisstopo (see Table ). Most information is available from https://www.opendata.swiss (last access: 18 May 2026), the FOEN Hydro-Service (https://www.hydrodaten.admin.ch (last access: 18 May 2026), or can be downloaded and inspected via https://www.map.geo.admin.ch (last access: 18 May 2026) (Swisstopo). Direct links to the datasets are provided below and in Sect. 8, Code and data availabilty.

The digital soil suitability maps of Switzerland provide information on a set of different soil characteristics assessed on 25 different geological and geomorphological units which are further discriminated by different landscape elements depending on aspect, slope and bedrock. The maps were first assessed in 1980 and revised in 2000 . The different soil characteristics include soil wetness (https://opendata.swiss/de/dataset/digitale-bodeneignungskarte-der-schweiz-vernassung, last access: 18 May 2026), soil depth (https://opendata.swiss/de/dataset/digitale-bodeneignungskarte-der-schweiz-grundigkeit, last access: 18 May 2026), permeability (https://opendata.swiss/de/dataset/digitale-bodeneignungskarte-der-schweiz- wasserdurchlassigkeit, last access: 18 May 2026), water storage capacity (https://opendata.swiss/de/dataset/ digitale-bodeneignungskarte-der-schweiz-wasserspeichervermogen, last access: 18 May 2026), nutrient content (https://opendata.swiss/de/dataset/digitale-bodeneignungskarte-der-schweiz-nahrstoffspeichervermogen, last access: 18 May 2026) and skeletal content (https://opendata.swiss/de/dataset/digitale-bodeneignungskarte-der-schweiz-skelettgehalt, last access: 18 May 2026).

The hydro-geological map of Switzerland provides information on groundwater resources in Switzerland , including information on aquifer type (loose or solid rock), aquifer genesis and aquifer productivity. The map was originally produced and published for the Hydrological Atlas of Switzerland (HADES, https://hydrologischeratlas.ch/, last access: 18 May 2026). The hydro-geological information was further complemented with the lithological map for Switzerland (produced by Swisstopo), which provides a general overview of dominant rock type classes (loose, sedimentary and crystalline rock). The maps are available via opendata.swiss (https://opendata.swiss/en, last access: 18 May 2026) (hydrogeological map, https://opendata.swiss/de/dataset/hydrogeologische-karte-der-schweiz-grundwasservorkommen-1-500000, last access: 18 May 2026; lithological map https://opendata.swiss/de/dataset/lithologisch-petrografische-karte-der-schweiz-gesteinklassierung-1-500000, last access: 18 May 2026) or can also be accessed via the Hydrological Service of the FOEN (https://www.bafu.admin.ch/bafu/de/home/themen/wasser/zustand/karten/geodaten.html, last access: 18 May 2026). The number of springs and swallow holes in karstic regions provides additional information related to aquifers and the contribution of subsurface water storage. The layer provides main discharge source locations in karstic regions and is available via opendata.swiss (https://opendata.swiss/de/dataset/quellen-und-schwinden-in-karstgebieten, last access: 18 May 2026) (produced by FOEN). The swissTLM3D Hydrography provides topological information on the different water bodies of Switzerland (including flowing and stagnant waters) and originates from the swissTLM3D dataset provided by and accessible via Swisstopo (https://www.swisstopo.admin.ch/de/landschaftsmodell-swisstlm3d#swissTLM3D---Download, last access: 18 May 2026).

The biogeographic regions of Switzerland provide six regions differentiated by similarity of flora, fauna, bryophytes and ornithological information as well as homogeneous surface water catchments . Biogeographic (eco-)regions often correspond well to catchment groups with similar streamflow regime types and are therefore frequently used for catchment regionalization e.g.,. The biogeographic regions are available via opendata.swiss (https://opendata.swiss/de/dataset/biogeographische-regionen-der-schweiz-ch, last access: 18 May 2026).

Finally, general information on the gauging stations and streamflow time series (availability and homogeneity) were provided as accompanying (meta-)data by the FOEN. Time series homogeneity was derived by the FOEN using breakpoint tests following the method of . Breakpoints are identified by partitioning the time series based on the number of potential breakpoints and subsequent modeling of the time series by piecewise linear regression see. The optimal breakpoints are found by minimization of the sum of squared residuals. Resulting breakpoints are indicative to changes in the mean annual 7 d mean flow (M7Q) and were further plausibilized by the FOEN based on catchment meta information and known (potentially) relevant anthropogenic influences such as the construction of (reservoir) dams, hydropower and wastewater treatment plants for more information see. General station information includes catchment area, mean catchment height (elevation), glaciation percentage, outlet coordinates and streamflow regime type (among others) (see Figs.  and ). Catchment outlines (polygons provided by the FOEN) and catchment outlets (point shapes) are provided in the coordinate system CH1903/LV03 (EPSG:21781).

