Rodney Thesis Development

From GeoMod

test pages
another test
Image:Abminages.zip ref test

• Blog
• Agent based modeling resources
• Literature Review Agent Based Modeling
• Literature Review Pedestrian Archeology
• Literature Review Sensitivity Analysis
• Literature Review Optimization
• Agent Operation Diagram
• Model Agent Properties
• Chertman Code
• Simulation

retired pages ABM

Sensitivity Analysis of: A model to simulate the procurement of lithic raw material assemblages for paleolithic archaeological sites using a random agent.

Contents 1 Introduction 2 Objectives 2.1 Hypothesis 3 Discussion 3.1 Agent based modeling 3.2 Sensitivity Analysis 3.3 Optimization 4 Methodology 4.1 Hardware: 4.2 Software: 4.2.1 Environment: 4.2.2 Chertman 4.2.2.1 Agent Transition Rules: 4.2.2.2 Agent Navigation: 4.2.2.3 Toolstone Procurement : 4.2.2.4 Model Output: 4.3 Data: 4.4 Sensitivity Analysis 4.5 Simulation Annealing 4.5.1 Site Info, move: 4.5.2 Procedure: 5 Significance of research 6 Operation: 6.1 Proposed Time line: 6.2 Budget: 7 References: 7.1 Bibliography

Introduction

The industrial efforts of paleolithic humans are evident in the production of stone tools from their environment. Remnants of tools produced from chert, a sedimentary rock composed partially of very small crystals of silica, have been cataloged as assemblages. The composition of tool stone assemblages reflects a mixture of stone type, and the associated masses of these types.

The patterns of these toolstone assemblages in respect to composition may offer clues into the lives of paleolithic humans. What motivations of paleolithic hunter gatherers actions are evidenced in the toolstone assemblages? Did the paleolithic hunter-gatherer groups remain in close proximity to assemblage sites, or were their activities concentrated, or dispersed as patterns? The composition of the assemblages may offer insight. A high concentration of a type of chert sourced far from the assemblage site may serve as an indication of the high utility of that specific type. The acquisition of high grade tool material may outweigh the expenditure necessary to procure chert sourced far from an assemblage. Conversely,a high percentage composition of chert sourced near an assemblage could indicate a lesser degree of mobility for hunter-gatherers. Evidence of trading networks may be supported in the presence of material sourced far from an assemblage.

A model to simulate the procurement of lithic raw material assemblages for paleolithic archaeological sites using a random agent (Cole et al. 2005) has been implemented. An agent based simulation best describes the operation of the model. Agent based simulation is a modeling approach unique in its characteristic for generation of large scale patterns as the result of multiple and independent small scale interactions. A model agent uses a set of pre-defined behaviors which are expressed in operation as a response to either internal, a timer for example, or external, the model environment, conditions.In the case of the toolstone procurement model affectionately dubbed "Chertman" an example of the agent based approach is the representation of the desire of the agent to return to the assemblage site.In our case the model agent expresses a desire to return "home". This is achieved by altering this "desire" once ther preset conditions which simulate the limits of a foraging expedition are reached. The model agent changes its preset response to the environment by altering the probability of choosing a path oriented towards "home."

A simulation is a necessary balance between complexity for robustness and simplicity for efficiency. A robust model would prescribe including all possible explanatory variables in the simulation.It would prove to be prohibitively expensive in time,cpu cycles, model development,and computer resources. A very simple model may prove deficient in explanatory power fail to explain significant interactions.As a compromise simulating the most significant behaviors of the paleolithic hunting party promotes a robust solution and provides efficiency, through modeling fewer variables.

The random agent model is in an early stage of development. A necessary step in the implementation of the toolstone procurement model is the discovery of the significance of individual input parameters. The significance of input parameters may be determined through a sensitivity analysis. Sensitivity analysis allows for the decomposition of individual model input parameters contribution to output.What variables contribute the most to a change in output composition and masses? In the Chertman model will we find that the slope of the terrain upon which the model operates is more important than the distance from home limits of a foraging trip? Sensitivity Analysis will also allow the range of variables to be refined. Realistic limits to the range of input variables are to be discovered during this development phase. As an example we may find an increase to the model agents ability to collect toolstone, an increase in payload,or mass, may either produce little change in model output, an insensitive variable,or change model output by a factor, a very sensitive variable. The discovery of insensitive variable will allow for their exclusion from the model, or their fixation at a nominal value. This again contributes to model simplicity. The discovery of very sensitive variables may require a redefinition of the variable.For example if we find the the model agent is very sensitive to a simulated stream crossing by the hunter gatherer agent. That is variation in the predefined ability to cross streams greatly changes model output.

