Antarctic meteorites threatened by climate warming

More than 60% of meteorite finds on Earth originate from Antarctica. Using a data-driven analysis that identifies meteorite-rich sites in Antarctica, we show climate warming causes many extraterrestrial rocks to be lost from the surface by melting into the ice sheet. At present, approximately 5,000 meteorites become inaccessible per year (versus ~1,000 finds per year) and, independent of the emissions scenario, ~24% will be lost by 2050, potentially rising to ∼76% by 2100 under a high-emissions scenario.


Supplementary Figures and Tables
Figure S1: Projected number of meteorites at the ice sheet surface until 2100 under two emissions scenarios.The piecewise linear function is fitted using linear least squares, forcing negative slopes throughout the century.The estimated slopes of the piecewise linear function represent the loss rates for periods of 20 years (Figure 2A).Differences between SSP1-2.6 and SSP5-8.5 before mid-century are related to the internal variability of the climate model (see "3. Climate model projections" in Supplementary Materials).The projections under the different emissions scenarios start to deviate in 2052, when the difference of estimated meteorites between the low-emissions and high-emissions scenario becomes larger than 2% of the estimated meteorites in the low-emissions scenario.The data consists of near-surface air temperatures, averaged over the entire globe, from the Community Earth System Model 2 (CESM2) using the high-emissions scenario (SSP5-8.5) 1 .The comparison to pre-industrial levels (defined as 1850-1900 in the literature 2 ) has been made in the same manner as for the near-maximum temperature changes over time (Figure S5).We fitted a 2 nd order polynomial through the data from 2020 until the end of the century and shifted the polynomial (and the here shown modelled temperatures) so that in 2020 the global air temperatures are +1.1 °C with respect to pre-industrial values (https://berkeleyearth.org/data/).From the 2 nd order polynomial fit through the modeled temperature changes, we estimated the years that correspond to a +1.5°C, +2.0°C, +2.5°C, and other warming levels relative to pre-industrial levels.From this smooth fit, we also estimated the temperature increases for each year, from which we constructed the x-axis of Figure 2B that displays the global air temperature increase.

Figure S4: Histogram of surface elevation at predicted meteorite finding locations
(excluding known locations where meteorites have been collected).The surface elevation is extracted from the Digital Elevation Model (DEM) observed by the TanDEM-X (TerraSAR-X add-on for Digital Elevation Measurement) satellite at a resolution of 90 meter 3 , where the ellipsoidal heights of the DEM are transformed to orthometric heights using approximated geoid heights provided in the data package Quantarctica 4,5 .We obtained the values at the gridded positive classified observation locations through bilinear interpolation.The values of 2100 consist of the average per bin of the two scenarios: (i) where new meteorites can appear with respect to 2020 and (ii) where no new meteorites can appear (see Methods).The rising temperatures will lead to a shift of the lower elevation limit at which meteorites can be found.The reported threshold of 1500 meters, which has been previously used to select potential meteorite stranding zones 6 , corresponds to the present-day 8 th percentile of predicted meteorite locations.This 8 th percentile will be about 200 meters higher by the end of the century under a high-emissions scenario.Table S1: Projected meteorite losses in dense collection areas listed by the Meteoritical Society 7 .Table indicates the official name of the area, the predicted number of meteorites per square kilometre (representative for 2020) in the area as delineated in the data provided by the Meteoritical Society ("-" corresponds to no predicted meteorites within the outlined area), and the year in which 50% of the total number of meteorites (including those already collected) are projected to be lost.NB, the number of meteorites per square kilometre is not a direct output of the classifier, which just predicts a probability on the presence meteorites 8 .We use the quantity here to evaluate the accordance between the outlined dense collection areas and our predictions.Hence, the estimated meteorite densities reflect both errors in the predictions and in the outlined dense collection area and should not be used to prioritize certain regions.

Uncertainties
We identify three possible limitations to the estimated numbers (e.g., Figure 2) that could potentially increase the meteorite loss significantly (i.e., we consider our loss estimates to be conservative).Firstly, the climate model output does not directly indicate an increase in heatwave-like events as would be expected in a warmer climate and has been demonstrated with climate model projections 9 .The projected near-maximum surface temperatures (here defined as the 99 th percentile of the 19-year distribution of 8-daily mean surface temperatures) over areas where meteorites have been found shifts only with 0.74 degrees for every 1 degree that the yearly average surface temperature over the Antarctic continent increases (r=0.993; Figure S6).This relationship could potentially be explained by a cap of surface temperatures in MAR at 0°C (by definition, if the temperature is above 0°C, the excess energy is used for melt, and the temperature returns to 0°C).So, over time, there must be more places where the surface temperature can no longer rise beyond 0°C.Next, we do not consider potential changes of the exposure of blue ice.In the high-elevation blue ice areas where meteorites are typically found, the variability in the extent of blue ice is determined by temporal (and spatial) variability in the surface mass balance 10 .Increases in surface mass balance reduce the exposed area of ice in time spans of weeks, while decreases in surface mass balance will result in sublimation of the firn layer until blue ice is exposed, a process that takes much longer 11 and results in limited expansion of the blue ice, as the observed horizontal mass-balance in blue ice areas is steep 10 .Hence, if the blue ice extent at high elevation (i.e., where sublimation is the dominant ablative process) changes, it is more likely that blue ice becomes snow-covered, causing meteorites exposed at the surface to become invisible 10,12 .Moreover, it is unlikely that young (new) blue ice areas contain a high concentration of meteorites as the build-up of such a concentration takes tens to hundreds of thousands of years 13 .Another limitation that is not captured in the estimated range is related to the fact that the meteorite classification algorithm does not consider a time componentwe assume here that its predictions are valid for the end of a 19-year interval of surface temperature observations.However, many of the meteorite finds used to train the algorithm are older than this 19-year period, where we know that surface temperatures were lower.This results in a classifier that is lenient regarding surface temperatures, i.e., with a hypothetical classifier that considers the time-component, the upper bound of the estimated loss range would be even higher.Moreover, with rising temperatures in the past decades, the build-up of meteorite concentrations could also be affected, because meteorites can sink before emerging to the surface.As meteorites get close to the surface (20-50 cm), the rock is affected by sunlight that penetrates through the ice 14 .This effect is enhanced in meteorites that are metal-rich (i.e., iron meteorites) and might explain the relatively small share of iron meteorites in the Antarctic collection, leading to the hypothesis of a hidden layer of iron meteorites under the surface of meteorite stranding zones 14,15 .

