Investigating the Coastline Paradox 1: Introduction As a runner and a frequent traveller, I often research whether the islands I am travelling to are small enough that I could run around them in a reasonable amount of time. I was born in Bali, in the country of Indonesia, an archipelagic country filled with thousands of little islands. Whilst we do not have an impressive landmass, my country has a lot of coastlines. Whilst area is easily measured in acres, trying to systematically measure a coastlines length appears to be more difficult than one may first assume. From my research I discovered that the value for various coastline’s lengths can vary significantly from site to site For Indonesia, the CIA World Factbook measures the length as 54,720km[13] whilst the World Resources Institute states it as 95,181km[14], a massive difference. At first I wanted to investigate how geographers approach measuring the length of a coastline length. Does someone just run around the island with a GPS? Can it be done through Google maps? However, after further investigation, I quickly realized how this problem is more complicated than I first imagined. 2: The Coastline Paradox 2.1 What is the Coastline Paradox? Anyone that has been to a beach will know that there is no clear line between the ocean and the land, as it is instead constantly changing due to the nature of the ocean. In search for a way to measure the length of a coastline, problems arise. If you attempt to measure the distance by walking around the island; where do you start and which path do you take? Coastlines naturally change overtime due to do the weather and erosion. An estimate of a coastline is generally found by approximating complex curves as straight lines. However, this approximation actually leads to another issue, which is that the size of the coastline increases as you decrease the size of the ruler. As the ruler becomes really small, the length of the measured coastline approaches infinity, as more complexity is added. This paradox is known as the Coastline Paradox and was first recorded by Lewis Fry Richardson and expanded on by Benoit Mandelbrot in his paper How Long Is the Coast of Britain? [2] in 1967. In it, Mandelbrot explained the problem with trying to map coastlines: Seacoast shapes are examples of highly involved curves such that each of their portions can in a statistical sense can be considered a reduced-scale image of the whole. This property will be referred to as "statistical self-similarity.” To speak of a length for such figures is usually meaningless... As even finer features are taken into account, the total measured length increases, and there is usually no clear-cut gap or crossover, between the realm of geography and details with which geography need not be concerned.” [2] 2.2 Aim of IA In this IA, I will, firstly, investigate the concept of the fractal dimension and, secondly, attempt to answer the question: What are the fractal dimensions of the Indonesian islands; Sulawesi and West Nusa Tenggara? These are both islands which appear to have complicated coastlines. I was also tempted to pick Sulawesi because the island just experienced an earthquake and tsunami. Events like these highlight how useless the measurement of coastline “length” is, as the shape of an island edge is changing all the time. 3: Investigating the Concept of the Fractal Dimension 3.1 The Complexity of the Bali’s Coastline As a preliminary study, I decided to investigate this phenomenon by looking at Bali’s coastline using Google Maps. As shown if Figure 2, Bali’s coastline does appear to embody the statistical self symmetry that Mandlebrot described. Fig 1. The Island of Bali at different magnifications to show the increasing complexity of the coastline From these images, there appears to be no point at which the island’s coastline “smoothens” out, as each zoom reveals more complexity. This presents a problem for someone who is trying to approximate the length of a coastline using a ruler, as it becomes impossible to pick a point at which the complexity can be simplified as a straight line. The conclusion is that coastlines cannot be analysed effectively using ideas from Euclidean geometry, where if you zoom in enough curved lines are essentially straight lines whose lengths can be easily measured (Fig 2). This was indeed Mandelbrots conclusion. Eddie Woo wrote that Mandelbrot found this “Euclidean vision of the world deeply unsettling” because “reality is not made up of smooth unbroken lines. It is filled with jagged edges and bumpy surfaces, lines that have been broken and split into millions of different pieces and have little resemblance to Euclid’s divine shapes”[4]. Fig 2 Circle the demonstrates the main idea in Euclidean geometry where curves are approximated to show straight lines [4] Mandelbrot proposed a new way of describing a coastline, in which he introduced the idea of the fractal dimension. The fractal dimension assumes an alternative definition for the concept of a dimension. Mandelbrot proposes that such shapes with “infinite” parameters can be interpreted using the fractional dimensions in which a shape can have a non-whole number dimension. (e.g 1.585 dimensions). For a coastline, the number would fall between 1 and 2 where the closer the number is to 2, the more jagged line. Smooth coastlines such as the one of South Africa are said to have really smooth coastlines and thus have a fractal dimension of 1.02 [4]. 