Today, satellite operators—including NASA—must launch and then maintain their satellites in a risky environment that is the result of a combination of threats. One of those threats is the presence of natural and artificial debris in the near-Earth space environment, and the increasing potential for high-impact events (e.g. on-orbit collisions) makes spacecraft even more vulnerable (Box 9.1).
Mitigating the risk of debris events to operational satellites requires different approaches based on the type of debris. Protecting against natural debris (meteoroids) must be addressed via passive techniques (e.g., shielding), since it is currently impossible to track these objects and predict their course to enable an operator to take evasive action. The same is true for small artificial debris, which cannot be tracked with today’s space surveillance networks.
The U.S. Space Surveillance Network (SSN) can track objects only down to about 10 cm in LEO (below 2,000 km) and 1 m in geosynchronous Earth orbit (GEO). The Joint Space Operations Center (JSpOC) currently tracks more than 22,000 objects larger than these thresholds. NASA, however, estimates that half a million objects larger than 1 cm reside in LEO—the size object that some portions of the International Space Station (ISS) are shielded to withstand. Typical satellites are not shielded even this well, and untracked objects larger than 5 mm can cause substantial damage, with relative velocities as high as 15 km/s in LEO, but with an average relative velocity in LEO of about 10 km/s. The relative velocity for collisions in GEO is much smaller, with an average of 500 m/s. Until better space surveillance tracking capabilities are operational, the threat of collisions with these untracked objects cannot be mitigated. An upgrade to the current U.S. Air Force (USAF) Space Fence is planned, and Phase 2 is in progress. The new fence will more accurately track objects as small as 5 cm in diameter in LEO.
To mitigate the risk of collision with a cataloged object, NASA performs two processes. Conjunction assessment risk analysis (CARA; at GSFC for robotic spacecraft)/conjunction assessment (CA; at JSC for the ISS and the space shuttle) is the process performed for mitigating the risk of an operational satellite colliding with a cataloged object, and launch collision avoidance (COLA) is the process performed to try to prevent a collision with a cataloged object during launch.
For objects that can be tracked, it is possible to make decisions to maneuver operational satellites (if they are able to maneuver) to reduce the chance of collision with a cataloged space object. CARA/CA is the effort to
On-Orbit Collision of Iridium 33 and Cosmos 2251
On February 10, 2009, the satellite communications company Iridium lost contact with one of its spacecraft, Iridium 33. Earlier that day, Iridium had received a prediction of a close approach of 584 m (1,916 ft) between Iridium 33 and another orbiting spacecraft, the non-operational Russian communications satellite Cosmos 2251. Iridium had received close approach reports before, and the one on February 10 was not particularly alarming or deemed a “top predicted close approach” compared to other predicted close-approach events for that week. Nevertheless, at the time the close approach was predicted to occur above northern Siberia, Iridium abruptly stopped receiving telemetry from its spacecraft.
The destruction of Iridium 33 was confirmed when the U.S. Space Surveillance Network (SSN) detected debris clouds in the orbits of both Iridium 33 and Cosmos 2251, marking the first payload-to-payload collision in the history of spaceflight. (See Figure 9.1.1.) The collision of Iridium 33 and Cosmos 2251 added an additional 2,181 trackable pieces of debris to the approximately 19,000 objects larger than 10 cm already in orbit in 2009.1 (See Figure 9.1.2.) Today, more than 22,000 pieces of debris are being tracked,2 plus an estimated population of approximately 500,000 particles between 1 and 10 cm, and more than tens of millions of particles smaller than 1 cm orbiting Earth.3 Some of the debris from that collision has reentered Earth’s atmosphere: as of May 2011, the SSN had cataloged 547 pieces of debris associated with Iridium 33 (down from 594 originally) and 1,488 pieces of debris associated with Cosmos 2251 (down from 1,587 originally).4 Particles smaller than 10 cm are very difficult to track reliably with current capabilities.5
Iridium created its own collision analysis process during the initial development and launch phase of the Iridium constellation of satellites. Although the data available for tracking a satellite’s position are the best the U.S. government can offer, those data were not designed to be used for conjunction analysis, although they are being used for that purpose today. Prior to the Iridium–Cosmos collision, Iridium had never made an on-orbit maneuver with one of its satellites in response to a close approach prediction. Between February 2009 and February 2011, Iridium made 41 collision mitigation maneuvers based on 46 close approach warnings;6 however, all of these actions were integrated into normal constellation maintenance actions and so had little impact on Iridium operations.