The final stage in model implementation is optimization of the model. What variables at what values best contribute to the validation of the model against known assemblage data. How do we maximize the covariance between the known masses and chert types from assemblage data? Simulated annealing may provide us an answer. Through successive model runs the covariance between model output and site observations will be compared between model runs the value of model inputs then will be moved or incrementally adjusted in the direction of higher covariance until convergence upon one or more best fit solutions.

Objectives

It is the purpose of this exercise to develop A model to simulate the procurement of lithic raw material assemblages for paleolithic archaeological sites using a random agent. The development of the model is focused upon predefined two distinct goals. First the sensitivity analysis of the model input parameters, Second the performance of optimization technique simulation annealing

Hypothesis

The covariance between the toolstone model and assemblage observations is significant.

Discussion

Agent based modeling

Agent based modeling is a rapidly developing field that is still in its infancy. The focus of this section is twofold. First, to refine the definition of an agent based model (ABM), and second to identify some of the key research trends in the field. Exploration begins by discussing the foundations of the key technologies used by agent based models.

Operating in a new niche within geography, bottom up modeling of spatial phenomenon ,by cellular automata and agent based models is filling a gap in geographic knowledge. Primary applications for cellular automata and agent based models can be found in pedestrian modeling, transportation simulation, and land use characterization. Each of these model types differ in their scale and suitability for implementation, by either cellular automata or agent based modeling. Pedestrian modeling operates at the finest scale of these model types and are very suitable for agent based models. At the street level micro scale interactions with the environment are important to safety planners, and shopkeepers among them. Transportation simulations operate on the scale of tens of miles and are suitable for either agent based or cellular automata the latter being more prevalent due to its earlier development and lower overall use of computer resources. Their importance is easily revealed at rush hour and may determine if a daily commute comes to a standstill. Land use characterization is most suitable for cellular automata both because of the immovable nature of the features being modeled, and the large scales which may be in order of hundreds of hectares. Predictions of urban sprawl which reveal micro scale influences far into the future are relevant to land use models which use cellular automata. Despite their the differences in scale or implementation, either cellular or agent based models share many similarities. First among these are the necessary definition of transition rules defining state change characteristics of the model. In the case of agent based pedestrian models transition criteria may focus upon decision making based upon social interaction. In the case of transport simulation decision making rules may be derived from fluid flow dynamics. In the example of land use characterization either policy decisions or information gathered by remote sensing may provide the frame work from which state transition rules are derived. These transition rules may be derived from very informal or rigorously scientific methods. Among the most informal being visual comparison of simulation results to observed patterns. Rigorous methods such as gravity weighted regression, or the implementation of neural networks may also be common. Second, model validation techniques follow a similar pattern of transition rule generation and are produced to the suitability of the model maintainer. Variation in qualitative or quantitative outputs may be tuned by trial and error, inside the black box of neural networks or through spatial statistics. Despite their short history cellular automata and agent based modeling are and will likely for the future gain in performance and application to many aspects of geography.