Climate model projections
We use the Modèle Atmosphérique Régional (MAR) 16 to project surface temperature changes.MAR is a polar-oriented regional climate model frequently used over the Greenland and Antarctic ice sheets.The model is forced every six hours at its boundaries by the surface pressure, air temperature and humidity, wind (u,v) components, sea surface temperature and sea ice of the Community Earth System Model 2 (CESM2) using a low-emissions and a highemissions scenario (SSP1-2.6 and SSP5-8.5, respectively).CESM2 has special parametrisations to better represent polar climates 17 .Although CESM2 has a high equilibrium climate sensitivity, it enables us to explore a large range of climate warming.
Since none of the climate models, including MAR, correctly simulate the location of all the blues ice areas, we forced their location in the model by prescribing an albedo representative of blue ice areas.This albedo is based on a blue ice index (percentage of blue ice for each 35km pixel of MAR) and represents a linear transition between no blue ice (albedo of 0.92, representing fresh snow) and 100% blue ice (albedo of 0.55, representing blue ice in MAR).The blue ice extent is based on rasterizing blue ice outlines (provided in the data package Quantarctica 5,18 ) to a 200-meter-resolution grid.We fixed the albedo values to not change over time, even in the event of fresh snowfall or melting.Prescribing the albedo makes it possible to calculate the energy balance and the resulting surface temperature representative of blue ice.However, the fixed albedo prevents the incorporation of albedo-feedback processes in MAR, which would enhance the surface warming significantly 19 , implying that the estimated surface temperatures are probably a lower bound of potential increases.We extract the temperature anomalies (see Methods) for each 450-meter pixel used in the meteorite-stranding-zone classifier by bilinear interpolation.Despite the fairly coarse resolution of MAR, the modelled surface temperatures compare well to the observed temperatures that were used to construct the meteorite-zones classifier, i.e., the 99 th percentile of the 19-year (2001-2020) distribution of 8-daily surface temperatures.The good correlation between the modelled and observed temperatures (r=0.90), and the near-one-to-one slope in Figure S7 indicates that the anomalies that we add to the observations are not biased toward either end of the temperature range.

Figure S2 :
Figure S2: Correlation between the global air temperature increase and the Antarctic meteorite loss for the lower-bound scenario and the upper-bound (see Figure 2, Figure S1 and Methods).These values are computed by comparing the temperatures and the number of meteorites at any possible interval between 2020 and 2100, resulting in 3240 datapoints.The trends are computed by a linear least squares estimation.

Figure S3 :
Figure S3: Global near-surface air temperature changes over time with respect to preindustrial values.The data consists of near-surface air temperatures, averaged over the entire globe, from the Community Earth System Model 2 (CESM2) using the high-emissions scenario (SSP5-8.5)1 .The comparison to pre-industrial levels (defined as 1850-1900 in the literature 2 ) has been made in the same manner as for the near-maximum temperature changes over time (FigureS5).We fitted a 2 nd order polynomial through the data from 2020 until the end of the century and shifted the polynomial (and the here shown modelled temperatures) so that in 2020 the global air temperatures are +1.1 °C with respect to pre-industrial values (https://berkeleyearth.org/data/).From the 2 nd order polynomial fit through the modeled temperature changes, we estimated the years that correspond to a +1.5°C, +2.0°C, +2.5°C, and other warming levels relative to pre-industrial levels.From this smooth fit, we also estimated the temperature increases for each year, from which we constructed the x-axis of Figure2Bthat displays the global air temperature increase.

Figure S5 :
Figure S5: Near-maximum temperature changes over time with respect to 2020.The years correspond to the last year of a 19-year period (i.e., 2020 corresponds to the 99 th percentile of 8-daily surface temperatures in the period 2001-2020).

Figure S6 :
Figure S6: Near-maximum temperatures compared to average temperatures.Correlation between the near-maximum surface temperatures at all known meteorite finding locations 7 (yaxis) and the Antarctic-wide yearly average surface temperatures (x-axis).

Figure S7 :
Figure S7: Comparison between modelled and observed temperatures over all places used for the classification of meteorite stranding zones (i.e., places where meteorites have been found, and potential meteorite-finding places consisting of blue ice areas and their near vicinity, represented by a 1-km buffer around blue ice outlines 18 ).