A 15-yr Circum-Antarctic Iceberg Calving Dataset Derived from Continuous Satellite Observations

. Iceberg calving is the main process that facilitates the dynamic mass loss of ice sheets into the ocean, which accounts for approximately half of the mass loss of the Antarctic ice sheet. Fine-scale calving variability observations can help reveal the calving mechanisms and identify the principal processes that influence how the changing climate affects global sea level 20 through the ice-shelf buttressing effect on the Antarctic ice sheet. Iceberg calving from entire ice shelves for short time intervals, or from specific ice shelves for long time intervals, has been monitored before, but there is still a lack of consistent, long-term, and high-precision records on independent calving events for all of the Antarctic ice shelves. In this study, a 15-yr annual iceberg-calving product measuring every independent calving event larger than 1 km 2 over all of the Antarctic ice shelves that occurred from August 2005 to August 2020 was developed based on 16 years of continuous satellite observations. First, the 25 2 the Ronne-Filchner and Ross ice shelves, advanced with low calving frequency, while small- and medium-sized ice shelves retreated and calved more frequently. Iceberg calving of ice shelves is most prevalent in West Antarctica, followed by the 95 Antarctic Peninsula and Wilkes Land in East Antarctica. The annual iceberg calving event dataset of Antarctic ice shelves provides consistent and precise calving observations with the longest time coverage. The dataset provides multi-dimensional variables for each independent calving event that can be used to study detailed spatial-temporal variations in Antarctic iceberg calving. The dataset can also be used to study ice-sheet mass balance, calving mechanisms, and responses of iceberg calving to climate change. The dataset is shared via National Tibetan Plateau Data Center, and entitled “Annual iceberg calving dataset 100 of the Antarctic ice shelves (2005-2020)” with DOI: 10.11888/Glacio.tpdc.271250. In addition, the average annual calving rate of 18.4±6.7 Gt/yr of the calving events smaller than 1 km 2 of the Antarctic ice shelves, as well as the calving rate of 166.7±15.2 Gt/yr of the marine-terminating glaciers, were estimated. The authors obtained comprehensive, detailed iceberg calving observations at different scales through image matching and feature tracking, which made it possible to investigate calving patterns and mechanisms. Their work laid the foundation for 140 the subsequent exploration of the physical triggers of small and large calving events (Medrzycka 2016) and revealed the "self-organized critical systems" of glaciers and ice sheets at different calving scales (Åström 2014). The long-term and high-precision remote sensing observation of circum-Antarctic independent calving events not only describes the spatial and temporal features of iceberg calving but also provides fundamental data for further investigating calving mechanisms and estimating ice-shelf mass balance in response to climate change. In this study, we identify annual calving events through a combination of a velocity-based ice shelf front edge simulation and semiautomatic annual iceberg calving extraction. We further acquire the calved-area outline, location, year of occurrence, area, thickness, volume, mass, and recurrence interval of each calving event. Building on this, we develop a circum-Antarctic iceberg calving dataset. The dataset spans August 2005 to August 2020. Using this product, we analyse the spatial and temporal characteristics of iceberg calving for the last 15 years.


