(571e) Improved Assessment of Personal Exposure to Chemicals Using Agent Based Modelling (ABM) Coupled with Multi-Sensors Networks

Innovation
in sensors technologies creates possibilities to collect environmental data at
an unprecedented depth and breadth. With the advent of GIS, GPS to track
individuals, and personal environmental monitoring, undertaking such analyses
throughout an individual’s routine, or even lifetime is now possible. Finding
the right balance between limited amounts of high-quality data from
standardized environmental monitoring campaigns and large amounts of moderate
quality data by sensor networks can transform the way we understand and
interact with our environment. Measuring, though, personal exposure directly
requires a large number of people and therefore is often not feasible due to
time and financial constraints. Considering the substantial technical and
ethical hurdles involved in collecting individual data for whole populations,
this study introduces a first approach of simulating human movement and
interaction behaviour, using Agent Based Modelling (ABM).

ΑΒΜ is a simulation
technique that allows us to explore and understand phenomena, where independent
entities interact together, forming an emergent whole. While the direct
representation of individuals’ actions is organisationally difficult,
ΑΒΜ simplifies this process by managing information at the level
of the autonomous decision-makers, called “agents”. These heterogeneous actors have
personal attributes and are programmed to react and act in their environment
while following a set of behavioural rules. By simulating actions and
interactions at the individual level, the diversity that exists among agents
can be detected, as rise is given to the behaviour of the system as a whole.

After a thorough
examination of the available ABM software tools, the GAMA platform was chosen,
allowing to execute large-scale simulations and to instantiate agents from
various format datasets. GAMA, like most ABM platform, is based on an
object-oriented programming language, with agents being naturally implemented
as objects, therefore structures that hold both data and procedures. The
incorporation of added functionalities from Geographic Information Systems
(GIS) software libraries, provides with important geospatial analytical
capabilities. Specifically, GAMA is structured on a Java-based rich modelling
language, GAML, which allows complex systems to be defined, integrating at the
same time geographical vector data and entities of different scales. Over a
particular period of simulated time, agents inspect their surroundings and make
decisions. Their actions might imply a relevant spatial grain, which can be
visualised and explored. A city scale ABM was developed for the urban area of Thessaloniki,
Greece. Population statistics, road and buildings
networks data were transformed into human, road and building agents,
respectively. Time-use survey outputs were associated with human agent behavioural
rules, aiming to model representative to real-world routines. The agents’
behavioural rules, expressed as utility functions, vary, but only
parametrically, not structurally. The culturally varying personal attributes
(such as age, gender, level of education) enter into these functions, but the
algebraic form remains fixed, as does the agent’s practice of maximising the
function. An event-triggering utility function has been
established in order to always assign the next activity of a human agent,
during a simulated day. Moreover, a utility function was developed that
encourages human agents to interact with each-other in commute and
leisure-related activities. Knowledge of human agent
specific attributes by other human agents provides a signal that acts to enable
or prevent interaction from occurring. In order to further inform and validate
the model, time-geography of exposure data, derived from a personal multi-sensors campaign on 150 households of urban
Thessaloniki, was used. Overall, as a prevalence of an agent-specific
decision-making and based on the distance between point of departure and the
targeted destination, virtual individuals of different
sociodemographic backgrounds, use different means of transportation and follow a different sequence and types of activities. Behaviours that were not explicitly programmed into the model’s
code, arise through the human agents’ interactions enabling the examination of
expected or unexpected emerging behaviours from the bottom-up. At the end of a model run, spatiotemporal trajectories
are coupled with spatially resolved pollution levels.

In this study, personal
exposure, expressed as inhalation adjusted exposure, was evaluated by assigning
PM10 and PM2.5 concentrations to human agents based on their coordinates, the type
of location and intensity of the encountered activities.
Figure 1 showcases the PM2.5 exposure profile, expressed as exposure time
series and daily intake dose per body weight, of a randomly picked human agent.

Figure
1. ABM exposure
profile. Human agent 42: 49-year-old male, married, with a full-time job,
post-secondary education, medium income.

ABM results
indicated that PM2.5 inhalation adjusted exposure between housemates can differ
by 32.2% whereas exposure between two neighbours can vary by as much as 77%,
due to the prevalence of different behaviours. Findings
from the compatibility check between the ABM virtual exposure data and the real
exposure profiles extracted from the sensors campaign indicated that the
difference between the median values in box plots of the same subgroup of (virtual
vs. real) population does not exceed the range of 5-12%. This demonstrates that
the ABM-derived emergent trajectories are compatible to human spatiotemporal
behaviours, which is of crucial importance since space-time and activity
information is a key determinant of personal exposure.

This study
provides details of a new methodology that permits the cost-effective
construction of refined time-activity diaries and daily exposure profiles,
taking into account different microenvironments and sociodemographic characteristics.
By modelling the heterogeneous routine of human agents, our ABM can produce
detailed information related to the societal system examined and can generate data
that could be used to fill in the gaps that exist in traditional datasets. This
method signifies a step forward, away from the earlier static nature approaches
of urban modelling, where total population is divided into homogeneous
subpopulations.

The
establishment of an ABM approach that integrates Socio-Economic Status (SES)
indicators with the capacity for aggregation and analysis at various levels of
population size, leads to an exposure assessment model especially useful for
vulnerable groups of population, such as children, the elderly and people
living in hot spot areas. This is an opportunity to capture evidence for cases
where specific subgroups of population are disproportionately exposed to higher
levels of environmental risk than other parts of the same society. This is,
therefore, a method where cases that exhibit environmental injustice can be
detected and interpreted through an artificial society type model. Moreover, the
proposed method can be used for evaluating the probable impacts of different
public health policies prior to implementation reducing, therefore, the time
and expense required to identify efficient measures.