Subject: Mathematics

Extended Essay

Modelling of Kessler Syndrome

How can space debris in low Earth orbit be modelled using statistics?

Word Count: 2822

1. Introduction 1

2. Modeling Orbital Debris in LEO 2

2.1 Linear Regression Model - - - - - - - - - - - - - - - - - - - - - - - - - - 4

a. Residual Analysis of Linear Model - - - - - - - - - - - - - - - - - - 7

b. Hypothesis Testing - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 8

2.2 Piecewise Model - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 9

a. Cubic Regression Model - - - - - - - - - - - - - - - - - - - - - - - - 10

b. Piecewise Graph - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 13

3. Conclusion
Introduction

On October 4th, 1957, the Soviet Union successfully launched Sputnik 1, using their R-7

launch vehicle to deliver the first artificial satellite into low Earth orbit. As of 2011, “over four

thousand rockets have sent more than six thousand payloads into orbit, greatly improving the

world's capacity to retrieve, transmit, and share information.” (Chen, 2011) This already large

figure has continued to grow rapidly in the past decade, due to a renaissance of the space

industry. William Welser IV, the director of Engineering and Applied Sciences at the RAND

Corporation, described this phenomenon in a 2015 article. He writes that the democratization of

the space industry “has made outer space accessible to not only the global superpowers and large

multinationals, but to developing countries, start-ups, universities, and even high schools”

(Welser, 2015). SpaceX, a company utilising reusable rockets to maintain a rapid launch

cadence, has launched over 1,700 of their Starlink satellites as of August 2021 (Sheetz, 2021).

This satellite constellation provides broadband internet access to more remote regions of the

world. As our dependence on infrastructure in low earth orbit (LEO) increases, it is imperative

that we are mindful of any debris left undiscarded after a launch.

To counteract the effects of Earth’s gravitational pull, satellites in orbit must be moving

horizontally at extremely high speeds. As an example, the International Space Station, to which

the universal principles of orbital mechanics apply, orbits 400 kilometers above the Earth and

travels tangentially to the Earth’s surface at a rate of almost eight kilometers per second. It is

estimated that the average relative velocity of space debris and a satellite during a collision is ten

kilometers per second (Chen, 2011). These high speeds allow for very small particles to release

enormous amounts of energy in the event of a collision, breaking satellites into many smaller

pieces.

1
 If a satellite becomes obsolete or malfunctions, its orbit will slowly decay as a result of

atmospheric drag. There are a number of factors that influence the rate of orbital decay, such as

the mass and surface area of a satellite. At lower orbital altitudes (200km-600km), without

correction maneuvers, satellites’ orbits will decay in a timespan of about a few months to a few

years (NASA Goddard Space Flight Center). Geostationary orbits of around 36,000km in altitude

are useful for weather and communications satellites because they allow a satellite to remain in

one position above the Earth. Because of the negligible air density at this altitude, satellites will

remain there indefinitely. These defunct satellites, as well as spent rocket stages and other

orbiting objects, are collectively known as space debris.

On February 10, 2011 the Iridium-33 communications satellite collided with the retired

Cosmos 2251, a Russian military satellite, at nearly ten kilometers a second. From this one

collision, nearly 1,900 new pieces of space debris were created, revealing just how quickly these

figures can grow. According to a study conducted by NASA in early 2021, “Millions of pieces of

orbital debris exist in low Earth orbit (LEO) - at least 26,000 the size of a softball or larger that

could destroy a satellite on impact; ove 500,000 the size of a marble big enough to cause damage

to spacecraft or satellites; and over 100 million the size of a grain of salt that could puncture a

spacesuit” (NASA, 2021).

In 1978, NASA scientist Donald J. Kessler proposed a scenario in which low Earth orbit

becomes inaccessible due to the density of space debris. The ramifications of “Kessler

syndrome” would be catastrophic. Large networks of weather, communication, and global

positioning system satellites would be at risk of destruction, and humanity would be trapped on

Earth for hundreds or thousands of years. Even if all space launches were stopped today, it is

likely that the exponential growth of debris could render certain common orbits, such as

2
geostationary orbit, inaccessible. It is also entirely possible that the number of sequential

collisions required to reach the tipping point will never be reached. I will collect data and attempt

to determine a relationship between satellite launch rate and debris in orbit. I will then examine if

enough collisions are possible to result in Kessler syndrome becoming a reality.

