The increase in air temperature in the Arctic region occurs nearly four times as fast as the global average over recent decades (Rantanen et al., 2021). An ensemble of climate models (CMIP6) projects continued 21st century Arctic warming more than double that of the global mean (Cai et al., 2021). Along with air temperature, precipitation is projected to increase due to changes in surface evaporation, atmospheric water vapor content and moisture transport that influences precipitation rates (Bintanja and Andry, 2017; McCrystall et al., 2021). The increase in precipitation with surface air temperature is also considerably higher in the Arctic (4.5%/°C) compared to the global average (1.6–1.9%/°C) (Bintanja and Selten, 2014; Bintanja and Andry, 2017). In addition, global climate models project an increase in the ratio of liquid to total precipitation in the Arctic as more precipitation is expected to fall as liquid precipitation with higher air temperatures. In recent decades, increases in temperature and liquid precipitation have been found in Greenland (Hanna et al., 2020; Niwano et al., 2021; Huai et al., 2022). Average air temperatures in Greenland have increased in the period 1991–2019 by 4.4°C in winter, 2.7°C in spring and 1.7°C in summer (Hanna et al., 2020). This increase has partly been attributed to increasing frequency and intensity of Greenlandic atmospheric blocking events since the 1990s (Delhasse et al., 2020; Hanna et al., 2020). On the Greenland Ice Sheet (GrIS), average annual rainfall increased by 25% in the last decade (2010–2019) compared to the 1981–2010 reference period (Niwano et al., 2021).

In this study we make a detailed assessment of past climatic changes on Ammassalik island, located on the south-east coast of Greenland. Assessing precipitation trends in south-east Greenland is particularly relevant because it receives the highest amounts of precipitation compared to other regions in Greenland (Berdahl et al., 2018). Additionally, there are many small glaciers and ice caps (GIC) located on the island. GICs in Greenland are known to be particularly sensitive to the ongoing atmospheric changes (Bjørk et al., 2012; Noël et al., 2017; Khan et al., 2022). Another advantage of Ammassalik island is that it has a unique coverage and quality of a network of automatic weather stations (AWS), compared to other places in Greenland. While these observations provide essential information about the local climate, the challenges of observing precipitation in remote (Arctic) regions should be considered. One difficulty is that maintenance of weather stations is expensive and labor-intensive, which can lead to extended gaps in the observed time series. Additionally, there are several environmental factors that strongly impact the robustness of measurements from precipitation gauges. Wind-induced undercatch is particularly relevant in the Ammassalik island, due to the complex topography and characteristic strong winds present in the region. Wetting and evaporating losses contribute to systematic underestimation (Yang et al., 1999). Since systematic errors regarding solid precipitation are greater than those for liquid precipitation, precipitation measurements are more accurate in the summer season. In this study we combine a regional climate model with AWS observations. The model data helps to bridge the gap of sparse observational data. We cross-check these two data sources and use them to provide a more complete perspective on climate (and in particular, precipitation) trends at Ammassalik island.

Ammassalik island (65.73 °N, 37.62 ′W) is located on the south-east coast of Greenland, just south of the Arctic circle and covers approximately 772 km² (Figure 1). Tasiilaq is the largest settlement (1913 inhabitants (per 1 January 2022) (StatBank Greenland, 2022) and is located on the south side of the island. The west coast of Ammassalik island borders the Sermilik fjord (10–15 km wide, 85 km long), which separates the island from the mainland and connects the GrIS with the Irminger Sea (S. H. Mernild, Malmros, et al., 2012). The distance between the GrIS and the Ammassalik region is about 40 km. In the east, the Ammassalik islands borders the Denmark Strait, which separates Greenland and Iceland. This area is of climatological importance, since it is the thermodynamic transition zone between the North Atlantic Ocean and the Arctic Ocean (Mernild S. et al., 2012). The Ammassalik region is characterized by strong alpine relief in the eastern part of the island (Mernild et al., 2006), where altitudes can reach approximately 1,000 m a.s.l. Mittivakkat glacier is the largest and the most studied glacier across the island. In a study of climate conditions in the Mittivakkat glacier catchment, temperature at 2-meter (T_2m), wind speed, relative humidity and summer precipitation were shown to be increasing, whereas winter precipitation and total annual precipitation have decreased slightly over the period 1989–2006 (Mernild et al., 2008). T_2m was -0.9°C (1900–2008) and mean annual corrected precipitation at the AWS in Tasiilaq was approximately 1,200 mm water equivalent/year (w.e./year) (corrected after (Allerup et al., 1998; 2000)) (Mernild S. et al., 2012). The relative occurrence of solid, mixed and liquid precipitation has been estimated based on air temperatures (Hanssen-bauer, 2003), and was found to be 68%, 14% and 18% respectively in the period 1999–2006 (Mernild et al., 2008). Glacier thinning and glacier recession has been observed in the broader region of south-east Greenland (Bjørk et al., 2012; Mernild S. H. et al., 2012). The mass change from Mittivakkat glacier was -1.0 +/- 0.7 m w.e./year in the period 1996–2012 (Mernild S. H. et al., 2012; Yde et al., 2014). Within this time period, 14 out of the 16 years had a negative surface mass balance, with record mass losses in 2009/2010 and 2010/2011 (∼2.2 and ∼2.5 m w.e./year respectively) (Mernild S. H. et al., 2012). A more recent mass balance record based on measurements at the AWS gives an accelerated average mass balance rate of -2.1 m ice eq./year (2009–2019) (Fausto et al., 2020).