Introduction
The ice shelves surrounding Antarctica's coastline play an important role in the stability of the Antarctic ice sheet and its 105 mass balance. Iceberg calving is a process whereby the ice from a glacier or ice-shelf frontal edge is stripped away and enters the ocean. Iceberg calving accounts for approximately half of the net mass loss of all Antarctic ice shelves (Rignot et al., 2013;Depoorter et al., 2013). Enhanced iceberg calving can indirectly lead to ice shelf instability, which accelerates the outflow of tributary glaciers into the ocean, causing sea level rise (Berthier et al., 2012;Furst et al., 2016;Rignot et al., 2004). In-depth studies of the calving process are essential to accurately predict the impact of future climate change on ice shelves/sheets and 110 sea levels.
Model simulations and remote sensing observations are two major tools used to study iceberg calving. The former focus on simulating the dynamic process of a calving front in response to atmospheric and oceanic forcings and stress within ice sheets. Different models are used to understand the evolution and changes of ice shelves (Hill et al., 2018;Lovell et al., 2017;Luckman et al., 2015;Miles et al., 2017). The latter focus on the monitoring and quantitative assessment of calved areas 115 using remotely sensed data, which can be assimilated into ice sheet models to further improve the accuracy of model simulations (Massom et al., 2018;Pattyn and Morlighem, 2020).
Research on remotely sensed iceberg calving monitoring can be classified as having three main focuses: (1) observations of specific ice shelves or glaciers with high spatial resolution data, e.g., long-term monitoring of the Pine Island Glacier, Mertz Glacier Tongue, and Amery Ice Shelf (Bindschadler, 2002;Massom et al., 2015;Zhao et al., 2014); (2) observations made of 120 larger regions with lower spatial and temporal resolution data, e.g., calving monitoring along the Antarctic Peninsula and Ross Sea coast (Cook et al., 2005;Cook and Vaughan, 2010;Fountain et al., 2017); and (3) circum-Antarctic calving front observations of specific years based on satellite image mosaics of the Antarctic coastline (Liu and Jezek, 2004;Liu et al., 2015;Scambos et al., 2007;Yu et al., 2019). The first two types of studies achieve the precise monitoring of calving events in specific ice shelves or small areas while the third type quantitatively assesses iceberg calving at the continental scale. Liu et 125 al. (Liu et al., 2015)extracted 579 independent calving events for six years from the Envisat ASAR circum-Antarctic mosaic. 3 The authors obtained comprehensive, detailed iceberg calving observations at different scales through image matching and feature tracking, which made it possible to investigate calving patterns and mechanisms. Their work laid the foundation for 140 the subsequent exploration of the physical triggers of small and large calving events (Medrzycka et al., 2016) and revealed the "self-organized critical systems" of glaciers and ice sheets at different calving scales (Åström et al., 2014).
The long-term and high-precision remote sensing observation of circum-Antarctic independent calving events not only describes the spatial and temporal features of iceberg calving but also provides fundamental data for further investigating calving mechanisms and estimating ice-shelf mass balance in response to climate change. In this study, we identify annual 145 calving events through a combination of a velocity-based ice shelf front edge simulation and semiautomatic annual iceberg calving extraction. We further acquire the calved-area outline, location, year of occurrence, area, thickness, volume, mass, and recurrence interval of each calving event. Building on this, we develop a circum-Antarctic iceberg calving dataset. The dataset spans August 2005 to August 2020. Using this product, we analyse the spatial and temporal characteristics of iceberg calving for the last 15 years. 150 2 Data

Satellite imagery
Considering the relatively low calving frequencies measured in August of each year (Liu et al., 2013) and the time limitations of available satellite images, we define the annual calving recurrence interval as running from August of the current year to August of the following year. We know that it is difficult to create such a circum-Antarctic iceberg calving dataset 155 based on a single satellite platform. To continuously monitor Antarctic iceberg calving for 2005 to 2020, multisource remotely sensed data are used in this study. We prioritize using SAR (Synthetic Aperture Radar) images for early August each year given that their quality is minimally affected by polar nights and cloudy days. For periods and areas for which SAR data are not available, optical images for close dates are used instead. Satellite images used in the development of this product include

Supplementary datasets
Additional remote sensing data were also used to facilitate product development and analyses. MEaSURE InSAR (interferometric synthetic aperture radar)-based Antarctica ice velocity map version 2 (Rignot et al., 2011;Mouginot et al., 2012)    are used for calving mass calculation for marine-terminated glaciers. MEaSUREs Antarctic Boundaries Version 2 (Rignot et al., 2013) is used for the ice shelf delineation and spatial analysis of the calving distribution. Two ice thickness datasets (Bedmap 2 and Bedmachine) Fretwell et al., 2012) are used for calving thickness extraction and calving mass calculation for both calvings from ice shelves and marine-terminated glaciers. 175 The Reference Elevation Model for Antarctica (REMA) (Howat et al., 2019) is used for the uncertainty evaluation of the extracted thickness. The Antarctic daily surface melt dataset (Picard and Fily, 2006) is used to analyse the response of iceberg calving to ice sheet surface melting.
Detailed descriptions of each remote sensing product used are presented in Table 2.