3.2 What is a fractal dimension? Mandelbrot saw the definition of dimension as; an n dimensional shape is one that when scaled by a factor 1 n x , the shape creates x copies of itself. For example, when a square plane (n=2) is scaled by a factor 2 1 3 , it creates 9 copies of itself, which is 3 . That power of 2 is its dimension. Fig. 4 summarises this for the three dimensions we normally associate with when thinking in terms of regular Euclidean geometry. Fig. 3 Summary of how fractal dimensions are derived [1] This can be represented through the equation: D ( r1 ) = N r = scale f actor D = shapes dimension N = number of copies of itself created Using log laws, this can be rearranged to find the D: D ( r1 ) = N D log ( 1r ) = log N D log 1 r D= D= = log N log N log 1r log N log 1r This equation suggests an alternative way of looking at dimensions and challenges the common intuition of what a dimension is. 3.3 Calculating the Dimensions of Sierpinksi’s Triangle and Koch Snowflake In order to deepen my understanding of fractal dimensions, I will attempt to investigate the fractal dimensions of various man-made fractals, the most common examples being the Koch Snowflake and the Sierpinski Triangle. In order to find the fractal dimensions, the shapes will be scaled down by a given scale factor, and then counted for the number of copies of itself that were created. Fractal Dimension of the Sierpinski Triangle Fig 4 Sierpinski's Triangle [6] D= D= D= log N log 1r log N log r log 3 log 11 ( ) 2 D= log 3 log 2 D ≈ 1.585 Fractal Dimension of the Koch Snowflake Fig 5 Koch Snowflake scaled down by a third [7] D= D= log N log 1r log 4 log 11 2 D= log 4 log2 D ≈ 1.262 3.4 The Connections of Fractals with Coastlines The Koch Snowflake has a “finite” area with an “infinite” perimeter. These properties are similar to the coastline which has a finite area and a perimeter that approaches infinity. Rather than measuring the length, the fractal dimension is able to measure the complexity of the edge. Thus, rather than attempting to estimate the length of the island's coastlines, I will attempt to calculate the fractal dimension of two different Indonesian islands, using two different techniques; the Hausdorff Method [9] and the Box Plotting technique [10]. Before that, I will attempt to find the fractal dimension of Britain, to check that my methods work. This value was calculated in Mandelbrots paper (≈ 1.25) [2]. If I am able to achieve a fractal dimension similar to his, it will indicate that my method is an effective way of calculating fractal dimensions, as well as give me a sense of my degree of error in carrying out the method. I will trial two different methods of calculating the fractal dimension, and evaluate from there which method is the most suitable for calculating the fractal dimension of 2 Indonesian Islands. 4.1: The Hausdorff Method This method for measuring the fractal dimensions of shapes with complicated edges was introduced by Felix Hausdorff [9] and was used by Mandelbrot when he was trying to determine the fractal dimension of the coast of Great Britain. It involves using a ruler with a length G and approximating several straight lines between points on a coastline. The final length of coastline will be the length of the ruler (G), multiplied by the number of rulers it takes to cover the entire coastline (n). In his paper, Mandelbrot described how “L depends greatly upon G”. As you make G smaller, you are able to pick up on more finer details, and thus L increases. How much the length of the approximated coastline increases and the length of the ruler decreases is related by the equation: L = M G1−D [2] L = length of coastline M = proportionality constant G = length of ruler D = f ractal dimension Using log laws this equation can be turned into the form y = mx + c : L = M G1−D log L = log M G1−D log L = log M + log G1−D log L = log M + (1 − D)log G (1) This can be plotted as: y = log L m = (1 − D) (2) Thus D = 1 − m (2) x = log G c = log M 4.2: The Hausdorff Method Applied to Britain’s Coastline Using photoshop, I placed rulers along Britain's coastline, each time using different lengths (G), and developed these four figures. Then I counted the number of rulers placed around the coastline to find L, the length of the coastline. L is thus equal to the length of the ruler multiplied by n, the number