See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/338698928

Introduction to Geomathematics

Presentation · January 2020

CITATIONS READS

0 2,672

1 author:

Osama Rahil Shaltami

University of Benghazi
226 PUBLICATIONS 2,051 CITATIONS

SEE PROFILE

All content following this page was uploaded by Osama Rahil Shaltami on 20 January 2020.

The user has requested enhancement of the downloaded file.

LECTURE FOR UNDERGRADUATE
STUDENTS

INTRODUCTION TO
GEOMATHEMATICS

Dr. Osama Shaltami

Department of Earth Sciences
Faculty of Science, Benghazi University, Libya
Geomathematics or Mathematical Geophysics is the
application of mathematical intuition to solve problems in
Geophysics. The most complicated problem in Geophysics is the
solution of the three dimensional inverse problem, where
observational constraints are used to infer physical properties.
The inverse procedure is much more sophisticated than the
normal direct computation of what should be observed from a
physical system. The estimation procedure is often dubbed the
inversion strategy (also called the inverse problem) as the
procedure is intended to estimate from a set of observations the
circumstances that produced them. The Inverse Process is thus
the converse of the classical scientific method.
 Applications
Terrestrial Tomography
An important research area that utilises inverse methods is
seismic tomography, a technique for imaging the subsurface of
the Earth using seismic waves. Traditionally seismic waves
produced by earthquakes or anthropogenic seismic sources (e.g.,
explosives, marine air guns) were used.
 Crystallography
Crystallography is one of the traditional areas of geology that use
mathematics. Crystallographers make use of linear algebra by
using the Metrical Matrix. The Metrical Matrix uses the basis
vectors of the unit cell dimensions to find the volume of a unit cell,
d-spacings, the angle between two planes, the angle between
atoms, and the bond length. Miller's Index is also helpful in the
application of the Metrical Matrix. Brag's equation is also useful
when using an electron microscope to be able to show
relationship between light diffraction angles, wavelength, and the
d-spacings within a sample.
 Geophysics
There are many applications in geophysics which include gravity,
magnetic, seismic, electric, electromagnetic, resistivity,
radioactivity, induced polarization, and well logging. Gravity and
magnetic methods share similar characteristics because they're
measuring small changes in the gravitational field based on the
density of the rocks in that area. While similar gravity fields tend to
be more uniform and smooth compared to magnetic fields. Gravity
is used often for oil exploration and seismic can also be used, but
it is often significantly more expensive. Seismic is used more than
most geophysics techniques because of its ability to penetrate,
its resolution, and its accuracy.
 Geomorphology
Many applications of mathematics in geomorphology are related
to water. In the soil aspect things like Darcy's law, Stoke's law, and
porosity are used.
 Glaciology
Mathematics in Glaciology consists of theoretical, experimental,
and modeling. It usually covers glaciers, sea ice, waterflow, and
the land under the glacier.
Isostacy
Explanations for lowered gravity
over mountain belts
Back to isostacy- The ideas we’ve been playing around with
must have occurred to Airy. You can see the analogy between
ice and water in his conceptualization of mountain highlands
being compensated by deep mountain roots shown below.
Other examples of isostatic
computations
Another possibility
 B C
A

The product of density and thickness must

remain constant in the Pratt model.
At A 2.9 x 40 = 116
At B C x 42 = 116 C=2.76
At C C x 50 = 116 C=2.32
Some expected differences in the
mass balance equations
 Segment II
More about Isostacy
 Recent Sedimentation Record - North Sea
16000
The general trend of
age versus depth
14000

12000

Age (years)
10000

8000

6000

4000

2000

0
10 510 1010 1510 2010

Depth (cm)
Problem 2.11

i) sedimentation rate from 10510 to 1490 years ago was 0.0429cm/yr

or 23.29 yr/cm. This value is computed directly from the endpoints
defining sedimentation during that period of time
ii) The trend of age versus depth was approximated by the line
shown above in the figure. Its slope is ~ 2.24 yrs/cm which
translates into a sedimentation rate of 0.447 cm/yr.
iii) Assuming that sedimentation continues at the rate of 0.0429 cm/yr,
it will take an additional 459.9 years to deposit the remaining 19.75 cm
of sediment. Thus the age at the surface or the time when sedimention
ceased is 1490 - 459.9 years or approximately 1030 years ago.
 200