Processes of direct observation of annual independent calving event
An annual calving event occurs when an independent calved area has an outline that does not overlap or is spatially adjacent to other calving events occurring in the same year (even if it occurs on the same ice shelf), namely, the topology 195 requires nonoverlapping and nonadjacent annual calved-area polygons for the specific year. Data generation involves the following three steps: preprocessing the data, extracting iceberg calving, and acquiring attributes ( Figure 1). Each of these steps is discussed in the following sections. Besides, the consistency of multisource satellite imagery used in monitoring annual iceberg calving has been validated.

Iceberg calving extraction of independent calving events
To create the annual iceberg calving dataset for the Antarctic ice shelves, we simulated the expansion of the ice-shelf frontal edge and detected the calved areas based on satellite images. It is worth mentioning again that our iceberg calving extraction only included calving from ice shelves but did not include marine-terminating glaciers, and the boundaries of ice shelves are referenced from the MEaSUREs Antarctic Boundaries Version 2 released by NSIDC. We first manually digitalized 225 the ice-shelf frontal line in August 2005August , 2010, and 2015 as the input benchmark coastline. Then, the following steps were iterated for the extraction of each annual calving cycle with the methodology divided into two overarching tasks: velocitybased ice shelf front edge simulation and semiautomatic annual iceberg calving extraction (Qi et al., 2020). a) Velocity-based ice shelf front edge simulation. We converted the vertices of the input coastline to obtain the set of coastline feature points for a specific year. Based on the velocity at the position of each coastline point, we calculated the 230 movement of feature points over the duration of the given year. By lining up the moved feature points sequentially, a new coastline was derived, namely, the simulated coastline of the next year, as shown with yellow lines in Figure 3.
Additionally, we conducted a controlled experiment on the impact of different ice velocity products while simulating the next-year coastline. Fifty points on the high-flowing Pine Island Glacier were randomly selected as samples. We simulate their 11-yr movement using both the average ice velocity map   For the non-calving area, theoretically, the simulated coastline should fit the real coastline shown in a remotely sensed image well, but due to the geographical bias of images and errors of the ice velocity product, some deviations between the directly obtained simulated coastline and actual coastline may occur. Therefore, before extraction, we first checked and rectified the simulated coastline to ensure that it fits the actual coastline in non-calving areas. After manual correction, the extraction results were found to be of good accuracy. 245 b) Iceberg calving extraction. We manually rectified the simulated coastline to ensure that after rectifying, it fit the real coastline shown in the corresponding satellite images. Then, we obtained the actual coastline for the next year, which is shown as the red line in Figure 3. We extracted the enclosed area between the simulated coastline and the actual coastline to acquire the calved area (the blue area in Figure 3). After extracting for one annual calving cycle, we checked topological relations at the continental scale for this year. We ensured that calved-area polygons did not intersect with each other and then obtained 250 vectors for each calved area for the given year.
This iceberg calving extraction method employs a simple process and broad applications. The actual coastline modified from last year's extraction can be used as the input coastline of the next year's extraction; thus, we can provide time-continuous iceberg calving monitoring and effectively avoid repetition and omission errors. Additionally, the semiautomatic operation offers incomparable precision and efficiency, greatly reducing the postprocessing workload. 255

Attribute acquisition of independent calving events 260
For individual calving events, attributes include the area, calving scale, average thickness, mass, calving recurrence interval, and uncertainties of relevant parameters. Therefore, the acquisition of calved area and calved mass, uncertainties, and recurrence intervals are discussed in the following sections.