Given the limited detailed knowledge of past precipitation changes in coastal regions in Greenland, the objectives of this study are to 1) establish a quality-controlled database of precipitation and temperature data from Ammassalik island from a combined use of observational and model data, 2) conduct a time series analysis of these data, 3) with particular focus on spatiotemporal patterns of the ratio of liquid to total precipitation and the 4) altitude of the rain-to-snow boundary.

The Regional Atmospheric Climate Model (RACMO2) was developed by the Royal Netherlands Meteorological Institute (KNMI) (van Meijgaard et al., 2008). In this study, we use the polar version of this model, RACMO2.3p2, (hereafter RACMO2) that was developed by the Institute for Marine and Atmospheric research (IMAU), Utrecht University, in order to better simulate the evolution of surface mass balance over the ice sheets of Greenland, Antarctica and other glaciated regions (Noël et al., 2018). In this study, we use RACMO2 at 5.5 km resolution (RACMO2-5.5 km) (hereafter RACMO2-5.5 km) and the statistically downscaled product at 1 km (RACMO2-1 km) (hereafter RACMO2-1.0 km) (Noël et al., 2019). The forcing was done on a three-hourly basis by ERA-40 (1958–1978), ERA-Interim (1979–1989) and ERA-5 (1990–2021). The topography in RACMO2-5.5 km and RACMO2-1.0 km are from the GIMP DEM at 90 m downsampled to 5.5 km and 1.0 km (Howat et al., 2014).

RACMO2-1 km is well suited to resolve the steep surface mass balance (SMB) gradients in the topographically complex regions in which GICs are often situated (Noël et al., 2019), including Ammassalik island. This product extends to the GrIS and the GICs located outside of the ice sheet margin. The cloud physics scheme which determines partitioning of precipitation is described in (Wessem et al., 2014).

The Danish Meteorological Institute (DMI) operates several AWS in the coastal regions of Greenland. In this study we use several time series of variables observed at the AWS in Tasiilaq (TAS), located at 37.64 °W and 65.61°N (54 m a.s.l.) (Cappelen et al., 2021). A daily time series of T_2m (1958–2020) (DMI-TAS-daily T_2m) has been created from the available min. and max. T_2m according to the formula: D a i l y   m e a n   T 2 m = daily   min ⁡ . T 2 m + daily   max ⁡ . T 2 m 2 (1)

In addition, we use the monthly time series of T_2m (1958–2020) (DMI-TAS-monthly T_2m). For precipitation, we use a daily time series of accumulated precipitation (1958–2020) (DMI-TAS daily precip). Several gaps (mainly in 1980 and 2013–2014) exist in the precipitation observations at TAS (Cappelen et al., 2021). Minor inhomogeneities could also still exist in this dataset due to a small change in station location on 15 August 2015 as well as the installation of a Pluvio automatic raingauge at this date (Cappelen et al., 2021).

A monthly time series is also available (1958–2020) (DMI-TAS monthly precip). Metadata on the time series as well as further information about the sensors installed at the AWS are available in the historical data collection report from DMI (Cappelen et al., 2021).

The programme for Monitoring the Greenland Ice Sheet (PROMICE) (Fausto and Van As, 2021) has been measuring climate data since 2007 at several locations on the GrIS as well as on GIC on the ice sheet margins. The AWS at Mittivakkat (MIT) is located at 65.69°N and 37.83°W at an elevation of 440 m a.s.l. (Fausto et al., 2020). The station has been installed on 3 May 2009. From this dataset we use daily time series of air temperature (2009–2020) (GEUS-MIT daily T_2m). The temperature sensor is located at a height of approximately 2.6 meter above the surface, although the actual height varies with the snow height.

A table with information about all temperature sensors and precipitation gauges used in DMI-TAS and GEUS-MIT, as well as accuracy values, is given in the Supplementary Table S1.

From both RACMO2 datasets, a time series of data for Ammassalik island has been selected based on the minimum and maximum longitude and latitude of the entire island. Atmospheric variables from DMI-TAS and GEUS-MIT were extracted from RACMO2-5.5 km and RACMO2-1.0 km respectively using the nearest grid point. The corresponding model elevation is 68 and 443 m a.s.l. for TAS and MIT respectively. Quality control of the observed data has been done by DMI and GEUS for their respective datasets. RACMO2-1.0 km solid precipitation has been restricted to 1958–2015 due to a data gap in the model output thereafter. Daily solid precipitation from this model output consists almost solely of values <1 mm in the 2015–2021 period in our study region.

