Caja PDF

Comparta fácilmente sus documentos PDF con sus contactos, la web y las redes sociales.

Compartir un archivo PDF Gestor de archivos Caja de instrumento Buscar PDF Ayuda Contáctenos



Kursk simulation .pdf



Nombre del archivo original: Kursk simulation.pdf
Título: /tardir/tiffs/a384597.tiff

Este documento en formato PDF 1.4 fue generado por LuraDocument PDF Compressor Server 5.7.66.46 (original: tif_convert_2_ps()) / Recoded by LuraDocument PDF v2.46, y fue enviado en caja-pdf.es el 21/08/2014 a las 15:46, desde la dirección IP 65.93.x.x. La página de descarga de documentos ha sido vista 3549 veces.
Tamaño del archivo: 6.8 MB (187 páginas).
Privacidad: archivo público




Descargar el documento PDF









Vista previa del documento


NAVAL POSTGRADUATE SCHOOL
MONTEREY, CALIFORNIA

THESIS
FITTING FIREPOWER SCORE MODELS
TO THE BATTLE OF KURSK DATA
By
Ramazan Gozel
September 2000

Thesis Advisor:
Second Reader:

Thomas W. Lucas
Jeffrey Appleget

Approved for public release; distribution is unlimited.

20001128 083
miß QUALTPT is^mm 4

REPORT DOCUMENTATION PAGE

Form Approved OMB No. 0704-0188

Public reporting burden for this collection of information is estimated to average 1 hour per response, including the
time for reviewing instruction, searching existing data sources, gathering and maintaining the data needed, and
completing and reviewing the collection of information. Send comments regarding this burden estimate or any
other aspect of this collection of information, including suggestions for reducing this burden, to Washington
headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite
1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project
(0704-0188) Washington DC 20503.
3. REPORT TYPE AND DATES COVERED
1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE
Master's Thesis
September 2000
5.
FUNDING NUMBERS
4. TITLE AND SUBTITLE: Fitting Firepower Score Models To The Battle
Of Kursk Data
6. AUTHOR(S) Gozel, Ramazan
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESSEES)
Naval Postgraduate School
Monterey, CA 93943-5000
9. SPONSORING/ MONITORING AGENCY NAME(S) AND ADDRESS(ES)
N/A

i. PERFORMING ORGANIZATION
REPORT NUMBER
10. SPONSORING / MONITORING
AGENCY REPORT NUMBER

11. SUPPLEMENTARY NOTES The views expressed in this thesis are those of the author and do not reflect the official
policy or position of the Department of Defense or the U.S. Government.
12b. DISTRIBUTION CODE
12a. DISTRIBUTION / AVADLABBLITY STATEMENT
Approved for public release; distribution is unlimited.
13. ABSTRACT (maximum 200 words)
This thesis applies several Firepower Score attrition algorithms to real data. These algorithms are used in
highly aggregated combat models to predict attrition and movement rates. The quality of the available historical
data for validation of attrition models is poor. Most accessible battle data contain only starting sizes and casualties,
sometimes only for one side. A detailed database of the Battle of Kursk of World War U, the largest tank battle in
history, has recently been developed by Dupuy Institute (TDI). The data is two-sided, time phased (daily), highly
detailed, and covers 15 days of the campaign. According to combat engagement intensity, three different data sets
are extracted from the Battle of Kursk data. RAND's Situational Force Scoring, Dupuy's QJM and the ATLAS
ground attrition algorithms are applied to these data sets. Fitted versus actual personnel and weapon losses are
analyzed for the different approaches and data sets. None of the models fits better in all cases. In all of the models
and for both sides, the Fighting Combat Unit Data set gives the best fit. All the models tend to overestimates battle
casualties, particularly for the Germans.
15. NUMBER OF
PAGES 190

14. SUBJECT TERMS
Combat Modeling, Simulation, Attrition, Validation, Firepower Scores, Battle of Kursk

16. PRICE CODE
17. SECURITY
CLASSIFICATION OF
REPORT
Unclassified
NSN 7540-01-280-5500

18. SECURITY
CLASSD7ICATION OF THIS
PAGE
Unclassified

19. SECURITY
CLASSIFICATION OF
ABSTRACT
Unclassified

20. LIMITATION
OF ABSTRACT
UL

Standard Form 298 (Rev. 2-89)
Prescribed by ANSI Std. 239-18

THIS PAGE INTENTIONALLY LEFT BLANK

Approved for public release; distribution is unlimited
FITTING FIREPOWER SCORE MODELS TO THE BATTLE OF KURSK DATA
Ramazan Gozel
First Lieutenant, Turkish Army
B. S., Turkish Army Academy, 1994
Submitted in partial fulfillment of the
requirements for the degree of
MASTER OF SCIENCE IN MODELING, VIRTUAL ENVIRONMENTS,
AND SIMULATION
from the
NAVAL POSTGRADUATE SCHOOL
September 2000

42^»

Author:

Ramazan Gozel
Approved by:

^^v, A^

/ y^^Oym^y
Thomas W. Lucas, Thesis Advisor

fey Appleget, Second Reader

Rudy Darken, Academic Associate
Modeling, Virtual Environments, and Simulation
AcademiÄffiroup

Department Name Mfchaelj^ydar Chair
Modeling, Virtual Environments, and Simulation
Academic Group

m

THIS PAGE INTENTIONALLY LEFT BLANK

IV

ABSTRACT

This thesis applies several Firepower Score attrition algorithms to real data. These
algorithms are used in highly aggregated combat models to predict attrition and
movement rates. The quality of the available historical data for validation of attrition
models is poor. Most accessible battle data contain only starting sizes and casualties,
sometimes only for one side. A detailed database of the Battle of Kursk of World War II,
the largest tank battle in history, has recently been developed by Dupuy Institute (TDI).
The data is two-sided, time phased (daily), highly detailed, and covers 15 days of the
campaign. According to combat engagement intensity, three different data sets are
extracted from the Battle of Kursk data. RAND's Situational Force Scoring, Dupuy's
QJM and the ATLAS ground attrition algorithms are applied to these data sets. Fitted
versus actual personnel and weapon losses are analyzed for the different approaches and
data sets. None of the models fits better in all cases. In all of the models and for both
sides, the Fighting Combat Unit Data set gives the best fit. All the models tend to
overestimates battle casualties, particularly for the Germans.

THIS PAGE INTENTIONALLY LEFT BLANK

VI

TABLE OF CONTENTS

I.

INTRODUCTION
A.
COMBAT MODELING
B.
ATTRITION
C.
FIREPOWER SCORE APPROACH
D.
THESIS OUTLINE

1
1
3
4
6

II.

PREVIOUS VALIDATION STUDIES ON COMBAT MODELING
A.
PREVIOUS STUDIES WITH LANCHESTER EQUATIONS
B.
PREVIOUS STUDIES WITH FIREPOWER
SCORE APPROACHES
C.
STUDY METHODOLOGY

7
7
9
12

HI.

HISTORY AND DATA ON THE BATTLE OF KURSK
13
A.
A SHORT HISTORY ON THE BATTLE OF KURSK
13
B.
DATA ON THE BATTLE OF KURSK
16
1.
Creation and Scope of the Kursk Database
16
2.
Limitations and Timeframe of the Kursk Database
17
3.
Assumptions for the Kursk Database
17
C.
METHODOLOGY USED FOR THE EXTRACTION OF THE
DATA
17
1.
Personnel Data...
.
.
.......—........
»18
2.
Weapons Data............
...
........
..
. 19
a.
Classification of German Weapon Types
20
b.
Classification of Soviet Weapon Types
22
3.
Unit Activities and Combat Postures
.
.........—.—.. 24
4.
AH Combat Units Data (ACUD)
29
5.
Contact Combat Units Data (CCUD)
31
6.
Fighting Combat Units Data (FCUD)
33
D.
STATISTICAL COMPARISONS OF
PERSONNEL AND WEAPONS
35
1.
Personnel Statistics
35
a.
Onhand Personnel.
35
b.
Personnel Casualties
38
2.
Tank Statistics
42
a.
Onhand Tanks
42
b.
Tank Losses
45

IV.