Calved area and calved mass
After acquiring vectors of the calved area polygons, we calculated their areas under polar projection. Then, these values 265 were divided into four different scales: small-scale (1-10 km 2 ), medium-scale (10-100 km 2 ), large-scale (100-1,000 km 2 ), and extra-large-scale values (>1,000 km 2 ). We further obtained the average thickness of each calved area from the Antarctic ice thickness products (Bedmap 2 and Bedmachine). First, we masked out the ice-shelf zone thickness in Bedmap 2 and Bedmachine. Second, we extracted the average thickness of each calving event from the masked ice thickness through step 1.
Then, we checked the average thickness of all calving events. For missed and abnormal values (results show that they only 270 account for a small proportion of the total), we moved the polygon backward along the ice flow to the calving front where there is thickness data coverage. After that, we re-extracted the average thickness of those calving events to make sure they are given appropriate thickness.
Based on area and thickness, the calving mass ( ) was calculated from Eq. (1): where stands for the calving area and represents the average thickness of the calved area. The standard value of ice density = 917 / 3 was used for the calculation.

Uncertainty assessment
The uncertainties involved in the calculation of calving mass based on Eq.
(1) include errors of calving area measurement, thickness extraction, and ice density. The uncertainty of the calving area is determined by the accuracy of the extraction method. 280 Thickness uncertainty should be theoretically affected by top surface elevation measurements and firn depth correction; in reality, there are also uncertainties in thickness changes with time, according to hydrostatic equilibrium assumptions, and in the offsets in locations during extraction. In this section, we evaluate the main uncertainties encountered during the development of the annual iceberg calving dataset.  (Qi et al., 2020); therefore, the uncertainty of the calving area ( ) can be calculated from Eq. (2): where represents the perimeter of each calving event (km).
b) Thickness uncertainty. The ice-shelf thickness dataset used in this product is derived from the hydrostatic equilibrium 295 , which is written as Eq. (3): where denotes ice-shelf thickness. is the top surface elevation, namely, the height of the snow top. is firn depth correction, and = 1,027 / 3 is the density of seawater.
Therefore, thickness uncertainty ( ) can be evaluated from Eq. (4): 300 where and represent the average thickness and average surface elevation of the calved area, respectively. is the uncertainty of the calved-area surface elevation, is the uncertainty of firn depth correction, and and represent the uncertainty of ice and seawater density, respectively.
For the calculations, 917 kg/m 3 is used for and 1,027 kg/m 3 is used for , and their uncertainties and are 305 valued at 5 kg/m 3 (Griggs et al., 2011). was obtained from REMA with typical elevation errors of less than 1 m (Howat et al., 2019). Firn depth correction and its uncertainty were calculated from regional climate model RACMO2/ANT with a ratio accounting for 8% (Pritchard et al., 2012). c) Calving mass uncertainty. The calving mass of our dataset is derived from three components unrelated to and independent of each other. Thus, we used synthetic standard uncertainty to evaluate its accuracy. The mass deviation of a 310 single calving event ( ) is as follows Eq. (5), and the mass deviation for the year cycle ( ) can be calculated from Eq. (6): where and are the mass and area of individual calving events, respectively. N is the number of years, and n is the total frequency of calving events that occurred in N years. 315

Recurrence interval
Calving recurrence means that a calving event with the same spatial scale reoccurs at the same calving front (Liu et al., 2015), which are usually thought to be part of the natural cycle of advance and retreat of ice shelves. The recurrence interval of a calving event, a measurement of the natural calving cycle, is defined as the year interval between the two recurrence calving events. To acquire this attribute, we performed the following work. First, we get the perimeter of each calving polygon 320 through the function "Calculate Geometry" in ArcMap. Based on that, we calculated the average perimeter of all calving events at the same scale for 15 years. We defined the Buffer radii as half of the average perimeters at different scales rounded upwards to the nearest integer. The specific values used for this dataset are shown in Table 3. After that, we used the function "Feature to Point" in ArcMap to get the center points of each individual calving polygon.
For an input polygon, the location of the output point will be determined as its center of gravity. Then, we build buffers for each calving center point based on the radii calculated in the previous steps. For each calving event, we count the number of calving center points with the same scale that falls into its buffer. For buffers that fall into more than two points, the calving recurrence interval is defined as the total number of years (15) divided by the exact number of calving center points falling 330 within. For buffers with only one point, the calving recurrence interval is defined as the greater value of time intervals between these calving events and boundary years (2005 or 2020).