of rulers placed around the coastline. Table 1 Length of Ruler (G) (k 200 100 50 25 log (G) 2.301 2.000 1.699 1.298 Number of rulers (n) 12 32 72 172 Length coastline (L) (=G*n) (km) 2400 3200 3600 4300 Log (L) 3.380 3.505 3.556 3.633 Solving for Fractal Dimension D = m − 1 (2) From the graph, m =− 0, 2694 D = 1 − (− 0.2694) D = 1.2694 D ≈ 1.27 4.3 Evaluating the Accuracy of My Application of the Hausdorff Method The value I obtained had a difference of 0.02 compared to Mandelbrots value of 1.25[2]. This is only a 1.6% error. This difference is due to the nature of the method. The placement of rulers along the coastline is approximated by eye. A person other than me would place the rulers in a different location. Based off how rulers are placed, the number of rulers that can fit along the coastline can change, resulting in a different fractal dimension. Thus, in my opinion, is an error of 1.6% is insignificant given the method I utilised. In addition, the high R2 (0.96852) also suggests that my placement of the rulers was precise and that the method was followed accurately. However I will still trial the Box Method to see if it can bring me even closer to Mandelbrot’s value. 5.1 The Box Counting Method The second method I will use to get an estimate of the fractal dimension is the Box Counting method, also known as the Minkowski–Bouligand method[10]. This method is similar to the last, as it considers how much detail is added when the scale factor is changed. The procedure involves superimposing grids of different calibre (size) over an image and then recording the number of blocks the coastline passes through for each calibre. With a finer grid, more detail is supposed to be picked up, and the method can be modeled based on the equation: N = csD [10] N = number of boxes c = proportionality constant s = scale f actor D = f ractal dimension Using logs this equation can be rearranged in to the form y = mx + c . log N = log csD log N = log c + logsD log N = log c + D logs log N = D logs + log c y = log N m=D x = logs c = log c The fractal dimension D will thus be the gradient of this graph. 5.2 Attempting Box Plot Method The most effective method would be to perform the method using programing to create a really fine grid whose boxes can be counted accurately by the computer. However I do not possess the skills to do so, but still attempted the method using photoshop. I began by placing a grid over the island of Britain and then highlighted all the boxes in green which touched the coast line. For each successive plot, I shrunk each box down by a scale down factor of 2. Then I counted the number of boxes by eye. In order to keep track of the amount of boxes, I would double highlight a box every ten counts. Table 2 Scaled down 0 Factor (S) 2 4 8 log (S) 1.863 2.326 2.704 73 212 506 1.863 2.326 2.704 Number Boxes (N) Log (N) 1.477 of 30 1.477 As m = D , the fractal dimension is calculated as 1.376. 5.3 Evaluation of Box Counting Method Whilst 1.38 is within a 10% error of Mandelbrots value 1.25[2], it is not nearly as accurate as the Hausdorff method. There are indeed reasons that can explain this error. The use of the boxplotting method becomes more accurate as a finer grid is used. For me, the problem with using a finer grid is that it increases the number of boxes that need to be accounted for, and increases the possibility in accidentally miscounting the number of boxes. This problem can be solved through computer programming. My selection of grid size was probably too small to accurately approximate the fractal dimension, and I have no experience in programming to have the skills that would allow me to generate data using a finer graph. The process of counting the boxes is more tedious and inaccurate than the Hausdorff method, and thus it is the Hausdorff method that I will utilize in order to find the fractal dimension of two Indonesian islands. 6.Determining the fractal dimensions of 2 Indonesian Islands Using the Hausdorff method, the fractal dimension was calculated for two Indonesian islands. 6.1 Sulawesi Table 3 Length of Ruler (G) (k 150 100 50 25 log (G) 1.477 1.863 2.326 2.704 Number of rulers (n) 26 40 85.5 181 Length coastline (L) (=G*n) (km) 3900 4000 4425 4525 Log (L) 3.591 3.602 3.646 3.656 Solving for Fractal Dimension D = m − 1 (2) From the graph, m =− 0.09 D = 1 − (− 0.09) D = 1.09 6.2 West Nusa Tenggara Table 4 Length of Ruler (G) (km) 50 25 12.5 6.25 log (G) 1.699 1.398 1.097 0.796 Number of rulers (n) 15 33 82 170 Length coastline (L) (=G*n) (km) 750 825 1025 1062.5 Log (L) 2.875 2.916 3.010 3.026 Solving for Fractal Dimension D = m − 1 (2) From the graph, m =− 0.18121 D = 1 − (− 0.18121) D = 1.18121 D ≈ 1.18 7: Discussion and Conclusion To conclude, I found that Sulawesi and West Nusa Tenggara have a fractal dimension of 1.09 and 1.18 respectively. My computation of the coastline of Britain was close to 1.6% accuracy. I