Note that the digital soil suitability maps, swissTLM3D hydrography, biogeographic regions of Switzerland as well as information on springs and swallow holes in karst regions are restricted to Swiss national territory. Catchments with a significant area outside of Switzerland should be treated with caution regarding descriptive variables extracted from these datasets (see Sect.  for a comprehensive overview on extracted descriptors). The hydrogeological and lithological maps of Switzerland to a large extent also cover areas outside of Switzerland. Only catchments of the Rhine (catchments 2091, 2143, 2288, 2289) and Wiese (catchment 2199) are not entirely covered. However, with a coverage of at least > 94 %, descriptors extracted from these datasets may still prove valuable for these catchments.

Methodological details on the extraction and preparation of catchment descriptors are presented in Sect. .

Based on the spatially gridded hydro-meteorological input products (see Sect. ), catchment-level time series were extracted using the R-packages terra and exactextractr . First, the hourly ERA5-Land data was aggregated to daily resolution following the standards used by the MeteoSwiss spatial climate analyses (e.g., RhiresD and TabsD). For this, instantaneous variables and variables representing accumulations or fluxes are distinguished. For instantaneous variables, we provide daily average values. For variables representing accumulations and fluxes, we provide daily sums. Flux variables (mainly precipitation and evapotranspiration) are aggregated using the same temporal convention as the RhiresD precipitation sums, i.e., from 06:00 UTC (day) to 06:00 UTC (day + 1) see. Instantaneous variables and ERA5-Land temperature were averaged from 00:00 to 00:00 UTC again following the convention used in equivalent MeteoSwiss products e.g., TabsD;. Daily catchment-average time series were then extracted by using the catchment outlines (polygons) provided by the FOEN. Units were homogenized across time series. Both units and full standard variable names are listed in Table  (glossary).

The length of the time series depends on the dataset that they were derived from (see Table  for details). Streamflow time series are provided for three different catchment-specific time periods: (1) the original time series (entire period), (2) the time series of the most recent gap-free time-period and (3) the most recent homogeneous time series (in case of significant and plausible breakpoints; otherwise equal to the gap-free time series) (see Fig. ). The breakpoint information is provided by FOEN (see Sect. ). Information on the start of the streamflow monitoring by water level sensor is also provided. The streamflow data should only be considered reliable after the initialization of a sensor. In case of no breakpoints the gap-free period is equal to the homogeneous period. The homogeneous period is usually the shortest (e.g., in case of breakpoints or water level sensor initialization; see for example catchment 2349 in Fig. ). In the case of gaps but no breakpoints, both the homogeneous and the gap-free periods are identical (see, i.e., catchments 2239, 2386 and 2368 in Fig. ). Indicators and (non-)standardized (drought/deficit) indices derived from the hydro-meteorological time series are available for the longest common period of all contributing variables.

(Potential) Cumulative water deficits (CWD and PCWD) are non-standardized indicators tracking evaporation-driven deficits in the (potential) water balance. CWD and PCWD were derived from the daily water balance indicator time series (see Sect. ) using the cwd R-package . A deficit starts when the water balance is negative (i.e., P-E<0) and is accumulated as long as the deficit remains uncompensated (deficit > 0). Note that no surplus information is tracked. Once the deficit is compensated, the values remain at zero (CWD =0). Example time series are shown for both CWD and PCWD in Fig. k, l. In some cases (especially for P–PET based only on ERA5-Land variables, i.e., tp - pev), PCWDs are not compensated each year and can persist over multiple years. Both CWDs and PCWDs are hence also provided on a yearly calculation basis (annual reset on 31 December). Non-standardized indices preserve units (here millimetres) and are physically interpretable in terms of absolute deficit amounts. Cumulative water deficits do not rely on a predetermined calculation time window, which allows the user to track both deficits accumulated over short periods in time (below one month) and deficits accumulated over very long periods.