Consistency validation of multisource satellite imagery
As mentioned above, a single satellite platform cannot accommodate long-time-series observations of circum-Antarctic calving events. Thus, multisource remotely sensed data are used in this study. To check whether the results derived from 335 different sensors are similar, especially for the results derived from optical sensors and SAR, we performed the following verification.
For the year for which we have both SAR and optical images, we extracted circum-Antarctic annual iceberg calving using the same method based on different sources of remotely sensed imagery. We chose to repeat the calving extraction for 2016/17 through Terra/Aqua MODIS imagery and to compare it to the contemporaneous extraction results for our dataset derived from 340 Sentinel-1 SAR imagery. We define area differences as the calving area obtained from MODIS subtracted from that obtained from SAR, and we define the calving perimeter as the calved-area perimeter obtained from SAR. Then, we analyse the area differences of the same calving events and calculate error-equivalent perimeter widths by dividing the area differences by the calving perimeter. 3.6 Estimation of the less than 1 km 2 calving from the Antarctic ice shelves

Estimation method
Considering the huge workload and relatively small calving area contributing to the total calving area, we estimated the 360 annual calving area and mass of the less than 1 km 2 calving of Antarctic ice shelves using the following equation: where of 0.22 is the area ratio between the 0.1-1 km 2 calving and 1-10 km 2 calving estimated by Qi et al. (2020)， <1 2 and 1−10 2 are the calved area of the less than 1 km 2 caving and the 1-10 km 2 , <1 2 and 1−10 2 are the 365 calved mass of the less than 1 km 2 caving and the 1-10 km 2 , we have neglected higher-order terms in the expansion.

Uncertainty assessment
The area uncertainty <1 2 and the mass uncertainty <1 2 of the less than 1 km 2 caving are calculated as follows:

Estimation method
The calving rate of the marine-terminating glaciers is equal to the ice flux along their grounding lines. Ice flux comprises 375 the flux gate width multiplied by ice velocity and ice thickness at the grounding line. The ice velocity and the ice thickness vary considerably from grounding line positions. Therefore, the grounding line is normally discretized to calculate the ice flux of each flux gate. Then, the calving rate − of the marine-terminating glaciers are calculated as follows: where is the equivalent ice thickness of the flux gate , ⃑ is the ice velocity along the ice flow direction, ⃑ is 380 the fluxgate width along the ice flow direction, and is the density of ice (917 kg/m 3 ).

Uncertainty assessment
The calving mass uncertainty 13 where and stand for the uncertainties of ice thickness and ice velocity. For the calculations, 100 m is used for 385 (Rignot, 2008;Rignot et al., 2011), and 17 m/yr is used for and . 917 kg/m 3 is used for , and 5 kg/m 3 for its uncertainty (Griggs et al., 2011).

Consistency of multi-source satellite imagery
We extracted a total of 220 calving events from MODIS for 2016/17 covering a total area of 9,064.6 km 2 . As shown in 390 Table 4, both the total number of calving events and the total calved area are slightly lower than those derived from SAR imagery. The numbers of calving events at different scales extracted from the two sources of satellite images are similar. The frequency error mainly originates from small-scale calving, although it accounts for a small percentage of the total area. The calved area derived from MODIS at all four scales is underestimated compared with that from SAR, which might be a result of lower image quality for cloudy areas. 395 The area of individual calving events extracted by MODIS is generally smaller. As the calving scale increases, errors caused by different data sources account for a lower percentage of the total calved area (Figure 4 (a)(b)(c)). The errorequivalent perimeter widths generally exhibit a normal distribution with a standard deviation of 0.15 km and a mean value of 400 -0.06 km (Figure 4 (d)). Based on this, the errors introduced by multisource satellite data are acceptable.

