Solving a mystery of Greenland’s melting ice sheet—what happens when models miss refreezing?

In the summer of 2015, just before officially starting my PhD, I found myself in southwest Greenland, standing on the surface of the Greenland Ice Sheet beside my PhD advisor and a small team of researchers. Our mission: to measure how much meltwater was flowing over the ice sheet surface and, eventually, into surrounding oceans. At that time, I expected the ice to resemble a solid, smooth hockey rink, rigid and impermeable. What I actually saw completely changed my perspective, and became central to my PhD work.

Instead of solid ice, the bare, snow‑free surface was surprisingly porous and sponge‑like, full of tiny pockets and cracks that could clearly hold meltwater. This “weathering crust” was low‑density, saturated with water, and dotted with cryoconite holes—small depressions filled with dark sediment and microbes that further enhance melting by absorbing more sunlight. It was far from the solid slab of ice traditionally represented in climate models.

After returning home and analyzing our field measurements of meltwater runoff, it quickly became apparent that something wasn’t matching up. The observed meltwater runoff—the water actually flowing over the ice sheet toward surrounding oceans—was consistently lower than what climate models predicted. We immediately suspected that the models were oversimplifying the ice surface, treating it as impermeable, meaning they instantly sent all meltwater straight into the ocean. But clearly, from what I’d seen, some meltwater was being retained, perhaps even refreezing within the ice itself.

This observation became the core hypothesis for my PhD: Could the models’ assumptions about Greenland’s bare ice be missing a critical piece of the puzzle?

Digging deeper into Greenland’s ice

In summer 2016, we returned to Greenland, this time with more instrumentation and more determination. Our goal was ambitious: to record meltwater runoff continuously every hour, day and night, for a full week. Through the bright Arctic summer days and freezing twilight, we measured how much water flowed over the ice, capturing a first‑of‑its‑kind snapshot of how the ice sheet behaves around the clock.

Back home, the data showed something intriguing yet confusing. One of the climate models dramatically overestimated our observed meltwater runoff, just as we’d suspected, while two others seemed to align closely with our measurements. Initially, this seemed to disprove my hypothesis. But when we dove deeper into the details, a surprising pattern emerged.

The hidden puzzle piece: albedo and energy

Albedo, or the reflectivity of the ice surface, turned out to be a crucial factor. Ice with a higher albedo (lighter, brighter surfaces) reflects more sunlight, absorbing less energy, and thus melting less. Conversely, darker ice absorbs more sunlight, leading to more melting. We noticed the models varied significantly in their representation of albedo, and therefore in how much meltwater they generated in the first place.

Surprisingly, the model that overestimated runoff was actually capturing albedo correctly, meaning it correctly predicted how much melt energy the ice surface was absorbing. In contrast, the two models that appeared accurate in their runoff estimates had overestimated the ice’s reflectivity (albedo), underestimating absorbed sunlight, and thus, generating too little meltwater from the start. They appeared accurate in their meltwater runoff predictions but were effectively getting the “right answer” for the wrong reasons.

This realization was a critical turning point: it showed that no existing model was accurately representing both the fundamental energy balance (driven predominately by ice reflectivity and hence sunlight absorption during summer) and the actual amount of meltwater runoff. Clearly, a missing process—potentially the nighttime refreezing of meltwater—needed to be accounted for.

Building a better model

Determined to unravel this puzzle, I painstakingly developed a custom numerical model designed specifically to capture the complexities we had observed. Rather than treating the ice like a parking lot that instantly drains water away, our model represented it as porous—slowly holding, draining, and, if the weather dictated, refreezing water each night when temperatures dipped below freezing.

Running the model on a sub‑hourly timestep proved essential. Even in the height of Greenland’s summer, nighttime temperatures frequently fall below freezing. When they do, some of the meltwater stored in the porous ice refreezes. Although these individual nightly refreezing events are small, they quickly add up over an entire melt season.

The results were illuminating. By including meltwater storage and nighttime refreezing, the model could accurately simulate both the fundamental energy‑driven melt production and the actual observed runoff. It finally provided the comprehensive explanation we’d been seeking, reconciling the previously confusing results from different climate models.

From Greenland to global implications

Expanding our analysis regionally, we estimated that this refreezing could reduce Greenland’s runoff by up to 15%—a significant amount given the enormous volumes of meltwater involved. Incorporating this refreezing mechanism into climate models isn’t just about improving scientific accuracy. It has real-world implications for forecasting how quickly Greenland’s meltwater runoff could raise global sea levels.

This discovery doesn’t mean current climate models are incorrect about current or future sea‑level rise estimates. For example, we know from multiple lines of evidence, including direct observations from satellites, that Greenland is losing more mass each year than it gains, directly causing sea levels to rise. Climate model predictions are consistent with these observations. Instead, it reveals an opportunity where models can improve their estimates of the specific components of ice sheet mass changes, offering a path toward both better forecasts and attribution of observed trends. By recognizing and integrating this overlooked process, we hope to help refine predictions of sea‑level rise, which will help to inform communities worldwide about how quickly and significantly their coasts might be affected.

Reflections and looking ahead

This experience taught me about the power of combining meticulous fieldwork with detailed modeling. It showed how initial assumptions about something seemingly straightforward—like just how “solid” glacier ice really is—can overlook subtle yet significant complexities. By challenging assumptions and exploring beneath the surface (quite literally, in this case), we unveiled an important new piece of the Greenland Ice Sheet puzzle.

As Greenland continues to warm, accurately modeling meltwater runoff, and factoring in complex processes like refreezing, will become increasingly important. Our study shows the necessity of continuously improving models based on real‑world observations, ensuring that future predictions remain accurate and reliable. The Greenland Ice Sheet, far from being a solid slab of frozen water, is a dynamic, complex environment. By embracing these complexities, we can better prepare for the challenges ahead.