Main

Historically, the solar wind1 has been categorized as ‘fast’ or ‘slow’ wind. The fast wind is generally described as having speeds greater than 500 km s−1 and originating from the interiors of coronal holes formed by open magnetic flux. The slow wind is classified as wind with speeds less than 500 km s−1 and is associated with the coronal streamer belt2. In general, slow solar wind is characterized its high degree of internal structure and variability, coronal elemental compositions and ionic compositions that indicate a hotter and denser source region than the fast wind3,4,5,6,7. Interplanetary observations reveal two components of slow wind, one that is characterized by its high degree of Alfvénic fluctuations similar to the fast wind and the other with lower Alfvénicity8.

During solar minimum the coronal streamer belt and its associated slow wind streams remain concentrated around the solar equatorial regions, but during periods of solar activity, small coronal holes, which quite commonly form in the unipolar remnants of active regions confined to low latitudes, can reshape the streamer belt, extending it to high latitudes and distorting the central heliospheric current sheet (HCS). These complex streamer belt topologies must create a web of magnetic separatrices within the corona, termed the S-web9, but how exactly the coronal streamer belt and this coronal S-web relate to the origin of the slow solar wind is still a subject of active debate.

Super-radial expansion of the coronal holes themselves is thought to be an important source of slow solar wind10,11,12,13,14,15. Magnetohydrodynamic (MHD) wave turbulence could drive the slow wind along open magnetic fields in coronal holes16. In this scenario, variations in expansion factors and footpoint field strengths in open field regions are thought to play a role, at least in part, in the observed compositional difference between fast and slow winds11. The spatial variability of slow solar wind is attributed to large-scale events such as coronal mass ejections and streamer blobs that propagate through the background slow solar wind11.

At the same time, interchange reconnection between open and closed magnetic fields17 and S-web reconnection (that is, at quasi-separatrix layers and separatrices) created by low-latitude coronal holes18 are also commonly invoked drivers of the slow solar wind and/or its internal structure. In this scenario, at least a component of the structured slow wind is attributed to reconnection that is inherently dynamic and transient. Such models also predict that S-web dynamics are particularly important within the middle corona, a transitional regime where magnetic topology changes from being predominantly closed to open with altitude. In streamer belt regions, this transition typically occurs over heights of 1.5 solar radii (from the Sun’s centre; 1 R☉ ≈ 700 Mm). Magnetic reconnection in the middle corona could release plasma from the closed magnetic structures into the solar wind5,18. However, the middle-coronal manifestation of reconnection dynamics associated with coronal holes and active regions, and the extent to which they drive slow solar wind structures, remain poorly understood19.

Results

We used images from a special off-pointing campaign of the Geostationary Operational Environmental Satellite’s (GOES) Solar Ultraviolet Imager (SUVI)20,21 that directly observed the middle corona to identify coronal dynamics associated with the highly structured slow solar wind. The extended field of view of SUVI allowed us to trace extreme-ultraviolet (EUV) features up to a height of about 2.7 R☉, into the middle corona (Methods). Using its EUV 195 Å passband, SUVI observed a pair of roughly east–west aligned coronal holes near the equator with an embedded (decaying) active region. During the course of one month of observations, we observed this coronal hole–active region (CH–AR) system as it rotated across the solar disk and subsequently when it appeared at the west and east limbs. The limb view of the system revealed a highly structured middle corona with elongated features, which we term a ‘coronal web’, that protrudes radially outward in the plane of sky, with a large latitudinal extent (Fig. 1 and Extended Data Fig. 1).

Fig. 1: Magnetic driver of a slow solar wind stream.
figure 1

a, SUVI 195 Å and LASCO composite image showing the corona over the west limb. The arrow points to an outflowing solar wind stream. The dashed and solid arcs are at 1 and 2.65 solar radii from the Sun centre. b, Difference image obtained by subtracting two composite images with timestamps as indicated. c, A sequence of four difference images in a field of view covered by the white rectangle in panel b. The star symbols guide the eye to follow a solar wind stream. d, A time sequence of nine SUVI images (field of view marked by a box in panel c). The slanted arrows point to interacting structures in the middle corona (Supplementary Video 1). In panels c and d, time increases from left to right and top to bottom.

