Surface Flux Transport Models

• Surface Flux Transport models simulate the evolution of the Sun's photospheric magnetic field by combining advection, diffusion, and active region flux emergence.
• They quantify the impacts of differential rotation, meridional flow, and supergranular turbulence to predict polar field buildup and solar cycle characteristics.
• Model calibration involves adjusting key parameters such as diffusivity, meridional flow speed, and tilt angles, addressing challenges in linking large-scale dynamics with small-scale structures.

Surface Flux Transport (SFT) models are kinematic models for the evolution of the Sun’s photospheric radial magnetic field $B_r$. In their classical form, they treat the surface field as a passive, radial field that is advected by differential rotation and meridional flow, dispersed by supergranular turbulence represented as an effective diffusivity, and replenished by newly emerging active regions. Their principal scientific role is to explain and predict the redistribution of emerged magnetic flux over the solar surface and, in particular, the buildup of the polar field or axial dipole near cycle end, a quantity closely connected with open flux, coronal boundary conditions, and solar-cycle prediction (Yeates et al., 2023, Wang et al., 2 Jun 2025).

1. Classical formulation and physical assumptions

The classical SFT equation evolves $B_r(\theta,\phi,t)$ on the spherical surface through horizontal advection, turbulent diffusion, and a source term: $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$ or, in explicit spherical form,

$$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$

In this representation, differential rotation shears magnetic patterns longitudinally, meridional flow advects flux poleward, diffusion parameterizes the random walk induced by supergranulation, and $S$ injects newly emerged flux (Yeates et al., 2023).

For the longitude-averaged field $\langle B_r\rangle$, the differential-rotation term vanishes, leaving a one-dimensional advection–diffusion problem in latitude. This reduction is central to analyses of polar-field formation and to axisymmetric calibrations of transport parameters. The axial dipole strength is the $\ell=1,m=0$ spherical-harmonic coefficient,

$$b_{1,0}(t)=\frac{3}{4\pi}\int_0^{2\pi}\int_0^\pi B_r\cos\theta\sin\theta\,d\theta\,d\phi,$$

and it functions as the standard large-scale proxy for the global polar field (Yeates et al., 2023).

The classical model is phenomenological but has a mean-field MHD justification. Its standard simplifying assumptions are that the radial field dominates near the photosphere, horizontal transport can be treated kinematically, supergranulation can be represented by a constant diffusivity, large-scale flows are steady and axisymmetric, and active-region emergence is prescribed externally. A useful control parameter is the magnetic Reynolds number

$\mathrm{Rm}=\frac{R_\odot U_0}{\eta},$

which measures the competition between advection and diffusion (Yeates et al., 2023).

2. Source-term physics and dipole generation

In predictive SFT, the source term is usually the dominant ingredient. In the classical bipolar magnetic region (BMR) representation, each emergence is characterized by flux, latitude, longitude, polarity separation, and tilt angle. The initial axial dipole contribution of a simple BMR is approximately

$$b_{1,0}(t_{\rm em}) \approx -\frac{3\Phi_0}{4\pi R_\odot^2}\rho_0 \sin\gamma_0 \sin\theta_0,$$

so the initial dipole scales with total flux, polarity separation, tilt, and emergence latitude (Yeates et al., 2023).

This source sensitivity is reinforced by historical-cycle modeling. A model spanning cycles 15–21 incorporated cycle-dependent sunspot-group tilt angles using

$B_r(\theta,\phi,t)$0

with $B_r(\theta,\phi,t)$1 determined from Mount Wilson Observatory and Kodaikanal Observatory data. In that framework, the observed anti-correlation between cycle strength and mean tilt angle was essential for reproducing the empirically derived time evolution of the solar open magnetic flux and the timing of polar-field reversals. The same study found that the polar field maxima near cycle minima correlate with the amplitude of the following sunspot cycle with $B_r(\theta,\phi,t)$2, rather than with the preceding cycle (Cameron et al., 2010).

In that model, a further scaling $B_r(\theta,\phi,t)$3 with reference value $B_r(\theta,\phi,t)$4 was introduced to represent the effect of observed inflows into active regions, which reduce polarity separation and hence reduce the amount of flux transported poleward. This suggests that the effective source term in SFT is not set only by Joy’s law in isolation, but also by near-source transport effects that can modify the latitudinal separation relevant to dipole production (Cameron et al., 2010).

The same source sensitivity appears in local emergence studies. Tests on 17 emerging active regions indicate that SFT assumptions become valid after about $B_r(\theta,\phi,t)$5 of the peak total unsigned flux has emerged. Before that point, emergence dynamics remain too important for a passive-transport description; after that point, the field evolution is well described by advection plus diffusion, with diffusivities in the range $B_r(\theta,\phi,t)$6–$B_r(\theta,\phi,t)$7 consistent with the first five days of post-emergence evolution (Gottschling et al., 2021).