150

Concentration (C)
100

50

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Liquid Fraction (F)

Problem 2.12

Given that C = CoF(d-1) where Co is the initial concentration of the element

in the liquid before crystalization, F is the fraction of the liquid remaining, and
d is the dis tribution coefficient. In the pres ent example we solve for C when
F = 0.5, and d = 6.5

o 5.5
C = C F(d-1) = 200*(05)
which yields C = 4.4194
The gravity anomaly map shown here indicates that the mountainous region is associated with an
extensive negative gravity anomaly (deep blue colors). This large regional scale gravity anomaly
is believed to be associated with thickening of the crust beneath the area. The low density crustal
root compensates for the mass of extensive mountain ranges that cover this region. Isostatic
equilibrium is achieved through thickening of the low-density mountain root.
Solving isostatic equilibrium problems

cl   m d  c L   m D

cl  md  c e  l  r   m D
On Tuesday, from the foregoing
starting point, we derived a couple
basic relationships governing the
isostatic equilibrium processes.
These included:


e h
m

Where m represents the density of the

mantle,  = m - c (where c is the
density of the crust), and h represents
crustal thickening (r + e).
And -
 c 
r   e
  m  c 
or
 c 
r  e
  

from which we must also have

  
e r
 c 
In Class Problem: A 500m deep depression on the
earth's surface fills with sandstone of density 2.2
gm/cm3. Assume that the empty basin is in isostatic
equilibrium and that normal crustal thickness in
surrounding areas is 20km. Calculate the thickness
of sediment that must be deposited in the basin to
completely fill it. (Use crustal and mantle densities of
2.8 and 3.3 gm/cm3, respectively.)
Hint: Compute the initial thickness of the crust
beneath the empty basin and assume that the crustal
thickness beneath the basin does not change.
 20 c  b  s  l c  r m
Recall that on Tuesday we
showed that l=16.7km - hence

3.3c  b s  r m
We also showed that
r = 20-e-l-b .. thus

3.3c  b  s  (3.3  e  b) m
After rearrangement

3.3c  b  s  (3.3  e  b) m

3.3( c  m )  b(  s  m )

or
3.3(  c   m )
b
( s  m )
&
b  1.5km
Recall that since l = 16.7km and lt does not
change as the basin is filled, we now have the
depth to the base of the crust in the rifted
region (b + l = 18.2km), after isostatic
equilibrium has been re-established. The base
of the crust now rests 1.8km above the base of
the continental crust in the surrounding un-
deformed area.
Recall, that when the basin was empty (0.5km
deep) the crust extended down to 17.2km and r
(the antiroot) was 2.8.
It took 1.5km of sediment to fill our half-
kilometer deep basin!
As sediment is deposited, the basin floor
gradually drops to maintain isostatic
equilibrium.

Does this really happen?

Conodant alteration indices from this

area of the Appalachians indicate that
rocks currently exposed at the surface
were once buried beneath 3km of
sediment.
Let’s examine the dynamics of this process using
EXCEL. Pick up the EXCEL file Isostacy1.xls from
my shared directory.
Basin Filling Process

0.6
Depth to Top of Sedim ent

0.5
0.4
Deposits

0.3
0.2
0.1
0
-0.1
-0.2
0 0.5 1 1.5 2
Thickness of Deposited Sedim ents
Take Home Problem: A mountain range 4km high is in
isostatic equilibrium. (a) During a period of erosion, a
2 km thickness of material is removed from the
mountain. When the new isostatic equilibrium is
achieved, how high are the mountains? (b) How high
would they be if 10 km of material were eroded away?
(c) How much material must be eroded to bring the
mountains down to sea level? (Use crustal and mantle
densities of 2.8 and 3.3 gm/cm3.)