APPLICATION OF DIFFERENT METHODOLOGIES TO THE DATA ON
THE BATTLE OF KURSK
49
A.
APPLICATION OF ATLAS GROUND ATTRITION MODEL
49
1.
Determining Firepower Score Values
50
Vll

B.

C.

V.

2.
The Application of the ATLAS Ground Attrition
Method to the All Combat Units Data (ACUD)
52
a.
Data
53
b.
Combat Power
53
c.
Computation of Force Ratio
54
d.
Casualty Rates
55
e.
Distribution of Combat Power Casualty Rates
58
f.
Results
55
3.
The Application of the ATLAS Ground Attrition Method to
Contact Combat Units Data (CCUD)
72
a.
Data
73
b.
Combat Power and Force Ratio
73
c.
Casualty Rates and its Distribution
74
d.
Results
78
4.
The Application of the ATLAS Ground Attrition Method to
Fighting Combat Units Data (FCUD)
83
a.
Data
84
b.
Combat Power and Force Ratio
84
c.
Casualty Rates and its Distribution
84
d.
Results
87
5.
The Application of the ATLAS Ground Attrition Method with
Air Sortie Data
92
a.
Data
92
b.
Combat Power
94
c.
Casualty Rates and its Distribution
95
d.
Results

95
6.
The Application of the ATLAS Ground Attrition Method
Using Different Firepower Score Values
99
APPLICATION OF RAND'S SITUATIONAL FORCE SCORING
METHODOLOGY
103
1.
Varying Asset Strength
105
2.
Determining Shortage Category Multipliers
106
3.
Combat Assessment
107
4.
Casualty Distribution
109
5.
Results
HO
6.
FLOT Movement Rates
119
APPLICATION OF THE QUANTIFIED JUDGMENT MODEL
(QJM) METHODOLOGY
123
1.
Personnel Attrition
124
2.
Weapon Losses
130
a.
Tank Loss Rate
130
b.
Artillery Loss Rate
737
3.
Results
132

CONCLUSIONS AND RECOMMENDATIONS
A.
CONCLUSIONS
Vlll

143
143

B.

RECOMMENDATIONS

150

LIST OF REFERENCES

153

INITIAL DISTRIBUTION LIST

155

IX

THIS PAGE INTENTIONALLY LEFT BLANK

LIST OF FIGURES

Figure 3.1.
Figure 3.2.
Figure 3.3.

Figure 3.4.
Figure 3.5.
Figure 3.6.
Figure 3.7
Figure 3.8.
Figure 3.9
Figure 3.10.
Figure 3.11
Figure 3.12.
Figure 3.13.
Figure 3.14.
Figure 3.15.
Figure 3.16.
Figure 3.17.
Figure 3.18.
Figure 3.19.
Figure 3.20.

Operation Zitadelle [From Ref. 17:p. 134]
15
Line Units in Contact. The vast majority of German units are in contact
each day while the Soviets have an increasing percentage
26
Line Units Attacking. Except for days 10, 14, and 15, the Germans have a
higher attacking percentage. Although on day 9 (12 July), the Germans
have a slightly higher ratio; based on historical attributes, it is assumed
that the Soviets are attacking on this day
27
Daily Percentages of German Units in Each Combat Posture. The vast
majority of units are in an attacking status
28
Daily Percentages of Soviet Units in Each Combat Posture. The vast
majority of units are in prepared in a hasty defense posture
29
German Onhand Personnel in Contact/Not in Contact. Almost all German
personnel are in contact with an average of 93 percentages
36
Soviet Onhand Personnel in Contact/Not in Contact. The ratio of contact
personnel strength increases until the eight day
37
German Onhand Personnel in Fighting/Not Fighting. On the first day, the
fighting personnel ratio is less than 40 percent
37
Soviet Onhand Personnel in Fighting/Not Fighting. On the first day, the
Soviets were not engaged with the enemy
38
German Personnel Losses in Contact/Not in Contact. Almost all German
personnel losses come from the contact units
39
Soviet Personnel Losses in Contact/Not in Contact. The majority of the
Soviet personnel losses come from the contact units
40
German Personnel Loss Ratios for Fighting/Not Fighting Combat Units in
Contact. Except for the 13th and 14th days, almost all the German losses
come from the fighting units
40
Soviet Personnel Loss Ratios for Fighting/Not Fighting Combat Units in
Contact. The Soviets do not have any losses in the first day for the
Fighting units
41
Comparison of Daily Personnel Losses in ACUD for Both Forces
41
Comparison of Daily Personnel Losses in FCUD for Both Forces
42
German Onhand Tanks in Contact/Not in Contact. Almost all German
personnel are in contact with an average of 90 percent
43
Soviets Onhand Tanks in Contact/Not in Contact. The ratio of contact tank
numbers increase until the eight day
44
German Tank Strength Ratios for Fighting/Not Fighting Combat Units in
Contact. On days 1 and 13, the fighting tank strength ratio is less than 40
percent
44
Soviet Tank Strength Ratios for Fighting/Not Fighting Combat Units in
Contact. On the first day, all Soviet tank strength is in not fighting status. ...45
Comparison of Daily Tank Strength for the Germans and the Soviets
46
XI

Figure 3.21.
Figure 3.22.
Figure 3.23.
Figure 4.1a.

Comparison of Daily Tank Strength in FCUD for Both Forces
47
Comparison of Tank Losses of Both Forces in ACUD
47
Comparison of Tank Losses of Both Forces in FCUD
48
The German Combat Power Casualty Percentages Versus Tank Loss
Percentages
:
59
Figure 4.1b. The Soviet Combat Power Casualty Percentages Versus Personnel Loss
Percentages
60
Figure 4.2.
Fitted Versus Actual for German Combat Power Losses Applying the
ATLAS Attrition Process to All Combat Unit Data. The model
overestimates casualties on most days of the battle
68
Figure 4.3.
Fitted Versus Actual for German Personnel Losses. The model
overestimates battle casualties except for the first day
69
Figure 4.4.
Fitted Versus Actual Tank Losses. The trend of the model is very
plausible. There is no significant outlier
69
Figure 4.5.
Fitted Versus Actual for Soviet Combat Power Losses. The peak four
points are the days when the Soviets attack. The model mostly
underestimates the battle, with the last two days being a noticeable
exception
70
Figure 4.6.
Fitted Versus Actual Soviet Personnel Losses. Except the last three days,
the model underestimates the battle. Day 8 is the bloodiest day of the
battle
70
Figure 4.7.
Fitted Versus Actual Soviet Tank Losses. The model underestimates battle
casualties except for the last three days. Again, day 8 is the heaviest tank
battle in history
71
Figure 4.8.
Fitted Versus Actual for the German Combat Power Losses Applying
ATLAS Attrition Process to Contact Combat Unit Data. The model
overestimates on most days except for the second, fifth and eighth days
80
Figure 4.9.
Fitted Versus Actual for the German Personnel Losses. The model
overestimates battle casualties except for the first and eighth day
80
Figure 4.10. Fitted Versus Actual Tank Losses. The model catches the battle trend, but
overestimates battle casualties towards the end. There is no significant
outlier
81
Figure 4.11. Fitted Versus Real for Soviet Combat Power Losses. The peak four points
are the days when the Soviets attack. The model mostly underestimates
battle casualties
81
Figure 4.12. Fitted Versus Real the Soviet Personnel Losses. Except for the last two
days, the model underestimates battle casualties. Day 8 is the bloodiest day
of the battle
82
Figure 4.13. Fitted Versus Real Soviet Tank Losses. The model underestimates the
battle except for the last three days. Again, day 8 is the heaviest tank battle
in history
82
Figure 4.14. Fitted Versus Actual for the German Combat Power Losses Applying
ATLAS Attrition Process to Fighting Combat Units Data. On days 2, 5,
and 9, the model fits the battle very well
89
xu

Figure 4.15.