Subsets of the different model time series have been created for all meteorological seasons (DJF, MAM, JJA, SON) as well as for every month. Yearly and seasonal averages of T_2m and accumulated precipitation were created from the daily dataset. Long term trends in yearly T_2m and precipitation data have then been statistically assessed using the non-parametric Mann-Kendall statistical trend (M-K test) (Mann, 1945; Kendall, 1975; Mahmud, 2019). We use a critical value to detect significant trends with a 0.05 significance level. The slope corresponds to the Theil-Sen estimator and is reported as direction and magnitude of the change per decade.

We use the term rain-to-snow boundary (RSB) to identify the altitude where the shift from liquid precipitation to solid precipitation takes place. We assess monthly changes in the RSB in this region. Our analysis is conducted with the dataset covering the Ammassalik island domain in the model (see data preparation). For each day within each timeseries, regions with daily liquid and/or solid precipitation of more than 1 mm w.e. were selected. The RSB is then defined as follows:

• On days with only liquid precipitation, the RSB is defined as the highest altitude within the selected model domain where liquid precipitation occurs.

Days with both solid and liquid precipitation <1 mm in all grid cells are considered precipitation-free and thus do not get considered in the analysis of the RSB. For each month, an average of daily values of the estimated RSB was calculated. The M-K test was performed on yearly time series for each month of the year. At the same time, annual trends were also calculated for the number of days with precipitation above the 1 mm threshold within each month.

To support our use of RACMO2, we evaluate its performance for temperature and precipitation at different time steps at the two locations with AWS. From the observed and modelled data, monthly and annual means/totals were calculated for those months with full data coverage on the daily time scale for both data types. Root mean square error (RMSE) and Pearson correlation coefficient (R) were calculated for T_2m at MIT, T_2m at TAS and total precipitation at TAS.

Precipitation phase trends based on the ratio of modelled liquid to total precipitation were analyzed for the individual summer months and over different altitude ranges in 200 m bins. The covered area within a certain altitude range decreases with increasing altitude. As with the other trend analyses, the M-K test was used in order to assess the presence of statistically significant trends in the time series. Sen’s slope values of the M-K test at different altitude bins are for months with detected statistical trends.

In Figure 2 we show that modelled and measured T_2m at MIT have Pearson correlation coefficients of 0.95, 0.99 and 0.95 on the daily (Figure 2A), monthly (Figure 2D) and annual (Figure 2G) time scale respectively. These values are slightly lower on at TAS (0.84, 0.91. 0.41) on the daily (Figure 2B), monthly (Figure 2E) and annual (Figure 2H) time scale respectively. The low correlation value for T_2m at TAS at the annual time scale is mainly due to larger than average absolute differences in the period 1979–1989 (see section 4.2.1). A relatively low correlation (R = 0.38) exists for daily precipitation values at TAS (Figure 2C), whereas this value is higher for the monthly (0.80) (Figure 2F) and annual scale (0.71) (Figure 2I). We highlight the high RMSE (1,016 mm w.e.) for annual precipitation at TAS. On average, the difference between modelled and observed annual precipitation does not get stronger or weaker depending on the precipitation amount. On the contrary, we do see a larger difference between relatively higher observed and modelled precipitation totals on the monthly time scale.

Figure 6 shows monthly averages of the modelled rain/snow boundary altitude (RSB), based on RACMO2-1.0 km (GIC). July, August and September all show significant increases in the modelled RSB, while the other months of the year show non-significant trends. The slope values are +24.5 m/decade (July) (Figure 6A), +23.3 m/decade (August) (Figure 6B) and +20.0 m/decade (September) (Figure 6C).

The average altitude of the RSB during precipitation events is 652 m a.s.l. in summer. Given the fact that this altitude is increasing (1958–2016), the precipitation phase is shifting towards more liquid precipitation at this altitude and further increases in temperature and hence the liquid precipitation ratio are projected, a likely conclusion is that Mittivakkat glacier (as well as other GIC on Ammassalik island) will receive significantly more liquid precipitation in the future. The largest glacier-covered areas on Ammassalik island are located between 400–800 m a.s.l. The average RSB has remained inside this elevation range. If the same trend continues in the future, the average RSB will increase towards higher elevations with smaller GIC cover, which increases the likeliness of liquid precipitation on the glaciers on Ammassalik island. While the specific effect of increased liquid precipitation at marginal glaciers is not well known, it could potentially be an important factor for accelerated melt. Liquid precipitation events related to cyclonic weather conditions lead to amplified runoff on the GrIS, as the drainage system of the ice sheet can less efficiently drain the meltwater during this time (Doyle et al., 2015). Given the identified summer and autumn trends in precipitation in our study region, we hypothesize that the same mechanism could potentially contribute to enhanced runoff on GIC on Ammassalik island.