Modelling Orbital Debris in LEO

To better understand how space debris has increased over the past sixty years of

spaceflight, it is beneficial to visualise this relationship using a graph. Figure 1 shows data

obtained from NASA’s LEGEND (LEO-to-GEO Environmental Debris model), which is an

evolutionary, three dimensional model used for the study of long-term debris environmental

projection. (National Aeronautics and Space Administration, 2019). Collisions with objects in

LEO larger than ten centimeters would result in a catastrophic loss of vehicle or crew (NASA’s

Orbital Debris Program Office, 2019). This graph features a clear positive trend, showing that

debris in LEO has increased at a relatively stable rate since 1958.

Figure 1

3
Linear Regression Model

Linear regression analysis is useful when attempting to model a relationship between the

independent and dependent variables. A linear equation can be fitted to the data in the form

𝑦 = β0 + β1𝑥, where β0is the y-intercept and β1 is slope. The following equations will be used

to arrive at the values of these two coefficients.

2 2
2 (Σ𝑥) 2 (Σ𝑦) Σ𝑥Σ𝑦
𝑆𝑥 = Σ𝑥 − 𝑛
𝑆𝑦 = Σ𝑦 − 𝑛
𝑆𝑥𝑦 = Σ𝑥𝑦 − 𝑛

𝑆𝑥𝑦 𝑆𝑦
𝑟 = 𝑆𝑥𝑆𝑦
β1 = 𝑟( 𝑆𝑥 )

To find β0, the point (𝑥, 𝑦) will be substituted into the equation written in point slope form.

𝑦 − 𝑦 = β1(𝑥 − 𝑥)

Calculations:

2 2
2 (Σ𝑥) (1953)
𝑆𝑥 = Σ𝑥 − 𝑛
= 81375 − 62
≈ 140. 9

2 2
2 (Σ𝑦) (546023)
𝑆𝑦 = Σ𝑦 − 𝑛
= 6916407559 − 62
≈ 45909. 5

Σ𝑥Σ𝑦 (1953)(546023)
𝑆𝑥𝑦 = Σ𝑥𝑦 − 𝑛
= 23561758 − 62
= 6362033. 5

𝑆𝑥𝑦 6362033.5
𝑟= 𝑆𝑥𝑆𝑦
= (140.9)(45909.5)
≈ 0. 984

𝑆𝑦 45909.5
β1 = 𝑟( 𝑆𝑥 ) =. 984( 140.9
) ≈ 320. 5

4
 Using the point (𝑥, 𝑦), or (31.5, 8806.8), the

formula for a trendline can be written using

point-slope form:

𝑦 − 8806. 5 = 320. 5(𝑥 − 31. 5)

𝑦 = 320. 5𝑥 − 1289. 25

Figure 2 displays the data used for

calculations shown on the previous page.

2 2
Columns “xy,” “𝑥 , " and “𝑦 , " as well as the

last summation row are included to make

evident where values used in calculations were

obtained.

For the sake of simplicity, the independent

variable will represent the number of years

after 1957. The dependent variable is

represented by the number of pieces of space

debris in low Earth orbit that are larger than ten

centimeters.

Figure 2

5
Figure 3

Figure 3 graphs a linear model for this data, using the equation

𝑦 = 320. 5𝑥 − 1289. 25. The slope of the linear model, or β0 predicts that the number of

objects larger than 10 cm in low Earth orbit increases by about 320.5 every year. A correlation

coefficient or r value of 0.984 indicates a strong positive linear relationship between objects in

LEO greater than ten centimeters in diameter and years since 1957. The negative y-intercept of

-1289.25 indicates there is error in this model, as there cannot be a negative number of objects in

orbit.

Visually, the linear model fits the data well for most x values. However, near x=50, the

data seems to deviate from the model in a patterned manner.

6
Residual Analysis of Linear Model

To examine the extent to which this model accounts for variation in the data, a residual

analysis can be performed by using the equation 𝑟 = 𝑥 − 𝑥0where r is the residual, x is the data

value, and 𝑥0is the value prediction by the linear model 𝑦 = 320. 5𝑥 − 1289. 25

Figure 4

Figure 4 graphs residuals from the linear model 𝑦 − 8806. 5 = 320. 5(𝑥 − 31. 5)

against years after 1957. The seemingly random grouping close to the x-axis for domain values

(4, 28) indicates that this linear model is a good fit for the measured data. This linear model fails

to account for variation in the data around x=42 to x=57 where a more distinct pattern can be

observed. This result suggests that a linear model does not fit the data in this region of the

domain. To further examine the fitness of this linear model, a hypothesis test can be conducted.