Figure 4.16.
Figure 4.17.
Figure 4.18.
Figure 4.19.

Figure 4.20.
Figure 4.21.

Figure 4.22.

Figure 4.23.

Figure 4.24

Figure 4.25.

Figure 4.26.
Figure 4.27.

Figure 4.28.

Figure 4.29.

Fitted Versus Actual for the German Personnel Losses. The model
overestimates battle casualties except for the first, eighth and the ninth
days
90
Fitted Versus Actual Tank Losses. The trend of the model is fairly good.
There is no significant outlier
90
Fitted Versus Real for Soviet Combat Power Losses. The model mostly
underestimates battle casualties
91
Fitted Versus Real Soviet Personnel Losses. The model underestimates
battle casualties, but gives a better fit towards the end of the battle
91
Fitted Versus Real Soviet Tank Losses. The model underestimates battle
casualties until the ninth day and it fits the battle fairly well on the last four
days. 92
Total Sorties in Each Aircraft Role
93
Fitted Versus Actual for the German Combat Power Losses with Air
Sorties in the FCUD Data Set. The figure has the same pattern as Figure
4.2. The model's trend is good, but overestimates in some parts, especially
towards the end
98
Fitted Versus Actual Soviet Combat Power Losses with Air Sorties in the
FCUD Data Set. The figure has the same pattern as Figure 4.5. The model
underestimates battle casualties
98
Estimated Versus Actual German Combat Power Losses with Air Sorties
Using Bracken's Weights for the FCUD Data Set. The model
overestimates battle casualties specifically after the 8th day of the battle.... 102
Estimated Versus Actual Soviet Combat Power Losses with Air Sorties
Using Bracken's Weights for the FCUD Data Set. The model
underestimates most of the battle except for the last four days
103
Estimated Versus Actual German Combat Power Losses in the ACUD
Data. The model dramatically overestimates the casualties until the eight
day. The high overestimation during the first four days is due to the lower
German force ratios
113
Estimated Versus Real German Armor Losses in the ACUD Data. The
general pattern of the model is good
114
Estimated Versus Real Soviet Combat Power Losses in ACUD Data. The
model overestimates the battle on days 8, 9, 13 and 14. On the other days,
on which the Soviets were the defender, the general pattern is very
plausible
114
Estimated Versus Real Soviet Armor Losses in the ACUD Data. The
model generally underestimates casualties, however it overestimates on
days 8 and 9 when the Soviets attacked. Also, the model caught the spike
in casual ties
115
Estimated Versus Actual German Combat Power Losses in the CCUD
Data. The model overestimates the battle on most days but as a whole, the
pattern is not bad
115
Xlll

Figure 4.30.

Figure 4.31.

Figure 4.32.

Figure 4.33.
Figure 4.34.
Figure 4.35.

Figure 4.36.

Figure 4.37.
Figure 4.38.

Figure 4.39.

Figure 4.40.
Figure 4.41.

Figure 4.42.

Figure 4.43.

Figure 4.44.

Estimated Versus Actual German Armor Losses in the CCUD Data.
Although the model highly overestimates the casualties on the first two
days, the general pattern of the model is good
116
Estimated Versus Real Soviet Combat Power Losses in the CCUD Data.
The model overestimates the battle on days 8, 9, 13 and 14. However, the
model catches the trend
116
Estimated Versus Real Soviet Armor Losses in the CCUD Data. Although
the model underestimates the casualties except for the days on which the
Soviets attacked, the overall pattern is very plausible
117
Estimated Versus Actual German Combat Power Losses in the FCUD
Data. The general pattern of the model is good
in
Estimated Versus Actual German Armor Losses in the FCUD Data. The
model overestimates the battle except for the last three days
118
Estimated Versus Real Soviet Combat Power Losses in the FCUD Data.
The model overestimates the battle on days 8, 9, 13 and 14. On the other
days, the overall pattern is good
118
Estimated Versus Real Soviet Armor Losses in the FCUD Data. The
model fits well on the 8th day of the battle. Although the model
underestimates the casualties during the first days of the battle, the general
pattern is good
119
Estimated Versus Real FMR for the Germans in the ACUD Data Set. The
model fits fairly well except for the 6th day
121
Estimated Versus Real FMR for the Germans in the CCUD Data Set. The
model highly overestimates the battle during the first four days. However,
towards the end, the fit is fairly good
122
Estimated Versus Real FMR for the Germans in the FCUD Data Set. The
model overestimates the battle for each day. Again, during the first days,
the overestimation is very high
122
Estimated Versus Real German Combat Power Losses in the ACUD Data
Set. The model underestimates the battle for the first 5 days
134
Estimated Versus Real German Personnel Losses in the ACUD Data Set.
The model underestimates the first 3 days, and overestimates the last 4
days. There is not any significant outlier which is supported by its
R2 value. The general pattern of the model is fairly good
135
Estimated Versus Actual German Tank Losses in the ACUD Data Set. On
most days, the model underestimates the battle. However, the general trend
is good
135
Estimated Versus Actual Soviet Combat Power Losses in the ACUD Data
Set. Although, the model overestimates the battle on the 9th day and for the
last two days, the general pattern is not bad
136
Estimated Versus Actual Soviet Personnel Losses in the ACUD Data Set.
The model mostly underestimates the battle except for the last four days.
Overall, the trend of the model is good
136

XIV

Figure 4.45.
Figure 4.46.
Figure 4.47.
Figure 4.48.
Figure
4.49.
•■a"

Estimated Versus Actual Soviet Tank Losses in the ACUD Data Set. The
model overestimates the battle during the whole campaign. It has a similar
pattern with the combat power figure
137
Estimated Versus Real German Combat Power Losses in the CCUD Data
Set. The model underestimates the battle for the first 5 days. It has the
same pattern as the one in the ACUD data set
137
Estimated Versus Actual Soviet Combat Power Losses in the CCUD Data
Set. The model overestimates the battle on the 9th day and for the last four
days. Except for the last days, the general pattern is not bad
138
Estimated Versus Real German Combat Power Losses in the FCUD Data
Set. The model mostly underestimates the battle, but it gives a good fit
towards the end
138
Estimated Versus Actual Soviet Combat Power Losses in the FCUD Data
Set. The model overestimates the battle on the 9th day and for the last two
days. Overall, the general pattern is not bad
139

XV

THIS PAGE INTENTIONALLY LEFT BLANK

XVI

LIST OF TABLES

Table 3.1.
Table 3.2.
Table 3.3.

Table 3.4.

Table 3.5.
Table 3.6.
Table 3.7.
Table 3.8.
Table 3.9.
Table 3.10.
Table 3.11.
Table 3.12.
Table 4.1.
Table 4.2.
Table 4.3.

Table 4.4.

Table 4.5.

Table 4.6.
Table 4.7.