Examination of the definition and direction of the field increases our understanding of this element modeling in GIScience. The advent of ubiquitous data sources, the emergence of complexity sciences, are coalescing with research from other disciplines along with an increase in computational power of computers to allow for the emergence of agent based modeling as an alternative to other modeling traditions (Batty 2001). A model agent is a finite state machine.A finite state machine is composed of preset behaviors. Thus the machine output is conditional on current environmental conditions and a set of transition rules.The set of rules which operate each individual agent are mathematically derived, and many are based upon the work of Helbing's description of pedestrian characteristics. In building a mathematical model for the movement of pedestrians, one has to assume that these decisions are not completely random, but show certain regularities. A pedestrian wants to move in a most convenient way, tries to minimize delays when having to avoid obstacles and other pedestrians, intends to take an optimal path and to walk with the minimal velocity allowing to reach a destination at a certain time for example (Helbing 1991). (Driving) the process is the desire for the agent to reach a destination. (Navigation algorithms, use mathematical relationships such as DeLaunay triangles and minimum distance segments to define the agent navigation strategies (Lamarche et al.). Agent navigation examples include giving limited field of vision, known in the literature as an isovist, to pedestrian agents (Batty 1999). (Other implementations of navigation algorithms may include DeLaunay triangles and minimum distance segments, of simulation topology as agent navigation strategies (Lamarche et al.)may be derived through the use of artificial neural networks, regression analysis or through modeling of discrete choices. Discrete choice models which unlike (regression and neural networks) always make the best choice are being introduced. Discrete choice models in general, and random utility models in particular, are disaggregate behavioral models designed to forecast the behavior of individuals in choice situations. They assume that each alternative in a choice experiment can be associated with a value called utility. The alternative with the highest utility is selected. The utility of each alternative is a latent variable which is modeled as a random variable depends on the attributes of the alternative (and the qualities of the agent, such as Lurbano) the socio-economic characteristics of the decision-maker (Antonini et al. 2005). Agents enter the simulation, make route choice decisions, based upon the present conditions and interact with the environment. The individual behavior of the agents will results in collective patterns (, the focus the study -delete There is a subtle distinction between the collective behavior patterns of interacting agents and the repeated simulation of a single agent on the landscape. Lurbano). The patterns created by agents of this type first came to the forefront or artificial life simulations in 1986 with the implementation of 'Boids' by Craig Reynolds. This model of simulated bird flight was the genesis of the collective behaviors now referred to in the literature as flocking and swarming(Reynolds 1987). Many animal groups such as fish schools and bird flocks clearly display structural order, with the behavior of the organisms so integrated that even though they may change shape and direction, they appear to move as a single coherent entity. Many of the collective behaviors exhibited by such groups can only be understood by considering the very large number of interactions among group members.(Couzin et al. 2002)

Agent based simulations and pedestrian simulations are very complex. The very nature of the field is drawn from the complexity sciences. However differences in model execution method leave room for some comparison. Often it is the variation in the number of modeled elements which differentiates ABM simulations. In the end the complexity of the model is left to the individual researcher. Advances in prepackaged software may lessen the divide between model implementations. The Swarm package and its offshoot Repast offer some modularization an offer some limited interaction with GIS packages like GRASS. Python and its developing extensions promise the built in analysis capability's of SAS and R (define your acronyms Lurbano) with the flexibility of a full programming language.

Agent based pedestrian modeling is both a field in its infancy, and rapidly developing. The focus of this report is twofold. First, to refine the definition of an agent based pedestrian model. Second to identify some of the key research trends in the field. Exploration begins by discussing the foundations of the key technologies used by agent based pedestrian models. What, when,where, and the how of ABM pedestrians forms the outline. Followed by the who, and what next in the field to describe the potential development of these techniques and technologies to maturity. By examining both the definition and direction of the field our understanding of these element modeling in GIScience.

Why agent based modeling; what attractive utilitarian function does it serve? Agent-based models have emerged as a powerful alternative to more aggregate and more geometric approaches to spatial modeling for many reasons...fine-scale data on locational geometry, on the disposition of activities in cities, and on flows of walkers are becoming available from various sources: remote-sensing, automatic counting, geo-demographic surveys for marketing purposes, and the integration of hitherto separate data sets in the public and private sectors across a range of spatial scales...fourth and most important, new ways of articulating social systems by using ideas from complexity theory have developed over the last 15 years and these are beginning to merge with similar developments in far-from-equilibrium physics, thus giving this field a sharper edge than anything in the social sciences hitherto. The power of this new paradigm cannot be underestimated... The notion of systems emerging from the bottom-upon which is a basic tenet of complexity theory became possible computationally only when enough individual actors could be simulated to provide predictions of more global structure from local action. Pedestrian modeling fits this notion of agent-based modeling almost literally.[Batty,2001] The time is right for the emergence of these technologies hither there to impossible in implementation.