3. Transport parameters, nonlinear feedbacks, and calibration

A central issue in SFT is that multiple parameter combinations can reproduce similar large-scale diagnostics. The weak polar field at the end of cycle 23 is a canonical example. A calibrated model reproduced the amplitude and reversal timing of the polar fields reasonably well between 1976 and 2002, but by the end of cycle 23 the simulated polar fields were about a factor of two too strong, and the open flux at the cycle-23 minimum was much higher than observed. A $B_r(\theta,\phi,t)$8 reduction in sunspot number weakened the modeled polar field but made the open flux too low at solar maximum. A $B_r(\theta,\phi,t)$9 decrease in mean tilt angle also yielded a weak polar field, but delayed polar-field reversal by about $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$0 years relative to observations. By contrast, a $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$1 increase in meridional-flow speed, from $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$2 to $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$3, brought both the polar-field amplitude and reversal time into good agreement with the observations (Jiang et al., 2011).

Systematic optimization studies using idealized average-cycle source terms reached a different but related conclusion: without a significant decay term in the SFT equation, specifically for $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$4 yr, the global dipole moment reverses too late in the cycle for all tested meridional-flow profiles and parameters. In that class of models, an allowed domain exists for $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$5 values in the $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$6–$$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$7 yr range, and higher diffusivities of $$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$8–$$\frac{\partial B_r}{\partial t}+\nabla_h\cdot(\mathbf{u}_h B_r)=\eta \nabla_h^2 B_r + S,$$9 are generally preferred, though some lower-$$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$0 solutions remain admissible (Petrovay et al., 2019).

Recent parameter-space studies have extended this calibration problem by introducing nonlinear source feedbacks. Tilt quenching (TQ) and latitude quenching (LQ) both reduce the admissible parameter domain, with LQ exerting the stronger influence. Their combined action produces a pronounced saturation or “ceiling” that limits axial dipole amplification, while a finite flux-decay timescale $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$1–$$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$2 yr further narrows the admissible domain within that parameterized source formulation (Alhosani et al., 30 May 2026).

The treatment of inflows illustrates that nonlinear feedbacks need not act monotonically. In a point-concentration SFT model, converging inflows toward active regions enhanced flux cancellation and usually reduced the axial dipole moment by limiting the latitudinal separation of the polarities. However, for very small initial tilt, the same inflows could increase the latitudinal separation and increase the axial dipole moment, even generating a tilt from an initially east–west aligned BMR (Martin-Belda et al., 2015).

4. Observational tests, spectral validation, and cycle reconstruction

SFT has been tested against observations over a wide range of scales. At local scales, observed surface flows from local correlation tracking can reproduce the evolution of young active-region flux once most emergence is complete, and the buffeting of the field by supergranulation can be described as a diffusion process. In those first five post-emergence days, converging inflows around active regions are not important for the total unsigned flux evolution, because their increase of cancellation is balanced by decreased transport away from the region (Gottschling et al., 2021).

At cycle scales, data-assimilative SFT has achieved high-fidelity reconstructions. A continuous simulation from May 2010 to November 2024, driven by assimilated observed active regions from the ARISE database, reproduced the axial dipole strength, polar-field reversal timing, and magnetic butterfly diagram in good agreement with SDO/HMI observations. The axial dipole agreement reached a correlation coefficient $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$3, and these results were obtained without incorporating radial diffusion or cyclic variations in meridional-flow speed. In that reconstruction, the anomalous southern poleward transport pattern in cycle 24 was traced primarily to intermittent active-region emergence, with about $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$4 of the total unsigned southern-hemisphere flux contributed by active regions emerging during Carrington Rotations 2141–2160 (Wang et al., 2 Jun 2025).

Spectral validation has refined the empirical scope of the model. Comparisons between observed and simulated magnetic power spectra show that SFT reproduces the observed spectra well at spherical-harmonic degrees $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$5, but diverges progressively at smaller spatial scales $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$6. Power at $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$7 is primarily determined by the magnetic flux sources, whereas power at $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$8 is more sensitive to transport parameters. In particular, the relative strengths of the $$\frac{\partial B_r}{\partial t} +\frac{1}{R_\odot\sin\theta}\frac{\partial}{\partial\theta}\!\Big(\sin\theta\,u_\theta B_r\Big) +\Omega(\theta)\frac{\partial B_r}{\partial\phi} = \frac{\eta}{R_\odot^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial B_r}{\partial\theta}\right) +\frac{\eta}{R_\odot^2\sin^2\theta}\frac{\partial^2 B_r}{\partial\phi^2} +S.$$9 and $S$0 multipoles suggest that the poleward flow above about $S$1 latitudes is very weak (Luo et al., 7 Sep 2025).

These results support a specific interpretation of the classical model. On large scales, SFT is quantitatively robust. On supergranular and network scales, the standard diffusion closure becomes inadequate. This suggests that large-scale dipole evolution and small-scale magnetic-network structure should not be treated as equivalent validation targets.