Daily German Onhand Personnel and Weapon Data. All Combat Units are
included
29
Daily Soviet Onhand Personnel and Weapon Data. All Combat Units are
included
30
Daily German Personnel and Weapon Losses. Notice that on the first day
the losses are very small. Almost all losses show a descending pattern
indicating that the intensity of the battle is decreasing
30
Daily Soviet Personnel and Weapon Losses. Notice that on the first day
losses are very small, like the German losses. Almost all losses show a
descending pattern except for the ninth day of the battle. This day is the
bloodiest tank battle in history
31
Daily German CCUD Onhand Personnel and Weapon Data
31
Daily Soviet CCUD Onhand Personnel and Weapon Data
32
Daily German CCUD Personnel and Weapon Losses
32
Daily Soviet CCUD Personnel and Weapon Losses
33
Daily German FCUD Onhand Personnel and Weapon Data
33
Daily Soviet FCUD Onhand Personnel and Weapon Data
34
Daily German FCUD Personnel and Weapon Losses
34
Daily Soviet FCUD Personnel and Weapon Losses
35
RAND's Firepower Score Values. The weapon categories are not
comprehensive. More details can be found in [Ref. 7:p. 88]
52
Firepower Score Values Used for the Nine Weapon Groups in KDB.
These scores are computed relative to RAND's firepower scores
52
Daily German Combat Power Values for Personnel and Weapon Type.
The last column shows the aggregate combat power of the Germans on
each day. Notice that the artillery combat power values do not change
much
55
Daily Soviet Combat Power Values for Personnel and Weapon Type. The
last column shows the aggregate combat power of the Soviets on each day.
The tank combat power values decrease dramatically during the campaign. 55
This Table Presents the Combat Power of Both Forces, Attacking Side,
Defenders Combat Posture, and Attackers Force Ratio. Notice that for
each day the Soviet combat power is significantly greater than the German
combat power
56
The Variables Used in Equations 4.1 and 4.2 to Estimate Casualties
57
Estimated Attacker Combat Power Casualty Rates are Always Higher than
the Defenders. For the first five days casualty rates are very close for both
sides. The Soviets have nearly the same estimated casualty rate for the
days they attack. Likewise, the Germans have consistent estimated
casualty rates on the days they defend
57

xvu

Table 4.8.
Table 4.9.
Table 4.10.
Table 4.11.
Table 4.12.

Table 4.13.
Table 4.14.
Table 4.15.
Table 4.16.

Table 4.17.
Table 4.18.
Table 4.19.
Table 4.20.

Table 4.21.

Table 4.22.

Table 4.23
Table 4.24.

Table 4.25.

Table 4.26.

German Regression Results for All Combat Unit Data. Only the RKTL
weapons result is not significant
61
Soviet Regression Results for Contact Combat Unit Data. The results are
not significant only for AA weapons
61
The Regression Results for the Hypothesis that the German and Soviet
Losses are the Same for the ACUD Data Set
63
Estimated Daily German Loss Percentages. Due to rounding off, some
days seem to have the same rate. Note the high estimated tank losses
65
Estimated Daily Soviet Loss Percentages. Due to rounding off, some days
seem to have the same rate. The Soviets have higher estimated casualty
rates when they attack
66
2
2
R Values of Personnel Casualties and Weapon Losses. Tank R value for
the Germans indicates a better fit
67
The p-Values from the Wilcoxon Signed-Rank Test for ACUD Data Set.
The non-significant values are highlighted
67
Estimated Daily German and Soviet Combat Power. Notice that the range
of force ratio is [0.63-1.45]
73
Attacker Casualty Rates are Always Higher than the Defenders. The
Soviets have lower estimated casualty rates than the Germans except on
the days they attack
75
German Regression Results for Contact Combat Unit Data. The results are
not significant only for RKTL weapons
75
Soviet Regression Results for Contact Combat Unit Data. The results are
not significant only for RKTL and AA weapon classes
76
The Regression Results for the Hypothesis that the German and the Soviet
Losses are the Same for the CCUD Data Set
76
Estimated Daily German Loss Percentages. Due to rounding off, some
days seem to have the same rate. A Large amount of tank losses is
estimated
77
Estimated Daily Soviet Loss Percentages. Due to rounding off, some days
seem to have the same rate. The Soviets have higher estimated casualty
rates when they attack
77
2
R Values of Personnel Casualties and Weapon Losses. Tank and combat
power values for the Germans and the APC value of the Soviets indicate a
better fit
79
The p-Values from the Wilcoxon Signed-Rank Test for CCUD Data Set.
The non-significant values are highlighted
79
Estimated Daily the German and the Soviet Combat Power. Notice that the
range of force ratio is [0.30-2.56]. On the first day, the force ratio is very
close to the 3-1 traditional attacker force ratio
85
Attacker's Estimated Casualty Rates are Always Higher than the
Defenders. The Soviets have lower estimated casualty rates than the
Germans except for the days they attack
86
German Regression Results for Contact Combat Unit Data. The results are
not significant only for RKTL weapons
86
xviii

Table 4.27.
Table 4.28.
Table 4.29.
Table 4.30.
Table 4.31.
Table 4.32.
Table 4.33.

Table 4.34.
Table 4.35.
Table 4.36.
Table 4.37.
Table 4.38.

Table 4.39
Table 4.40.
Table 4.41.
Table 4.42.
Table 4.43.

Soviet Regression Results for Contact Combat Unit Data. The results are
not significant only for RKTL and AA weapon classes
87
The Regression Results for the Hypothesis that the German and Soviet
Losses are the Same for the FCUD Data Set
87
2
R Values of Personnel Casualties and Weapon Losses. Tank, personnel,
APC, ATH, Flame/MG, and combat power values for the Germans and the
APC value indicate a better fit. The APC losses fit better for both sides 88
The p-Values from the Wilcoxon Signed-Rank Test for the FCUD Data
Set. The non-significant values are highlighted
88
The Daily Number of Ground Attack and Bombing Air Sorties for Both
Sides. The Germans generated more ground attacks during the first two
days
94
Firepower Score Values for 11 Weapon Groups. These scores are
computed relative to RAND's firepower score values
95
2
R Values for both Sides in Different Data Sets with Air Sorties. The APC
values are almost positive in all models for both sides. The only positive
personnel value is seen for the Germans in the FCUD data set. The
German combat power values are positive in all three data sets
96
The p-Values from the Wilcoxon Signed-Rank Test for all Data Sets for
Both Sides. The non-significant values are highlighted
96
Firepower Score Values for 11 Weapon Groups. Tank, APC, and Artillery
Values are from Bracken's Study. Others are computed relative to
RAND's values
100
R2 Values for Both Sides in Different Data Sets with Air Sorties Using
Bracken's Weights
100
The-p-Values from the Wilcoxon Signed-Rank Test for all Data Sets for
Both Sides. The significant values are highlighted
101
Force Ratio, Estimated Combat Casualty Rates for Each Side in ACUD
Data Set. Notice that for the first four days, the German estimated casualty
rates are very high. The Soviets have higher estimated casualty rates on the
days they attacked
109
2
R Values for both Sides in the Three Data Sets. The value of German
armor in the ACUD data and the combat power value in FCUD data are
very plausible. Other values are very poor, mostly negative and very low. .111
The p-Values from the Wilcoxon Signed-Rank Test for all Data Sets for
both Sides. The non-significant values are highlighted
111
Daily Average German Northbound Progress Increased Every Day Except
for the 13th Day (17 July)
120
Estimated and Real FMR for the Germans. FMR denotes the FLOT
movement rate. The days are the actual dates of the battle..
120
Daily the German Combat Power and Factors Used in its Calculation. The
last three columns show the German combat power ratio for all three data
sets. The Pg and Ps denote the combat power of the Germans and Soviets
respectively.
126
XIX

Table 4.44.

Table 4.45.
Table 4.46.
Table 4.47.
Table 4.48.
Table 4.49.
Table 5.1.

Daily the Soviet Combat Power and Factors Used in its Calculation. Ts
last three columns show the Soviet combat power ratio for all three das
sets. The Pg and Ps denote the combat power of the Germans and Soviet
respectively
;
Daily the German Factors used in Equation 4.7
1
Daily the Soviet Factors Used in Equation 4.7
12
The Values of the Factors Used in Equations 4.11 and 4.12 for all of the
Data Sets
131
2
The R Values for both Sides in the Three Data Sets
132
The p-Values from the Wilcoxon Signed-Rank Test for all of the Data Sets
for both Sides. The significant values are highlighted
133
Results of All the Models Investigated in Chapter TV
149

XX

LIST OF SYMBOLS, ACRONYMS AND ABBREVIATIONS

FPI: Fire Power Index .
FR: Force Ratio
FEB A: Forward Edge of the Battle Area
FLOT: Forward Line of Troops
ACUD: All Combat Unit Data
CCUD: Contact Combat Unit Data
FCUD: Fighting Combat Unit Data
QJM: Quantified Judgement Models
SFS: Situational Force Scoring
TNDM: Tactical Numerical Deterministic Model
TDI: The Dupuy Institute
CAA: US Army Concepts Analysis Agency
ARCAS: The Ardennes Campaign Simulation Study
KDB: Kursk Data Base
KOSAVE: The Kursk Operation Simulation and Validation Exercise
OH: On hand
HQ: Headquarters
KIA: Killed In Action
WIA: Wounded In Action
CMIA: Captued/Missing In Action
DNBI: Disease/Nonbattle Injuries
SP: Self Propelled
ARTY: Artillery
APC: Armored Personnel Carrier
RKTL: Rocket Launcher
ATH: Heavy Antitank Weapon
LTH: Light Antitank Weapon
MTR: Mortar
xxi



Flame/MG: Flame-throwers and heavy machineguns.