What is an agent based model pedestrian model? Microscopic Pedestrian Simulation Model is computer simulation model of pedestrian movement where every pedestrian in the model is treated as individual. Most of pedestrian researches have been done on macroscopic level.[Kardi et.al,2000] An agent based model may be seen as akin to a robot acting in a simulated environment. Agents like robots are autonomous within their environment and respond to their with a series of preset behaviors. The idea of an agent has been linked to that of a finite state machine. A finite state machine is a model of behavior composed of states, transitions and actions. A state stores information about the past, i.e. it reflects the input changes from the system start to the present moment. A transition indicates a state change and is described by a condition that would need to be fulfilled to enable the transition. An action is a description of an activity that is to be performed at a given moment. [Wiki]The action of a finite state machine follow Markov processes. To quote a fellow student a Markov process operates on the principle that the “The present determines future.�? What does this mean? All information on past conditions of the finite state machine are unknown. Only the present conditions determine the output within the range of preset possibilities.

What is the agent within the reviewed simulations? In this case the agent is an individual pedestrian. The abstraction of which for the is a set of rules which direct the action of the modeled pedestrian within the simulation. The set of rules which operate each individual agent are mathematically derived, and most are based upon the work of Helbing's description of pedestrian characteristics. In building a mathematical model for the movement of pedestrians, one has to assume that these decisions are not completely random, but show certain regularities instead...A pedestrian wants to move in a most convenient way, tries to minimize delays when having to avoid obstacles and other pedestrians, intends to take an optimal path and to walk with the minimal velocity allowing to reach a destination at a certain time, etc.[Helbing,1998] This model describes the motion of a pedestrian agents as the sum of terms. Outlining the process is the desire for the agent to reach a destination. Navigation within the simulation may be achieved by many methods. Employed examples have includes giving limited 'vision' to pedestrian agents. Information about a limited space is available to an agent. Other implementations of navigation algorithms may include DeLaunay triangles and minimum distance segments, or as simple as a random variation in the orientation towards a desired position. Thus follows with the orientation of the behavior of the agent toward its destination. Included in Helbing's outline are descriptions of a social forces model acting upon the agent. The agent wants to avoid collision with other pedestrians and make forward progress by adjusting heading or velocity. Concluding the rule sets to describe pedestrian interaction under the formation of groups defined by their attractive, and repulsive effects on the individual agent. The resultant action of the sum of terms is the Markov process of the agent.`

How are the state transition rules derived? Variations in methods exist from the lifelike recreations of motion in feature film animation and video games to rigorous statistically derived models. In commenting upon the author of the model made a notable example of parameter estimation without statistical rigor. The weights (i.e,. parameters associated with these variables) are often assigned arbitrarily. They are rarely estimated statically.[Huff,2003] Examples of transition rules developed on these parameters can been seen in the motion of animated bats in the films based upon their fictional human namesake. Transition rules may be derived through the use of neural networks. Neural networks are a set of n connections with an invisible middle layer. Input characteristics, the conditions of the agents simulated environment are trained to specific behaviors as outputs. Finally traditional statistical methods are also employed to determine the conditions within which weights operate.

What is the simulated environment? The simulated environment includes familiar elements those areas accessible to pedestrian movement, streets and walkways, areas inaccessible to pedestrian movement, barriers and buildings. The uncommon elements of the simulated environment are again abstractions. Often it is the case that the demographic characteristics of pedestrians need to be modeled. These may include but are not limited to physical attributes such as average velocity, and may include more abstract notions such as preference. Preference are described as variations from the Helbing model and modeled as gradient surfaces often with distance decay properties.

