A geomagnetic storm is a period of hours to days in which energy from the solar wind enters the magnetosphere fast enough to build an intense ring current and to depress the horizontal magnetic field at low latitudes. The accepted definition ties the storm to a sustained interval of southward interplanetary magnetic field (Gonzalez et al., 1994), in keeping with the reconnection picture first set out by Dungey (1961). The practical effects include induced currents in power grids, errors in satellite navigation, increased drag on spacecraft and damage to electronics, and their economic weight has been set out in a widely cited workshop report (National Research Council, 2008).

Forecasting rests on observing the chain of events that precedes the storm. That chain starts at the Sun and ends in the ionosphere, and at present it is watched by very few instruments. The solar wind arriving at Earth has for decades been measured from the neighbourhood of the first Lagrange point, L1, about one and a half million kilometres sunward, by spacecraft such as the Advanced Composition Explorer (Stone et al., 1998). An international road map for space weather research named the fragility and sparseness of this observing system as a leading concern (Schrijver et al., 2015).

Small satellites offer a different way of building the system: many modest spacecraft in place of a few elaborate ones. This review asks what such constellations could contribute to the monitoring of storm precursors, what has been shown in orbit, and what stands in the way.

The review is narrative. It draws on the standard literature of solar-terrestrial physics, on mission description papers for multi-spacecraft and small satellite projects, and on assessments published by scientific academies. It does not claim to be exhaustive, and the selection reflects a judgement about which sources best represent each topic. The term small satellite is used broadly for spacecraft from CubeSats of a few kilograms up to microsatellites of roughly a hundred kilograms. No new analysis of data is presented.

The word precursor is used loosely in the field. It is helpful to order the candidates by where they are observed, since position along the Sun-Earth line fixes the warning time. Table 1 gives such an ordering, and the paragraphs below follow it.

Space physics has used small formations for a quarter of a century. The four Cluster spacecraft flew as a tetrahedron so that spatial gradients could be separated from changes in time (Escoubet et al., 2001). The five THEMIS probes were placed on orbits that line up along the magnetotail to time the sequence of events in substorms (Angelopoulos, 2008). Magnetospheric Multiscale took the tetrahedron down to electron scales (Burch et al., 2016), and the three Swarm satellites survey the geomagnetic field and its current systems from low Earth orbit (Friis-Christensen et al., 2006). These are science missions with exacting instruments, and none was designed for continuous monitoring, but they established the methods by which several spacecraft are combined into one measurement.

The clearest precedent for monitoring with many modest sensors came from an unexpected quarter. The engineering magnetometers on the Iridium communications satellites, more than sixty spacecraft in polar orbits, were shown to detect the magnetic signatures of the field-aligned currents (Anderson et al., 2000), and methods were developed to assemble their readings into global maps (Waters et al., 2001). The instruments are far less precise than a science magnetometer. Their number and distribution compensate, and the result is a view of the current systems that no single precise spacecraft can give.

The CubeSat standard, built around a ten-centimetre cube, was introduced as an educational device (Heidt et al., 2000). Within about a decade such spacecraft were returning publishable measurements. A student-built CubeSat measured energetic electrons and protons in low Earth orbit through geomagnetic storms (Li et al., 2013), and a pair of CubeSats carried electric field probes and magnetometers to study the storm-time ionosphere (Fish et al., 2014). A review of the platform's capabilities documents steady gains in pointing, power, data rate and propulsion (Poghosyan and Golkar, 2017), and a study by the United States National Academies concluded that CubeSats can deliver high-value science, with constellations named as a particular strength (National Academies of Sciences, Engineering, and Medicine, 2016).

Three difficulties recur in the literature for the instruments that matter here. The first is magnetic cleanliness. A small spacecraft places reaction wheels, batteries and current loops within centimetres of the sensor, so a boom, a gradiometer arrangement of two sensors, or careful characterisation of spacecraft fields is needed to reach the accuracy that solar wind work requires. The second is the thermal and radiation environment beyond low Earth orbit, for which most commercial components were not designed. The third is communications. A CubeSat near L1 cannot return continuous real-time data with a small antenna and a few watts unless the ground segment is correspondingly large.

Proposed architectures can be arranged by distance from Earth, as in Figure 1. Each layer answers a different weakness of the present system, and the layers are complementary and not alternatives.

Continuous monitoring is an operational requirement and differs from a science campaign. It needs replacement spacecraft on a schedule, real-time downlink, and data whose quality is known at the moment of use. Small satellites fit the first need well, since short lifetimes are tolerable when replacements are cheap. The second is a matter of ground infrastructure. The third is the least developed. A constellation of inexpensive magnetometers has as many offsets, gains and alignment errors as it has sensors, and these drift. Science constellations solve the problem with extensive calibration effort per spacecraft; an operational constellation of dozens would need automated cross-calibration between neighbours and against a small number of reference instruments.

Use of the data is the other half of the subject. Global magnetohydrodynamic models driven by upstream measurements, such as the Space Weather Modeling Framework (Tóth et al., 2005), currently take a single time series as their boundary condition. Multi-point upstream data would let that boundary condition vary across the inflow, and dense low-altitude data would give the models something to be corrected against while a storm develops. Neither capability is routine.

The review points to four gaps. There is no agreed statement of how accurate a small upstream magnetometer must be for its data to improve a forecast, and without one the engineering trade between boom length, spacecraft cleanliness and number of spacecraft cannot be closed. Studies that withhold or degrade real upstream data and measure the loss of forecast skill would supply it. Cross-calibration methods for large sets of low-cost sensors need to be developed and published as openly as the instruments themselves. The value of the proposed early precursors, energetic particles and cosmic ray anisotropy, has been argued from selected events, and their false alarm rates over long periods are not well established. And the organisational question of who operates and replenishes a monitoring constellation over decades has received less attention than its design.

Storm precursors are observed today at too few points and with too little redundancy. The literature reviewed here supports a measured view of what small satellites can do about it. In low Earth orbit they are ready, and the principle that many modest sensors can outperform one excellent sensor for mapping current systems has been shown in practice. In the outer magnetosphere they are plausible. Upstream of Earth, where the warning is earned, they face real constraints of magnetic cleanliness, communications and station-keeping, and they are best seen for now as companions to a conventional monitor. Across all layers the limiting factor is likely to be calibration and continuity of operations.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.