Agent-based modeling P. Michael Link, Jürgen Scheffran CliSAP Research Group Climate Change and Security Institute of Geography, Universität Hamburg Models of Human-Environment Interaction Lecture 9, June 10, 2015 p. 1
Outline 1. Fundamentals of agent-based modeling 2. Examples of ABM I: fisheries in times of climate change 3. The NetLogo software package 4. Examples of ABM II: the NetLogo model library p. 2
Basic types of dynamic mathematical models x(t): System state at time t Dx(t) = x(t+1)-x(t): System change at time t Dx(t) = f(x,t): dynamic system Dx(t) = f(x,u,t): dynamic system with control variable u Dx(t) = f(x,u 1,u 2,t): dynamic game with control variables u 1,u 2 of two agents 1 and 2 Dx(t) = f(x,u 1,,u n,t): agent-based model and social networks with control variables u 1,,u n of multiple agents 1,,n p. 3
From artificial intelligence to artificial societies Artificial Intelligence: Recognition of rules and patterns in the environment Artificial Life: Behavioral rules and patterns in evolution und biology Artifical Societies: Application of behavioral rules and patterns in the social environment Conway's Game of Life: A set of rules for cellular automata which move in a two-dimensional grid. Each cell changes its state in dependence of the state of the eight next neighbors: cells become alive for exactly three living neighbors, cells die otherwise. Cellular automata: Discrete dynamical systems composed by arrangements of cells that behave like an automaton in a finite state. All interactions are local wheras the next state is a function of the current state of the cell and its neighbors. Relevant variables are the radius of relevant neighbor cells and the number of possible states of a cell. p. 4
Artificial societies Artificial societies: computational laboratories to grow'' social structures to discover fundamental local or micro mechanisms and generate macroscopic social structures and collective behaviors of interest. (Epstein/Axtell 1997) Modeling techniques for the study of human social phenomena, including trade, migration, group formation, combat, interaction with an environment, transmission of culture, propagation of disease, and population dynamics. Basic elements of Artificial societies: Agents: main acting units of artificial societies, having internal states and behavioral rules, including the ability to move around and interact. Environment or space: e.g. landscape/lattice of renewable resources that agents consume and metabolize, changed by agents Rules of behavior: for the agents and sites of the environment, e.g. simple movement rule to find the site richest in resources. p. 5
Agent based modeling (ABM) Autonomous agents capable to interact with each other and the environment according to rules of behavior. Agent: object in a computer program that encapsulates a particular behaviour when interacting with other agents within an environment. The behaviour may be simple or complex; deterministic, stochastic or adaptive; and the system as a whole may be homogeneous (all agents are of the same type) or heterogeneous (more than one type of agent present). (Hood 2003) Cognitive capabilities: perceive signals, react, act, making decisions, etc. according to a set of rules (Conte/Castelfranchi 1995): beliefs: what agents think to know about the world (experience, perception) goals: what agents would like to achieve (desired states) intents: which specific actions will agents undertake to achieve the desires. p. 6
Agent based modeling: agent criteria With regard to their action capabilities, agents can be autonomous: they act independently of any controlling agency; social: they interact with other agents; communicative: with other agents explicitly via some language; pro-active: they are driven by goals and objectives; reactive and adaptive: observe and respond to environmental changes rule-based: they can follow a well-defined and logical set of decision rules. p. 7
Schelling s agent model of segregation Thomas C. Schelling s 1971 pathbreaking early study on the emergence of racial segregation in cities: Instead of full understanding of the highly complex outcomes of processes, decision rules represent behavior small number of individual actors. Small preference for neighbors of same color could lead to total segregation. Coins on graph paper to demonstrate the theory by placing pennies and nickels in different patterns on the "board" and then moving them one by one if they were in an "unhappy" situation. Simulation models are very good at incorporating time and space, especially when tied to a geographic information system. p. 8
Schelling s agent model of segregation Initial condition of Schelling's experiment (left), stable segregated pattern after several iterations (right) p. 9
Outline 1. Fundamentals of agent-based modeling 2. Examples of ABM I: fisheries in times of climate change 3. The NetLogo software package 4. Examples of ABM II: the NetLogo model library p. 10
Global ocean circulation (Hall & Behl, 2006) p. 11
