Today’s blog post was sent in by Stijn Lemmens, an analyst working in ESA’s Space Debris Office at ESOC in Darmstadt, Germany.
We are now within days of the projected Tiangong-1 reentry. One question that ESA’s space debris team has been receiving frequently is, why is the actual reentry date and time remain so uncertain – even as the predicted time window has been shrinking steadily (and is now considerably smaller) since our original estimates were posted here back in January.
Animation of successive Tiangong-1 radar images acquired by the Tracking and Imaging Radar (TIRA) operated by Fraunhofer FHR, Wachtberg (near Bonn), Germany. Actual images shown at left together with an animated computer model to illustrate motion of the craft. TIRA data are being used in the ESA-hosted reentry tracking campaign. More details via the FHR website.
As can be seen in our regularly updated time-window plots, the uncertainty on the predicted reentry epoch (time and date) – in other words, on the remaining orbital lifetime – has been running at about plus or minus 20%, and the actual predicted reentry date has been subject to quite some variability.
This variability is largely driven by inaccuracy in forecasting the density of the lower layers of the thermosphere. This is the upper level of Earth’s atmosphere, producing the drag which is responsible for the decay of the orbit, which in turn depends on a number of atmospheric models and actual space weather activity.
These atmospheric models are applied to many aspects of spaceflight.
They provide a fast algorithm to calculate the temperatures, densities and chemical composition of the atmosphere at any given time and position around the Earth. These values are indispensable for the orbit-determination and orbit-prediction aspects of any mission. The models are based on the ‘energy balance’ physics of the atmosphere, just like climate models for Earth science, and are calibrated based on satellite measurements to provide averaged solutions.
Four models are commonly used for spaceflight. These are known as:
In order for these models to do their orbital determination and prediction jobs, they require so-called ‘proxies’ to describe how space weather affects spaceflight, e.g. how much energy the Sun is sending to Earth and how these radiation and particle streams interact with the near-Earth environment.
Commonly used proxies include the energy flux received in the radio spectrum (the ‘F10.7 Index’), the Sun spot number (SSN), and current level of disturbance in Earth’s geomagnetic field (‘Ap Index’). More detailed information can be found via the Space Debris Office technical website – see Section 4. The variations in space-weather proxies result in large changes in the predicted reentry time.
Foreseeing the future
Statistical and machine learning techniques can be used to make such forecasts on the time scale of days to years, but in reality, it has proven to be difficult to capture this in predictive models. Forecasts on shorter timescales, days to hours, tend to be a little better but require essentially a large dose of supercomputing and human ingenuity.
Furthermore, when it comes to spaceflight in the lower thermosphere, dedicated satellite observations that can be used to calibrate the reentry models (coming from missions such as ESA’s low-orbiting GOCE satellite) are rare and hence all models need to extrapolate to a large degree.
The sum of these effects – along with the core fact that even decaying objects in orbit are still in motion around Earth at upwards of 7.8 km per second – is what makes uncontrolled reentries so difficult to forecast, and imply that even seven hours prior to reentry, a given prediction can still have an uncertainty on the final impact point of one full revolution around the Earth.
Breaking up is easy to do but hard to predict
Predictions closer to reentry are possible, but only when the object is observed, which cannot be guaranteed as the object would need to pass over a sensor – like a telescope or radar. Even assuming perfect knowledge, any potential fragments generated following reentry breakup would spread out somewhat randomly over a ground track on average hundreds of kilometres long and a few tens of kilometres wide (which is why the risk of hitting a person on the ground is very, very low).
The interplay between the atmosphere models and other effects such as the object’s relative facing position, or attitude (which increases or decreases its surface area exposed to the atmosphere), cannot be unambiguously decoupled based on trajectory data alone (see more details, for example, here).
In a nutshell: a drag coefficient as derived from space surveillance for Tiangong-1 is the result of averaging the observed decay behaviour of the space station over days against an averaged atmosphere model. It is not not an instantaneous measurement. Note also that radar image data (see animation above) have so far confirmed a continuing and increasing rotation rate for Tiangong-1.
Direct observations of the attitude of re-entering objects from the ground are complicated by the fact that continuous observation times are limited to generally ten minutes or less (before the object orbits out of sight below the horizon).
Hence lower rotation periods can only be extrapolated at best. Sparse sensor coverage generally leads to many gaps, which need to be covered by predicting the attitude evolution during the object’s passage from one sensor to another, which again requires a significant amount of detailed knowledge of the object, especially when the atmosphere has a significant effect, as it does during re-entries.
Data sources: radars, ‘scopes and lasers
A standard ‘triumvirate’ of sensors is used to tackle this task for targets in a variety of orbits: radar, optical telescopes and lasers; all have their benefits and drawbacks with regard to attitude determination. Radars can provide frequent coverage but the attitude needs to be reconstructed from the Doppler information where the response is difficult to interpret – this applies to both Radar Cross Section (RCS) and Inverse Synthetic Aperture (ISAR) techniques.
Optical telescopes unfortunately require propitious illumination conditions on the target, but are generally suited for establishing the long-term trending of the rotational motion. And lasers require either high power or reflective surfaces on the target, but when combined with accurate positional knowledge and surface information these can enable the reconstruction of the full attitude motion with a minimum of ambiguity.
The best way of reconstructing shifting attitude is to combine data from all three sensor types to resolve the individual ambiguity. ESA is pioneering this field of ‘fused attitude surveillance’ as well as further developing sensor capabilities. An introduction to the subject is available here.