Water-balance based non-standardized drought indices are widely in use in ecohydrological, land-atmosphere interaction research and catchment-memory studies both with and without temporal resets see e.g.,. Being more strongly tied to the actual physical water availability, non-compensated CWDs may provide valuable information on carry-over effects in multi-year drought contexts and/or long-term shifts in climatic water balance . PCWDs in contrast are based on potential water balance and (absolute) carry-over deficits should hence be treated with caution. CWD and PCWD time series which are annually reset provide complementary year-to-year information, which may better align in contexts of annual low-flow statistics and allows for a year-to-year comparison across years and catchments.

Standardized (drought) indices depict the anomaly of a deficit over a fixed retrospective time window (e.g., 1 month). The hydro-meteorological indicator time series is first aggregated over the given window and then transformed to a standard normal distribution by fitting a suitable candidate distribution . Standardized indices therefore provide information on both anomalously dry and wet conditions, which are often defined by thresholds corresponding to standard deviations (SD). As such, values below -1 SD indicate drier than normal conditions (moderate droughts), while values above +1 SD indicate wetter than normal conditions (moderate wetness) . The HYD-RESPONSES dataset provides daily time series for three standardized (drought) indices: the Standardized Precipitation Index SPI,, the Snowmelt and Rain Index SMRI,, and the Standardized Precipitation Evaporation Index SPEI,. The SPI represents deficits driven by precipitation only (derived from P), while the SMRI tracks deficits in liquid water input originating from both rainfall and snowmelt (derived from P+ΔSWE) accounting for seasonal snowfall and snowmelt dynamics . The SPEI represents deficits driven by evaporative demand (derived from P-PET) and hence indirectly accounts for temperature effects . Daily time series for all three indices (SPI, SPEI, SMRI) are provided for aggregation windows ranging from 1–24 months (31–730 d). Exemplary SPI and SMRI time series for all aggregation windows are shown in Fig. g,h.

All indices were calculated using the SCI R-package with custom modifications accounting for the daily time series resolution. All candidate distributions provided within the SCI R-package (gamma, genlog, gumbel, lnorm, norm, gev, pe3, weibull) were tested for suitability. The distributions were fitted for each day of the year (DOY) based on the reference period 1991–2020. This results in a fit for each DOY derived from the same (window of) values for each distribution. Monthly SPI fits (SPI-1) are for example based on the 30 daily values up to the specific DOY for each of the 30 years in the reference period 1991–2020. The suitability of candidate distributions was assessed based on three indicators: the Shapiro-Wilks normality tests p-values;, the number of flags returned by the fitting function fitSCI see SCI R-package; , and the number of missing and/or implausible values. Implausible values are defined as values above or below ±3 SD following . Estimating more extreme standardized index values from a 30-year climatology requires substantial extrapolation of the fitted distribution and is therefore associated with large uncertainty, particularly given the strong temporal autocorrelation of drought indices. Values beyond ±3 correspond to events with return periods far exceeding the length of the reference record and cannot be robustly quantified see.

The returned flags in distribution parameter fitting were mainly related to convergence issues (non-convergence) flag 3, see SCI R-package,. Without a valid fit, the transformation to standardized index values is not possible resulting in missing values on the flagged DOYs in all time series years. As in , one best-fitting distribution (over all DOYs) is chosen for all catchments to allow for catchment comparability. The distribution was selected among the distributions satisfying the following conditions: (1) the transformed values are not significantly different from a normal distribution for the majority of catchments (p-values >0.05 for at least 75 % of the catchments), (2) fewer than 5 DOYs flagged and (3) fewer than 50 implausible and/or missing values in the transformed time series (combined consideration of missing values due to flags and unrealistically high/low values). The distribution selection procedure is illustrated for the SPEI in Fig. . The results of the Shapiro-Wilks tests (p-values) and information on missing/implausible values and flags are also provided in the HYD-RESPONSES dataset and can be used to identify catchments with non-satisfying properties within the overall best-fitting distribution (see Fig. ). Following , values of all standardized (drought) indices time series were constrained to the interval [-3, 3] SD.

The Gamma distribution was chosen for the SPI for all variables (RhiresD, ERA5-Land), which is consistent with other studies and recommendations of the World Meteorological Organization (WMO) . The SMRI was fitted by the genlog (lnorm) distribution for the snow-corrected precipitation series based on SPASS (ERA5-Land and OSHD). For the SPEI, the genlog distribution was found to perform best across time scales (see Fig. ).