5. Coupling to dynamos, three-dimensional transport, and stellar applications

SFT is also used as an empirical constraint on deeper and more general magnetic-transport models. In comparisons with flux transport dynamo (FTD) models, the longitude-averaged SFT solution serves as the near-surface benchmark. In that context, an FTD model with a vertical-field upper boundary can match SFT only if downward pumping is strong enough to give a near-surface magnetic Reynolds number of about $S$2, corresponding to pumping of order $S$3 in the boundary layer. An FTD model with a potential-field upper boundary was unable to match the SFT results (Cameron et al., 2012).

Three-dimensional Babcock–Leighton modeling complicates the standard diffusion picture. A 3D flux-transport/Babcock–Leighton model with explicit observed surface convection found that the large-scale transport effect of convection could be approximated by a turbulent diffusivity of order $S$4 or $S$5, close to standard SFT values. However, the explicitly computed turbulent electromotive force bore little resemblance to a diffusive flux, and the convective models produced mixed-polarity bands in the polar regions that have no counterpart in solar observations (Hazra et al., 2018).

Outside solar applications, SFT forms the surface component of composite stellar-activity models. In one class of models, a deep-seated dynamo determines the probability density for flux tube eruption, buoyant rise determines the emergence latitudes and tilt angles, and SFT then evolves the emerged BMRs across the stellar surface (Işik et al., 2012). In forward modeling of Sun-like stars with higher rotation rates, the SFT stage translates rotation-dependent emergence patterns into spot distributions and polar caps: at $S$6, the equatorial inactive gap reaches a half-width of $S$7, the maximum spot coverage reaches $S$8, and polar spots can form by accumulation of follower-polarity flux from decaying bipolar regions (Işık et al., 2018).

6. Data assimilation, far-side reconstruction, and current methodological directions

A major operational use of SFT is the construction of synchronic full-Sun magnetic maps from incomplete observations. Far-side emergence is a persistent limitation. The combined surface flux transport and helioseismic Far-side Active Region Model (FARM) addresses this by converting helioseismic phase-shift maps into approximate magnetic source terms for insertion into an SFT model. From 2010 to 2024, it modeled 859 active regions with an average total unsigned flux of $S$9 Mx and an average area of $\langle B_r\rangle$0; $\langle B_r\rangle$1 were found to have an anti-Hale configuration and were manually corrected. Including these far-side active regions produced a substantial improvement in agreement between modeled open-field areas and EUV observations (Yang et al., 2024).

A related approach uses STEREO He II $\langle B_r\rangle$2 intensity as a far-side emergence proxy in the Advective Flux Transport model. That study estimated that during solar cycle 24 maximum, $\langle B_r\rangle$3–$\langle B_r\rangle$4 Mx of flux is missing from SFT models that do not include far-side data. It also concluded that while $\langle B_r\rangle$5 data can improve surface-flux models, they are insufficient by themselves to produce a complete picture without direct magnetic observations from magnetographs (Upton et al., 2024).

Methodological diversification has accelerated. Different SFT implementations can yield substantially different coronal and heliospheric predictions, and one comparative study found that the differences between predicted solar-wind properties, open-field footpoints, and coronal nulls are dominated by model assumptions and implementation rather than by the choice of a particular realization of the random convective evolution (Barnes et al., 2023). This suggests that transport closure, source treatment, and assimilation strategy are now major systematic uncertainties.

On the computational side, the open-source Flux Transport program HipFT provides a modular SFT solver with advection, diffusion, and data assimilation, supports ensembles of realizations in a single run, and is designed for multi-CPU and multi-GPU execution (Caplan et al., 10 Jan 2025). Physics-Informed Neural Networks have also been applied to one- and two-dimensional SFT, where a mesh-independent PINN solver was reported to reproduce the observed polar magnetic field with better flux conservation; in analytic tests, its error lay in the range $\langle B_r\rangle$6–$\langle B_r\rangle$7 (Athalathil et al., 2024). In an observing-system simulation experiment, sequential Ensemble Kalman Filter assimilation into a one-dimensional SFT model was shown to infer the peak meridional flow speed from synthetic high-latitude magnetic observations and use the posterior flow estimate to drive subsequent forecasts (Dash et al., 2024).

The current literature therefore does not support a single uniform picture of SFT. One recurrent controversy concerns radial decay: parameterized average-cycle models often prefer finite decay timescales, whereas data-assimilative active-region simulations can reproduce recent observations without radial diffusion (Petrovay et al., 2019, Wang et al., 2 Jun 2025). Another concerns the treatment of supergranulation: large-scale evolution is well captured by an effective diffusivity, but explicit convection and spectral tests both indicate that the diffusion approximation is incomplete at smaller scales (Hazra et al., 2018, Luo et al., 7 Sep 2025). A plausible implication is that future SFT development will continue to separate large-scale predictive skill from small-scale transport realism, with the source term, high-latitude flow, and far-side emergence remaining the dominant frontiers.