General model features two fish species (cod and capelin) that interact via predation both stocks are harvested commercially fishermen can either follow an adaptive or profit-maximizing harvesting strategy the fleet sizes may vary depending on the economic success of the fisheries management measures such as total allowable catches limit the amount of fish harvested p. 12
Model structure trawlers coastal fisheries environmental change purse seine fisheries fishing effort fishing effort recruitment recruitment fishing effort fixed costs, variable costs fixed costs, variable costs cod: age 0 capelin: age 0 fixed costs, variable costs cod: age 2 capelin: age 1 total catch total catch total catch cod: age 6 capelin: age 2 revenue revenue cod: age 15 capelin: age 5 revenue profits profits spawning stock spawning stock cod stock capelin stock profits (Link, 2006) p. 13
The fisheries in the model (1): profit maximization Harvest: h q n v e sat,, sia,, sat,, it, it, i Costs: e it, i i it, Revenue: r P w h sit,, si, sa, siat,,, sa, Objective: t 0 y t t0 i e s,, i t es,, i t t t 0 h = harvest q = catchability coefficient n = no. of fish v = no. of vessels e = fleet utilization Ψ = total costs θ = variable costs φ = fixed costs r = revenue P = market price of fish δ = discount rate y = optimization period w = fish weight π = profit per fishing period Π = overall profit p. 14
The fisheries in the model (2): adaptive harvesting strategy logistic growth is assumed actual growth rate is estimated fishermen try to obtain MSY fishing effort is set accordingly target catch and actual catch are compared fishing effort for next period is adjusted new information on actual growth rate is incorporated (Link, 2006) speed of learning can vary p. 15
Environmental change in the model environmental change directly influences fish stock development recruitment success of both species depends on water temperature at time of spawning and on spawning stock biomass survival rates of young age classes depend on strength of THC p. 16
Development of overturning near the Norwegian Sea (Link, 2006) p. 17
Mar/Apr temperature change near the Lofoten Islands (Link, 2006) p. 18
Influence of circulation strength on survival rates (Vikebø et al., 2005) p. 19
Development of the cod stock biomass (Link, 2006) p. 20 4-year profit maximization adaptive harvesting
Development of cod catches (Link, 2006) p. 21 4-year profit maximization adaptive harvesting
Developments of profits of the cod fishery (Link, 2006) p. 22 4-year profit maximization adaptive harvesting
Main conclusions from the model simulations (1) Warming of the Norwegian Sea/Barents Sea region has a positive impact on the stocks of key fish species due to increased occurrences of strong recruitment year classes. On the other hand, a THC collapse negatively affects stock dynamics as the youngest age classes experience lower natural survival rates. Both fisheries remain profitable regardless of the harvesting strategy if the THC only weakens and later recovers. A shutdown of the THC necessitates the complete closure of the cod fishery. p. 23
Main conclusions from the model simulations (2) In times of stable hydrographic conditions, the adaptive harvesting strategy leads to higher returns from fishing. When environmental conditions become more variable, adaptive harvesting is less successful than profit maximization because of the time lag associated with learning. In times of insecure stock development, the fleet types with the highest catch efficiency are favored. If the harvesting strategy allows for longer-term planning and deferments of catches, smaller and more cost effective vessels increase in importance. p. 24
Main conclusions from the model simulations (3) If a shutdown of the THC occurs as a consequence of global warming, the socioeconomic impacts will affect many different sectors and countries all over the world. While some countries may suffer considerably from a THC breakdown, it is not catastrophic on a global scale. A THC collapse would do great harm to the important fishery of Arcto-Norwegian cod. Gains in other fisheries are unlikely to offset losses of the cod fishery. Only continuous management measures can prevent the fishery from depleting the stocks in times of high variability in environmental conditions. p. 25
Outline 1. Fundamentals of agent-based modeling 2. Examples of ABM I: fisheries in times of climate change 3. The NetLogo software package 4. Examples of ABM II: the NetLogo model library p. 26
The NetLogo software package NetLogo is a free software package It has been developed by Northwestern University. Current version of the software is 5.2.0. Runs on Windows, Macintosh, and Linux Computers. Software comes with a large model library with sample models. Technical features of NetLogo are described p. 27
Outline 1. Fundamentals of agent-based modeling 2. Examples of ABM I: fisheries in times of climate change 3. The NetLogo software package 4. Examples of ABM II: the NetLogo model library p. 28
Examples from the NetLogo Model Library The following models from the NetLogo Model Library were shown in class (available at https://ccl.northwestern.edu/netlogo): Life Climate Change Fire Segregation p. 29