Towards critical white ice conditions in lakes under global warming – Nature.com

Ice quality observations

During the IceBlitz campaign, the total lake ice thickness on lakes across the Northern Hemisphere varied from 1 to 103 cm, with a median of 29 cm (mean ± one standard deviation: 32 ± 18 cm). Most measurements were performed in January and February (n = 53 and 58, respectively), a few weeks after ice-on in December or January. Many lakes were already ice-free in April, decreasing the number of sampling occasions during that month to ten. In our data set, the thickest ice layers were usually observed in February or March, depending on how long the ice cover lasted.

The observed ice thickness during the 2020/2021 IceBlitz campaign was generally lower than we found reported in the literature (Table 1). Out of 117 lake ice thickness observations from 18 lakes performed in winters prior to the sampling year (Table 1, Fig. 1a), only 21 observations demonstrated an ice thickness less than 29 cm, and the median across all 117 observations from the literature was 42 cm (mean ± one standard deviation: 45 ± 16 cm). There are many possible reasons why previous observations from other lakes deviate from our observations during the very warm winter in 2020/2021 (Fig. 1b), such as geographical location, lake size, lake depth, sampling date, degree of global warming, morphological and hydrothermal conditions. It is, however, rather well documented that many lakes in the world presently experience a trend of thinner ice e.g.,20,21,22, suggesting that the thinner ice conditions during our sampling campaign are a result of rapidly increasing winter air temperatures in the Northern Hemisphere during the past years (Fig. 1b). Increased winter air temperatures are most likely also the reason why we found a significantly decreased total ice thickness in the Mozhaysk Reservoir in 2011–2021 compared to 1971–1979 (median: 37 and 53 cm, respectively; Wilcoxon signed-rank test: p < 0.0001, n = 210 and 169, respectively). Because winter air temperatures in the Northern Hemisphere are projected to further increase23, we expect that the trend towards thinner ice will continue into the future.

The ice on IceBlitz lakes commonly consisted of both a black and a white ice layer where the black ice layer was the dominant ice layer when all 167 measurements were taken into consideration (mean ± one standard deviation: 70 ± 28%). There was, however, a clear seasonal trend in the ice layer composition with a substantial increase in the thickness and the proportion of white ice towards late winter and spring (Fig. 2a, b). From being mainly absent in December 2020, the median white ice thickness across the Northern Hemisphere increased to 3 cm (13% of the total ice thickness) in January 2021, 10 cm (33% of the total ice thickness) in February 2021, and finally 23 cm (52% of the total ice thickness) in March 2021 (Fig. 2a, b). Thus, white ice became the dominant ice type for the 31 seasonally frozen lakes located across the Northern Hemisphere during the time before ice-off. An increase in the thickness and proportion of white ice from the beginning towards the end of ice cover was observed for all lakes with available seasonal data (see Fig. 2c for an example) except for Hobart Lake, USA, where the white ice layer slightly decreased from 9 to 4–8 cm from January to February 2021. The range of spatial variation in the accumulation of white ice on a lake during one sampling occasion remained below 8 cm even when 12 different sites on a lake were visited. Only once the in-lake difference in the white ice layer reached 21 cm on Lake Vendyurskoe. Thus, the seasonal variation in the thickness of the white ice layer commonly exceeded the observed in-lake variation assessed during one sampling event.

Fig. 2: Seasonal accumulation of white ice in lakes.
figure 2

a, b Boxplots showing the seasonal development of the thickness of white ice and the percentage of white ice observed in 31 Northern Hemisphere lakes during the IceBlitz sampling campaign in 2020/2021. Boxplots depict the minimum, first quartile, median, third quartile, and maximum. April values are not shown because too few measurements were available from that month. Colors represent the lake site-specific mean air temperature anomaly during December through March in 2020/2021 relative to the base period 1951–1980. Except for the lakes located on the Kola Peninsula (Northwest Europe), all lakes experienced warmer than normal winter air temperatures during the IceBlitz campaign. Eight lakes even had 3 °C warmer air temperatures compared to 1951–1980. c Seasonal development of total ice thickness and the thickness of black and white ice in Lake Oulujärvi, Finland during 2020/2021. The orange shaded area marks the time period when air temperatures varied around the freezing point, which is relevant for the formation of white ice. d Simplified, typical winter air temperature (T) curves representing a cold (blue line) and a warm (red line) winter (data taken from Weyhenmeyer et al.26). The number of days when air temperatures vary around the freezing point increases when the seasonal cycle of winter air temperatures falling below 0 °C flattens during a warm year26. In our conceptual figure, the number of days when air temperatures vary around the freezing point corresponds to ~15 days during a warm winter (marked in orange) compared to ~8 days during a cold winter (marked in blue).