AA Antiaircraft Weapons



WEI/WUV: Weapon Effectiveness Index, Weighted Unit Value
KV: Killer-Victim

XXll

THIS PAGE INTENTIONALLY LEFT BLANK

XXlll

EXECUTIVE SUMMARY

"War is a matter of vital importance to the state; the province of life or death; the
road to survival or ruin. It is mandatory that it be thoroughly studied." Sun Tzu, one of the
most famous military thinkers in history, described war and pointed out the importance of
the studies of war over 1500 years ago in his book The Art of War [Ref. 1].
Since the dawn of history scientists, researchers, and the military have tried to
develop fundamental laws or theories that explain the interactions of military forces in
combat and the outcomes of battles. Combat models are widely used in battle planning,
wartime operations, force sizing, human resource planning, logistics planning, national
policy analysis, and the decision process for the acquisition of weapon systems [Ref. 4:p.
6-7].
Combat models are categorized into two groups according to the level of
representation of the combatants. The first group is high-resolution combat models in
which each combat vehicle or soldier is explicitly represented as an entity. The second
group is low-resolution (aggregated) combat models, which represents battalion and
higher units as an entity. The issue of aggregation has been addressed by many authors
over the years. Many of these studies on attrition methodology include the subjects of
Lanchester's Law, data sources, such as historical vs. engineering vs. empirical, and a
combination of the two which are scoring systems such as Firepower Scores. The primary
focus of this thesis is the validity of the Firepower Score approach to attrition in lowresolution (aggregated) combat models.

XXIV

Attrition is a reduction in the number of personnel, weapons, and equipment in a
military unit, organization, or force [Ref. 3:p. 1]. Combat attrition is one of the most
important aspects of combat modeling. Combat attrition is the only combat process for
which well-developed mathematical theories exist [Ref. 6:p. 4-1]. Even though attrition is
the most studied combat process, there is no agreement on the best way to model it. The
main reason is the lack of real data, which can be used to validate combat attrition
models.
Aggregated attrition process models can be categorized into two basic types that
correspond to the two basic entity aggregation patterns-homogeneous and heterogeneous
[Ref. 6:p. 4-2]. The basic idea of homogenous force ratio attrition models is to aggregate
all the individual combatants in a unit into a scalar measure of the unit's combat power
[Ref. 6:p. 4-3]. The Firepower score approach is used in, aggregated, large-scale, combat
simulations as the primary descriptor of what a combat unit is worth [Ref. 6:p. 2-5]. The
ratio of attacker to defender combat power is used to determine the casualties for both
sides.
The quality of the available historical data for validation of attrition models is
very poor. The most accessible battle data contain only starting sizes and casualties and
sometimes only for one side [Ref. 5:p. 470]. Recently, more data has become available.
Improved database management and computing power have helped in gathering such
data.
Detailed databases on the Battle of Kursk, the largest tank battle in history, and
the Ardennes Campaign of World War II, have recently been developed. Both data sets

XXV

are two-sided, time-phased (daily) and detailed. Hartley and Helmbod pointed out that
unless we are able to procure time-phased detailed data, we will not be able to validate
any attrition model [Ref. 5:p. 89]. In this thesis, we focus on time-phased, highly detailed,
two-sided data.
Most of the past empirical validation studies have focused on the Lanchester
Equations, which were introduced by the English engineer, Frederick William Lanchester.
These studies include the works of Bracken [Ref. 8] and Flicker [Ref. 9] on the Ardennes
Campaign of World War E, Clemens [Ref. 10] and Turkes [Ref. 11] on the Battle of
Kursk of World War n, and Hartley and Helmbold [Ref. 12] on the Inchon-Seoul
Campaign of the Korean War. These works are among the few quantitative studies that
use daily force size data for real battles.
Besides the Lanchester equations, another approach for combat attrition is models
that use force ratio in their structure. This thesis focuses on aggregate attrition
methodologies that use combat power ratio to compute the casualties of the forces. Unlike
the Lanchester equations, there is no study in the literature that used firepower score
attrition models on real data in which force sizes are available day by day for both sides.
One of the interesting aspects of the Battle of Kursk is the engagement
percentages of the forces. The Germans had a considerably larger portion of their forces
in contact. This suggests that the German force may have been subjected to more fatigue
than the Soviets. With an average of 97 percent of its heavy mechanized force on the
front lines, the Germans, unlike the Soviets, had no reserves to use [Ref. 14:p. 9-4].
Depending on combat engagement intensity, this study extracts three different data sets

XXVI

from the data on the Battle of Kursk in the KOS AVE report.
The first data set, all combat units data (ACUD), includes all combat units:
contact, out of contact, active, and inactive, including HQ above division level. The
second data set, contact combat units data (CCUD), includes only combat units those are
in contact with the enemy: units in contact fighting and not fighting, HQ above division
level is excluded [Ref. 14:p. 5-9]. The third data set, fighting combat units data (FCUD),
includes only combat units that are in contact and fighting with the enemy: HQ above
division level are excluded [Ref. 14:p. 5-9].
In this research, three Firepower score models are applied to the three data sets
that are extracted from the data on the Battle of Kursk. The first model is the ATLAS
ground attrition model, which is used in the TACWAR simulation [Ref. 24]. The second
model is RAND's SFS model, which was proposed in 1991 and is used in the JICM
simulation [Ref. 25]. The last model is the simplified QJM model, developed by Trevor
Dupuy. Instead of focusing only on one model and investigating it in detail, the
applicability of the three primary firepower score models to real data is investigated. In
addition, some insight is given about the attrition processes and other factors used in
aggregated combat models.
The key findings from this research include:


Of all the models looked at, when combat power losses are considered, the
ATLAS model with the air sortie data fits best.



Generally, the models overestimate the attacker's casualties during the
battle.



Overall, all of the models fit better for the Germans than the Soviets. In his
study [Ref. 11], Turkes also found that his models fit better for the
Germans.
xxvn

In all of the models and for both sides, the FCUD data set gives the best
fit.
One of the difficulties with aggregated combat attrition models that use
force ratio is the need to determine the attacking side. It is always not very
easy to determine the attacking side.
Prior to a battle, it is difficult to determine factors such as intensity and
nationality factors.
One of the problems with traditional force ratio models is that the loss
rates in each weapons category are the same as the combat power casualty
rate. For instance, if the combat casualty rate is 4 percent, then each
weapons category will take 4 percent losses. However, this does not match
either the historical facts or the results from the higher-resolution combat
models. In this thesis, the linear regression analysis is used to determine
how to allocate the combat power casualty rates to the different weapon
groups.
Due to the general overestimation of the German casualties and the
underestimation of the Soviet casualties, anything that improves the force
ratio with respect to the Germans improves the quality of the fits.
Anything that added to German effectiveness or cut Soviet effectiveness
could improve the quality of the fits.
Using different firepower scores, like Bracken's weights, does not give a
better fit except for the Soviet values in the CCUD data set, which is
slightly better. Much more work is needed to find the best firepower
scores, such as optimization of the score values, and sensitivity analysis.
The FLOT movement rate is only computed in the SFS model. The
ATLAS model has also look up tables to compute the movement rates
according to the force ratio, terrain, and combat postures. However, in the
tables, the force ratio threshold is higher than the ones computed for this
battle. The force ratio of the Germans in the ATLAS model is very low.
As a result, it was not possible to compute the FLOT movement rates in
the ATLAS model.
This analysis is based on observational census data of the Battle of Kursk
of World War n, and may not generalize, since it is not a random sample
of a larger population. The outcome of a battle cannot be precisely
determined with the use of combat models. They might provide insights
into future battles between adversaries. Besides being used to gain insight
into the battles, which occurred in the past, they should help in making
better decisions by enabling the decision-maker to compare the different
alternatives by using various combat model techniques [Ref. ll:p 145].