Full size image

This coronal web exhibited complex dynamics. In Fig. 1a,b, we show a SUVI snapshot up to a height of about 2.7 R☉ (bounded by solid white arc), from the period when the CH–AR system was at the west limb. Above this height, the SUVI EUV image is paired with a co-temporal white-light coronagraphic image from the Large Angle Spectroscopic Coronagraph (LASCO)22 onboard the Solar and Heliospheric Observatory (SOHO) spacecraft (Methods). We point to an outflowing stream on the west limb above 2.65 R☉, observed with SOHO/LASCO (arrow in Fig. 1a,b). Examining the image sequence reveals that similar streams are repeatedly driven in that region (Fig. 1c highlights two such events). Underlying these streams, we found a system of interacting magnetic structures in the middle corona that are clearly linked to the CH–AR system in SUVI images. Magnetic structures that are apparently open initially approach each other and interact to form (transient) closed loops; two such interactions associated with stream events in Fig. 1c are displayed in Fig. 1d (Supplementary Video 1). Three other examples of solar wind streams emerging over interacting middle-coronal structures are presented in Supplementary Figs. 1–3 (Supplementary Videos 2–4). We found that the speeds of these outflows are consistent with speeds of slow solar wind streams (Supplementary Fig. 4 and Supplementary Information). Such streams are generally considered to be passive tracers of the slow solar wind23,24. We also found that these types of event are quite prevalent in the coronal web. Over the course of five days when the system crossed the west limb, we observed persistent interaction and continual rearrangement of the coronal web features and emergence of highly structured and variable slow solar wind streams over their tips (Supplementary Video 5). We observed similar (plane of sky) complex dynamics in the middle-coronal web and the emergence of slow solar wind streams also when the same system crossed the east limb (Supplementary Fig. 5 and Supplementary Video 6).

Based on low-coronal EUV images and photospheric magnetic field data from the Solar Dynamics Observatory (SDO)25, we found that the coronal web is associated with a highly warped HCS (Methods, Extended Data Figs. 2 and 3 and Supplementary Information). To decipher the complexity of the latitudinally extended complex coronal web and to investigate how the CH–AR system is topologically related to the global magnetic field, we employ advanced global three-dimensional (3D) MHD coronal models driven at the lower boundary by synoptic observations of photospheric magnetic fields26,27 (Methods). We visualize the skeleton of the global magnetic field using the squashing factor, Q, which employs field-line mappings to identify the quasi- and true-separatrix surfaces where Q » 1 (ref. 28). In particular, the synoptic map of the signed logQ at 3 R☉, coloured blue (red) by the inward (outward) direction of magnetic field, illustrates complex S-web features29 that map from the coronal holes at the surface to the open middle corona and warped HCS above the active region (Fig. 2a and Extended Data Figs. 4 and 5). This signed logQ map shows longitudinally extended, prominent complex S-web features surrounding the active region. These include a warped global polarity inversion line along with a pair of northeast and southwest pseudostreamer arcs on larger scales, with embedded cellular pockets of smaller scale features that are formed by separatrices in the mapping due to presence of distinct small-scale flux concentrations at the surface within the larger open field region (Supplementary Information).

Fig. 2: Imprints of the S-web in the observed coronal web.
figure 2

a, Synoptic signed logQ map at 3 R☉, spanning the CH–AR system (field of view marked in Extended Data Figs. 3 and 4). Red (blue) shaded regions represent outward- (inward-) oriented magnetic field. The arrows point to the prominent S-web features. The four different symbols along the vertical lines (marking the Carrington longitudes at the west limb) identify the S-web features that are labelled in b and c. b, Solar west-limb map along line b in panel a. Overlaid red (blue) curves represent outward (inward) magnetic structures at those locations, outlining the magnetic topology of helmet streamers (H-S) and pseudostreamers (P-S) in the plane of sky. c, Solar west-limb map along line c in a in the same style as b. d, A bird’s-eye-view illustration of the H-S and P-S features. Closed magnetic field (black solid curves), the polarity inversion line (yellow solid and dashed curves) and prominent pseudostreamer arcs (red and blue solid and dashed curves) are marked. The observed coronal web is bounded by these topological structures. The lighter and darker shaded regions on the grey solar surface represent magnetic polarities (exaggerated for clarity) (Supplementary Video 5).