2 Government Accountability Office, Space Acquisitions: Development and Oversight Challenges in Delivering Improved Space Situational Awareness Capabilities, Report to the Subcommittee on Strategic Forces, Committee on Armed Services, House of Representatives, GAO-11-545, Washington, D.C., May 2011.
5 J. Lyver, NASA, presentation at the Workshop to Identify Gaps and Possible Directions for NASA’s Micrometeoroid and Orbital Debris Programs, March 9, 2011, National Research Council, Washington, D.C.
6 J. Campbell, Iridium, presentation at the Workshop to Identify Gaps and Possible Directions for NASA’s Micrometeoroid and Orbital Debris Programs, March 9, 2011 National Research Council, Washington, D.C.
FIGURE 9.1.1 View of Iridium 33 and Cosmos 2251 orbits and debris from each spacecraft 180 minutes after their collision with one another. What should be one dot for each orbit is now thousands. SOURCE: Courtesy of T.S. Kelso, CelesTrak.com.
FIGURE 9.1.2 Screen shot from AGI Viewer 9 file of current Iridium constellation and collision debris clouds. SOURCE: Courtesy of T.S. Kelso, CelesTrak.com.
determine the orbits of all known objects relative to the operational satellite population in order to facilitate decision making regarding whether to take preemptive action to avoid potential collisions. To perform CARA/CA, a satellite operator must know not only the orbit of their satellites and the potential collision threats, but also the uncertainty associated with those estimated orbits. This information can then be used to determine future close approaches and flag those that exceed a certain threshold (e.g., range at closest approach or probability of collision). When a close approach is identified that exceeds the operator’s threshold, the operator must know whether the potential threat is another operational satellite or a piece of debris (such as a spacecraft fragment or a dead satellite). If the potential threat is an operational satellite, which may be capable of maneuvering, the operator will also need to know how to contact the other operator to collaborate on a course of action and ensure that they do not unwittingly make the situation worse.
The computation of the probability of collision, Pc, is a critical part of CARA because it is that value that is the primary basis for the decision about whether or not to make a collision avoidance maneuver. The two key assumptions in the calculation of Pc for objects in LEO are that (1) the relative motion of the two objects at conjunction is rectilinear, and (2) the position uncertainty distributions of the objects at conjunction are Gaussian. The first assumption is valid as the time interval of concern is about 0.25 s. The uncertainty (covariance) at conjunction is obtained by propagating the covariance from epoch (i.e., the time when covariance was calculated) to the conjunction time. This propagation is based on a linearization of the equations of motion about the reference (estimated) orbit. CARA is performed for conjunctions 6 to 7 days into the future. There is evidence that the Gaussian assumption may not be valid beyond 2 to 3 days into the future1 as a result of neglected nonlinearities in the propagation of the covariance. The quantitative impact of this non-Gaussian behavior on Pc is not known.
NASA currently performs CARA for about 50 missions as part of normal satellite operations and CA for the ISS and previously for the space shuttle. Although NASA and its partners (e.g., ESA) have the best data for those 50 satellites—including planned maneuvers—they do not have an independent means to perform orbit determination for the thousands of other objects that might pose a collision threat to their satellites. These data are available only by means of the JSpOC, but the data are not released to NASA or other satellite operators (for several reasons, including national security), with the exception of two-line element sets (TLEs), which are released only for the public catalog (more than 16,000 objects) and not for objects that have an unknown source (another ~7,000 objects, which still pose a threat).
Although TLEs are available to NASA and the public, several problems exist with these data. First, to propagate these data correctly, access is required to the same orbital model (Simplified General Perturbations 4 or SGP4) used to generate the data. However, the Air Force Space Command (AFSPC) has marked this model as “Export Controlled” and has not released it to the public since 1980. Without the changes made to this model since then, it is possible to have errors on the order of 1,000 km in GEO.
Second, although these data are generally good enough for their intended purpose to maintain tracking by the U.S. Space Surveillance Network (SSN), the associated uncertainties (which are not quantified or provided with the TLE data, but are on the order of hundreds of meters to kilometers in LEO and kilometers to tens of kilometers in GEO) contribute to high false-alarm rates and discourage efforts to perform CARA. The large uncertainties are the result of the nature of the SSN tracking that does not take into account maneuvers when computing the TLEs. It is not unusual for it to take a week or more to detect maneuvers and update the associated orbits. It is also common to see objects cross-tagged (switched) within the GEO population, since the noncooperative tracking has no independent way to know which object is which.