There are actually 4 parts to this problem - we must

first determine the starting equilibrium conditions
before doing solving for (a).
The preceding questions emphasize the dynamic aspects
of the problem. A more complete representation of the
balance between root and mountain is shown below. Also
refer to the EXCEL file on my shared directory.
Isostatic Response to Erosion

25
Mountain Elevation &
Mountain Root (km)

20
Root Extent (km)
15

10
Mountain Elevation(km)
5

0
0 5 10 15 20 25 30
Amount Eroded (km)
A few more comments on Isostacy
 B C
A

The product of density and thickness must

remain constant in the Pratt model.
At A 2.9 x 40 = 116
At B C x 42 = 116 C=2.76

At C C x 50 = 116 C=2.32
Statistics
 y = mx + b

The best fit line is a line which minimizes the difference between the estimated and
actual values of y.
 is the estimate of yi.
ŷi
yˆi  mxi  b
We want to minimize these differences for all yi.
 .. and the best way to do this is to minimize the sum of the squares of these departures.
Mathematically the sum of the square of the departures or differences is

N
  yi  yi 
2
ˆ
i 1

N 2
  yi  mxi  b 
i 1
Let the sum of these squared differences = D.

How can we minimize D?

Remember, when you want to find the minimum of something you compute its derivative (its
tangents) and set the derivative equal to 0, i.e., find a tangent to the curve whose slope is
zero.

Where is the minimum of the function

y  ( x  a) 2 ?
 N 2
Given
D    yi  mxi  b 
i 1

there are two ways we could minimize this expression - one with respect to the
slope m - and the other with respect to the intercept b.
 The end result -

The intercept

b  y  mx

n n n
n xi yi   xi  yi
i 1 i 1 i 1
m
n(n  1) s 2
Where s2=variance of x
 It also turns out that

covariancexy
m
variancex

Where the covariance between x and y is

1 n
sxy  
n  1 i 1
( xi  x )( yi  y )
or

s xy
m
s 2x
Back to statistics - remember the pebble mass
distribution? 224 322 353 384
242 324 355 386
256 324 355 389
256 326 355 389
265 327 357 393
269 329 358 394
277 330 359 394
283 331 359 395
283 331 364 397
Pebble masses collected from beach A 283 331 366 400
0.40 284 334 367 401
0.35
287 335 368 403
290 338 369 403
0.30
294 338 370 403
301 338 370 407
Probability

0.25
301 340 371 408
0.20
302 340 373 409
0.15 303 341 374 420
307 342 374 422
0.10
307 342 375 423
0.05 311 343 379 432
0.00
314 346 380 433
150 200 250 300 350 400 450 500 550 317 346 383 435
Mass (grams) 318 350 384 450
318 352 384 454
 Pebble masses collected from beach A
0.40

0.35

0.30

Probability
0.25

0.20

0.15

0.10

0.05

0.00
150 200 250 300 350 400 450 500 550

Mass (grams)

The probability of occurrence of specific values in a sample often takes on that bell-shaped
Gaussian-like curve, as illustrated by the pebble mass data.
 Probability Distribution of Pebble Masses

0.01

0.008

Probability
0.006 Series1
0.004 Series2

0.002

0
0 200 400 600 800
Pebble Mass (grams)

The Gaussian (normal) distribution of pebble masses looked a bit different from the probability distribution
we derived directly from the sample, but it provided a close approximation of the sample probabilities.
 Equivalent Gaussian Distribution of Pebble Masses

0.01
0.008
Probability
0.006 Series1
0.004 Series2

0.002
0
0 200 400 600 800
Pebble Mass (grams)

Range (g) Measured Range Gaussian (normal)

probability (multiple of s) probability
201-250 0.02 -3.10 to -2.06 0.019
251-300 0.12 -2.06 to -1.02 0.134
301-350 0.35 -1.02 to 0.02 0.354
351-400 0.36 0.02 to 1.06 0.347
401-450 0.14 1.06 to 2.10 0.127
451-500 0.01 2.10 to 3.13 0.017
The pebble mass data represents just one of a nearly infinite number of possible
samples that could be drawn from the parent population of pebble masses.

We obtained one estimate of the population mean and this estimate is almost
certainly incorrect.

What might additional pebble mass samples look like?