Full size image

As the CH–AR system crossed the west limb, we first observed the signature of the polarity inversion line in the form of a bipolar helmet streamer northwest of the active region. Similarly, in the southwest, we observed a unipolar pseudostreamer (Fig. 2b; symbols on line b in Fig. 2a). Then we observed this configuration gradually changing to a unipolar pseudostreamer in the northeast and a bipolar helmet streamer in the southeast with solar rotation (Fig. 2c; symbols on line c in Fig. 2a). More importantly, over this entire period (about five days), SUVI recorded complex coronal web activity over the CH–AR system, along with observations of solar wind streams in LASCO (Supplementary Video 5). Similar middle-coronal dynamics are observed at the east limb (Supplementary Fig. 5 and Supplementary Video 6). Its observed persistence suggest that the coronal web is not only extended latitudinally (Extended Data Fig. 1) but also spans longitudinally over the entire CH–AR system on either side of the HCS (Fig. 2d).

Thus, our SUVI observations captured direct imprints and dynamics of this S-web in the middle corona. For instance, consider the wind streams presented in Fig. 1. Those outflows emerge when a pair of middle-coronal structures approach each other. By comparing the timing of these outflows in Supplementary Video 5, we found that the middle-coronal structures interact at the cusp of the southwest pseudostreamer. Similarly, wind streams in Supplementary Figs. 1–3 emerge from the cusps of the HCS. Models suggest that streamer and pseudostreamer cusps are sites of persistent reconnection30,31. The observed interaction and continual rearrangement of the coronal web features at these cusps are consistent with persistent reconnection, as predicted by S-web models. Although reconnection at streamer cusps in the middle corona has been inferred in other observational studies32,33 and modelled in three dimensions30,31, the observations presented here represent imaging signatures of coronal web dynamics and their direct and persistent effects. Our observations suggest that the coronal web is a direct manifestation of the full breadth of S-web in the middle corona. The S-web reconnection dynamics modulate and drive the structure of slow solar wind through prevalent reconnection9,18.

A volume render of logQ highlights the boundaries of individual flux domains projected into the image plane, revealing the existence of substantial magnetic complexity within the CH–AR system (Fig. 3a and Supplementary Video 7). The ecliptic view of the 3D volume render of logQ with the CH–AR system at the west limb does closely reproduce elongated magnetic topological structures associated with the observed coronal web, confined to northern and southern bright (pseudo-)streamers (Fig. 3b and Supplementary Video 8). The synthetic EUV emission from the inner to middle corona and the white-light emission in the extended corona (Fig. 3c) are in general agreement with structures that we observed with the SUVI–LASCO combination (Fig. 1a). Moreover, radial velocity sliced at 3 R☉ over the large-scale HCS crossing and the pseudostreamer arcs in the MHD model also quantitatively agree with the observed speeds of wind streams emerging from those topological features (Supplementary Figs. 4 and 6 and Supplementary Information). Thus, the observationally driven MHD model provides credence to our interpretation of the existence of the complex coronal web whose dynamics correlate to the release of wind streams.

Fig. 3: Global magnetic skeleton and coronal structures based on 3D MHD simulations.
figure 3

a, 3D volume render of squashing factor (Q) from Earth’s perspective on 2018-08-10 14:00 ut. Field of view is ±1.4 R☉ (Supplementary Video 7). b, 3D Q volume render for Earth’s perspective on 2018-08-17 02:00 ut. Field of view is 0–6 R☉ in X, ±3 R☉ in Y (Supplementary Video 8). c, Forward-modelled AIA 193 Å (brown) and polarized brightness (red) unsharp mask-filtered composite for the same field of view and view as in b. The yellow and magenta arrows point to the northeast and southwest streamers. The green arrow points to the top or cusp of closed coronal loops.

