资源环境科技发展态势分析平台

Estimating the Greenland ice sheet surface mass balance (SMB) is an important component of current and future projections of sea level rise. Given the lack of in situ information, imperfect models, and underutilized remote sensing data, it is critical to combine the available observations with a physically based model to better characterize the spatial and temporal variation of the Greenland ice sheet SMB. This work proposes a data assimilation framework that yields SMB estimates that benefit from a state-of-the-art snowpack model (Crocus) and a 16-day albedo product. Comparison of our results against in situ SMB measurements from the Kangerlussuaq transect shows that assimilation of 16-day albedo product reduces the root-mean-square error of the posterior estimates of SMB from 1,240 millimeter water equivalent per year (mmWE/yr) to 230 mmWE/yr and reduces the bias from 1,140 mmWE/yr to -20 mmWE/yr.

Plain Language Summary Diagnosing the surface mass balance (SMB) of the Greenland ice sheet (GrIS) is a critical objective, which, despite its importance, continues to contain large uncertainties from significant errors in modeled precipitation as well as errors related to subgrid process representation. This work uses a data assimilation framework (which has not been used in estimations of the GrIS SMB) and a satellite-derived 16-day albedo product to improve the estimates of the SMB on the southern GrIS. We used the Kangerlussuaq transect (K-transect) point-scale SMB measurements to validate our results over the 2009-2010 hydrological year. The data assimilation technique (i.e., particle batch smoother) reduces the spatial root-mean-square error of SMB over the K-transect stations by 82% from 1,240 millimeter water equivalent (mmWE) to 231 mmWE and bias of the estimates by 98% from 1,142 mmWE to -20 mmWE. It was shown that this methodology has the potential to resolve the spatial variability of the surface processes along the K-transect stations and surface albedo that is not resolved by the model at a resolution of 25 km. The modularity of the algorithm makes it possible to combine nearly any land surface model with different satellite-based or ground-based measurements.