SPI, SPEI and SMRI provide complementary (and standardized) information on hydroclimatic variability and drought-related processes facilitating integrated analyses of drought development and propagation and allowing consistent comparisons across catchments, regions and climates . Standardized indices are frequently employed in drought monitoring and early warning systems DEWS;, drought propagation analysis or as proxy for various storage processes . Example use cases for drought event analysis and catchment response patterns are provided in Appendix and .

Climatologies and anomalies are provided for all time series, including the time series of variables directly extracted from spatially gridded data products with no modifications except for spatial and temporal aggregation where required (see Sect. ), derived indicators (Sect. ), standardized drought indices (Sect. ), and cumulative water deficits (Sect.  and ).

Both climatologies and anomalies are based on the reference period 1991–2020. The climatology is provided for two variants: (i) using moving windows and (ii) for fixed time windows. The variants are available at the following time scales: daily (only i), monthly (both), seasonal (both), and annual (only ii). The moving window climatology was calculated by using a moving window of 31 d (day-15, day=0, day+15) for the monthly, a 3-month window (91 d) for the seasonal and a 6-month (183 d) window for the extended season time scale. The moving window climatology is calculated for DOYs 1–366 with NA-values set for 29 February in the case of non-leap years. Example time series for monthly anomalies in 2 m temperature, precipitation, evaporation, soil moisture, snow water equivalent, snowmelt and (potential) cumulative water deficits are shown in Fig.  a–f, m–n.

The regular climatology is available for monthly, seasonal (DJF, MAM, JJA, SON), extended season (May–October, November–March) and annual time scales. Using the moving window climatology, standardized anomalies have been derived by first subtracting the climatological mean (μ) and then dividing by the climatological standard deviation (σ) (also known as z-scores). The following climatological statistics are provided: minimum, maximum, mean, median, standard deviation, 5th, 25th, 75th and 95th percentiles. For the 7 d average streamflow series (M7Q) we also provide the 2nd, 10th and 15th percentiles which are frequently used in streamflow drought analysis and monitoring see e.g.,.

Time series of cumulative streamflow deficits (CQD) were calculated based on negative streamflow anomalies (drought phases) by using the same procedure as for cumulative water deficits (see Sect. ). CQD time series are provided for both fixed and variable threshold definitions. Fixed thresholds (e.g., a constant percentile threshold) are used for critical flow levels that do not change seasonally (e.g., directly linked to physcial/actual low-flow or water scarcity situations) whereas variable thresholds (e.g., seasonally varying percentiles) account for seasonality and changing flow regimes, allowing drought phases and deficits to be identified relative to expected (seasonal) conditions “anomalies”,. Hence, variable threshold definitions are often used to analyse seasonally varying streamflow (drought) generating processes or to understand drought propagation mechanisms . For the fixed threshold definition, daily M7Q anomalies were derived for events exceeding the Q347 threshold, defined as the daily flow rate exceeded for 347 d yr⁻¹ (i.e., the 347 d exceedance flow, roughly corresponding to the 5th streamflow percentile; see Sect. ). For the variable threshold definition, daily M7Q anomalies were calculated for the following monthly (31 d) and seasonal (91 d) percentiles: 2nd, 5th, 10th, 15th, 25th, 50th (median) and mean. Cumulative deficits are physically interpretable and in the case of cumulative water deficits [mm] and streamflow deficits [m³ s⁻¹] also physically comparable in terms of total runoff depth [mm]. Figure j shows CQDs for both fixed and variable threshold definitions for the year 2022 for catchment 2034 – Broye, (Payerne, Caserne d'aviation).

We define drought events as coherent phases of non-zero deficits for cumulative deficits (CWD, PCWD and CQD) and as negative M7Q-based streamflow anomalies for streamflow droughts. Streamflow drought phases were extracted for the same percentiles and time scales as used for CQDs (see Sect. ), namely for monthly (31 d) and seasonal (91 d) percentiles: 2nd, 5th, 10th, 15th, 25th, 50th (median) and the mean. Streamflow events were also extracted for the fixed (yearly) Q347 threshold (see Sect.  and ).