7
Hypothesis Testing

A hypothesis test will be useful in determining how atypical the result is, and if the

chance variation is within reason. To begin, a hypothesis 𝐻1must be established: If there is a

significant linear relationship between the independent and dependent variables, then obtaining

the sample data would not be too unlikely. The null hypothesis 𝐻0states that there is no

significant linear relationship between the two variables. At significance level, α, of 0.05, the

null hypothesis will be rejected if there is a less than 5% chance of obtaining data as extreme as

the sample data if the null hypothesis were true (La Trobe University).

𝐻0: β1 = 0

𝐻1: β1 ≠ 0

To find standard error, 𝑆 ,the following equation can be used where 𝑦𝑖 is the value of the
β1

dependent variable for observation i, 𝑦𝑖is estimated value of the dependent variable for

observation i, 𝑥𝑖 is the observed value of the independent variable for observation i, 𝑥 is the mean

of the independent variable, and n is the number of observations.

2
Σ(𝑦𝑖−𝑦𝑖) (61,823,128.71)
2

(𝑛−2) (60)
𝑆 = =
β1 Σ(𝑥𝑖−𝑥)
2
(19,855.5)
2

𝑆 = 402.0
β1

8
 Test statistic t is defined by the following equation:

β1 320.5
𝑡= 𝑆
= 402.0
= 0. 7973
β1

This test statistic can then be compared to a critical value corresponding with a

significance level, α, of 0.05, and 60 degrees of freedom, df. If the absolute value of test statistic

t is greater than the critical value, the null hypothesis is rejected. If t is less than the critical value,

the null hypothesis is not rejected (Stat Trek). A test statistic of 1.671 was obtained from a t

distribution table in a statistics textbook.

|0. 7973| < 1. 671

Therefore, we fail to reject the null hypothesis, and can conclude that the sample data

does not provide sufficient evidence to support that there is a relationship between the number of

years after 1957 and the quantity of debris in low Earth orbit greater than ten centimeters in

diameter.

Piecewise Model

When examining the linear model near domain values of x=50, it was noticed that the

data began to deviate significantly from the values predicted by the linear regression. The pattern

observed on the residual analysis graph (Figure 4) indicates a non-linear relationship might better

model this data. To account for this, the data can be modeled using a piecewise function. A linear

model will be used when 𝑥 < 50, as the random pattern observed in the residual analysis

indicates this model is appropriate. A cubic model will be used when 𝑥 ≥ 50, as the data first

increases, then decreases, and then increases again.

9
Cubic Regression Model

Serving a similar function to its linear counterpart, cubic regressions are useful when

modeling a cubic relationship between an independent and dependent variable. A cubic equation

3 2
in the form 𝑦 = 𝑎𝑥 + 𝑏𝑥 + 𝑐𝑥 + 𝑑 can be fitted to the data. Coefficients a, b, c, and d, will

be calculated with matrices using the least squares method.

Therefore, the data can be modeled by the following cubic equation:

1 3 3 2 5 6
𝑦 = (1. 22⋅10 )𝑥 − (2. 06⋅10 )𝑥 + (1. 17⋅10 )𝑥 − (2. 17⋅10 )

10
Figure 5

1 3 3 2 5 6
𝑦 = (1. 22⋅10 )𝑥 − (2. 06⋅10 )𝑥 + (1. 17⋅10 )𝑥 − (2. 17⋅10 )

2
𝑅 = 0. 924

The cubic equation graphed in Figure 5 seems to more accurately represent the data in the

2
domain interval 50 ≤ 𝑥 ≤ 62. A coefficient of determination or 𝑅 value of 0.924 was obtained

using a graphing calculator and indicates a strong cubic relationship between the independent

and dependent variables. To further test the fitness of this model, a residual analysis can be

performed. The process to create a residual plot for a cubic model is identical to that of the linear

model. Residuals can be calculated using the equation 𝑟 = 𝑥 − 𝑥0where r is the residual, x is the

data value, and 𝑥0is the value prediction by the cubic model

1 3 3 2 5 6
𝑦 = (1. 22⋅10 )𝑥 − (2. 06⋅10 )𝑥 + (1. 17⋅10 )𝑥 − (2. 17⋅10 ).