Full size image

The long lifetime of the system allowed us to probe the region from a different viewpoint using the Sun-orbiting STEREO-A, which was roughly in quadrature with respect to the Sun–Earth line during the SUVI campaign (Methods and Extended Data Fig. 6). By combining data from Solar Terrestrial Relations Observatory-Ahead’s (STEREO-A) extreme ultraviolet imager (EUVI)34, outer visible-light coronagraph (COR-2) and the inner visible-light heliospheric imager (HI-1)35, we found imprints of the complex coronal web over the CH–AR system extending into the heliosphere. Figure 4a and the associated Supplementary Video 9 demonstrate the close resemblance between highly structured slow solar wind streams escaping into the heliosphere and the S-web-driven wind streams that we observed with the SUVI and LASCO combination. Due to the lack of an extended field of view, the EUVI did not directly image the coronal web that we observed with SUVI, demonstrating that the SUVI extended field-of-view observations provide a crucial missing link between middle-coronal S-web dynamics and the highly structured slow solar wind observations.

Discussion

Our observations provide important close-to-the-Sun context to the measurements of the highly structured nature of slow solar wind emerging from an equatorial coronal hole recorded by the Parker Solar Probe (PSP) during its first perihelion on 2018-11-06 (ref. 7). The western coronal hole that we discussed here is the same region to which the PSP was magnetically connected during its perihelion at a later date7. A similar north–south aligned polarity inversion line separated the western and eastern coronal holes and the active region, which was decaying at that time (Methods and Extended Data Figs. 3 and 6). Some 10 days before PSP’s encounter, instruments on STEREO-A recorded inner coronal and heliospheric structures similar to those found during the SUVI observations (Fig. 4b and Supplementary Video 10). During its perihelion, the PSP observed ubiquitous radial velocity spikes in the solar wind associated with switchbacks, local reversals of radial magnetic fields7. Such switchbacks are linked to the inherently Alfvénic slow wind originating from rapidly expanding small coronal holes7,15. Switchbacks could also form in situ in the expanding solar wind36. At the same time, models predict that interchange reconnection between open and closed magnetic fields, such as in the S-web, could generate and launch Alfvén waves30 and magnetic flux ropes37, which in turn may further evolve to form switchbacks.

Fig. 4: Heliospheric connection.
figure 4

Solar wind structures seen in the outer corona (COR-2; blue) and heliosphere (HI-1; red) and their association with the low-latitude CH–AR system (EUVI; green) (Extended Data Fig. 6). a, Closer in time to the SUVI observations (Supplementary Video 9). b, Closer in time to the first perihelion of the PSP (the symbol marks the approximate plane-of-sky position of the PSP during its perihelion on 2018-11-06; Supplementary Video 10). The position angle increases in the counter-clockwise direction from solar north.

Full size image

Both photospheric and low-coronal observations point to ubiquitous small-scale ephemeral magnetic activity in and around active regions and in coronal holes38,39. Such small-scale magnetic activity will lead to separatrices even in open field coronal holes near active regions. Our MHD model does reveal more cellular and smaller separatrices on either side of the active region (Fig. 2a), which result from small-scale magnetic concentrations within the observed coronal holes (Methods). Although the origin of switchbacks still remains elusive, our SUVI observations and MHD model hint that the S-web interchange reconnection could play a role in the process. Recent observations do hint at such a close link between interchange reconnection and switchbacks40,41.

Systems of low-latitude coronal holes surrounding active regions are quite common during periods of solar activity13. The CH–AR system presented here is well-isolated and thus allowed us to observe the full breadth of S-web dynamics in the middle corona. Because such systems are common, based on our SUVI observations we suggest that a complex coronal web is quite common, too.

Our findings lend support to reconnection-based slow solar wind models9,17,18 and identify two critical components necessary for further study of the coronal S-web. First, depending on the alignment of remote observatories and the orientation of the global polarity inversion line, the S-web imprints that we clearly observed with SUVI may not be apparent in the plane of sky due to projection effects, but that does not necessarily imply that they are absent. Moreover, single vantage point middle-coronal observations do not provide a detailed 3D view of the reconnection. In addition, uncertainties that arise due to the superposition of structures in the middle corona cannot be resolved in such single vantage point observations. Here a polar mission targeting the middle corona in combination with an ecliptic view could, one day, reveal the 3D structure of the S-web. Second, as demonstrated by Figs. 1 and 4, above the closed field regions the S-web separatrices are more established and less likely to change due to topology. A direct consequence of this can be seen in the predominantly radial flows that are ubiquitous beyond 3–4 R☉. Therefore, without direct images of the middle corona, important and more complex S-web dynamics associated with the solar wind, which are evidently more frequent in this regime as compared with the outer corona, will continue to go undetected. Additionally, because the middle corona provides direct access to the global topological skeleton of the Sun’s magnetic fields (Fig. 2), observing this regime provides crucial information on how plasma upflows that originate low in the corona eventually make their way through the relatively complex topology in the middle corona and into the solar wind.