White ice formation

White ice is commonly formed when snow accumulates on ice, melts, and refreezes24. White ice also forms when rain falls on the snow layer to form slush, which subsequently can freeze and turn into white ice or when the snow load is sufficient to force lake water to the ice surface through cracks in the ice matrix9,25. Melting and refreezing of ice, slush and snow as well as rainfall are conditions that typically occur when air temperatures vary around the freezing point, i.e., usually during the time before ice-off. We attribute the ice layer composition change towards spring to changes in the seasonality of winter air temperatures. When winters become warmer, the seasonal cycle of air temperatures below the freezing point flattens, similar to the flattening that occurs towards warmer geographical regions26. A decrease in the winter air temperature amplitude implies that the number of days with air temperatures varying around the freezing point increases (Fig. 2d), prolonging the formation of white ice and increasing its proportion. This lake ice quality response to warmer winter air temperatures is increasingly pronounced towards warmer geographical regions26. Under continued global warming, lakes located in those warmer geographical regions are most likely increasingly exposed to air temperatures varying around the freezing point, resulting in an increased prevalence of white ice, in particular during the critical time before ice-off.

White ice and ice stability

An increase in the proportion of white ice can jeopardize the use of seasonally ice-covered lakes for subsistence, recreation, transportation and other purposes,27,28. It has already been noted that most fatal winter drownings occur just prior to ice-off29, coinciding with the time when ice layers on lakes commonly become dominated by white ice. Gold15 has developed a method to estimate the bearing strength of ice depending on the total thickness and ice quality:

$$P={A;cdot ;}{H}^{2}$$


where P is the allowable load in kg, A is the bearing strength that varies between 3.5 and 30 kg cm−2 depending on ice quality (30 kg cm−2 for cold, black ice conditions of high quality, i.e., no snow ice, no cracks, no bubbles etc. and 3.5 kg cm−2 for low quality ice16), and H is the total ice thickness in cm. Although Eq. (1) with A = 3.5 kg cm−2 is used for general ice safety guidelines30, Gold15 described some cases in which the predicted load was not supported by the ice layer. White ice conditions at air temperatures varying around the freezing point (0 °C) might be responsible for ice collapse despite a sufficient theoretical total ice thickness. Laboratory experiments have shown a reduction of the flexural strength of ice (which is proportional to the bearing strength A) by 51% for white ice compared to black ice at a temperature of −0.5 °C16. Based on these results, we suggest a modification of Gold’s equation during days when air temperatures vary around the freezing point:

$$P=frac{A,cdot {H}^{2}}{2},cdot ,left(1+frac{100-%{white},{ice}}{100}right)$$


where P is the allowable load in kg, A corresponds to 3.5 kg cm−2 (see also Eq. (1)), H is the total ice thickness in cm and % white ice is the proportion of white ice in the ice layer. Further adjustments of Eq. (2) might be needed when air temperatures stay above 0 °C for 24 h or more because under those conditions ice usually loses additional strength15,31. It also important to keep in mind that Eq. (2) will not apply to ice conditions just before ice-off when the entire ice matrix begins to deteriorate.

Applying Eq. (2), an individual weighing 100 kg might safely walk on ice if there is a white ice layer of at least 8 cm thickness (Fig. 3a). Thus, the recommended 10 cm of total ice thickness on lakes, which most webpages give as an ice safety guideline e.g.,32, seems generally suitable for one single person to safely walk on ice, as this ice thickness allows for a load between 175 and 1753 kg depending on ice quality (Fig. 3a). It is, however, important to keep in mind that usually more than one person walks on the ice at a given time. For example, the weight of a group of three people often exceeds 175 kg, which is the allowable load for 10 cm of pure white ice conditions (Fig. 3a). Consequently, groups of people are at higher risk to fall through ice under complete white ice conditions, despite an ice thickness of 10 cm or more. We observed such complete white ice conditions on eight occasions in three different countries during the IceBlitz campaign, i.e., in Canada (Lake Wilcox), Switzerland (Lake Nero) and Sweden (Lake Erken). All 100% white ice conditions occurred in February 2021, when the total ice thickness in the three lakes varied between 2 and 38 cm, with a median of 12.5 cm. In central and northern Sweden, February is usually the month when people can safely walk on ice, and rarely are any fatal winter drownings reported. In February 2021, however, ten people died by falling through ice which represented the highest winter drowning death rate in Sweden during the month of February since the records began in 200033. The ice situation in February 2021 demonstrates that behavioral adaptations to a warmer world are needed. Ice conditions that traditionally have been safe during winters of the past will become unsafe in the future. One strategy to make people more aware of the increasing risk of unstable ice conditions in a warmer world might be to spread information on revised ice safety guidelines highlighted above. To keep the update of the ice safety guidelines as simple as possible a rule of a thumb could be to double the presently used ice thickness guideline.