An event starts on the first day values fall below the threshold value and lasts until values exceed the threshold again (see Fig. ). For each variant, the event time series consists of consecutively numbered event phases and information on the event duration since the start (i.e, an event with a duration of 5 d is represented in the time series as: “1 1 1 1 1” (event phase number), “1 2 3 4 5” (duration since start)). Additional event characteristics (e.g., lowest value during a phase) can easily be derived by the user in combination with the indicator time series. A minor pooling of hydrological drought events is introduced by using the 7 d average streamflow (M7Q) time series which merges closely successing and potentially dependent individual events to one single event as a result of the smoothing of large day-to-day fluctuations . Streamflow drought events based on both fixed and variable threshold definitions were used for the event shadings in Fig. .

Spatially non-overlapping polygon datasets (e.g., soil suitability maps) typically provide categorized values for variable-specific classes (e.g., soil depth classes are shallow, medium, deep, very deep). To extract catchment-level information, polygon-based information was first rasterized to a spatial grid identical to the MeteoSwiss spatial climate analyses grid products (in both extent and resolution). The rasterization was done by using the rasterize function of the terra R-package . Each grid cell only contains the value of the category with the largest overlap. The percentage overlap with the catchment area was then assessed for all variable-specific classes by using the exact_extract function (as for time series) and adjusting the aggregation function to fractions “frac”; see. Catchment area overlap fractions are provided for all categories. Descriptors with multiple classes can also be reduced to a single dominant category represented by the largest percentage overlap (“proportion”). An example is shown for the biogeographic regions in Fig. a. Reducing a specific catchment descriptor to one single (dominant) category (e.g., derived via largest overlap percentage) may however lead to a loss in explanatory power as the category with the largest overlap may not necessarily be the most representative or most influential for streamflow (drought) analysis.

Two descriptive variables related to catchment shape and drainage were derived in R by using the catchment outlines, namely the basin shape index (BSI) and drainage density. The HYD-RESPONSES dataset provides two BSI variants. The first variant is derived based on a ratio between area and basin length (A/L2) known as form factor and the second variant is based on a ratio between the catchment area and the area of the circle with the smallest radius encircling the entire catchment (Acatch/Acircle) known as circularity ratio . Both indices range from 0 to 1. Both are frequently used (also in combination) as morphometric catchment indicators see e.g.,. Albeit providing similar information, the form factor is primarily controlled by basin length and hence provides information on catchment elongation, while the circularity ratio is more sensitive to basin shape accounting for complex/irregular shapes resulting in larger areas for more information on basin shape indices see. The drainage density denotes the ratio between the catchment area and the total length of streamflow channels both natural and stormwater drainage infrastructure;. The drainage density was calculated by using the swissTLM3D hydrography dataset (see Sect. 8 for a download link). Both indices (BSI and drainage density) are frequently used in flood-related studies but may also provide valuable information during low-flow periods as high-intensity precipitation events are a relevant factor for (streamflow) drought recovery . Further, also the overlap percentage with the Swiss territory (swissBOUNDARIES3D, see Sect. 8 for a download link) is provided for each catchment and can be used to exclude catchments with significant portions outside of Switzerland which goes along with a limited coverage in both hydro-meteorological and catchment descriptor input datasets (see Sect.  and ) for ca. 12.5 % of catchments (see Sect. ). Information on karstic sources and sinks is provided as the number of sources and swallow holes per catchment and km².

Several indices related to streamflow characteristics (low flow, responsiveness, baseflow and flow stability) are provided in the HYD-RESPONSES dataset. The Q347 is a low flow index used as the basis for water abstraction restrictions in Switzerland and corresponds to the daily flow rate exceeded for 347 d yr⁻¹. The Q347 was derived from the flow duration curve (FDC) by using the hydroTSM R-package and corresponds roughly to the 5th streamflow percentile (95th percentile of 365 d ≈ 347, hence Q347). The baseflow index BFI; is a widely used index linked to multiple catchment characteristics such as aquifer type, productivity and soil characteristics. The BFI provides information on the (base-)flow sustained during dry periods e.g., by subsurface storages;. The BFI was derived using the baseflow function of the lfstat R-package and is shown in Fig. . introduced the delayed-flow index (DFI) which breaks down the BFI into individual hydrograph components. The components include fast, intermediate, slow and base responses and potentially reflect various storage processes contributing to the overall streamflow response (e.g., snowmelt and groundwater). The DFI was derived by using the delayedflow R-package https://modche.github.io/delayedflow/, last access: 18 May 2026; see also. The last two indices related to streamflow behaviour are the “flashiness” or R-B-index which represents the ratio of the sum of day-to-day streamflow changes divided by the total streamflow and the flow-stability index which relates the mean annual minimum flows (MAM_q) to the mean annual flow (MQ; MAMq/MQ). The remaining catchment descriptors were derived from the extracted hydro-meteorological time series and/or their respective climatology. Information on average precipitation, temperature, evaporation, snow water equivalent, streamflow, the fraction of precipitation falling as snow and the runoff fraction (Q/P) are provided on yearly scales for identifying broad climatic (i.e., water balance) and physiographic controls on hydrological behavior. Finally, monthly Pardé coefficients (PCs) are provided which indicate the contribution of monthly mean streamflow to the annual mean streamflow.