Using the long-duration EUV image sequence of the solar corona up to 2.7 R☉, complemented by observationally driven advanced 3D MHD coronal models, we identified a complex coronal web as a direct imprint of the S-web. While the speed of the bulk of the slow solar wind itself could be closely governed by the expansion factors of coronal holes11,12,13, our observations suggest that S-web dynamics imprint magnetic topological structures on the slow wind and play a role in releasing hotter plasma with coronal compositions into the slow wind5. The dynamic evolution of the observed coronal web thus presents direct observational evidence for a key and integral role of the S-web in driving the highly structured slow solar wind.

Our SUVI observations are more limited in both time and spatial resolution than would be needed to fully resolve the origins of slow solar wind and switchbacks. To this end, our exploratory measurements of the time-dependent middle corona lay a clear path forward. They demonstrate the need for a more systematic and detailed EUV observations of the middle-coronal region in close coordination with outer coronal and in situ measurements. Future and planned missions, including NASA’s (the National Aeronautics and Space Administration’s) in-development Polarimeter to Unify the Corona and Heliosphere mission42, will partially address this need for instruments operating from the Earth’s perspective. Additional missions proposed and in development that will explore the middle corona specifically, such as ECCCO43, the Sun Coronal Ejection Tracker44 and PROBA3/ASPIICS45, among others, will help to fully connect the global corona–heliosphere system. Combined with 3D information we can derive from the PSP46, the Solar Orbiter’s Metis47 and Extreme-Ultraviolet Imager48 away from the Sun–Earth line and new spectral capabilities in several proposed missions, we can address these concerns to fully close these questions.

Methods

Observations

We used observations from four different spacecrafts to gain a broader view of coronal sources and to trace the origins of the highly structured slow solar wind. These observations span a range of distances from the Sun, covering features from low altitudes closer to the solar surface to distances exceeding 20 times the solar radius (1 R☉ ≈ 700 Mm).

GOES SUVI data

The GOES SUVI data we used come from a deep-exposure offpoint campaign run between 2018-08-07 and 2018-09-13 (ref. 20) with 2 EUV filters: 171 Å and 195 Å. The SUVI observations are three-panel mosaic images, with a central Sun-centred image and two eastward and westward off-pointed side panels49. SUVI was not designed to be operated in off-pointed mode, so some additional processing is required to reduce image noise, remove background stray light and reduce the large-scale brightness gradient (up to a factor of 106 between the brightest and faintest structures) to permit the visualization of the whole field of view in a single, uniformly scaled image. The resulting images have a field of view out to about 5 R☉, at an image scale of 5″ per pixel and a cadence of around 20 minutes. Because of the substantial decrease in EUV emission at large heights and the substantial contribution of noise resulting from stray light, there is negligible structure detectable above 3.5–4 R☉ in these observations. Additional details about the processing, contents and limitations of these observations appear in ref. 20.

In Extended Data Fig. 1, we present the overview of the SUVI coronal observations recorded by its 195 Å filter. The period of SUVI observations covers a part of the decay phase of the previous solar activity cycle (number 24). During the course of approximately one month of observations, SUVI observed multiple coronal holes and active regions. Among these features are a pair of coronal holes roughly aligned along the east–west orientation. The eastern coronal hole appears to be an extension of the northern polar coronal hole and the western coronal hole is more isolated. A decaying active region (National Oceanic and Atmospheric Administration number 12711) is embedded in this pair of coronal holes. Based on the observations from the SDO, we identified that these two coronal holes and the active region lived for at least 5 solar rotations (that is, about 135 days). In the lower panels of Extended Data Fig. 1, we mark these features when they appear on the west limb (panel b) and then on the east limb (panel c) after rotating through the far side of the Sun.

