Austrian observations of snow depth date back to 1895 and are thus among the longest available quantitative snow information from hydrometeorological networks worldwide. It is well known that such long‐term observations are prone to inhomogeneities, which may not only affect climatologies and trends, but derived products used in research or practice. While the reliability of available methods for detecting breaks in snow time series has been shown before and could also be confirmed by our work, we focused on improving the adjustment method. Conventional methods often refer to the median of difference or quotient series (INTERP), whereas our proposed method also uses a quantile‐wise adjustment (InterpQM), which is useful to minimize a bias on the tails of the frequency distribution. We demonstrated the success of the new method by using Swiss parallel snow depth observations. Errors of the analysed indicators could be reduced in 68% of the cases when compared with INTERP. The results were best for large snow depths, being up to 19% better. Overall, InterpQM was better in 75% of validation cases for the daily large, 72% of all observations and 56% of mean seasonal snow depth cases. We describe the performed homogenization procedure in detail, including quality control, gap filling, homogeneity testing, break detection, calculation of and improvements to the adjustment method. Our results show that snow depth time series generally have a lower number of breaks compared with station data of other climate variables. This underlines their high quality, even if measuring snow presents challenges. Using Austrian snow depth series as an example, the effects of the new adjustment method on trends were analysed using the Mann–Kendall and Sen's Slope. Homogenization may have a significant effect on derived trends: Two of the six adjusted series were changed from nonsignificant to significant and one vice versa.

Our current knowledge of spatial and temporal snow depth trends is based almost exclusively on time series of non-homogenised observational data. However, like other long-term series from observations, they are prone to inhomogeneities that can influence and even change trends if not taken into account. In order to assess the relevance of homogenisation for time-series analysis of daily snow depths, we investigated the effects of adjusting inhomogeneities in the extensive network of Swiss snow depth observations for trends and changes in extreme values of commonly used snow indices, such as snow days, seasonal averages or maximum snow depths in the period 1961–2021. Three homogenisation methods were compared for this task: Climatol and HOMER, which apply median-based adjustments, and the quantile-based interpQM. All three were run using the same input data with identical break points. We found that they agree well on trends of seasonal average snow depth, while differences are detectable for seasonal maxima and the corresponding extreme values. Differences between homogenised and non-homogenised series result mainly from the approach for generating reference series. The comparison of homogenised and original values for the 50-year return level of seasonal maximum snow depth showed that the quantile-based method had the smallest number of stations outside the 95 % confidence interval. Using a multiple-criteria approach, e.g. thresholds for series correlation (>0.7) as well as for vertical (<300 m) and horizontal (<100 km) distances, proved to be better suited than using correlation or distances alone. Overall, the homogenisation of snow depth series changed all positive trends for derived series of snow days to either no trend or negative trends and amplifying the negative mean trend, especially for stations >1500 m. The number of stations with a significant negative trend increased between 7 % and 21 % depending on the method, with the strongest changes occurring at high snow depths. The reduction in the 95 % confidence intervals of the absolute maximum snow depth of each station indicates a decrease in variation and an increase in confidence in the results.

Measurements of snow depth can vary dramatically over small distances, and as with any other meteorological variable, snow depth time series are affected by inhomogeneities or break points. Such inhomogeneities can arise due to e.g.; changes of instrumentation, changes to station location and observer practices, or changes in the local environment such as urbanisation or plant growth. In order to analyse and monitor variation in snow depth time series accurately, homogenised snow data series are required. In deriving such homogenised series, it is essential to understand the characteristics and impacts of inhomogeneities. Having applied some pre-selection criteria to identify candidate series, time series homogenization for 184 Swiss snow depth series was performed using ACMANT, Climatol, and HOMER, three state-of-the-art break detection algorithms. For the 91 year base period of 1931-2021, we investigated which method and set-up worked best for detecting breaks in this network of Swiss snow data series. The approach identified valid break points in 25% of the series, with HOMER identifying more valid breaks than either ACMANT or Climatol. By evaluating the network using multiple methods, there is more confidence that the results can be applied to snow time series with insufficient metadata or no immediately nearby reference stations in order to include them in future homogenisation efforts.

Abstract. Our current knowledge on snow depth trends is based almost exclusively on these non-homogenized data.Long-term observations of deposited snow are well suited as indicator of climate change. However, like all other long-term observations, they are prone to inhomogeneities that can influence and change trends if not taken into account. We investigated the effects of removing inhomogeneities in the large network of Swiss snow depth observations on trends and extreme values of commonly used snow indices, such as snow days, seasonal averages or maximum snow depth in the period 1961–2021. For this task, three homogenization methods were applied: Climatol and HOMER, which use a median based adjustment method, and interpQM, which applies quantile based adjustments. All three were run using the same break points and input data. This allowed us to investigate and quantify the effects of these methods on the homogenization results. We found that all three methods agree well on trends in seasonal average snow depth, while differences are visible for seasonal maximum snow depth and the corresponding extreme values. Here, the quantile-based method performed slightly better than the two median-based methods, as it had the smallest number of stations outside the 95 % confidence interval for 50-year return periods of maximum snow depth. These differences are mainly caused by the way the reference series are selected. The combination of a high minimum correlation (>0.7) and restrictions in vertical (<300 m) and horizontal (<100 km) distances proved to be better suited than only using correlations or distances respectively as criteria. The adjustments removed all positive trends for snow days in the original data and strengthened the negative mean trend, especially for stations >1500 m. In addition, the number of significant negative stations was increased between 7–21 %, with the strongest changes at higher snow depths.