Attribute uncertainties of independent calving events
We assessed the accuracy of the calved area, the calved-area thickness, and the calved mass attributes with Eq. (2), (4), (5), and (6). 420 The maximum area measurement uncertainty of a single calving event represented in this dataset was calculated as 30.7 km 2 with an annual average calved area uncertainty value of 14.3 km 2 and a standard deviation of 5.1 km 2 . The calved area uncertainty is mainly determined by the perimeter of each single calving event. In the case of the same area, a long and narrow calving area has higher uncertainty than a square calving area. Thickness uncertainty is mainly attributed to firn depth correction. For individual calving events, thickness uncertainty ranges from 1.0 m to 67.7 m with a mean value of 18.5 m and 425 a standard deviation of 9.1 m. The calved mass uncertainty is mainly determined by thickness uncertainty with a mean value of 29.5 Gt and a standard deviation of 23.6 Gt for 15 years, and its annual percentage fluctuates from 1.9% to 6.0% each year.

Number, calved area, and calved mass of independent calving events
We identify 1,975 annual calving events covering areas larger than 1 km 2 occurring in the circum-Antarctic ice shelves This fluctuating trend of calved mass is generally consistent with that of the calved area.

Calved area and calved mass of the less than 1 km 2 calving from the Antarctic ice shelves and the calving mass from 450 the marine-terminating glaciers
We assessed the annual calved area and calved mass of the less than 1 km 2 calving from the Antarctic ice shelves and the annual calving mass from the marine-terminating glaciers ( also take the calved mass of the marine-terminating glaciers into consideration by calculating the ice flux along grounding lines, which is about 166.7±15.2 Gt/yr. Therefore, the annual average calving rate of whole Antarctica is 955.4±51.4 Gt/yr.

Calving scale of independent calving events
The annual distributions of the number, total calved area, and total calved mass of calving events greater than 1 km 2 at different scales are shown in panels (a), (b), and (c) of Figure 5. Over the 15 years, the cumulative numbers of calving events 525 of small-, medium-, large-and extra-large-scale events accounted for 72.6%, 23.5%, 3.5%, and 0.3%, respectively, and frequencies increased exponentially as the scale decreased. The cumulative calved areas of the four different sizes accounted for 9.3%, 25.3%, 34.7%, and 30.6%, respectively. The distribution of calved mass is similar to that of the calved area.
The number of small-scale calving events accounts for a large percentage of total calving, especially in 2015/16-2019/20. The interannual variations in the number of small-scale calving events show obviously moderate variations. However, the area 530 and mass of small-scale calving remain relatively stable and low. As the calving scale increases, interannual variations in frequency become less significant; in contrast, interannual variations in area and mass become increasingly volatile. In some years, the number of calving events increased but calved area and mass remained stable or even decreased because more smallscale calving events made a limited contribution to the total calved mass and area. Thus, further studies must be conducted at different scales. 535