SOHO LASCO data

To extend the field of view covered by SUVI to larger heights, and to fill in gaps where the EUV emission revealed by it is insufficient to detect structure, we use data from the LASCO22 C2 coronagraph, which has a field of view that extends roughly to 6 R☉. We prepared these data using the standard LASCO data reduction tools distributed in the Interactive Data Language SolarSoft by the LASCO team. We improve signal-to-noise ratio in these images by stacking in time to reach an effective cadence of one hour. We used a custom calculated minimum background spanning the entire observation set to remove stray light and F-corona and an additional despiking step to suppress both cosmic rays and background stars. Due to the inherent substantial differences in the intrinsic brightness of the visible-light corona observed by LASCO and the EUV emission observed by SUVI, to merge these LASCO observations with SUVI data we also must rescale the data to approximately the same normalized dynamic range. We do this by applying a radial normalizing filter, derived from the median of the whole dataset over time. We then merge the datasets by pairing images on the basis of proximity in time. The resulting SUVI–LASCO composites show EUV emission up to about 2.7 R☉ and the visible-light corona above this. Additional details on the processing and merger of these datasets appear in ref. 20.

SDO data

The SDO25 provided lower-coronal and photospheric context of the CH–AR system. We have used the 171 Å, 193 Å and 211 Å EUV filter full-disk images from the Atmospheric Imaging Assembly (AIA)50 onboard the SDO to examine the general evolution of coronal structures over the course of one month from 2018-08-08 to 2018-09-09 (roughly covering the period of the SUVI off-pointing campaign). The surface magnetic structures were observed using the full-disk line-of-sight magnetic field maps obtained by the Helioseismic and Magnetic Imager (HMI)51 on the SDO. We retrieved these full-disk AIA and HMI data at a two-hour cadence from the Joint Science Operations Center, which are then processed using the standard AIA_PREP procedure available in SolarSoft.

We visually identified the system of eastern and western coronal hole pair and the embedded active region on 2018-08-10 and followed those points at the rate of Carrington rotation to demonstrate that it is indeed the same set of features that rotate to the west limb around 2018-08-16 and then reappear on the east limb after rotating through the far side of the Sun around 2018-08-31 (Extended Data Fig. 2). Closer to the eastern and western limbs, the line-of-sight component of surface magnetic field patches associated with the CH–AR system may not be apparent in the HMI maps. Here the system is better seen in the coronal images, in which the associated active region loops and coronal holes are evident. The active region embedded between the two coronal holes emerged on 2018-05-23 and received the National Oceanic and Atmospheric Administration number 12711 the next day. The active region is composed of leading negative-polarity magnetic fields and trailing positive-polarity magnetic fields. The western coronal hole formed within a region of predominantly negative-polarity magnetic field, whereas the eastern coronal hole formed in a region of predominantly positive-polarity magnetic field.

AIA synoptic maps

We complemented the full-disk SDO images with the AIA 171 Å, 193 Å and 211 Å EUV filter synoptic Carrington maps52 to investigate the general long-term evolution of the pair of coronal holes and the active region. Each synoptic map covers a full Carrington rotation (CR) and are in Carrington-longitude and sine-latitude projection. In particular, we used AIA synoptic maps from CR2206 through CR2211 (from 2018-07-09 to 2018-12-20). These CRs cover the period of the SUVI 2018 off-pointing campaign (2018-08-07 to 2018-09-13), as well as the first perihelion of the PSP on 2018-11-06 (ref. 7) (Extended Data Fig. 3).

The coronal hole pair exhibited evolution throughout the period spanning the six CRs that we considered, both in terms of their spatial morphology and general location with respect to their position that we visually determined on 2018-08-10, but remained clearly distinguishable. The active region, on the other hand, exhibited a continual decay from CR2206 through CR2210 and was nearly absent or completely decayed in CR2211. To illustrate these long-term changes in the lower-coronal features, we visually identified and marked their new positions in each CR. All the features migrated westward (that is, increasing Carrington longitudes). For example, the Carrington longitude of the western coronal hole in CR2206 was roughly 320°, whereas in CR2211 it was 350°.