believe that I applied the method in the same way for Suluwesi and West Nusa Tenggara, therefore, I will use this error in calculating the error of the islands fractal dimensions. Table 5 Fractal Dimension % Error Absolute Error Max Value Min Value Sulawesi 1.09 1.6 0.0174 1.1074 1.0726 West Nusa Tenggara 1.18 1.6 0.01888 1.1612 1.1988 Qualitatively, looking at West Nusa Tenggara, it does appear to have a more complex coastline which folds in on itself more than Sulawesi which thus explains the larger fractal dimension. Both have fractal dimensions lower than that of Great Britain, implying that they are smoother and less jagged than Britain's coastline. The Hausdorff method is done by eye, and is therefore not infallible. This is the reason for the error. An interesting way of expanding this IA would be to ask other people to calculate the fractal dimension using the Hausdorff method, and compare these results. This also could be expanded by using the Boxploting method, but done by utilizing a computer program rather than by hand. Additionally, I only found the fractal dimension of 2 coastlines, and Indonesia has 14,000 potential islands that could be analysed. I first met the concept of fractals in grade 6, when we learned about the Sierpinski triangles. I found them mathematically elegant and pleasing to draw. To learn how the concept of fractals is be applied mathematically to real life was fascinating for me. Also the ability to bring in logs and regression equations really taught me how different areas of math can be woven together to solve complicated problems. Through my investigation, I also discovered various ways in which the fractal dimensions is applied in different areas of science. The fractal dimension is an effective way of describing the roughness/complexity of real world objects. Grant Sanderson, a software engineer from Stanford said that a fractal dimension seems to be “the core differentiator between objects that arise naturally, and those that are just man made”[7]. Prof T. Bhaskara Reddy’s paper (2017) analysed how fractal dimensions can be used in plant morphology to “characterize the structure/architecture of medicinal leaves”[11]. Wolfgang Bauer’s paper (2010) described how the fractal dimension can be used in cancer cell analysis [12]. This academic exercise was a great introduction for me into calculating and interpreting the fractal dimension, a skill that I could possibly utilize later on in my academic pursuit. Bibliography [1] Lesmoir-Gordon, Nigel. Introducing Fractals: A Graphic Guide (Kindle Location 276). Icon Books Ltd. Kindle Edition. [2] Mandelbrot, B. (2018). How Long Is the Coast of Britain? Statistical Self-Similarity and Fractional Dimension. [online] Li.mit.edu. Available at: http://li.mit.edu/Stuff/CNSE/Paper/Mandelbrot67Science.pdf [Accessed 30 Sep. 2018]. [3] Chapter 4: Calculating Fractal Dimensions, www.wahl.org/fe/HTML_version/link/FE4W/c4.htm. [4] Woo, Eddie (2018) “Lightning Through Your Veins.” (p. 46-27).Woo's Wonderful World of Maths, MacmillanPan Macmillan Australia [5] Grant. “Fractals Are Typically Not Self-Similar.” 3Blue1Brown, 3Blue1Brown, 26 Jan. 2017, www.3blue1brown.com/videos-blog/2017/5/26/fractals-are-typically-not-self-similar. [6] Rosenman, Richard. “Sierpinski Triangle - Recursive Fractal Photoshop Plugin.” Richard Rosenman Advertising & Design, richardrosenman.com/shop/sierpinski-triangle/. [7] “Fractal Dimension Koch Snowflake.” Lindenmayer Fractals - Fractal Dimension - Koch Snowflake, www.math.ubc.ca/~cass/courses/m308-03b/projects-03b/skinner/ex-dimension-koch _snowflake.htm. [8] “British Isles Outline Map .” Maproom, maproom.net/shop/detailed-outline-map-of-britain-and-ireland-the-british-isles/. [9] Duvall, P "The Hausdorff Dimension of the Boundary of a Self-Similar Tile." J. London Math. Soc. 61, 649-760, 2000. [10] Karperien, A. “‘Box Counting.’” National Institutes of Health, U.S. Department of Health and Human Services, imagej.nih.gov/ij/plugins/fraclac/FLHelp/BoxCounting.htm. [11]Nithiyanandhan, K. and Bhaskara Reddy, T. (2019). Analysis of the Different Medicinal Leaf with Fractal Dimension. International Journal on Recent and Innovation Trends in Computing and Communication, [online] 5(5). Available at: https://pdfs.semanticscholar.org/ace4/cb11b90c9988d55f6cad8d899114eedef704.pdf [Accessed 14 Jun. 2019]. [12] Bauer, W. and D Mackenzie, C. (2019). Cancer Detection via Determination of Fractal Cell Dimension. [ebook] East Lansing. Available at: https://pdfs.semanticscholar.org/72ce/f199642ba0bd348e8f2f6ef447edd3b3ea3f.pdf [Accessed 14 Jun. 2019]. [13] Cia.gov. (2019). The World Factbook - Central Intelligence Agency. [online] Available at: https://www.cia.gov/library/publications/the-world-factbook/ [Accessed 20 Jun. 2019]. [14] Wri.org. (2019). World Resources Institute | Making Big Ideas Happen. [online] Available at: https://www.wri.org/ [Accessed 20 Jun. 2019].