The HYD-RESPONSES dataset addresses fundamental challenges in hydrological drought analyses by compiling and harmonizing multiple data sources into a coherent catchment-scale framework, enabling multi-variable drought analyses in Switzerland. Drought (deficit) indices derived from two high-resolution snow climatologies for Switzerland (SPASS, OSHD) also allow for in-depth quantitative analyses on the contribution of snow processes to cross-seasonal drought propagation in Alpine catchments . A combined used of standardized indices (SPI, SPEI, SMRI at 1–24 month scales) and non-standardized cumulative deficits (CWD, PCWD, CQD) facilitate multi-scale (drought) deficit and catchment response sensitivity assessments and allows for a concurrent anomaly-based and physically interpretable characterization of drought deficits . By providing time series for many relevant variables for drought monitoring precipiation, temperature, evaporation, soil moisture and streamflow; see e.g., at daily temporal resolution, the HYD-RESPONSES dataset may also be used for training machine learning models such as Random Forests RFs; e.g., or Long Short-Term Memory models LSTMs; see which have recently emerged as promising approach for rainfall-runoff modeling . Three example applications of the HYD-RESPONSES dataset are illustrated in Appendix , and .

Although the dataset was developed for Switzerland, the methodological framework – combining in-situ observations, gridded products, and reanalysis into catchment-scale time series – is transferable, with requirements that scale with data availability. Replication requires four essential components: (1) streamflow observations with defined catchment boundaries, (2) meteorological forcing data (precipitation, temperature), (3) snow information for mountain regions, and (4) static catchment descriptors (e.g., information on soils, geology, topography). While the first component is often limiting, the second component is decisive for the applicability of the dataset on specific use cases. Switzerland can leverage from high-density observational station networks resulting in high-quality spatially gridded hydro-meteorological products see e.g.,. With known biases in mind (especially over complex terrain), ERA5-Land is a viable alternative by providing temporally and physically consistent variables at sufficiently high spatial resolution for comparative catchment studies and machine learning applications aiming for generalizable results in regions where observational networks are less dense . For local operational drought management or absolute deficit quantification, reliable high-resolution observational products remain however preferable. Several recent developments address observational data limitations, including the Caravan global community dataset , rapidly advancing machine learning-based bias correction methods for downscaling reanalysis products such as ERA5-Land or advances in developing high-quality remote sensing-based products for soil moisture e.g. SMAP; see, snow e.g., ICESat-2, evaporation see e.g., and terrestrial water storage e.g., GRACE-FO;.

The HYD-RESPONSES time series are provided for product-specific periods, and the spatial coverage is restricted to Swiss territory for most of the higher resolution MeteoSwiss and SLF products (TabsD, TminD, TmaxD, SPASS, SrelD, OSHD) as well as many catchment descriptor input datasets. Full coverage over the entire hydrological Switzerland is only available for ERA5-Land (all variables) and the MeteoSwiss RhiresD product after 1992; see. Catchments with significant areas outside of the Swiss national borders – approximately 12.5 % of the catchments (see Sect. ) – may therefore be considered with caution or used solely for time series based on ERA5-Land variables only. Time series for standardized drought indices are provided only for the transformation variant based on the best-fitting distribution across all catchments to allow for comparison across catchments comparability see e.g.,. The best-fitting distribution may however vary across catchments and climates see e.g.,. The HYD-RESPONSES dataset therefore also provides information on fits, missing values, and flags which can be used to exclude catchments with unsatisfying fitting and transformation properties from analyses. Field- and feature-based catchment descriptors were aggregated at catchment level via summarization (e.g., karstic sources) or percentage overlaps (see Sect. ). Maximum percentage overlaps with catchment area may however only insufficiently account for spatial differentiation which could enhance the representation of factors most influential to streamflow evolution by accounting for spatial proximity to the stream/river courses .