Calving recurrence interval of independent calving events 560
The recurrence interval of calving provides additional qualitative information about the style of calving (Liu et al., 2015) and determines the suitable observation period for identifying ice shelf nonsteady-state behavior. For example, the rift-opening calving of the Amery Ice Shelf has reoccurred in 2019 since the last calving in 1963/64 (Li et al., 2020), detach along the boundary of isolated pre-existing rifts for decades. The observational records spanning many decades would be needed to determine its nonsteady-state behavior. In contrast, more frequent disintegration calving events are mainly caused by the hard 565 to observe rapid basal crevasse propagation (Liu et al., 2015). The calving front retreat associated with these frequent calving events can be robustly identified over a short observation period due to the shorter recurrence intervals. In other words, the calving events with shorter recurrence intervals are more sensitive to current climate change.
Figure 6 (a) shows the calving recurrence interval is little related to calving scales of caving. The two extra-large-scale (> 1,000 km 2 ) calving events reoccurred on the Thwaite Glacier during our observed period indicating its distinct retreat, while 570 the other four extra-large-scale events from the Larsen C, Wilkins, Totten, and Amery Ice shelves did not reoccur. Figure 6 (b) shows that 76% of the total number of calving events reoccurred during the observed period (i.e., their recurrence intervals of calving are less than 8 yr), which suggests that the annual calving number is likely to be an indicator of the response of calving 18 to climate change. Nearly half of the cumulative calved area from the events with the recurrence intervals greater than 8 yr (i.e., the events only occurred once during the observed period) suggests that the annual calved area is not suitable for identifying the nonsteady-state behaviors of some ice shelves. 580 In contrast, small-and medium-sized ice shelves, widely found along the circum-Antarctic coastline, exhibit a higher calving rate (Gt/yr). Among them, the Thwaites Ice Shelf, Pine Island Ice Shelf, and Getz Ice Shelf in West Antarctica show 605 calving rates of 108 Gt/yr, 91 Gt/yr, and 52 Gt/yr, respectively. These are followed by the Mertz Ice Shelf and Totten Ice Shelf in East Antarctica with calving rates of 52 Gt/yr and 35 Gt/yr, respectively. Notably, we detected calving events at Totten Ice Shelf every year during the observation period, unlike the average calving mass of the Mertz Ice Shelf is mainly contributed by an extra-large disintegration event covering more than 2,500 km 2 that occurred in February 2010.

