Chris BRUNSDON, Professor and Director of the National Center for Geocomputation at NUI Maynooth presented a brilliant overview of the links between Geocomputation and the social sciences. Three milestones structured his talk: big data, reproducible research and spatial statistics.

First, Chris BRUNSDON stressed how much the so-called “Big data” are not only large. They are also complex data. They provide new opportunities for social sciences, but specific issues have to be addressed:

• Experimental design: as Chris BRUNSDON says, “there often is none!” ;
• Metadata: there is a crucial need for documented metadata ;
• Openness in analysis, methods and algorithms, data and softwares.

Chris BRUNSDON illustrated his point with a quite surprising example: monitoring climate change from crowd sourcing first bloom dates of a specific plant, lilac in north America. Indeed, a rough and global analysis of the data produced suggest a delay in blooming over years, first bloom date getting later by about 1 day every 5 years, which is the opposite of expected effect (as global warming may anticipate the process instead of delaying it).
Introducing the spatial dimension reveals a large spatial variability in the blooming dates, something like an east-west gradient.
A random coefficient model, taking into account this spatial trend, therefore shows that first bloom date is getting earlier by about 1 day every 6 years. “Spatial is special” therefore, and should be taken into account!

In a second time, Chris BRUNSDON examined the question of “Big-Data” through the social networks. Based on the example of “Twitter”, he immediatly showed that the complexity both comes from the kind of data collected and from the way to obtain them. Geocomputation tools allow to analyze the links between “#” symbols to finally draw word clouds and perform graph-based analysis.
He took the example of “#Liverpool” that leads to a two headed cloud : one for the city and the second for the football club.
The use of social network data offers the perfect illustration of what we call “giant connected components” in graph theory.
Chris BRUNSDON concluded this part revealing one of his actual scientific concern: to use the geocoded tweets and study geographical patterns to see linkages.

After this brilliant introduction to some key challenges involved by big data in the social sciences, Chris BRUNSDON took the opportunity to deliver thought provoking ideas about reproducible research. For him, once published, most research are dead-ends in the sense that they are not directly reproducible. Chris presented various approaches allowing coupling data, and computational codes with publication.He suggested some possible good practices, favoring on the long term reproducibility of research. He also demonstrated interesting possibilities offered by Rstudio, using shiny render-plot function.

Finally, Chris BRUNSDON offered a remarquable introduction to spatial statistics, focusing on what makes spatial data so special:

• the so-called Modifiable Areal Unit Problem (MAUP) ;
• spatial dependency ;
• spatial heterogeneity ;

Chris BRUNSDON finally delivered a key message: a good model should be able to generate data sets that look like the actual data set. This is one very direct way of validating models.

Tim KOHLER – Regents Professor and Graduate Coordinator in Archaeology and Evolutionary Anthropology at the Washington State University (USA) and also member of the Science Board at the Santa Fe Institute – led a very exciting conference untitled “Reconstructing Past Dynamics by coupling spatio-temporal data analysis and agent-based modelling”.

Making perfectly echo to the conference of Chris BRUNSDON, Tim KOHLER showed that the term “Big Data” can also be employed in academic fields such as Archeology. If Chris BRUNSDON showed that the term Big data can refers both to size and complexity, Tim KOHLER prove that we can also talk about Big Data on a 10 000 individuals sample characterized by an incredible complexity. Tim is convinced of the needs for change in archeology. If fieldwork remains an essential prerequisite, new needs emerge in the field of spatial data analysis. He articulated his point around two axes: one “empirical / inferential” and the other “deductive”.

Based on his own experience of researcher, Tim KOHLER presented various results from the Village Ecodynamics Projet (VEP) led in Southwestern Colorado. Keeping in mind the Neolithic Demographic Transition theory, Tim wants to understand why an almost perfect agricultural sector has experienced phases of depopulation. Studying a 700 years sequence (from AD 600 to 1300), he showed us the importance of climatic perturbations in the local production (e.g. maize) such as phenomenons of competition between communities in conjunction with population growth to understand the quantitative evolution of the population and also its geographic distribution. Neolithic societies that functioned lived in world of rapid population growth. In a second example focused on the Mesa Verde region, Tim presented an analysis of 3621 habitation sites from surface evidence making a partition of each into one or more of 14 periods using Bayesian approach applied to surface count of ceramics.

After this dense and brilliant first part, Tim KOHLER presented the strengths of Agent-Based Modelling (ABM) applied to archeology. Drawing a brief history of ABM from the Swarm project headed by Chris Langton to Repast or Netlogo, he reminded us that the most interesting aspects of a system emerge from the interactions between agents who have a local knowledge. Tim insisted on the fundamental term of emergence. During this second part, he developed the importance of data driven models that can mix population and resources dynamics, competition for resources etc. The key point of these work is the reproduction of patterns that indirect observations (e.g. tree rings) could previously highlight.

Tim concluded his point humorously explaining that, in a way, his scientific position leads him to produce models appearing at the same time too simple for his empirical colleagues and too complex for his modeler colleagues.

The third day conference of the Spatial Systems and Dynamics Summer School was devoted to the Spatial Networks and Flows topics. Two brilliant speakers – from distinct disciplinary origins – highlighted this crucial topic based on the example of the city of Seoul (South-Korea). The first part of the speech was led by Keumsook LEE, Full Professor of Geography at the Department of Geography of Sungshin’s Women University and the second part has been conducted by Moo Young CHOI, Professor of physics at Seoul National University. During their two-headed presentation, they presented ideas and methods for analyzing spatial data networks and flows keeping in mind that “understanding time-space characteristics of intra-urban passenger flows provide insights in transportation and urban studies”.