Several known limitations are further related to the datasets used to compile the HYD-RESPONSES data. ERA5-Land is a state-of-the-art reanalysis product provided at a higher spatial resolution than the standard ERA5 reanalysis . The higher spatial resolution results in a better depiction of soil moisture, lakes, river discharge estimations, and the orographic enhancement of precipitation . ERA5 and ERA5-Land datasets however share most of the parameterizations, as ERA5-Land consists of output from the ECMWF land surface model driven by downscaled and elevation-corrected ERA5 data . Despite the advantage of higher spatial resolution over ERA5, a grid resolution of 9 km still has limitations over complex high-altitude terrain. The extracted time series related to snow water equivalent should be used with caution, as snow depth in ERA5-Land is of mixed quality depending on geographical location and altitude . showed that ERA5-Land overestimates SWE at high elevations with larger biases in the southern compared to the northern Alps. Higher-resolution datasets such as SPASS and OSHD should hence be preferred over ERA5-Land. Note however that all snow-related datasets have problems in representing small SWE amounts at low altitudes . Caution is also required when using the snow-corrected precipitation (water input) time series. The time series corrected by the ΔSWE series consider both snowfall (ΔSWE > 0) and snowmelt (ΔSWE < 0), while the correction based on snowmelt variables only accounts for snowmelt (smlt in ERA5-Land and romc in OSHD; see Tables , , and Sect. ). Snowmelt-corrected precipitation time series may therefore be of limited use during the main snow accumulation season but can still provide valuable information during the snowmelt season.

Another limitation of the ERA5-Land dataset is the parameterization of subgrid-scale processes and the representation of subsurface storages that affect evapotranspiration e.g., fixed maximum storage volume assumption; see. Key processes such as dynamic groundwater–vegetation interactions, irrigation withdrawals, and adaptive rooting strategies are hence not represented and may lead to biases in evapotranspiration responses . Although ERA5-Land compares more favorably with in situ soil moisture and evapotranspiration observations than previous reanalyses (e.g., ERA5), considerable discrepancies remain, especially in dry summers and in regions with heterogeneous land cover . Given these limitations, drought indicators based on ERA5-Land evapotranspiration should generally be interpreted with caution. This limitation is further compounded by the fact that the validation of long-term soil moisture and evapotranspiration remains challenging due to the scarcity of consistent observational datasets, particularly at high spatial resolution and over multi-decadal time scales . Many state-of-the-art evapotranspiration products are limited in temporal or spatial extent and can be affected by gaps and cloud contamination e.g., remote sensing-based products; see. ERA5-Land thus remains one of the few datasets providing spatially consistent and continuous long-term evapotranspiration estimates with sufficiently high spatial resolution over Switzerland.

Additional caution is warranted when using HYD-RESPONSES water balance time series (and indicators derived from them) when they were derived by combining ERA5-Land evapotranspiration with (snow-corrected) precipitation from independent data sources (RhiresD, OSHD, and SPASS). While ERA5-Land variables are internally consistent, the combination with independent data sources may lead to systematic biases in absolute deficit estimates. This limits the interpretability of absolute cumulative deficits but does not invalidate the approach for comparative, process-oriented drought analyses across regions and catchments (e.g., drought propagation, catchment response sensitivities). Relative measures of cumulative deficits, their temporal evolution, and their normalization through ratios (e.g., CWD/PCWD) can still provide valuable insights, even when absolute magnitudes are uncertain. In such contexts, relative anomalies, temporal evolution, and spatial patterns are more informative than absolute deficit magnitudes. Studies have further demonstrated coherent representation of major drought events (e.g., drought years 2003 and 2018) across datasets, which supports the usability of combined indicator time series when known limitations are adequately taken into account . Note that the HYD-RESPONSES dataset also provides complementary water balance and SPEI time series derived from ERA5-Land variables only, providing consistent metrics and opportunity for comparisons among data products. Guidance on the usage and reliability of all HYD-RESPONSES time series products is provided by a classification based on three reliability levels (see Sect.  and Table ). The levels are based on the origin of the underlying data, the extent to which variables rely on (model) assumptions, and the degree of processing applied to derive the hydro-meteorological time series.