Fig. 3: Lake ice conditions that are critical for ice stability and for the transmittance of photosynthetically active radiation.
figure 3

a Variation in the estimated allowable load on ice depending on total ice thickness and ice quality. Pure black ice conditions were modeled using Eq. (1) with A = 30 kg cm−2 and pure white ice conditions using Eq. (2) with A = 3.5 kg cm−2. The dashed line represents estimates of the allowable load using Eq. (1) with A = 3.5 kg cm−2, which is commonly used for ice safety guidelines. Black dots show estimates for the IceBlitz dataset using Eq. (2) with A = 3.5 kg cm−2. The red and gray shaded areas mark the allowable load for an ice thickness of 10 cm and less under pure white ice (red) and pure black ice (gray) conditions. b Daily mean (8.00 a.m. to 8.00 p.m.) under-ice irradiance (Edz) in Watts (W) m−2 in Lake Vendyurskoe during spring just before ice-off in relation to the thickness of snow and white ice (H) on the lake. Data were taken from Zdorovennova et al.36, measured during 1997–2020. Shown is the exponential decline of Edz with increasing H (black line). The red shaded area marks the light availability below a 10 cm thick snow and white ice layer on a lake.

Light transmittance through white ice and ecological effects

White ice conditions do not only affect the stability of the ice cover but also the light regime underneath. White ice is known to have a severalfold higher reflectance than black ice, thus only small amounts of photosynthetically active radiation penetrate through white ice34. The transmittance of light through a combined black and white ice/snow layer on lakes has previously been described by a two-layer model34:

$$tau left(Cright)=frac{tau (A)cdot tau (B)}{1-rho (A)cdot rho (B)}$$


where τ(C) is the transmittance of light through the combined black and white ice/snow layer, τ(A) is the transmittance of light through the upper white ice/snow layer, τ(B) is the transmittance of light through the lower black ice layer, ρ(A) is the reflectance of the upper white ice/snow layer and ρ(B) is the reflectance of the lower black ice layer. τ(C) approaches zero when snow covers the ice, as snow usually has a reflectance of close to one35. τ(C) also declines when the upper white ice layer becomes thicker36. Based on the data from Lake Vendyurskoe (1997–2020)36 we obtained the following model for underneath ice light conditions:

$${{Ed}}_{z}=acdot {e}^{-varepsilon cdot H}$$


where Edz is the 12-hour day time mean (8.00 a.m. to 8.00 p.m.) irradiance underneath ice in W m−2, a is a solar radiation dependent constant (in our example corresponding to a value of 103; see Fig. 2b), ε is the extinction of light with 0.09 W m−2 cm−1 and H is the thickness of the sum of snow and white ice on a lake in cm. When we included the thickness of black ice in H, the prediction of Edz did not improve (R2 = 0.45, p < 0.0001, n = 39 compared to R2 = 0.52, p < 0.0001, n = 39 when only white ice and snow was considered). We attribute the lack of model improvement to the fact that the thickness of black ice was unrelated to Edz in Lake Vendyurskoe (p > 0.05). These results indicate that Edz is mainly driven by the thickness of snow and white ice, where each cm of snow and white ice reduces the under-ice irradiance by approximately 9%. Further, by analyzing the inter-annual variability in Edz in Lake Vendyurskoe in spring, we found no trend over time in Edz despite a significant decrease in total ice thickness during this period (linear time series model: p > 0.05 and p = 0.007, respectively). Thus, under-ice light conditions did not improve despite thinner ice in spring. We suggest that these under-ice light conditions were driven by the thickness of white ice which did not show a significant decrease in spring from 1997 to 2020 (linear time series model: p > 0.05).

Low light conditions in spring caused by a white ice layer and/or snow on ice are critical for the development of primary producers and consumers as their growth and reproduction is dependent on light and convection underneath ice37,38. Most vulnerable to white ice conditions and snow on ice are photoautotrophs in lakes39, in particular non-motile photoautotrophs that are dependent on radiatively-driven convection to reach sufficient light (e.g., diatoms)40,41. Motile taxa can have an advantage during low light conditions under-ice13. Mixotrophy is another strategy to adapt to low-light under-ice conditions42. Further up in the food web zooplankton has developed survival strategies to low-light under-ice conditions by feeding on mixotrophic microorganisms and allochthonous organic material, but for growth and reproduction, a diet containing nutrient-rich photoautotrophs is still essential43. Thus, despite a variety of survival strategies, snow and white ice induced changes to the photoautotroph community cascade through the food web, with substantial consequences for microbial, zooplankton, and fish populations44,45.

Relevance and future perspective

Our sampling campaign focused on spatial and seasonal patterns. The most obvious and relevant result was the consistent buildup of white ice in spring until the time just before ice-off which we attribute to the increased time window when air temperatures vary around the freezing point. Under expected future global warming, many lakes in the world will become exposed to these prolonged freeze-thaw cycles. According to our study, such lakes will experience a shift in the ice layer composition towards white ice conditions. We conclude that white ice conditions in lakes need far more consideration than previously given. We suggest the use of Eq. (2) to re-examine ice safety guidelines to account for ice instability during white ice conditions. We also recommend considering the thickness of white ice as an important regulator of physical, chemical, and biological processes in lakes.


Please help keep this Site Going