A deep learning reconstruction of mass balance series for all glaciers in the French Alps: 1967-2015

Glacier mass balance (MB) data are crucial to understand and quantify the regional effects of climate on glaciers and the high-mountain water cycle, yet observations cover only a small fraction of glaciers in the world. We present a dataset of annual glacier-wide surface mass balance of all the glaciers in the French Alps for the 1967-2015 period. This dataset has been reconstructed using deep learning (i.e. a deep artificial neural network), based on direct MB observations and remote sensing annual estimates, meteorological reanalyses and topographical data from glacier inventories. The method’s validity was assessed through an extensive cross-validation against a dataset of 32 glaciers , with an estimated average error (RMSE) of 0.55 m.w.e. a-1, an explained variance (r2) of 75% and an average bias of -0.021 m.w.e. a-1. We estimate an average regional area-weighted glacier-wide MB of -0.71±0.21 (1 sigma) m.w.e. a-1 for the 1967-2015 period, with negative mass balances in the 1970s (-0.44 m.w.e. a-1), moderately negative in the 1980s (-0.16 m.w.e. a-1), and an increasing negative trend from the 1990s onwards, up to -1.34 m.w.e. a-1 in the 2010s. A comparison with ASTER-derived geodetic MB for the 2000-2015 period showed important differences with the photogrammetric geodetic MB used to train our model. When recalibrating our reconstructions with the new ASTER-derived geodetic MB, the estimated average regional area-weighted glacier-wide MB (1967-2015) is reduced to -0.64±0.21 (1 sigma) m.w.e. a-1. Following a topographical and regional analysis, we estimate that the massifs with the highest mass losses for the 1967-2015 period are the Chablais (-0.93 m.w.e. a-1), Champsaur and Haute-Maurienne (-0.86 m.w.e. a-1 both) and Ubaye ranges (-0.83 m.w.e. a-1), and the ones presenting the lowest mass losses are the Mont-Blanc (-0.69 m.w.e. a-1), Oisans and Haute-Tarentaise ranges (-0.75 m.w.e. a-1 both). This dataset provides relevant and timely data for studies in the fields of glaciology, hydrology and ecology in the French Alps, in need of regional or glacier-specific annual net glacier mass changes in glacierized catchments. The MB dataset is presented in two different formats: (a) A single netCDF file containing the MB reconstructions, the glacier RGI and GLIMS IDs and the glacier names. This file contains all the necessary information to correctly interact with the data, including some metadata with the authorship and data units. (b) A dataset comprised of multiple CSV files, one for each of the 661 glaciers from the 2003 glacier inventory (Gardent et al., 2014), named with its GLIMS ID and RGI ID with the following format: GLIMS-ID_RGI-ID_SMB.csv. Both indexes are used since some glaciers that split into multiple sub-glaciers do not have an RGI ID. Split glaciers have the GLIMS ID of their "parent" glacier and an RGI ID equal to 0. Every file contains one column for the year number between 1967 and 2015 and another column for the annual glacier-wide MB time series. Glaciers with remote sensing-derived estimates (Rabatel et al., 2016) include this information as an additional column. This allows the user to choose the source of data, with remote sensing data having lower uncertainties (0.35±0.06 () m.w.e. a-1 as estimated in Rabatel et al. (2016)). Columns are separated by semicolon (;).

冰川物质平衡（Glacier mass balance, MB）数据对于理解并量化气候对冰川及高山区水循环的区域影响至关重要，但现有观测仅覆盖全球极小部分冰川。我们构建了1967至2015年间法国阿尔卑斯山区所有冰川的年度全域表面物质平衡数据集。本数据集基于直接物质平衡观测、遥感年度估算数据、气象再分析数据以及冰川编目的地形数据，通过深度学习（deep learning，即深度人工神经网络（deep artificial neural network））重建得到。该方法的有效性通过针对32座冰川的大规模交叉验证得到评估，估算得到平均误差（均方根误差（Root Mean Square Error, RMSE））为0.55 米水当量·年⁻¹，解释方差（决定系数r²）为75%，平均偏差为-0.021 米水当量·年⁻¹。我们估算得到1967至2015年法国阿尔卑斯山区以面积加权的全域冰川平均表面物质平衡为-0.71±0.21（1σ）米水当量·年⁻¹，其中20世纪70年代物质平衡为负（-0.44 米水当量·年⁻¹），80年代呈中度负值（-0.16 米水当量·年⁻¹），90年代起负向趋势加剧，至2010年代达到-1.34 米水当量·年⁻¹。将本数据集与2000至2015年基于ASTER（Advanced Spaceborne Thermal Emission and Reflection Radiometer）的大地测量物质平衡数据进行对比后发现，其与本模型训练所用的摄影测量大地测量物质平衡结果存在显著差异。当采用新的ASTER大地测量物质平衡数据对重建结果进行重新校准后，1967至2015年区域面积加权平均冰川全域表面物质平衡的估算值降至-0.64±0.21（1σ）米水当量·年⁻¹。通过地形与区域分析，我们估算得到1967至2015年物质损失最严重的山块为沙布莱山（Chablais，-0.93 米水当量·年⁻¹）、尚博尔与上莫里耶讷山（Champsaur和Haute-Maurienne，均为-0.86 米水当量·年⁻¹）以及于拜山脉（Ubaye ranges，-0.83 米水当量·年⁻¹）；物质损失最小的山块为勃朗峰（Mont-Blanc，-0.69 米水当量·年⁻¹）、瓦桑与上塔朗泰斯山（Oisans和Haute-Tarentaise，均为-0.75 米水当量·年⁻¹）。本数据集可为法国阿尔卑斯山区冰川学、水文学与生态学领域的研究提供及时且相关的数据，此类研究亟需冰川覆盖流域的区域或单冰川年度净物质变化数据。本表面物质平衡数据集以两种格式提供：(a) 单个netCDF文件，包含物质平衡重建结果、冰川的RGI（Randolph Glacier Inventory）编号与GLIMS（Global Land Ice Measurements from Space）编号以及冰川名称，该文件包含与数据交互所需的全部信息，包括作者信息与数据单位等元数据；(b) 由多个CSV文件组成的数据集，对应2003年冰川编目（Gardent等，2014）中的661座冰川，每个文件以其GLIMS编号与RGI编号按以下格式命名：GLIMS-ID_RGI-ID_SMB.csv。由于部分分裂为多个子冰川的冰川并无RGI编号，因此同时使用两种编号，分裂冰川保留其“母冰川”的GLIMS编号，且RGI编号设为0。每个文件包含两列：一列对应1967至2015年的年份，另一列对应冰川全域年度表面物质平衡时间序列；对于具备遥感估算数据（Rabatel等，2016）的冰川，文件中会额外增加一列包含该类数据，这允许用户自主选择数据来源，其中遥感数据的不确定性更低（据Rabatel等(2016)估算为0.35±0.06 米水当量·年⁻¹）。列之间以分号（;）分隔。