Sea ice thermodynamics: the Stefan condition
Sea ice grows at the ice-ocean interface when the ocean heat flux is less than the conductive heat flux through the ice. The Stefan condition governs ice thickness $h$:
\[\rho_i L \frac{dh}{dt} = F_{cond} - F_{ocean}\]where $\rho_i = 917$ kg m⁻³ is ice density, $L = 3.34\times10^5$ J kg⁻¹ is latent heat of fusion. The conductive flux through ice of thermal conductivity $k_i = 2.0$ W m⁻¹ K⁻¹:
\[F_{cond} = k_i\frac{T_f - T_s}{h}\]where $T_f = -1.8°C$ is the freezing point and $T_s$ is the ice surface temperature (determined by energy balance at the top surface). Integrating the growth equation from rest ($h(0) = 0$) gives the Stefan law for ice growth in a simple freeze-down:
\[h(t) = \sqrt{\frac{2k_i}{\rho_i L}|T_f - T_{air}|t} \propto \sqrt{t}\]This square-root growth law reflects the insulating effect of thicker ice — thicker ice has lower conductive flux, slowing further growth. Ocean heat flux $F_{ocean}$ from turbulent mixing opposes freezing; in the Arctic Ocean $F_{ocean} \approx 2$–5 W m⁻².
Surface energy balance at the top of ice
At the ice surface, the energy balance determines $T_s$:
\[F_{sw}(1-\alpha) + F_{lw} - \sigma T_s^4 - F_{turb} - F_{cond} = 0\]Surface albedo is critical:
- Fresh snow: $\alpha \approx 0.85$
- Melting snow: $\alpha \approx 0.70$–0.75
- Bare multiyear ice: $\alpha \approx 0.50$–0.60
- Melt ponds: $\alpha \approx 0.20$–0.40
- Open ocean: $\alpha \approx 0.06$
Melt ponds form on first-year ice in spring and dramatically lower the area-averaged albedo, accelerating melt — a powerful positive feedback on the 1–10 km scale.
Ice-albedo feedback and the tipping point question
The ice-albedo feedback is the dominant driver of Arctic amplification. As sea ice retreats, dark ocean absorbs more shortwave radiation, warming the ocean, which melts more ice. The feedback parameter:
\[\lambda_{ia} = \frac{\partial F_{net}}{\partial T} = \frac{S_0}{4}\frac{\partial \bar\alpha}{\partial A_{ice}}\frac{\partial A_{ice}}{\partial T} > 0\]The debate over a sea ice tipping point centers on whether the summer ice minimum can disappear abruptly or gradually. GCM evidence suggests summer sea ice loss is reversible and approximately linear with global temperature — no tipping point — but the September ice minimum is extremely sensitive: ~3 m² per tonne of CO₂ emitted.
Sea ice dynamics: viscous-plastic rheology
Sea ice is not passive — it deforms under wind and ocean currents. The momentum equation for sea ice:
\[m\frac{D\mathbf{u}}{Dt} = \boldsymbol{\tau}_{air} + \boldsymbol{\tau}_{ocean} - mf\hat{k}\times\mathbf{u} - mg\nabla\eta + \nabla\cdot\boldsymbol{\sigma}\]where $\boldsymbol{\sigma}$ is the internal ice stress tensor, $\boldsymbol{\tau}{air}$ and $\boldsymbol{\tau}{ocean}$ are surface drags, $f$ is the Coriolis parameter, and $\eta$ is sea surface height. The viscous-plastic (VP) rheology (Hibler 1979) relates stress to strain rate via a yield curve. For strain rates $\dot{\boldsymbol{\varepsilon}}$ below the yield stress, ice behaves as a viscous fluid; at the yield surface it flows plastically:
\[\boldsymbol{\sigma} = 2\eta\dot{\boldsymbol{\varepsilon}} + (\zeta - \eta)\dot{\varepsilon}_{kk}\mathbf{I} - \frac{P}{2}\mathbf{I}\]with $\eta$ (shear viscosity) and $\zeta$ (bulk viscosity) determined by the ice pressure $P$ and strain rate magnitude. The elliptical yield curve defines the transition from elastic to plastic flow. The elastic-viscous-plastic (EVP) scheme (Hunke & Dukowicz 1997) replaces the computationally expensive implicit VP solver with subcycling over a pseudo-elastic relaxation, enabling efficient parallelization.
Arctic sea ice trends and multi-year vs first-year ice
Arctic September sea ice extent has declined at ~13% per decade since 1979 (NSIDC satellite records). Ice volume has declined even faster due to thinning:
| Period | September extent (10⁶ km²) | Mean thickness (m) |
|---|---|---|
| 1980s | ~7.0 | ~3.5 |
| 2000s | ~5.5 | ~2.0 |
| 2012 record low | 3.34 | ~1.5 |
| 2020s average | ~4.5 | ~1.3 |
The shift from multi-year ice (MYI, survived at least one melt season, thick, low salinity, high albedo) to first-year ice (FYI, thinner, saltier, lower albedo) represents a structural change. MYI fraction declined from ~75% in the 1980s to ~30% by 2020. First-year ice is more prone to complete summer melt, further accelerating the transition.
Ice-free Arctic projections
All CMIP6 models project an ice-free Arctic summer (September extent $< 10^6$ km²) before 2100 under SSP2-4.5 and higher. The threshold for ice-free conditions scales with cumulative CO₂ emissions:
\[A_{Sept} \approx A_0 - \beta \cdot \Delta T_{global}\]with $\beta \approx 3\times10^6$ km² K⁻¹. Ice-free conditions likely occur when global warming exceeds ~2°C above pre-industrial. Under SSP1-1.9, the ice-free period may be brief and reversed. The Bering Sea shows strong interannual variability, driven by atmospheric blocking (dipole anomaly pattern) that can shift September extent by ~1–2 million km² year-to-year. The Beaufort Gyre stores freshwater that, when released, could influence AMOC stability on decadal timescales — an ocean-ice-thermohaline coupling that is poorly constrained in models.