The JSpOC also has special perturbations (SP) data that are generally considered to have higher accuracy, but that are collected by the same network and processed with the same algorithms but using a higher-fidelity force model. These data do not consider the effects of maneuvers and can be impacted by cross-tagging. Also, although SP data do include covariance information, it is generally agreed that the amount is too small (meaning that the uncertainty is large) as a result of the way the SSN sensor observations are collected and sent to the JSpOC for
1 C. Sabol, T. Sukut, K. Hill, K.T. Alfriend, B. Wright, and P.W. Schumacher, “Linearized Covariance Generation and Propagation Analysis via Simple Monte Carlo Simulations,” Paper No. AAS 10-134, AAS/AIAA Space Flight Mechanics Conference, San Diego, Calif., February 14-17, American Astronautical Society, Springfield, Va., 2010.
processing, and the method used for determining the orbits from the observations. In addition, the probability of collision, Pc, is very sensitive to errors in the uncertainty near the threshold of the decision to maneuver, 10–4<Pc<10–3. An error of a factor of two in the uncertainty can change the probability of collision by at least two orders of magnitude.2 Release of the SP data occurs under very limited circumstances, and it would have to be released in near-real time for it to be useful to NASA. Consequently, NASA must rely on the JSpOC to screen its satellites and report close approaches based on what the JSpOC deems reasonable. Currently, the JSpOC looks out 5 to 7 days ahead, but reports only close approaches within 1 km total and 200 m radial for LEO objects and 5 km total for GEO objects 3 days ahead. The 3-day time horizon leaves little time for an operator to attempt to get better tracking data, make the necessary decision regarding collision avoidance measures, and then plan and conduct a maneuver if necessary.
The JSpOC provides only a state vector and covariance in its Orbit Conjunction Messages (OCM) to NASA when the JSpOC identifies a conjunction, but it is unclear where these data come from. Originally, the source was the SP data used in the analysis. However, even when it takes in ephemeris data from NASA, the JSpOC does not provide that information back in the OCM. Therefore, when the JSpOC does not use ephemeris data with maneuvers, it can report false conjunctions or miss real ones. The overall lack of transparency in the data and underlying processes serves to undermine confidence in the entire CARA process and increases the risk of collision for NASA missions.
NASA currently has little or no control over these data restrictions or the JSpOC processes used in the conjunction assessments. One potential recourse would be to participate in the non-profit Space Data Association’s Space Data Center (SDC), a collaborative effort in which 20 commercial and civil operators now share data on 300 satellites with the goal of improving the safety of space operations, including improving CARA. All SDC participants provide ephemeris for their satellites—including planned maneuvers—to be used in conjunction screening. Operator ephemeris data has been shown to be an order of magnitude more accurate than the SSN data (TLE data), due to its use of active tracking observations. That reduces the uncertainty volume by three orders of magnitude and helps make the problem much more manageable. In addition, the inclusion of maneuvers and lack of cross-tagging (since the satellite operators know which satellite is which) greatly reduce the number of false alarms and the chance of unnecessary or misguided actions. However, these data are available only for the satellites of the companies participating in the SDA. Hence, participation in the SDA can improve CARA only with the satellites participating in the SDC, but even with these satellites a probability of collision cannot be performed because covariance information is not provided.
It is important to understand how risk guides CARA efforts. Although the current CARA process and the data on which it relies can be improved, only a regular exercise of the process and identification of the shortcomings will make it an effective decision-making tool. As emphasized in An Introduction to Factor Analysis of Information Risk, “You can’t effectively and consistently manage what you can’t measure, and you can’t measure what you haven’t defined.”3 Without a clear understanding of how risk is defined, it cannot be appropriately managed.
Risk is the combination of the probability of an event with the consequences of that event. For example, the probability of a satellite being hit by a 1-mm piece of debris may be higher than its being hit by another satellite, but in the latter case the consequences for the survival of the satellites and the effects on the space environment would be worse (as demonstrated in the Iridium–Cosmos collision). Likewise, the consequences of not attempting to avoid a predictable collision could cause serious negative reactions among stockholders of a satellite operator or among U.S. taxpayers and Congress, who fund NASA’s budget. Reactions would be particularly bad if it were
2 K.T. Alfriend, M.R. Akella, D.-J. Lee, M.P. Wilkins, J. Frisbee, and J.L. Foster, Probability of collision error analysis, International Journal of Space Debris 1:21-35, 1999.