Discussion and conclusion
The annual iceberg calving dataset of the Antarctic ice shelves (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020) is the first that provides consistent and precise calving observations with the longest time span of 15 years. It not only directly reflects the quantitative characteristics and 620 spatial distribution of Antarctic iceberg calving, but it also provides multi-dimensional variables of each independent calving event. This dataset can be used as fundamental data for subsequent studies on ice-sheet mass balance, calving mechanisms, and their responses to climate change.
The interpretation of calving records spanning 12 orders of magnitude from 1 to 10 12 m 3 has demonstrated that the probability of calving events obeys a particular pattern whether they are small or large events-much like the Gutenberg-625 Richter law for earthquakes (Åström et al., 2014). Thus, the fine-scale and continuous observation of calving can be used to investigate how close particular glaciers are to their critical point, and thus how sensitive they may be to near-future changes in climatic and geometric conditions. However, finer-scale direct observation is greatly limited by the accessibility of highresolution remotely sensed imagery and significant manual overhead. Our observations provide records of calving volumes ranging from 10 8 to 10 12 m 3 of Antarctic ice shelves. 630 The calved-area uncertainty of our direct observation (Qi et al., 2020) is dependent on the spatial resolution of the imagery, uncertainty of velocity data, and the perimeter-to-area ratio of the calved area. In the case of the same area, a long and narrow calving area has higher uncertainty than a square calving area. The relatively low-spatial-resolution satellite imagery used in this work and the characteristic of a long and narrow calving area are the main reasons that this method is not suitable for highaccuracy calving observation of marine-terminating glaciers. The trade-off between workload and uncertainty reduction is 635 another consideration in choosing the minimum spatial scale of calving observation. With the calving scale decreasing from 100 km 2 , the number of annual calving events increases exponentially, which means that the monitoring workload also increases exponentially (Qi et al., 2020). Although direct calving observation has the minimum valid extraction area of 0.05 km 2 based on 75-m SAR resolution images (Qi et al., 2020), it is uneconomical to observe calving events of less than 1 km 2 using exponentially increasing manual workload to reduce slightly the uncertainty of the total calving-rate estimation. This is 640 why in the present work the calving area and mass of calving events of less than 1 km 2 of the Antarctic ice shelves were estimated based on observation-area ratio and direct observation of 1-10 km 2 calving events.
The total circum-Antarctic iceberg calving rate of 955.4±51.4 Gt/yr between 2005 and 2020 observed and estimated in the present study is less than the steady-state iceberg calving fluxes of 1,265 Gt/yr estimated by Rignot et al. (2013) and 1,321 Gt/yr estimated by Depoorter et al.(2013), respectively. The steady-state calving flux is the calving flux necessary to maintain 645 an assumed steady-state calving front for a given set of ice thicknesses and velocities along the ice-front gate (Rignot et al., 2013;Depoorter et al., 2013). Such "flux-gate" calving calculations for the marine-terminating glaciers are suitable. Our estimated calving rate of the marine-terminating glaciers, 166.7±15.2 Gt/yr, is very close to that reported by Rignot et al.(2013), i.e., 176 Gt/yr. However, such "flux-gate" calving calculations for ice shelves are inevitably biased as they underestimate iceberg calving for retreating ice shelves or overestimate it for advancing ice shelves. Our observed average calving rate of 650 770.3±29.5 Gt/yr from calving events larger than 1 km 2 between 2005 and 2020 is slightly greater than the average rate of 755 22 Gt/yr between 2005 and 2011 (Liu et al., 2015), which is contributed by two distinct high calving rates of 1,398.8 Gt/yr in 2015/16 and 1,832.6 Gt/yr in 2016/17, respectively. The average calving rate of 788.7±36.2 Gt/yr of all of the Antarctic ice shelves between 2005 and 2020 is the sum of 770.3±29.5 Gt/yr and the estimated average calving rate of 18.4±6.7 Gt/yr from calving events less than 1 km 2 , which is less than the steady state calving fluxes of 1,089 Gt/yr estimated by Rignot et al.(2013) 655 and 1,026 Gt/yr estimated by Liu et al.(2015), respectively. Thus, the Antarctic ice shelves are growing in extent.
Observations show that enhanced iceberg calvings have primarily been attributed to varying atmospheric and oceanic conditions (Shepherd et al., 2003;van den Broeke, 2005;Scambos et al., 2009;Braun and Humbert, 2009;Liu et al., 2015;Massom et al., 2018). Previous studies have revealed that ocean-driven thinning enhances iceberg calving and the retreat of Antarctic ice shelves based on the first record of all icebergs larger than 1 km 2 calving from all of the Antarctic ice shelves 660 between 2005 and 2011 (Liu et al., 2015). Here, the time series of this dataset has been extended from 6 to 15 years. The calving probability of Antarctic ice shelves indicated by the number of calving events has obvious inter-annual variation during our observation period ( Figure 6). Because 76% of the calving events have recurrence intervals of less than 8 yr, the annual variation of the number of calving events thus probably reflects the calving response to current climate variability. This provides an opportunity to examine the potential associations between iceberg calving and remote and local climate forcings. 665 Here, we show two examples from our preliminary analysis. First, Figure 9 (c) and (a) shows the relationship between the number of calving events and the oceanic Niño index. Remotely, El Niño leads to anomalous increases in sea surface temperature and Antarctic ice sheet temperature. We found that a strong El Niño might lead to an increase in the number of calving events of Antarctic ice shelves, and there has been intensified iceberg calving since then. Second, Figure 9 (c) and (b) shows the correlation between iceberg calving and ice sheet surface melting. Locally, atmospheric warming intensified ice 670 sheet surface melting, resulting in increased meltwater, which may trigger the expansion of rifts and crevasses and finally enhance iceberg calving. Based on this dataset, we found significant positive correlations between the maximum daily surface melting area and the number of calving events (r=0.76, p=0.003).

Conclusion and Data availability
The developed iceberg calving product applies a 15-yr calving distribution with year, length, area, scale, thickness, volume, mass, recurrence interval, and measurement uncertainty attributes for each calving event. The product applies an annual temporal resolution, and its spatial resolution is set to 1 km² . The dataset is stored in Shapefile format, shared via the National Tibetan Plateau Data Center (http://data.tpdc.ac.cn/en/), and entitled "Annual iceberg calving dataset of the Antarctic 685 ice shelves (2005-2020)" with DOI: 10.11888/Glacio.tpdc.271250 (Qi et al., 2021).