 Sample1 <x>=348.84 Sample2 <x>=350.6
25 20
20 15
N 15 N
10
10 These samples were drawn at random from
5
5 a parent population having mean 350.18
0 0 and variance of 2273.
200 300 400 500 200 300 400 500
Mass (grams) Mass (grams)
Sample 3 <x>=356.43 Sample 4 <x>=354.5
25 30
20 25
20
N 15 N
15
10
10
5 5
0 0
200 300 400 500 200 300 400 500
Mass (grams) Mass (grams)
Sample 5 <x>=348.42
20

15
N
10

0
200 300 400 500
Mass (grams)
 Sample1 <x>=348.84 Sample2 <x>=350.6
25 20
20 15 Note that each of the sample means
N 15 N differs from the population mean
10
10
5
5
0 0
200 300 400 500 200 300 400 500
Mass (grams) Mass (grams)
Sample 3 <x>=356.43 Sample 4 <x>=354.5
25 30
20 25
20
N 15 N Mean Variance Standard
15
10
10 deviation
5 5 348.84 2827.5 53.17
0 0 350.6 2192.59 46.82
200 300 400 500 200 300 400 500
Mass (grams) Mass (grams)
356.43 2124.63 46.09
354.5 1977.63 44.47
Sample 5 <x>=348.42 348.42 2611.3 51.1
20

15
N
10

0
200 300 400 500
Mass (grams)
The distribution of 35 means calculated from 35 samples drawn at random from
a parent population with assumed mean of 350.18 and variance of 2273 (s =
47.676).

Distribution of Means
12

10

N
6

0
330 335 340 345 350 355 360 365
Mass
 Distribution of Means
12

10

N
6

0
330 335 340 345 350 355 360 365
Mass

The mean of the above distribution of means is 350.45.

Their variance is 21.51 (i.e. standard deviation of 4.64).
The statistics of the distribution of means tells us something different from the statistics of
the individual samples.

The statistics of the distribution of means gives us information about the variability we can
anticipate in the mean value of 100-specimen samples.

Just as with the individual pebble masses observed in the sample, probabilities can also be
associated with the possibility of drawing a sample with a certain mean and standard
deviation.
 This is how it works -

You go out to your beach and take a bucket full of pebbles in one area and then go to another
part of the beach and collect another bucket full of pebbles.

You have two samples, and each has their own mean and standard deviation.

You ask the question - Is the mean determined for the one sample different from that determined for the
second sample?
To answer this question you use probabilities determined from the distribution of
means, not from those of an individual sample.

The means of the samples may differ by only 20 grams. If you look at the range of
individual masses which is around 225 grams, you might conclude that these two
samples are not really different.
 However, you are dealing with means derived from individual samples each consisting of
100 specimens.

The distribution of means is different from the distribution of specimens. The range of
possible means is much smaller.

Histogram of pebble masses Distribution of means

40 12

35
Number of Occurrences

10

Number of occurrences
30

8
25

20 6

15
4
10

5 2

0
0
200 250 300 350 400 450 500
200 250 300 350 400 450 500

Mass (grams) Mean Mass (grams)

Thus, when trying to estimate the possibility that two means come from the same parent
population, you need to examine probabilities based on the standard deviation of the
means and not those of the specimens.

Number of Number of Number of

standard Area standard Area standard Area
deviations deviations deviations
0.0 0.000 1.1 0.729 2.1 .964
0.1 0.080 1.2 0.770 2.2 .972
0.2 0.159 1.3 0.806 2.3 .979
0.3 0.236 1.4 0.838 2.4 .984
0.4 0.311 1.5 0.866 2.5 .988
0.5 0.383 1.6 0.890 2.6 .991
0.6 0.451 1.7 0.911 2.7 .993
0.7 0.516 1.8 0.928 2.8 .995
0.8 0.576 1.9 0.943 2.9 .996
0.9 0.632 2.0 0.954 3.0 .997
1.0 0.683
In the class example just presented we derived the mean and standard deviation of
35 samples drawn at random from a parent population having a standard deviation
of 47.7.

Recall that the standard deviation of means was only 4.64 and that this is just
about 1/10th the standard deviation of the sample.
This is the standard deviation of the sample means from the estimate of the true
mean. This standard deviation is referred to as the standard error.

The standard error, se, is estimated from the standard

deviation of the sample as -

se  sˆ / N
What is a significant difference?

To estimate the likelihood that a sample having a specific calculated mean and standard
deviation comes from a parent population with given mean and standard deviation, one has
to define some limiting probabilities.

There is some probability, for example, that you could draw a sample whose mean might be
10 standard deviations from the parent mean. It’s really small, but still possible.
What chance of being wrong will you accept?

This decision about how different the mean has to be in order to be considered
statistically different is actually somewhat arbitrary.

In most cases we are willing to accept a one in 20 chance of being wrong or a one in
100 chance of being wrong.