HMI synoptic maps for magnetic field extrapolations

The magnetic polarity inversion line will provide information on the general distribution and separation of domains of positive- and negative-polarity magnetic fields at a given height above the solar surface. This provides further insights into the observed (large-scale) coronal plasma structures that generally outline magnetic fields. To this end, we employed a potential field source surface (PFSS) technique to extrapolate coronal magnetic fields from the observed surface magnetic field maps53. As the lower boundary to the PFSS model, we used synoptic surface radial magnetic field maps archived at the Joint Science Operations Center (CRs2206–2211). Specifically, we used synoptic maps of the type hmi.synoptic_mr_polfil_720s, in which the polar fields for latitudes above 60° are inferred from annual observations of the poles when the respective north or south pole of the Sun is most tipped toward Earth54. More details about this data product are available at http://hmi.stanford.edu/QMap/. For these magnetic field extrapolations, we used a PFSS code that is distributed in the SolarSoft. We placed the source surface or the top boundary at a height of 2 R☉ (ref. 7). Above this height or source surface, magnetic fields are assumed to remain open. The contour where the radial magnetic field is zero at the source surface defines the polarity inversion line, which is shown as the yellow line in Extended Data Fig. 3.

STEREO data

During August 2018, the STEREO-A spacecraft55 was orbiting the Sun roughly in quadrature with respect to the Sun–Earth line at Earth ecliptic longitude of about −108° (in the heliocentric Earth ecliptic coordinate system). This rough quadrature means that the solar features in Earth’s line of sight would appear on the western limb of the Sun as seen from the vantage point of STEREO-A. The CH–AR system traversed the solar disk and the central meridian (from Earth’s perspective) between 2018-08-07 and 2018-08-12, the period that is roughly one week before when it rotated to the western limb as observed with both SUVI and the SDO (Extended Data Figs. 1 and 2). This enabled us to probe the heliospheric manifestation of the CH–AR system when it was approximately in Earth’s line of sight, using images obtained by a suite of instruments on the Sun Earth Connection Coronal and Heliospheric Investigation (SECCHI) observatory34 on STEREO-A.

In particular, we employed coronal images obtained with the 195 Å filter on the EUVI on SECCHI (which records EUV emission from coronal plasma around 1–2 MK in the inner corona up to a field of view of about 1.7 R☉). These images are combined with imaging data from SECCHI’s outer Lyot visible-light coronagraph (COR-2; covering the extended corona between 2.5 to 15 R☉) and the inner visible-light Heliospheric Imager (HI-1; covering the heliosphere between 15 to 84 R☉)35.

From the Virtual Solar Observatory, we retrieved the EUVI (195 Å filter; Extended Data Fig. 6) and COR-2 (double exposure total brightness) data covering the period from 2018-08-07 to 2018-08-12 at a cadence of 40 minutes. These data are processed using the standard SECCHI_PREP procedure that is available in SolarSoft. We extracted the K-corona signal from the COR-2 images by subtracting pixel-wise minimum intensity from the entire time period considered (this will essentially remove the F-corona signal from the images). We retrieved the level-2 F-corona-subtracted HI-1 images from the UK Solar System Data Centre (http://www.stereo.rl.ac.uk/). Besides the coronal signal, HI-1 also images background star field. We suppressed the intensity of these background stars with peak intensities greater than 3 × 10−13 times the mean solar brightness, using standard procedures available in SolarSoft.

STEREO observations during days leading up to the first perihelion of the PSP

About 10 days before the first perihelion of the PSP on 2018-11-06, the pair of coronal holes and the decaying active region appeared on the west limb from the vantage point of STEREO-A, at Earth ecliptic longitude of about −103°. We retrieved the EUVI, COR-2 and HI-1 data covering a five-day period from 2018-10-25 to 2018-10-29 from the Virtual Solar Observatory. These data are processed in a similar way as described in the previous paragraph (Extended Data Fig. 6).

STEREO composite maps

We created composite STEREO images by projecting EUVI, COR-2 and HI-1 data on to a grid of conformal radial coordinates and position angle, using the World Coordinate System information available in the headers of respective Flexible Image Transport System (FITS) files. In Fig. 4, the EUVI images are plotted in log-scale and the COR-2 and HI-1 images are scaled by the cube of radial distance56.

