Frontal ablation processes at marine‐terminating glaciers are challenging to observe and difficult to represent in numerical ice flow models, yet play critical roles in modulating ice sheet mass balance. Current ice sheet models typically rely on simple iceberg calving models to prescribe either terminus positions or iceberg calving rates, but the relative accuracies and uncertainties of these calving models remain largely unconstrained at the ice sheet scale. Here, we evaluate six published iceberg calving models against spatially and temporally diverse observations from 50 marine‐terminating outlet glaciers in Greenland. We seek the single model that best reproduces observed conditions across all glaciers, at all observation times, and with low sensitivity to calibration uncertainty. Five of six calving models can produce unbiased estimates of calving position or calving rate at the ice sheet scale. However, time series analysis reveals that, when using a single, optimized model parameter, rate‐predicting calving models frequently yield calving rate errors in excess of 10 m d⁻¹. In comparison, terminus position‐predicting calving models more accurately track observed changes in terminus position (remaining within ~1 km of the observed grounded terminus position). Overall, our results indicate that the crevasse depth calving model provides the best balance of high accuracy and low sensitivity to imperfect parameter calibration. While the crevasse depth model appears unlikely to capture the true controls on crevasse penetration, numerically, it reproduces observed terminus dynamics with high fidelity and should be considered a leading candidate for use in models of the Greenland Ice Sheet.

中文翻译：

---

根据格陵兰岛出口冰川的观测评估冰山崩塌模型

海洋终止冰川的额叶消融过程很难观察，并且难以在数值冰流模型中表示，但在调节冰盖质量平衡中起着关键作用。当前的冰盖模型通常依靠简单的冰山产犊模型来规定终点位置或冰山产犊率，但是这些产犊模型的相对精度和不确定性在冰盖规模上仍然不受很大限制。在这里，我们根据格陵兰岛的50个海洋终止出口冰川的时空多样性评估了六个已发布的冰山崩塌模型。我们寻求一个单一模型，该模型在所有观测时间都能最佳地再现所有冰川的观测条件，并且对校准不确定性的敏感性较低。六个产犊模型中的五个可以在冰盖尺度上产生产犊位置或产犊率的无偏估计。但是，时间序列分析显示，当使用单个优化的模型参数时，速率预测产犊模型经常会产生产犊率误差超过10 m d^-1。相比之下，终点位置预测产犊模型可以更准确地跟踪终点位置的观测变化（距离观测到的接地终点位置约1 km之内）。总体而言，我们的结果表明，缝隙深度划分模型可以对不完善的参数校准提供高精度和低灵敏度的最佳平衡。虽然裂隙深度模型似乎不太可能捕捉到对裂隙渗透的真正控制，但从数值上讲，它可以高保真度再现观察到的终点动态，因此应被认为是格陵兰冰盖模型使用的主要候选对象。