The chance we are willing to take is related to the “confidence limit” we choose.
The confidence limits used most often are 95% or 99%. The 95% confidence limit
gives us a one in 20 chance of being wrong. The confidence limit of 99% gives us
a 1 in 100 chance of being wrong.

The risk that we take is referred to as the alpha level.

 If our confidence limit is 95% our alpha level is 5% or 0.05.

If our confidence limit is 99% -

 is 1% or 0.01

Whatever your bias may be - whatever your desired result - you can’t go wrong in your
presentation if you clearly state the confidence limit you use.
 Number of Number of Number of
standard Area standard Area standard Area
deviations deviations deviations
0.0 0.000 1.1 0.729 2.1 .964
0.1 0.080 1.2 0.770 2.2 .972
0.2 0.159 1.3 0.806 2.3 .979
0.3 0.236 1.4 0.838 2.4 .984
0.4 0.311 1.5 0.866 2.5 .988
0.5 0.383 1.6 0.890 2.6 .991
0.6 0.451 1.7 0.911 2.7 .993
0.7 0.516 1.8 0.928 2.8 .995
0.8 0.576 1.9 0.943 2.9 .996
0.9 0.632 2.0 0.954 3.0 .997
1.0 0.683

In the above table of probabilities (areas under the normal distribution curve), you can see that
the 95% confidence limit extends out to 1.96 standard deviations from the mean.
The standard deviation to use when you are comparing means (are they the same or
different?) is the standard error, se.

Assuming that our standard error is 4.8 grams, then 1.96se corresponds to  9.41 grams.
Notice that the 5% probability of being wrong is equally divided into 2.5% of the area greater
than 9.41grams from the mean and less than 9.41 grams from the mean.
 You probably remember the discussions of one- and two-tailed tests.

The 95% probability is a two-tailed probability.

So if your interest is only to make the general statement that a particular mean lies
outside  1.96 standard deviations from the assumed population mean, your test is a
two-tailed test.

and your  is 0.05

If you wish to be more specific in your conclusion and say that the estimate is significantly
greater or less than the population mean, then your test is a one-tailed test.

then your  is 0.025

i. e. the probability of error in your one-tailed test is 2.5% rather than 5%.

 is 0.025 rather than 0.05

Using our example dataset, we have assumed that the parent population has a
mean of 350.18 grams, thus all means greater than 359.6 grams or less than
340.8 grams are considered to come from a different parent population - at the
95% confidence level.

Mean Variance Standard Note that the samples we

deviation drew at random from the
348.84 2827.5 53.17
parent population have means
350.6 2192.59 46.82
356.43 2124.63 46.09 which lie inside this range and
354.5 1977.63 44.47 are therefore not statistically
348.42 2611.3 51.1 different from the parent
population.
It is worth noting that - we could very easily have obtained a different sample having a
different mean and standard deviation. Remember that we designed our statistical test
assuming that the sample mean and standard deviation correspond to those of the
population.

Mean Variance Standard

deviation
348.84 2827.5 53.17
350.6 2192.59 46.82
356.43 2124.63 46.09
354.5 1977.63 44.47
348.42 2611.3 51.1
This would give us different confidence limits and slightly different answers. Even so, the
method provides a fairly objective quantitative basis for assessing statistical differences
between samples.

Mean Variance Standard 95%

deviation C. L.
348.84 2827.5 53.17 338.29 - 359.39
350.6 2192.59 46.82 341.31 - 359.89
356.43 2124.63 46.09 347.28 - 365.57
354.5 1977.63 44.47 345.68 - 363.33
348.42 2611.3 51.1 338.28 - 358.56
The method of testing we have just summarized is known as the z-test, because we use
the z-statistic to estimate probabilities, where

m2  m1
z
se
Remember the t-test?

Tests for significance can be improved if we account for the fact that estimates of the mean
derived from small samples are inherently sloppy estimates.
The t-test acknowledges this sloppiness and compensates for it by making the criterion for
significant-difference more stringent when the sample size is smaller.

The z-test and t-test yield similar results for relatively large samples - larger than 100 or so.
The 95% confidence limit for example, diverges considerably from 1.96 s for smaller
sample size. The effect of sample size (N) is expressed in terms of degrees of freedom
which is N-1.
Thanks
View publication stats