Description of 3D MHD coronal model

The 3D MHD coronal simulation used here was produced with the Magnetohydrodynamic Algorithm outside a Sphere (MAS) thermodynamic MHD global coronal model27,57. This model solves the resistive MHD equations on a global, non-uniform spherical mesh from 1–30 R☉. MAS also solves for non-ideal energy transport terms relevant to low-coronal hydrodynamics (coronal heating, radiative losses, electron heat conduction) as well as auxiliary equations that represent the propagation, reflection and dissipation of low-frequency Alfvénic fluctuations. Together, these form the MAS Wave-Turbulence-Driven (MAS-WTD) model for coronal heating27,58,59. For our purposes, the MAS-WTD model provides a volume-filling 3D magnetic field and plasma state that includes the balance between magnetic and hydrodynamic forces. Most importantly, this computation is not restricted to assumptions about the plasma beta or source-surface radius, which may be essential for understanding dynamics in the middle corona. Combined with forward modelling (Fig. 3c) and comparisons with SDO/AIA data (Extended Data Fig. 5), we use this simulation to reasonably assert that our physical inferences and magnetic field analysis, including the HCS location, active region cusp formation height and coronal hole properties (Fig. 3 and Extended Data Fig. 4) are relevant to the observations at hand.

The most important inputs to the MAS-WTD model are the magnetic boundary conditions at the surface, the computational mesh and the coronal heating model. The magnetic boundary conditions are based on the full-Sun SDO/HMI synoptic map for CR2207, which represents radial magnetic fields measured at the solar surface as derived from line-of-sight magnetograms spanning 2018-08-06 to 2018-09-02. This map is then filled using a custom pole-filling procedure to add high-resolution flux patches at the poles while matching the net flux in this region as derived from long-term observations27. The map is then multiplied by 1.4 (to roughly match the HMI vector B observations for which the MAS-WTD model was calibrated) and re-gridded using a flux-preserving algorithm to the final MAS mesh. For this case, we use a relatively high-resolution mesh in the radial (r), co-latitudinal (θ) and longitudinal (ϕ) directions (364 × 344 × 504 in r, θ, ϕ) with a smoothly varying but non-uniform angular resolution that focuses the majority of mesh points around the active region and coronal hole complex that is the subject of this study (∆ϕ = 0.005 rad here, which transitions to ∆ϕ = 0.03 rad on the opposite side of the Sun). The next step is to smooth the re-gridded map to match the mesh resolution. Because of the relatively weak fields on the Sun at this time, we experimented with using less smoothing than previous simulations. This helps bring out the high-resolution complexity in the plasma state and magnetic field mapping by preserving the parasitic polarities of small-scale network flux and ephemeral regions (Extended Data Fig. 4). Lastly, for the heating model, we employ the same basic WTD approach and formulation as in ref. 27, but with updated parameters from the 2019 MAS-WTD eclipse prediction. This newer parametrization improves the electron density distribution in the middle corona as inferred by a comparison with white-light observables60.

The forward-modelled images and comparisons with SDO/AIA imaging data in Extended Data Fig. 5 are computed using the SDO/AIA v.10 effective areas and the CHIANTI 8.0.2 spectral synthesis package61,62. Observed emission in the AIA 171 Å channel, particularly from the quiet Sun and coronal hole regions, is probably dominated by small-scale transition region heating processes at the base of the solar corona63, which could be attributed to radiation from plasma below 1 MK. Our MHD model does not capture these detailed small-scale heating events and the associated transition region emission. This is why there are differences in the observed and synthesized AIA 171 Å emission maps in the Extended Data Fig. 5 (particularly in the coronal hole regions). This limitation, however, does not affect our inference and interpretation of the global magnetic topology.

The forward-modelled composite in Fig. 3c uses a combination of forward-modelled SDO/AIA 193 Å emission and Thompson scattered polarized brightness, pB (ref. 64). Before compositing, each image is radially filtered and lightly unsharp masked at a fixed spatial scale. The radial filtering is done by dividing the raw observable at each pixel by a one-dimensional monotonic profile in radius multiplied by a radial power law. The profile for each observable is determined simply by forward-modelling the same observable using one-dimensional radial profiles of the average temperatures and densities from a related simulation. The 3D volumetric renderings of the squashing factor28 in Fig. 3a,b are based on a volumetric 3D magnetic field mapping done at twice the simulation resolution in each direction (8× overall). The rendered images are produced in the same manner as described in ref. 27.