This project aimed to improve understanding and prediction of surface melting on the Greenland ice sheet by investigating the past and future climate of this region with new resources.  By combining satellite remote-sensing of melting, a regional weather forecast model and a simple model for temperature-driven surface melting, we gained a better picture of future change under high and low warming scenarios.

Satellite remote-sensing provides 40-plus years of surface melting observations across the ice sheet and a look at variability (timing, spatial patterns) over decades.  As a reference, these data also support melt model skill testing.

We used our regional forecast model (Polar WRF) in both historical (late 20^th/early 21^st century) and future (2071-2080) timeframes to (1) test our melt prediction model against observations and (2) make future melt predictions.

Historical weather forecasts use global boundary data from reanalysis or global climate models (GCMs).  Future forecasts can only use GCMs.  To study two potential extremes, we used "high warming" (RCP 8.5) and "low warming" emissions scenarios.  The former includes emissions expected to increase global mean temperature by 5-6 degrees C over pre-industrial levels in 2100.  The latter limits the increase to 1.5 degrees C.  Because local changes differ from global averages, a regional model is needed to understand Greenland's futures.

Collectively, these forecasts represent a significant resource for the polar meteorology and climatology communities with 14 TB of model outputs archived at the Arctic Data Center.

We tested reanalysis, GCM and forecast model skill with daily average temperature observations from eleven Greenland Climate Network (GC-Net) automatic weather stations (AWS).  The benefits of the forecast model’s 15 km spatial resolution and improved polar-centric model physics were notable:  mean model biases at the AWS were reduced from 0.7-3.6 degrees C (reanalysis) and  -2.9 to 4.2 degrees C (GCM) to -1.4 to 1 degrees C in the forecast model.

We also compared future (2071-80) predictions to the historical observations to evaluate potential futures in GCM and forecast model data.  For "high warming", the forecast model predicts future July warming of 1.5-5.4 degrees C compared to the GCM's higher and narrower range of 4-6.1 degrees C.  Above-freezing temperatures are much more frequent at the lowest elevation sites while Summit inches into a regime with occasional melting more likely. Under "low warming", changes in the forecast model are more moderate (0.2-1.1 degrees C) but, with recent warming already at ~1 degrees C, some sites still exceed 1.5 degrees C.

Our melt-prediction model is based on a simple premise:  melting depends on temperature therefore temperature can be used to predict melt versus no-melt.  Specifically, the distributions of temperatures under melt and no-melt conditions can enable reliable melt occurrence prediction.

Temperature data are first split into melt and no-melt subsets (using the observations) each with an associated probability distribution (Figure 1).   Overlap produces three temperature ranges for melting predictions: low, high and transitional (where "melt" and "no melt" observations are seen for the same temperature). In the low (high) range, no observations of melting (no melting) are seen therefore melt probability is zero (one) and melt occurrence is "no melt" ("melt").

In the transitional range between the peaks of the "no melt" and "melt" temperature distributions, melt occurrence probability (0-1) is a linear interpolation of modeled temperature into this range.  Melt occurrence (zero or one) applies a threshold to the probability value. The threshold minimizes the difference between predicted and observed melt in the transitional range.

For example, if the modeled temperature interpolates to 0.7 and the threshold was calculated at 0.78, the probability of melt is recorded as 0.7 and, because 0.7 < 0.78, melt occurrence is set to zero ("no melt").

Comparisons of the two model datasets to observations (Figure 3) are qualitatively favorable with a similar spatial distribution of over- and underestimation of total melt days for the two datasets: too few melt days in the south and extreme northeast, too many in the northwest and the southeast coastline.

As expected under the significant global warming of RCP 8.5, surface melting is more extensive and frequent (Figure 3) with dramatic spatial changes in the north and notable increases of the melting boundary elevation around the entire ice sheet.  Changes under the +1.5 degrees C warming scenario are less dramatic but still nontrivial. Spatial extent and frequency still increase compared to present, possibly passing glaciologically significant thresholds.

We also supported, financially and academically, a successful Master’s degree in Hydrology by Ms. Margeaux Carter.  Carter’s research expanded this project into a new area: detecting subsurface liquid water on the ice sheet with satellite observations.  Specific targets were buried lakes and firn aquifers. In both cases, Ms. Carter successfully used satellite observations to reliably predict locations where subsurface liquid water should exist and developed a long-term record, complete in time and space, for these important phenomena.

Last Modified: 01/28/2021
Modified by: David B Reusch