3 J.A. Jones, An Introduction to Factor Analysis of Information Risk (FAIR), draft, Risk Management Insight, Columbus, Ohio, 2005, available at http://riskmanagementinsight.com/media/documents/FAIR_Introduction.pdf, accessed July 20, 2011.
discovered that the problem were manageable at low cost, but that the satellite operator chose to ignore the risk or that its efforts were confounded by unnecessary bureaucratic obstacles.
NASA and the U.S. National Space Policy
Under the new 2010 National Space Policy, NASA is required to take efforts to preserve the space environment—specifically, to “pursue research and development of technologies and techniques … to mitigate and remove on-orbit debris, reduce hazards, and increase understanding of the current and future debris environment.”4 NASA can and must perform CARA on its satellites to reduce the hazard to other satellites or the risk of creating more debris. NASA may take collaborative action—with organizations like the new Space Data Association—to promote transparency in CARA, promote best practices, and further reduce the likelihood of avoidable collisions that could impact national or economic security and further degrade the near-Earth space environment. These actions are all consistent with U.S. national space policy and would help NASA to more effectively and efficiently conduct CARA to support NASA space missions.
Finding: The computation of the probability of collision for use in an assessment of risk requires the uncertainty parameters in the orbits of the two objects at conjunction, and assumes that these uncertainties are represented by a Gaussian distribution. Research has shown that the uncertainty distribution typically is Gaussian for several days, but when propagating for more than 2 to 3 days it may no longer be Gaussian. In addition, the uncertainties provided by the JSpOC are known to be usually too small, and the probability of collision can be very sensitive to errors in the size of the uncertainty.
Recommendation: NASA should develop a research plan for (1) assessing the impact of inaccuracy in the uncertainty on computations of the probability of collision and on the ensuing risk assessment, and (2) improving the accuracy of the computation of the probability of collision, given the presence of these uncertainty errors.
Launch collision avoidance (COLA) (see item 16 in Box 12.1, Chapter 12) is the process of actively screening for potential collisions between a launch vehicle and known, tracked, on-orbit objects from liftoff through the end of the launch phase and subsequently taking action to avoid any unacceptable conjunctions. Range safety COLA applies to crewed or crewable space objects, and mission assurance COLA applies to uncrewed objects. COLA is performed by the Launch Services Program (LSP) at Kennedy Space Center. COLA is not required by LSP; it is performed at the request of the customer. GSFC has requested COLA for all of its missions for the primary purpose of protecting orbiting assets, not the satellite being launched. USAF instructions mandate the operational implementation of launch conjunction assessment and collision avoidance at AF-controlled ranges, and avoidance of crewed conjunctions is mandatory for safety.
LSP obtains launch COLA support from The Aerospace Corporation, which uses a probability-based tool called “Collision Vision” that has extensive heritage on NASA, USAF, and NRO (National Reconnaissance Office) launches. The use of the tool requires a trajectory ephemeris and covariance at key trajectory event milestones from the contractor. The launch COLA analysis is performed for all separated bodies up through 100 minutes after separation. The primary issues with COLA include the following:
4 National Space Policy of the United States of America, June 28, 2010, available at http://www.whitehouse.gov/sites/default/files/national_space_policy_6-28-10.pdf, accessed July 6, 2011.
• How long should screening last? The AFI 91-217 requires launch COLA screening until the objects have entered the space object catalog,5 and this process can take several days. However, due to the large uncertainties in the launch dispersions (deviations from a planned trajectory), the COLA methodology cannot be reasonably used except for a few orbits after launch. This leaves a large gap in time.
• Does one use probability-based screening or miss-distance screening? The probability of collision is a function of the miss distance, the direction of the miss distance relative to the trajectory, and the uncertainty. Thus, one cannot select a probability of collision and obtain a minimum miss distance. The problem with probability-based screening is that the large dispersions with the launch vehicle trajectories usually result in a probability of collision of less than 10–5, and this risk is much lower than other risks usually associated with launch activities. Of 676 conjunctions analyzed by NASA, only 1 percent had probabilities of collisions greater than 10–5. LSP currently uses the probability-based approach.
Finding: The large uncertainties in the launch dispersions (deviations from a planned trajectory) that yield a probability of collision of less than 10–5 translate to a very low return on investment in launch collision avoidance (COLA), and funds could probably be used more effectively in some other area of debris mitigation. However, in the event of a collision during launch, the political realities of potentially having done nothing probably mean that the use of COLA needs to continue, especially for crewed launches.
5 B. Beaver, “Launch Collision Avoidance for MASA ELV Missions,” presentation to the Committee for the Assessment of NASA’s Orbital Debris Programs, January 19, 2011, National Research Council, Washington, D.C.