XXVlll

THIS PAGE INTENTIONALLY LEFT BLANK

XXIX

ACKNOWLEDGMENTS

I would like to express my sincere thanks to Prof. Lucas and LTC Appleget for
their guidance and patience during the work in performing this thesis.
I would like to thank my family for their great support.

XXX

I.

INTRODUCTION

"War is a matter of vital importance to the state; the province of life or death; the
road to survival or ruin. It is mandatory that it be thoroughly studied." Sun Tzu, one of the
most famous military thinkers in history, described war and pointed out the importance of
the studies of war over 1500 years ago in his book The Art of War [Ref. 1]. Clausewitz
defined war in his book, On War [Ref. 2], as "War is thus an act of force to compel our
enemy to do our will." Throughout history, war has been a topic of analysis for scientists
and researchers.
A.

COMBAT MODELING
An understanding of combat phenomena is facilitated by using a hierarchy of

combat to describe combat events and aggregate them for analysis [Ref. 3:p. 153]. A
commonly accepted hierarchy of combat is as follows: War, already defined, is at the top
of the hierarchy. A campaign is a phase of war involving a series of battles related in time
and space with the aim of achieving a single, specific objective. A battle is combat
between major forces, each having opposing assigned or perceived operational business.
An engagement is combat between two forces from battalion to division level. An action
is combat between squad or battalion level. A duel is combat between two individuals.
Since the dawn of history scientists, researchers, and the military have tried to
develop fundamental laws or theories that explain the interactions of military forces in
combat and the outcomes of battles. Combat models are widely used in battle planning,
1

wartime operations, force sizing, human resource planning, logistics planning, national
policy analysis, and the decision process for the acquisition of weapon systems [Ref. 4:p.
6-7].
In Operation Desert Storm an existing theater-level simulation, Concept
Evaluation Model (CEM), was used by analysts at the US Army Concepts Analysis
Agency (CAA) [Ref. 5:p. 549] to simulate the Desert Storm scenario for assisting in warplanning and war-fighting efforts. The actual efforts of their studies became clear when a
letter from LTG Reimer to the Director of CAA, Mr. E.B. Vandiver, arrived in February
1991. The letter, in part, stated, "The analytical support you provided for Operation
Desert Shield and Desert Storm has been absolutely outstanding. [It was] used by the
Army Staff, the Joint Staff and our Army in Southwest Asia to prepare for war. The Army
leadership used it for discussion and briefings with key military and civilian leaders,
including the National Command Authority [Ref. 5:p. 559]."
Combat models are categorized into two groups according to the level of
representation of the combatants. The first group is high-resolution combat models in
which each combat vehicle or soldier is explicitly represented as an entity. The second
group is low-resolution (aggregated) combat models, which represents battalion and
higher units as an entity. The issue of aggregation has been addressed by many authors
over the years. Many of these studies on attrition methodology include the subjects of
Lanchester's Law, data sources, such as historical vs. engineering vs. empirical, and a
combination of the two which are scoring systems such as Firepower Scores. The primary

focus of this thesis is the validity of the Firepower Score approach to attrition in lowresolution (aggregated) combat models.
B.

ATTRITION
Attrition is a reduction in the number of personnel, weapons, and equipment in a

military unit, organization, or force [Ref. 3:p. 1]. Combat attrition is one of the most
important aspects of combat modeling. Combat attrition is the only combat process for
which well-developed mathematical theories exist [Ref. 6:p. 4-1]. Even though attrition is
the most studied combat process, there is no agreement on the best way to model it. The
main reason is the lack of real data, which can be used to validate combat attrition
models. It is useful to be able to predict combat attrition accurately in order to provide
estimates of requirements for the planning process for medical, logistics and personnel
training [Ref. 3:p. 2].
Aggregated attrition process models can be categorized into two basic types that
correspond to the two basic entity aggregation patterns—homogeneous and heterogeneous
[Ref. 6:p. 4-2]. A heterogeneous aggregated attrition model assesses the amount of
attrition caused by a weapon system class against each enemy weapon system class. Thus,
the interactions between different weapon groups (i.e., who kills who) are implemented in
this type of attrition processes. In a homogenous aggregated attrition process, all of the
weapon groups are aggregated with their weights into the combat power of a single unit.
Most homogeneous attrition models determine the amount of combat power attrition by

computing attacker to defender force ratios [Ref. 6:p 4-2]. The interactions between
different weapon groups are not considered in homogeneous attrition models.
C.

FIREPOWER SCORE APPROACH
The basic idea of homogenous force ratio attrition models is to aggregate all the

individual combatants in a unit into a scalar measure of the unit's combat power [Ref.
6:p. 4-3]. The Firepower score approach is used in, aggregated, large-scale, combat
simulations as the primary descriptor of what a combat unit is worth. [Ref. 6:p. 2-5]. The
ratio of attacker to defender combat power is used to determine the casualties for both
sides.
In the Firepower score approach, the combat power of a unit is computed by
summing the combat power value for each weapon system in the unit. In Parry's notes
[Ref. 6:p. 4-5] the combat power computation is given in a simple equation as follows:
Suppose that there are n different types of weapon system in a combat unit and that:
Xi: the number of weapons of type i in the unit [i=l,2,3.. .n]
Si: the firepower score value representing the combat power for each type i
weapon. Then, the firepower index of the aggregated unit is

FPi^xrs,

(i.i)

Finally, the force ratio is determined as:
FR = FPI(A)/FPI(D)

(1.2)

Where:
FPI(A): the firepower index of the attacking forces
FPI(D): the firepower index of the defender.
The force ratio gives a measure of relative combat power in the battle. The force
ratio in many aggregated combat models, such as TACWAR, is used to compute
casualties for both sides in a battle and to determine the FEBA (forward edge of the battle
area) or FLOT (forward line of troops) movement rates.
The method of determining the firepower scores is a very difficult problem. There
are several methods of computing firepower score values, such as military judgement and
experience (RAND's ground force scoring system [Ref. 7]), historical combat
performance derived from WWII and the Korean War, and results from high resolution
simulations (i.g., Anti-Potential-Potential Method) [Ref. 6:p. 2-6].
There is no published validation study in the literature using firepower score
approaches on real data in which force sizes are available day by day for both sides. This
thesis describes how different firepower score approaches fit to the data on the Battle of
Kursk. This data is time-phased, two-sided and very detailed. This study will help
analysts make better decisions and perhaps provide a better understanding of war by
adding to an understanding how combat models fit to real data. The next section presents
the outline of the thesis.

D.

THESIS OUTLINE
This thesis consists of five chapters. This first chapter introduces the general

concept of combat modeling and firepower score approaches used in the attrition process
of aggregated combat models. The second chapter gives a brief history of the Battle of
Kursk of World War II and analyzes the battle's data. Three different data sets are
extracted from the data on the Battle of Kursk according to the combat engagement
intensity. These data sets are all combat units data (ACUD), contact combat units data
(CCUD), and fighting combat units data (FCUD).
In the fourth chapter, three force ratio attrition models that use the firepower score
approach are applied to the three data sets described above. Chapter five presents the final
conclusions and recommendations based on the results and also recommends future areas
of study in combat modeling.

II.

A.