What is the overall theme behind this type of research? The observation of collective behavior of individuals. Many animal groups such as fish schools and bird flocks clearly display structural order, with the behavior of the organisms so integrated that even though they may change shape and direction, they appear to move as a single coherent entity. Many of the collective behaviors exhibited by such groups can only be understood by considering the very large number of interactions among group members.[I. D. Couzin et al.,2002] Pedestrian agents enter the simulation, make route choice decisions, based upon the present conditions and interact with the environment.. The individual behavior of the agents will results in collective patterns, the focus the study. The patterns created by agents of this type first came to the forefront or artificial life simulations in 1986 with the implementation of 'Boids' by Craig Reynolds. This model of simulated bird flight was the genesis of the collective behaviors now referred to in the literature as flocking and swarming. The simulated flock is an elaboration of a particle system, with the simulated birds being the particles. The aggregate motion of the simulated flock is created by a distributed behavioral model much like that at work in a natural flock; the birds choose their own course. Each simulated bird is implemented as an independent actor that navigates according to its local perception of the dynamic environment, the laws of simulated physics that rule its motion, and a set of behaviors programmed into it by the .animator.. The aggregate motion of the simulated flock is the result of the dense interaction of the relatively simple behaviors of the individual simulated birds.[Reynolds,1987]

What are the recent trends in research? GIScience is composed of hardware,software, data people and methods. There are no precise divisions between these terms. The abstraction of this set I believe is necessary in this case. So I have combined some of these terms as categories for analysis. Data collection and validation are synonymous in agent based modeling. Data must be collected to generate transition rules and analyzed against the results of the simulation. Remaining are advances and variation in the methods of model execution. We must begin by looking at advances in hardware utilization, data extraction, and modeling methodology within the field.

Trends in hardware adoption for data collection is our first focus. A recent survey of both the quantity and quality of pedestrian data resources has revealed both to be lacking in quality and quantity.[BTS00-02,2000] Pedestrian tracking is primary to the formation of transition rules for the individual agent and for the validation of the model. The collection of simulation geometry may be time consuming and inaccurate.To resolve these problems researchers are bringing remote sensing hardware to the micro scale environment. The use of both passive and active collection systems are alleviating these problems. A 2004 survey of pedestrian tracking systems outlines the optimal application of among available pedestrian tracking systems.[Chan et. al.,2005].Chan outlines the benefits and drawbacks of modern pedestrian tracking solution. Also of note are the use use of modified infra-red detectors to determine pedestrian velocities and vectors. The accurate measurement of pedestrian trajectories has been difficult to achieve in environments other than carefully controlled laboratories. Traditional techniques include direct observation and analysis of recorded video footage.[Armitag, et.al.,]The use of Lidar is becoming prevalent in pedestrian tracking applications. Each LD-A,laser range scanner, is controlled by a client computer, which gathers laser data, extracts moving points by background subtraction...integrates the moving points from all client computers to a global coordinate system, and tracks trajectories by identifying the patter of moving legs.[Zhao,Shibasaki]

Researchers are advancing the field by the adoption of new methods as well as technology. Computer vision is making inroads into pedestrian modeling. Pedestrian motions are being extracted from video images using techniques borrowed from artificial intelligence.[Teknomo,2002] In addition to advances in tracking and surveillance the methodology behind agent based pedestrian modeling is also evolving. Moving from simple to complex ever more attributes are being included within current models. Adding to present state of the art techniques in visualization the technology of video games is being added to pedestrian simulations. The WalkEd simulation uses the popular video game engine found in Quake. In addition research efforts are being made to extend the ability agents to determine their 'next move' Discrete choice model which unlike prior examples always make the best choice are being introduced. Discrete choice models in general, and random utility models in particular, are disaggregate behavioral models designed to forecast the behavior of individuals in choice situations. They assume that each alternative in a choice experiment can be associated with a value called utility. The alternative with the highest utility is selected. The utility of each alternative is a latent variable which is modeled as a random variable depending on the attributes of the alternative and the socio-economic characteristics of the decision- maker.[Antonini, et al]

Conclusion about agent based simulations and pedestrian simulations are very complex. The very nature of the field is drawn from the complexity sciences. However differences in model execution method leave room for some comparison. Often it is the variation in the number of modeled elements which differentiates ABM pedestrian simulations. In the end the complexity of the model is left to the individual researcher. Advances in prepackaged software may lessen the divide between model implementation. The Swarm package and its offshoot Repast offer some modularization an offer some limited interaction with GIS packages like GRASS. Python and its developing extensions promise the built in analysis capability's of SAS and R with the flexibility of a full programming language.

