Monthly Frontal Ablation at 147 Tidewater Glaciers in Svalbard (2015-2024)

Frontal ablation measurements have long been limited to annual and decadal temporal resolution, leaving a knowledge gap in the intra-annual and seasonal variability. To address the temporal evolution, improve the process understanding of frontal ablation, and provide much needed reference data for the modeling community, we present a frontal ablation dataset for 147 tidewater glaciers in Svalbard at a monthly temporal resolution. As proper frontal ablation computation requires frontal information (i.e. calving fronts), which is inherently time- and labor-intensive to digitize manually, we employ a multi-temporal deep learning segmentation model (Dreier et al., 2025, 2026) to automate the mapping of glacier termini using a Sentinel-1 SAR image time series from January 2015 to December 2024. The result is a dataset with 15,500 frontal ablation estimates (88% temporal coverage), derived from 15,647 monthly-averaged calving front segmentations. For months missing frontal information in a glacier’s time series, frontal area changes over the data gaps are linearly interpolated to produce an additional 1,840 estimates, giving 99% total temporal coverage. In addition, a monthly time series of ice velocity and discharge is provided, developed from ITS_LIVE velocity datacubes (Gardner et al., 2025) and existing, regional ice thickness products (Fürst et al., 2018; Hugonnet et al., 2021; Malz et al., 2021). Climatic mass balance (CMB) is estimated to correct for ice mass loss due to atmospheric processes between the fluxgate (where ice velocity and thickness are sampled) and the glacier terminus; this area is referred to as the glacier domain. We aggregate daily, modeled CMB outputs from the MAR regional climate model (Fettweis and Grailet, 2024) to a monthly resolution over the glacier domain. Two NetCDF files are published here; (1) the monthly frontal ablation time series for all 147 glaciers and (2) the monthly ice velocity and discharge time series for all glaciers. Data variables are named and descriptions/units are given (when applicable) in both files for additional context. A GeoPackage containing relevant vector geometries for training the segmentation model at Svalbard and producing the frontal ablation time series is also published. The layers include regional ice, ocean, and land zone labels encompassing Svalbard and its tidewater glaciers. The ice zone label was originally derived from the Randolph Glacier Inventory version 6 (RGI Consortium, 2017), but the ice extent is updated at the terminus of each studied glacier using manually digitized terminus polygons (also provided as a layer) from Kochtitzky et al. (2022). The 15,647 monthly-averaged fronts and temporally matching monthly glacier domain polygons are provided as well. Other layers include the fluxgates, fluxgate sampling points, ice flow orientation points (points forward of the fluxgate and behind the glacier terminus), and RGI basin sampling boxes used for the preprocessing of SAR imagery and postprocessing of the segmentation model’s predictions. The main RGI boxes are particularly important as they provide glacier metadata (name, RGI-ID, etc.) and a summary of decadal-averaged results per glacier, as well as a glacier surge inventory and velocity seasonality information for glaciers with strong seasonal flow regimes defined by the classifications of Moon et al. (2014). All 11 layers are assembled into a preset QGIS project for easy visualization in the 'svalbard_frontal_ablation.qgz' file. Below is a table defining the data variables in the ‘sval_rgi_boxes’ layer in the GeoPackage: Lansat_ID ID of Landsat-8 image used to map the 2019 terminus by Kochtitzky et al. (2022) RGIId Randolph Glacier Inventory version 6 identifier GLIMSId Global Land Ice Measurements from Space identifier Name Tidewater glacier name RGIId_v7 