11
Figure 6

Figure 6 graphs residuals from the cubic model against years after 1957. The pattern

appears almost entirely random, indicating a good fit for this model. The cubic function

1 3 3 2 5 6
𝑦 = (1. 22⋅10 )𝑥 − (2. 06⋅10 )𝑥 + (1. 17⋅10 )𝑥 − (2. 17⋅10 ) will serve as the second

sub-function in the piecewise model. As previously mentioned, the first sub-function will come

from a linear regression model for x values less than 50.

𝑦 = β1𝑥 + β0

𝑦 = 273. 8𝑥 − 435. 5

r = 0.976

12
Figure 7

Figure 7 graphs the piecewise function shown above. This model is a better fit for the

data than the solely linear model derived earlier, as it accounts for the patterned residuals from

50 ≤ 𝑥 ≤ 62.

Conclusion

Several different methods of modelling data have been discussed in previous sections, as

well as methods for testing accuracy of those models. Regressions allow for the construction of

both linear and non-linear models which can be fitted to a set of data points. Residual analysis

was conducted to test if those models were appropriate for the given data. A hypothesis test

determined there was a less than 5% chance of obtaining data as extreme as the sample data if

the relationship was truly linear.

13
 In terms of relating these findings to real-world applications, the sample data used was

limited in two significant ways. By only accounting for objects greater than ten centimeters in

diameter, all smaller objects were excluded. Smaller objects can still pose great risk to other

1 2
structures in orbit due to the nature of kinetic energy, represented by the equation 𝐾𝐸 = 2
𝑚𝑣 .

Simply put, because it is squared, velocity has a more significant impact on kinetic energy than

mass. The most dangerous pieces of space junk are too small to be tracked. Consequently,

collisions with these small but fast moving objects cannot be predicted or knowingly avoided.

Space debris included in sample data is in low Earth orbit, or orbits of less than 2,000

kilometers in altitude (National Aeronautics and Space Administration). While collisions with

space debris at these lower altitudes certainly pose a risk to human spaceflight and some

satellites, a far more concerning scenario is one in which geostationary orbit becomes

inaccessible. This orbit of about 36,000 kilometers is essential for weather, communication, and

global positioning system satellites.

Statistical analysis is useful when predicting the nature of growth of space debris.

14
 List of Sources

La Trobe University. “Hypothesis Tesing.” La Trobe University Library, 13 May 2020,

https://www.latrobe.edu.au/. Accessed 15 October 2021.

NASA. “NASA's Efforts to Mitigate the Risk Posed by Orbital Debris.” 2021,

https://oig.nasa.gov/docs/IG-21-011.pdf. Accessed 3 April 2021.

NASA Goddard Space Flight Center. “The Hubble Space Telescope Servicing Missions.”

Astrophysics Science Division, 2021,

https://asd.gsfc.nasa.gov/archive/hubble/missions/sm4.html. Accessed 10 October 2021.

National Aeronautics and Space Administration. “LEGEND : 3D/OD Evolutionary Model.”

Orbital Debris Program Office, 2019,

https://orbitaldebris.jsc.nasa.gov/modeling/legend.html#:~:text=LEGEND%20is%20a%2

0full%2Dscale,long%2Dterm%20debris%20environment%20projection. Accessed 18

August 2021.

National Aeronautics and Space Administration. “LEO Economy FAQs.” NASA, 19 November

2021, https://www.nasa.gov/leo-economy/faqs. Accessed 23 November 2021.

Sheetz, Michael. “SpaceX says Starlink has about 90,000 users as the internet service gains

subscribers.” CNBC, 3 August 2021,

https://www.cnbc.com/2021/08/03/spacex-starlink-satellite-internet-has-about-90000-use

rs.html. Accessed 8 August 2021.

Stat Trek. “Hypothesis Test for Regression Slopes.” Stat Trek, 2021,

https://stattrek.com/regression/slope-test.aspx. Accessed 15 October 2021.

15
Welser, William, and Dave Baiocchi. “The Democratization of Space: New Actors Need New

Rules.” Foreign Affairs, June 2015,

https://www.foreignaffairs.com/articles/space/2015-04-20/democratization-space.

16