In the end the addition of multiple space geometries and interactions within can be overwhelming to the novice researcher. I have found myself overwhelmed by the subject as well. It is possible that we have much to gain in Giscience as well as in other fields. The pedestrian agent is atomic and indivisible from a physical standpoint. It might serve us well to begin there and work our way up. I personally view the situation as an analog to quantum mechanics in chemistry and physics. GIS and Geography as disciplines are subject to the effects of scale at every level. Explanations of phenomenon at one scale do not necessarily hold true at the next. Inverse to this are agent based simulations. The difficulty is not operation at scale. Given enough computational recourses elements emergent at scale should still be perceivable as scale increases. The problem is with explanations of the phenomenon themselves as the products of sometimes random and fuzzy conditions. In this way we may have found our equivalent of Heisenberg's uncertainty simply stated total knowledge is impossible.

Sensitivity Analysis

What is the scope of sensitivity analysis in use in GIS and agent based modeling? An overview of specific applications of uncertainty and sensitivity analysis in GIS environment employed by other researchers(Crosetto Michele et al. 2001). Other model implementations include a demonstration of sensitivity analysis in a Everglades wetlands model in which model parameters were identified for inadequate precision(Fitz et al.1995). Sensitivity analysis in a multi-agent model has been employed in a stochastic combat simulation(Alam et al. 2004). Sensitivity testing of model input parameters has been demonstrated in an agricultural multi-agent simulation.(Berger et al. 2005) A multi-step procedural approach is introduced before specific methodology for sensitivity analysis is employed in a GIS environment(Crosetto Michele et al. 2001). The use of spatial interpolation and Monte-Carlo techniques for model parameterization in an complex example which included both the consideration of land cover, soil type and socio-economic variables. Also researchers must be aware of the scarcity of data for calibration and validation of simulation results(Berger et al. 2005). The one at a time or (OAT ) method of parameter variation was performed in the implementation of the SIMBAT combat simulation(Alam et al. 2004). An overview of the implementation of screening techniques characterizing method suitability and difficulty of implementation(Trocine et al. 2000). A comparison of sensitivity analysis approaches with emphasis on analysis of variance as the primary indicator of parameter sensitivity Computational methods of analysis Sobel and extended FAST are also detailed in application(Saltelli et al. 2002).

What are the pros and cons of the available methods for sensitivity analysis? The necessity of a linear response on input parameters was identified as a drawback to using regression coefficients as a measure of sensitivity. Computational efficiency is also a concern to modelers when the number of variables increases. Although applicable to nonlinear models rank based sensitivity measurements do not reveal the magnitude of differences in sensitivity between parameters (Saltelli et al. 2002). The drawbacks of the (OAT) screening methods through their inability to reveal parameter interactions were identified in the examination of a simulation(Alam et al. 2004).

Optimization

Methodology

The sensitivity analysis of the toolstone procurement model will be performed using the Morris one at a time(OAT) method of variation of model input parameters. The outline of the sensitivity analysis methodology will follow the GIScience approach of identifying contributing components of hardware, software, data and methods.

Hardware:

Off the shelf personal computer hardware will be used for the sensitivity analysis.</P>

Software:

Environment:

The toolstone procurement model is executed in the Python scripting language. The implementation of the sensitivity analysis will be performed in Python.

Chertman

Commonly referred to as Chertman: A model to simulate the procurement of lithic raw material assemblages for paleolithic archaeological sites using a random agent Stephen Cole and Lensyl Urbano was created to further investigation into archeology and modeling. The simulation consists of lithic toolstone assemblages, chert outcrops, the local terrain, and a random agent. The random agent functions as a lithic hunting party.

The Cherman model code and data files are available from the links below.
[[1]Chertman_package.zip]
[[2]Md5sum.txt]

Agent Transition Rules:

The transition rules for the Chertman model are implemented as a beliefs, desires and intentions (BDI) framework. The random agent desires to explore the local terrain, obtain toolstones, and return to the agents origin.

Agent Navigation:

Agent navigation is random on outbound environment exploration. The random agent potentially may occupy any space as the resultant vector of prior location, random direction, and simulation step distance. Random direction is one of the model input variables and may be reduced from 360 to 0 degrees, no direction. Local slope may prevent the random agent from occupying a new location through restriction on the maximum slope that the agent may traverse.