Randolph Glacier Inventory version 7 identifier RGIv7_surge_type Denotes the surge type tag included in RGI v7; 0 = No evidence, 1 = Possible, 2 = Probable, 3 = Observed, 9 = N/A Region Abbreviated tag for Svalbard’s distinct subregions (SS – South Spitsbergen, NW – Northwest Spitsbergen, NE – Northeast Spitsbergen, AF – Austfonna Ice Cap, VF – Vestfonna Ice Cap, BE - Barentsøya and Edgeøya, KV – Kvitøya) terminus_length_km Decadal terminus length of all available terminus geometries (averaged per year, then years averaged over decade) frontal_ablation Decadal frontal ablation rate in Gt a-1 frontal_ablation_unc Decadal frontal ablation rate uncertainty in Gt a-1 FA_rank Indicates the glacier’s rank of frontal ablation magnitude relative to all 147 glaciers, with 1 being the strongest and 147 the weakest peak_year Distinguishes the year with the strongest frontal ablation rate in the glacier’s time series total_percent Total contribution (%) to Svalbard’s total decadal frontal ablation rate ice_discharge Decadal ice discharge rate in Gt a-1 ice_discharge_unc Decadal ice discharge rate uncertainty in Gt a-1 tmc Decadal terminus mass change rate in Gt a-1 (TMC is defined as the decadal frontal area change – CMB) tmc_unc Decadal terminus mass change rate uncertainty in Gt a-1 surge Denotes surge classification type, if applicable (Strong or Mini) surge_period Denotes the interval of the active surge phase; * indicates uncertainty in the surge period due to an active phase onset or persistence before/beyond the time series range vel_seasonality Indicates the seasonal ice flow classification type (if applicable) defined by Moon et al. (2014); includes Type 1, Type 2, and Type 3长期以来，冰前端消融观测仅能达到年际及年代际时间分辨率，导致年内与季节尺度变异性的研究存在认知空白。为厘清其时间演化规律、加深对冰前端消融过程的理解，并为建模领域提供亟需的参考数据，本研究构建了斯瓦尔巴群岛147条潮汐冰川的月分辨率冰前端消融数据集。 由于准确计算冰前端消融需要获取冰前端（即崩解冰舌）信息，而手动数字化此类工作本质上耗时耗力，本研究采用多时相深度学习分割模型（Dreier等，2025、2026），基于2015年1月至2024年12月的哨兵一号合成孔径雷达（Sentinel-1 SAR）影像时间序列，自动提取冰川末端位置。本次研究共得到15500组冰前端消融估算结果（时间覆盖率达88%），其数据来源于15647幅月均崩解冰舌分割结果。针对冰川时间序列中缺失冰前端信息的月份，本研究通过对数据间隙内的冰前端面积变化进行线性插值，额外得到1840组估算结果，使整体时间覆盖率提升至99%。 此外，本数据集还提供了基于ITS_LIVE速度数据立方体（Gardner等，2025）以及现有区域冰厚产品（Fürst等，2018；Hugonnet等，2021；Malz等，2021）生成的月分辨率冰流速与冰通量时间序列。 本研究通过估算气候物质平衡（CMB），校正了通量门（即采集冰流速与冰厚的区域）至冰川末端之间区域（即冰川域）因大气过程导致的冰量损失。我们将MAR区域气候模型（Fettweis与Grailet，2024）生成的逐日CMB模拟结果，聚合为冰川域范围内的月分辨率数据。 本数据集包含两份NetCDF格式文件：（1）147条冰川的月分辨率冰前端消融时间序列文件；（2）所有冰川的月分辨率冰流速与冰通量时间序列文件。两份文件均对数据变量进行了命名，并在适用情况下提供了变量说明与单位，以补充数据背景。同时公开的还有一份地理数据包（GeoPackage），其中包含用于斯瓦尔巴地区分割模型训练与冰前端消融时间序列生成的相关矢量几何数据。该数据包包含覆盖斯瓦尔巴群岛及其潮汐冰川的区域冰、海洋与陆地区域标签图层。 冰区标签最初来源于第六版伦道夫冰川目录（Randolph Glacier Inventory，RGI联盟，2017），但本研究针对每条研究冰川的末端，使用Kochtitzky等（2022）手动数字化的末端多边形（同样作为一个图层提供）更新了冰体范围。数据集同时提供了15647幅月均冰前端数据，以及与时间匹配的月分辨率冰川域多边形数据。 其余图层包括通量门、通量门采样点、冰流方向点（位于通量门前方、冰川末端后方的点位），以及用于合成孔径雷达影像预处理与分割模型预测后处理的RGI流域采样框。主RGI采样框尤为重要，其包含冰川元数据（名称、RGI-ID等）、每条冰川的年代际平均结果汇总，以及基于Moon等（2014）分类标准定义的季节性强流速冰川的冰川跃动清单与流速季节特征信息。 全部11个图层被整合至预设的QGIS项目文件中，方便用户在"svalbard_frontal_ablation.qgz"文件中进行可视化操作。下表对地理数据包中"sval_rgi_boxes"图层的数据变量进行说明： Landsat_ID：Kochtitzky等（2022）用于绘制2019年冰川末端的Landsat-8影像ID RGIId：第六版伦道夫冰川目录标识符 GLIMSId：全球陆地冰空间测量（GLIMS）标识符 Name：潮汐冰川名称 RGIId_v7：第七版伦道夫冰川目录标识符 RGIv7_surge_type：标注RGI v7中的跃动类型标签：0=无证据，1=可能，2=很可能，3=已观测，9=无可用信息 Region：斯瓦尔巴群岛不同子区域的缩写标签（SS：南斯匹次卑尔根，NW：西北斯匹次卑尔根，NE：东北斯匹次卑尔根，AF：奥斯特福纳冰盖，VF：韦斯特福纳冰盖，BE：巴伦支岛与埃奇岛，KV：基特岛） terminus_length_km：所有可用末端几何数据的年代际冰川末端长度（按年平均后，再进行年代际平均） frontal_ablation：年代际冰前端消融速率，单位为Gt a⁻¹ frontal_ablation_unc：年代际冰前端消融速率不确定性，单位为Gt a⁻¹ FA_rank：相较于全部147条冰川，该冰川的冰前端消融量级排名，1为消融最强，147为消融最弱 peak_year：标注该冰川时间序列中冰前端消融速率最高的年份 total_percent：对斯瓦尔巴群岛年代际总冰前端消融速率的总贡献占比（%） ice_discharge：年代际冰通量速率，单位为Gt a⁻¹ ice_discharge_unc：年代际冰通量速率不确定性，单位为Gt a⁻¹ tmc：年代际冰前端质量变化速率，单位为Gt a⁻¹（TMC定义为年代际冰前端面积变化减去气候物质平衡） tmc_unc：年代际冰前端质量变化速率不确定性，单位为Gt a⁻¹ surge：标注冰川跃动分类类型（如适用），包括Strong（强跃动）与Mini（弱跃动） surge_period：标注冰川跃动活跃阶段的区间；*表示因跃动活跃阶段的起始或持续时间超出本数据集时间序列范围，导致跃动阶段存在不确定性 vel_seasonality：基于Moon等（2014）的分类标准，标注季节性冰流类型（如适用），包括Type 1、Type 2与Type 3