Knowledge concerning possible inhomogeneities in a data set is of key importance for any subsequent climatological analyses. Well-established relative homogenization methods developed for temperature and precipitation exist but have rarely been applied to snow-cover-related time series. We undertook a homogeneity assessment of Swiss monthly snow depth series by running and comparing the results from three well-established semi-automatic break point detection methods (ACMANT – Adapted Caussinus-Mestre Algorithm for Networks of Temperature series, Climatol – Climate Tools, and HOMER – HOMogenizaton softwarE in R). The multi-method approach allowed us to compare the different methods and to establish more robust results using a consensus of at least two change points in close proximity to each other. We investigated 184 series of various lengths between 1930 and 2021 and ranging from 200 to 2500 m a.s.l. and found 45 valid break points in 41 of the 184 series investigated, of which 71 % could be attributed to relocations or observer changes. Metadata are helpful but not sufficient for break point verification as more than 90 % of recorded events (relocation or observer change) did not lead to valid break points. Using a combined approach (two out of three methods) is highly beneficial as it increases the confidence in identified break points in contrast to any single method, with or without metadata.

Measurements of snow depth and snowfall can vary dramatically over small distances. However, it is not clear if this applies to all derived variables and is the same for all seasons. Almost all meteorological time series incorporate some sort of inhomogeneities. Complete metadata and existing “parallel” stations in close proximity are not always available. First, we analyse the impacts of local-scale variations based on a unique set of parallel manual snow measurements for the Swiss Alps consisting of 30 station pairs with up to 70 years of parallel data. Station pairs are mostly located in the same villages (or within 3km horizontal and 150m vertical distances). Seasonal analysis of derived snow climate indicators such as maximum seasonal snow depth, sum of new snow, or days with snow on the ground shows that largest differences occur in spring and the smallest ones are found in DJF and NDJFMA. Relative inter-pair differences (uncertainties) for days with snow on the ground (average snow depth) are below 15% for 90% (30%) . Second, in view of any homogenization efforts of snow data series, it is paramount to understand the impacts of inhomogeneities. Using state-of-the-art break detection algorithms, we strive to investigate which method works best for detecting breaks in snow data series. The results can then be used on time series with insufficient metadata or no neighbouring stations in order to include them in future homogenization processes. Furthermore, the knowledge about inhomogeneities and breakpoints paves the way for new applications such as the reliable combination of two parallel series into one single series.

Daily measurements of snow depth and snowfall can vary strongly over short distances. However, it is not clear if there is a seasonal dependence in these variations and how they impact common snow climate indicators based on mean values, as well as estimated return levels of extreme events based on maximum values. To analyse the impacts of local-scale variations we compiled a unique set of parallel snow measurements from the Swiss Alps consisting of 30 station pairs with up to 77 years of parallel data. Station pairs are usually located in the same villages (or within 3 km horizontal and 150 m vertical distances). Investigated snow climate indicators include average snow depth, maximum snow depth, sum of new snow, days with snow on the ground, days with snowfall, and snow onset and disappearance dates, which are calculated for various seasons (December to February (DJF), November to April (NDJFMA), and March to April (MA)). We computed relative and absolute error metrics for all these indicators at each station pair to demonstrate the potential variability. We found the largest relative inter-pair differences for all indicators in spring (MA) and the smallest in DJF. Furthermore, there is hardly any difference between DJF and NDJFMA, which show median variations of less than 5 % for all indicators. Local-scale variability ranges between less than 24 % (DJF) and less than 43 % (MA) for all indicators and 75 % of all station pairs. The highest percentage (90 %) of station pairs with variability of less than 15 % is observed for days with snow on the ground. The lowest percentage (30 %) of station pairs with variability of less than 15 % is observed for average snow depth. Median differences of snow disappearance dates are rather small (3 d) and similar to the ones found for snow onset dates (2 d). An analysis of potential sunshine duration could not explain the higher variabilities in spring. To analyse the impact of local-scale variations on the estimation of extreme events, 50-year return levels were quantified for maximum snow depth and maximum 3 d new snow sum, which are often used for avalanche prevention measures. The found return levels are within each other's 95 % confidence intervals for all (but three) station pairs, revealing no striking differences. The findings serve as an important basis for our understanding of variabilities of commonly used snow indicators and extremal indices. Knowledge about such variabilities in combination with break-detection methods is the groundwork in view of any homogenization efforts regarding snow time series.

Snow on the ground is an important climate variable which is normally measured either as snow depth or height of new snow. Like any other meteorological variable, manually measured snow is prone to local influences, changes in the environment or procedure of the measurements. In order to investigate the robustness of snow measurement series towards such non-climatic changes, a unique set of parallel manual snow measurements over 25 years from 23 station pairs between 490 and 1800 m a.s.l. was compiled. A sensitivity analysis based on typical snow climate indicators (e.g., mean snow depth, sum of new snow) from these parallel time series was carried out to find the most robust snow climate indicators for climatological analyses. Results show that there are only small differences in the sensitivity of the various snow climate indicators with regards to local changes. However, the indicators number of days with snow on the ground as well as the maximum snow depth are least affected by local influences and changes at station level. Median values of all station pairs reveal relative differences of about 7% for the number of days with snow cover and 11-16% for all other indicators. However, in extreme cases, the deviations within a single station pair can reach 25-40%.