PREVIOUS VALIDATION STUDIES ON COMBAT
MODELING

PREVIOUS STUDIES WITH LANCHESTER EQUATIONS
The quality of the available historical data for validation of attrition models is

very poor. The most accessible battle data contain only starting sizes and casualties and
sometimes only for one side [Ref. 5:p. 470]. Recently, more data has become available.
Improved database management and computing power have helped in gathering such
data.
Detailed databases on the Battle of Kursk, the largest tank battle in history, and
the Ardennes Campaign of World War II, have recently been developed. Both data sets
are two-sided, time-phased (daily) and detailed. Hartley and Helmbod pointed out that
unless we are able to procure time-phased detailed data, we will not be able to validate
any attrition model [Ref. 5:p. 89]. In this thesis, we focus on time-phased, highly detailed,
two-sided data.
Most of the past empirical validation studies have focused on the Lanchester
Equations, which were introduced by the English engineer, Frederick William Lanchester.
These studies include the works of Bracken [Ref. 8] and Flicker [Ref. 9] on the Ardennes
Campaign of World War E, Clemens [Ref. 10] and Turkes [Ref. 11] on the Battle of
Kursk of World War n, and Hartley and Helmbold [Ref. 12] on the Inchon-Seoul

7

Campaign of the Korean War. These works are among the few quantitative studies t
use daily force size data for real battles.
In his study [Ref. 8], Bracken found that the Lanchester linear model best fits the
Ardennes campaign data. Flicker [Ref.7] revisited Bracken's modeling of the Ardennes
campaign. In contrast to Bracken, Flicker found that the Lanchester linear and square
laws do not fit the data. He concludes that a new form of the Lanchester equations, with a
physical interpretation closest to Lanchester's logarithmic law, applies best.
Clemens [Ref. 10] applied the Lanchester Equations to the data on the Battle of
Kursk. Clemens used two estimation techniques: linear regression and Newton-Raphson
iteration. He concludes that neither the Lanchester linear nor the Lanchester square model
fit the data. The Lanchester logarithmic model fits better than the Lanchester linear and
square models.
Hartley and Helmbold's study [Ref. 12] focused on validating the homogeneous
Lanchester square law by using Inchon-Seoul Campaign data. Hartley and Helmbold use
three analysis techniques to examine the data: linear regression, the Akaike Information
Criterion (AIC) and Bozdogan's consistent AIC (CAIC). They find that the data do not fit
a constant coefficient Lanchester square law. They conclude that, by dividing the
campaign into three distinct battles that each battle's data can be fit to a constant
coefficient Lanchester square law, using separate coefficients for each battle.
In his study Turkes [Ref. 11] applies a total of 39 diverse models to the data on
the Battle of Kursk using different approaches. These approaches include applying the

8

methodologies of previous studies, using robust LTS (least trimmed squares) regression,
including the air sortie data of the battle, considering the battle in separate phases, fitting
basic Lanchester equations and using different weights [Ref 9]. He concludes that:

B.



None of the original Lanchester equations applies very well to the data on
the Battle of Kursk. The best fits are implausible.



The parameters derived from Bracken and Flicker's Ardennes studies do
not apply to the data on the Battle of Kursk. This implies that there are no
unique parameters that apply to all battles.



The Robust LTS regression method is the best analytic technique for
estimation of parameters.

PREVIOUS STUDIES WITH FIREPOWER SCORE APPROACHES
Besides the Lanchester equations, another approach for combat attrition is models

that use force ratio in their structure. This thesis focuses on aggregate attrition
methodologies that use combat power ratio to compute the casualties of the force. Unlike
the Lanchester equations, there is no study in the literature that used firepower score
attrition models on real data in which force sizes are available day by day for both sides.
In this research, the ATLAS ground attrition equations [Ref. 13], RAND's
Situational Force Scoring (SFS) [Ref. 7] and Dupuy's Quantified Judgement Models
(QJM) [Ref. 3] are applied to three data sets that are extracted from the data on the Battle
of Kursk.
The ATLAS theater level simulation uses a straightforward force ratio method.
The simplicity of its structure is one of the main attractions of the ATLAS model.
TACWAR is one of the simulations that use the ATLAS equations. In the combat
attrition process of the ATLAS model, the casualty rates are determined by using simple

equations for the attacker and the defender. The original casualty rates used in the
ATLAS model were derived from data on 37 division level engagements in World War II
and Korea [Ref. 6:p. 4-9]. Since the specific engagements are not documented, it is
unknown as to whether the division-level data includes the battle of Kursk—though it is
believed not to. If it is included, then the comparisons are not strictly independent.
However, since Kursk would be the only one of a large number of engagements (37) the
dependence will be very small. There is no published study on the validation of ATLAS
equations for real combat data.
The second method used in this study is RAND's SFS methodology, which was
proposed by Patrick Allen in 1991 [Ref. 7]. The SFS methodology has been developed to
better account for situation-dependent combined arm's effects in aggregate combat
models [Ref. 7:p. 1]. In the SFS method, the value of a weapon system is varied as a
function of the combat situation, defined by type of terrain and type of battle, and as a
function of shortages in the weapon mix in a given combat situation. The ratio of
attacking combat power to the defending combat power is defined as the "situationally
adjusted" or "modified" force ratio (MFR). The SFS's equations use this modified force
ratio to compute the casualty rates and FEB A (FLOT) movement rates.
The basis for the equations used in the SFS was documented in an unpublished
work by Paul Davis and Patrick Allen in the mid-1980s [Ref. 7:p. 41]. There has been no
effort to date to calibrate these equations or their parameters. Also, this method is not
applied to any real two-sided, daily combat data.

10

The last method used in this research is Dupuy's QJM model. In his book,
Attrition [Ref. 3], Dupuy presents simple equations to predict the personnel and material
losses of a military force. These equations are incorporated in the Quantified Judgment
Model (QJM) and the Tactical Numerical Deterministic Model (TNDM), both of which
were developed by Dupuy [Ref. 3:p. 104].
Dupuy applied his methodology retrospectively to a number of historical battles
from 1805 to 1973, with quite good results [Ref. 3.p. 113]. All the data used in his
examples contain only the starting and ending force strength and casualties mostly for
personnel and armor assets. At the end, he presents a scale (such as fair, excellent,
phenomenal) to show his subjective assessments of the relative quality of the forecasts or
estimates. For personnel estimates, the results are as follows [Ref. 3:p. 124]:


Total data sets:

25



Fair:

3



Good:

6



Excellent:

4



Phenomenal:

12

For armor estimates the results are as follows [Ref. 3:p. 124]:


Total data sets:

8



Good:

3



Excellent:

1



Phenomenal:

4

Overall, the average deviation for personnel estimates is 9.6%. The average
deviation for armor estimates is 9.0% [Ref. 3:p. 124].

11

C.

STUDY METHODOLOGY
This thesis applies the firepower score attrition models to the data on the Battle of

Kursk. The two main areas of interest are the quality of the fits and the insights provided
by the models. Different models will be compared and contrasted. The methodology used
in this thesis research consists of the following steps:
Arranging and setting up the data on hand to be useful for analysis.
Conducting a through analysis and interpretation of the data.
Extracting three data sets from the original data according to the combat
engagement status of the units.
Applying the ATLAS ground attrition equations to all data sets.
Evaluate the fit of the ATLAS equations apply to the three data sets.
Applying the SFS methodology to the three data sets.
Evaluate the fit of the SFS methodology to the data on the Battle of Kursk.
Evaluate the fit of the FLOT movement rates in the SFS methodology to
the Battle of Kursk.
Applying the QJM models to the Battle of Kursk.
Evaluate the fit of the QJM models apply to three data sets.
Distribute the combat power losses into different types of weapons.
Including the air sorties to the data.
Comparing and contrasting different methodologies.
Analyzing the results and conclusions of all the models.

12

III.

A.