Toolstone Procurement :

If a chert outcrop is encountered during agent navigation the acquisition of toolstone is determined as a pickup probability within a gravity decay gradient from the proximity of the toolstone outcrop.

Model Output:

The model output is the mass and composition of toolstone carried by the random agent as the end of the model run. N model runs are committed until the average mass carried by the model agent reaches equilibrium.

Data:

Reproduced from Chertman_Model_Instructions

Data for the sensitivity analysis was compiled for the creation of the toolstone procurement model.

1. Streams traced from maps in Geneste (1985).
2. Topography from 30-arc second Digital Elevation Model (DEM) downloaded from the USGS's EROS data center (http://edcdaac.usgs.gov/gtopo30/gtopo30.asp).
3. Slopes computed from DEM.
4. Chert outcrops from:
   1. P-Y Demars (1980), L'Utilisation du Silex au Paléolithique Supérieur: Choix, Approvisionnement, Circulation. CNRS.
   2. J-M Geneste (1985), Analyse Lithique des Industries Moust-eriennes du Perigord... Ph.D. thesis, Université de Bordeaux I.
   3. A. Turq (1992), Le Paléolithique Inférieur et Moyen Entre Le Vallées de la Dordogne et du Lot. Ph.D. thesis, U. of Bordeaux I.
5. Archaeological data: raw material composition (percentage by weight) from six assemblages: La-Côte III; Le Moustier H8; Caminade Est M3 and G; Le Flageolet I, IX and XI. From Cole (2002), Lithic Raw Material Exploitation Between 30,000 BP and 40,000 BP in the Perigord, France. Ph.D. dissertation, University of Washington.

Sensitivity Analysis

The sensitivity analysis is to be performed to define the model input parameter boundaries. Describe the individual contribution of model parameters to model output. Define necessary levels of model parameter precision. The specific method of model screening using the Morris (OAT) method. For a list of the sensitivity analysis parameters see Table1.

Simulation Annealing

Site Info, move:

la-cote 638932.963507464, 3310348.68552568 UTM zone30 lat 29.916007128753137,long. 1.5607863788656513

Procedure:

1. A base value for input parameters is randomly chosen
2. The values of the input parameters are sampled at the randomly chosen base value
3. One of the input parameters is varied by a factor delta
4. Delta is a fraction of the input parameter domain being even and greater than 1
5. Incrementation of parameter value is made for each individual input parameter until all parameters have been incremented by delta for all values within the the parameters range. For a total number of model runs defined by
   1. The number of independent input parameters,
   2. The precision of the model input parameters

Variable	Abbreviation	Unit	Min	Max
Direction of Movement	ndirs	degrees	0	360
Trip distance
Model agent movement distance	step_size	km	100	1200
Number of steps in simulation run	nsteps	scalar	100	100
Trip return direction bias	home_sickness	probability	0	1
Maximum load	max_load	kg	20	80
Amount added at outcrop	load_add	kg	.5	1.0
Rock type pickup	prob_pickup	probability	.25	1.0
Fractional loss of procurement during trip	loss	%	0.0001	0.15
probability of pickup from an outcrop	rng	surface gradient
Probability of crossing a stream	stream_perm	probability	0.25	1
Maximum slope traversable by model agent A function of maximum slope in model environment	l_slope	probability	1/l_slope	0

Significance of research

The sensitivity analysis of the toolstone procurement model contributes to modeling of systems of interaction between humans and their environment.

Operation:

Proposed Time line:

• May , 05
  • Complete Literature Review Agent Based Modeling
• December, 05
  • Complete an agent based pedestrian simulation in order to become familiar with software environment
• March, 06
  • Complete thesis outline
• April, 06
  • Begin Literature review of toolstone assemblages
• May, 06
  • Complete Thesis Proposal Draft
• Summer, 2006
  • Brown Bag Presentation of Thesis Proposal
  • Begin coding and testing of updated Chertman model
  • Perform Sensitivity Analysis

• Fall, 2006
  • Write Thesis
  • Complete Comprehensive Examination
  • Defend Thesis

Budget:

References:

Bibliography