HISTORY AND DATA ON THE BATTLE OF KURSK

A SHORT HISTORY ON THE BATTLE OF KURSK
In the spring of 1943, the Russo-German front was dominated by a salient located

to the north of Kharkov, to the south of Orel, and centered in the city of Kursk. The Kursk
salient had a frontage of 250 miles and 70 miles across its base [Ref. 14:p. 2-2].
In order to regain the initiative in Russia after the reverses in the winter campaign
of 1942-1943, and to strengthen the front line, Hitler decided to launch an offensive
operation known by the code name "Citadel (Zitadelle)" [Ref. 15:p 152] Through this
attack, Hitler wanted to considerably strengthen the front in the Belgorod-Orel area. The
Donets Basin was of great economic importance. Since the front line passed directly
along the eastern edge of the basin, Hitler considered it too insecure and vulnerable to
enemy attack [Ref. 15:p 153]. The German plan was a two-front attack on the Kursk
salient in a classic pincer operation.
Initially the attack was to be launched on 4 May 1943, however, the attack was
postponed until 5 July 1943. Postponement of the attack from May to July 1943
subsequently proved to be a great disadvantage to the Germans. Although Hitler argued
that the delay was necessary in view of an anticipated Allied attack on the Italian coast, it
was clear that the longer the Germans delayed, the more probable it was that the Soviets
would develop defenses to thwart the attack.

13

Operation Citadel was launched on July 5, 1943, see Figure 3.1. Using a massive
armor attack, General Model's 9th Army attacked the northern front of the salient, while
General Hoth's 4th Panzer Army attacked from the southern front [Ref. 14:p. 2-2].
German forces encountered heavy losses as they fell upon the prepared Soviet positions,
which contributed, to Germany's defeat in this campaign.
After an initial gain of a few miles in the first two days of the battle, the 4th Panzer
Army surged forward on 7 July, creating great damage and alarm among Soviet positions
[Ref. 14:p. 2-3]. Despite these heavy losses in men and armor, Soviet reinforcements
were sufficient to restrict the German gain to 25 miles by 12 July.
On 12 July, a German breakthrough attempt resulted in a major close quarters
tank battle near the town of Prokhorovka. This day was a turning point in the battle and
described in the The Battle of Kursk [Ref. 16] as " Immense in scope, ferocious in nature,
and epic in consequences, the Battle of Kursk witnessed (at Prokhorovka) one of the
largest tank engagements in world history and led to staggering losses." Unable to gain a
decisive victory, the Germans drew back into generally defensive postures after this
battle.
Hitler canceled Operation Citadel on 13 July, and later German attacks were
limited in scope. The Soviets began counterattacks on the southern front on 12 July but
shifted primarily defense postures by 14 July. The Soviet counteroffensive resumed on 18
July and they regained all of the ground lost in the theatre by July 23, 1943. [Ref. 14:p. 23]

14

1f1

;■
ELEVENTH

' V
V
\

)

OPERATION ZITADELLE AND THE
WITHDRAWAL TO THE HAGEN POSITION
5 duly ^18 August 1943
f " If ""71
••••«••
itiiiKiiv;
JO

f~

<^'

\

FOURTH TANK

.V

■ /

^f..-"

;#&*

THE fMUT ON * JUit»*3 ■.-'■■'
THETARTHCST ADVANCE O* OPERATION
ZITADCLL&; I* JUL 1943
.

©'■'■to'
©

20.

jO ■ 20 ;30

-30

*0

^^FIF Ty- S£ v*E.NTH_

«O 5« WJUDME^RS

Figure 3.1.

Operation Zitadelle [From Ref. 17:p. 134].

15

™"

\

^V.RJESE^.VE.-.'

it AD« »«
^ >&Mä.ES -

THE FOURTH ANO N«NTH ARMY TROHTS.

10

'■

FORTr-SEVENTH

B.

DATA ON THE BATTLE OF KURSK
This section presents the procedure for the extraction of data sets used in this

study. This process was the most difficult and time-consuming process of the study.
1.

Creation and Scope of the Kursk Database

In order to improve combat model credibility, the Ardennes Campaign Simulation
(ARCAS) Study [Ref. 18] compared a computerized combat model representation of the
World War II (WWII) 1944-1945 Ardennes Campaign with a database of historical
results from that campaign. This comparison was used to assess the accuracy of the
simulation model and to develop algorithmic changes [Ref. 14:p. 1-1].
Another comparative historical campaign is necessary to test the ARCAS
simulation. The Dupuy Institute (TDI), under contract to the US Army Concepts Analysis
Agency (CAA), collected historical data from forces on the southern front of the Battle of
Kursk, 5 July 1943 through 18 July 1943, from military archives in Germany and Russia
[Ref. 14:p. 1-1]. This data was reformatted as the Kursk DataBase (KDB). The Kursk
Operation Simulation and Validation Exercise (KOSAVE) Study, a follow-on effort to
the ARCAS Study of 1995, was initiated to compare simulated campaign results with
history [Ref. 14:p. 1-1].
The results and products of this simulation are presented in report CAA-SR-98-7
[Ref. 14:p. 1-1]. Additional supplemental data was released with this report on a CDROM [Ref. 19]. All of the data used in this research is extracted from this CD-ROM.
The KOSAVE report includes only the southern front of the World War E
(WWII) Battle of Kursk, as represented in the KDB historical data [Ref. 14:p. v]. Only
16

the results and data for combat units in the KDB are included. Non-combat support units
are not covered in this report.
2.

Limitations and Timeframe of the Kursk Database

In the KOSAVE report, results are not expressed in terms of specific weapon
types; instead, weapons are aggregated into categories or classes. Human factors such as
leadership, morale, fatigue, caution, and aggressiveness are not quantified. The timeframe
for the data is from 4 July 1943 through 18 July 1943.
3.

Assumptions for the Kursk Database

There are three primary assumptions made for the KDB database. The database
accurately represents the status and structure of forces in the southern front of the actual
Battle of Kursk. The personnel casualty and system kill criteria used to categorize KDB
casualty and weapon losses are sufficiently consistent with each other to allow
meaningful reporting and comparisons between combatants. The use of interpolation
techniques for gathering data between inconsistent reports in historical records to create a
complete set of daily report records in the KDB is reasonable [Ref. 14:p. 1-3].

C.

METHODOLOGY USED FOR THE EXTRACTION OF THE DATA
The data used throughout this study is extracted from the KOSAVE [Ref. 14]

report. All of the data used in this study are for combat units represented in the KDB.
Support units, such as bridging and logistic units, are excluded in the KOSAVE study.
Twenty-four primary German combat units are represented in the KDB. These 24
primary combat units are composed of 7 headquarters (HQ) units and 17 line units. The

17

17 line units are further partitioned into 8 infantry divisions (IDs), 5 Panzer divisions
(PzDs), and 4 Panzer Grenadier divisions (PzGrDs) [Ref. 14:p. 3-1]. The German HQ
units represented are corps and army level.
There are 67 primary Soviet Combat units represented in the KDB. These are 19
Soviet headquarters units and the 48 line units. The 48 line units are further partitioned
into 35 rifle divisions, 8 tank corps, 2 mechanized corps, 2 airborne divisions, and 1
detachment [Ref. 14:p. 1-3]. The Soviet HQ unit types are armies, rifle corps and the
Voronezh Front.
1.

Personnel Data

German and Soviet personnel strengths are used to represent the combat
manpower of combat units for both forces in the southern front Kursk Battle. The
personnel strengths are presented as "onhand" (OH) which represents the available
combat manpower.
Personnel strength losses are killed in action (KIA), wounded in action (WIA),
and captured/missing in action (CMIA). Disease and nonbattle injuries (DNBI) are not
counted as combat losses. Since DNBI are not caused directly by the enemy and only
combat units are taken into account, they are not considered as combat losses in this
study. Upon comparing this study's results with the previous studies, this classification
should be considered.

18


Documentos relacionados


Documento PDF small units tactics
Documento PDF kursk simulation
Documento PDF tank destroyer doctrine us
Documento PDF combat in the east
Documento PDF complete wargames handbook dunnigan 1
Documento PDF de angelo et al 2010 jwm track identification


Palabras claves relacionadas