Seoul suffers from severe congestion problems mainly due to: activities’ concentration, a large population, urban sprawl and the increase of car ownership. The aim of such studies would be for example to encourage a decrease in the automobile use and increase the public transportation share. The transportation cards of Seoul (T Cards) allow a tracking of time and location information of each passenger using the public transportation network. Keumsook LEE presented 4 steps of her analysis by explaining: the structural properties of the transportation system, the time-space characteristics of intra-urban passenger flows, and the dynamic visualization of passenger flows such as its dynamic complexity. Working mainly on passenger flows in the subway system, she analyzed her dataset by time-zone (morning/daytime/evening), by travel-time, by trip patterns and illustrated both origin and destination points. She then focused on the spatial distribution of passengers (depending on departure and arrival points). If the departures in the morning mostly come from residential areas, arrivals in the evening correspond to popular entertainment areas. Keumsook LEE finally proposed a dynamic visualization of passenger flows on a functional map by spatial sectors’ types. As concluding remarks, she presented distinct purposes of studies such as the development of various data-mining algorithms for exploring valuable traffic information from the big databases or the analysis of real-time population distribution patterns by integrating passenger flow data, land use data and smart media data.

The second part of the speech – by Moo Young CHOI – focused on Dynamic Complexity in transportation Networks. He introduced his point drawing an analogy with the biological systems such as social networks with the transportation networks. Moo Young CHOI defined the subway system as a functional (weighted) spatial network based on a geographical structure. He studied the dynamics of passenger flows by analyzing the Seoul Subway Network and making equations of time evolution of passengers’ distribution. To solve the complex issues in Seoul Transportation Networks, he both performed gravity model aiming at highlighting the flow decreases with distance (time) and power law correlations to analyze the bus network. Among many points, Moo Young CHOI showed that morning arrival sectors/evening departures sectors correspond to downtown business districts (fewer number of drastic changes); morning departures sectors/evening arrivals correspond to suburban residential areas; afternoon corresponds to stations, terminals and tourists spots. He showed clearly that evening is not the opposite of the morning, a proof of a diffusion phenomenon. In a second part, after talking about the criticality of the bus system, he presented an application of block renormalization to public transport stops. This coarse graining approach allows improving significantly models predictions by reducing the degrees of freedom of the system under study. Moo Young CHOI concluded his speech reminding key points such as the power-law distributions of maximum spanning-trees, the Log-normal and Weibull distributions of weights and strengths, the growth and the time-zone dependence of passenger flows.

One of the ultimate sentence of the conference summarized very well both presentations: “complex system approach applied successfully to transportation networks and passenger flows, revealing emergent complexity”.

Professor Lael PARROTT leads the Complex Environmental Systems Laboratory at University of British Columbia. She is also Director of the Okanagan Institute for Biodiversity, Resilience, and Ecosystem Services. She gave a very inspiring talk on the complexity of socio-environmental systems. Through various examples from physical, natural and social systems, she insisted on the idea that complexity requires a major paradigme shift, combining reductionist and holistic approches, and therefore an evolution not only of our methods but also of the way we think and act. She demonstrated this key issue with three different examples.

The first one concerns the management of maritime traffic in the saint Lawrence river estuary, which experiences 6000 cargo ships and 13000 whale seeing boats per year, while being a protected are for marine mammals. The 3MTSim model is a data-driven agent based model coupling a whale model, an environmental model and a social model, that was designed to become a decision support systems for management and planning in this specific context. The model allowed proposing new routes for cargo shipping and adapted speed reduction policy in the area.

The second example deals with forest management in Canada, as there is a new concern for the impact of forestry on endangered species. Lael PARROTT focused specifically on the case of Caribou, impacted by social (recreational) and economic (timber production) activities. The main objective is to find social, economic and ecological (woodland caribou) trade-offs. Again, a data-driven agent-based model helped investigating these critical issues.

As a third example, Lael PARROTT addressed the question of the futures of Okanagan. Three steps were followed: data analysis, ABM prototype in NetLogo and Final model development in Repast. She demonstrated the WaterDynamic prototype developed in NetLogo.

She finally raised the issue of sustainable development, each landscape providing essential goods and services: the challenge is to maintain them in the context of increasing human footprints. She concluded her talk by an open question: will complex systems approach help us attain sustainable and resilient landscapes?

Having a PhD in Mathematical Biophysics, Itzhak BENENSON is Full Professor and Head, Department of Geography, Tel-Aviv University. He is also Head of the Geosimulation and Spatial Analysis Lab.
Itzhak’ brilliant talk addressed a major issue: in an ever increasing interconnected and fast evolving world, more and more shaped by technologies, modeling and simulation take more and more importance as a way of understanding and managing socio-environmental systems.

Itzhak BENENSON focused on several examples.

Firstly, he demonstrated how much accessibility shape cities.

Accessibility is multi-scale phenomena:

• land use is the slowest component and can be assumed fixed ;
• transportation is an order parameter, i.e. a main driver ;
• individual utility is an outcome of people adaptation and can be averaged.

Public data have never been so accessible. Example: Fragile-success.rpa.org, open-street Map,…
These big data require many different operations before being usable for analysis. Then, they require often parallel or high performance computing. Itzhak compared accessibility to jobs estimation based on 10^6 buildings using parallel computing with the one estimated from traffic zones. The results obtained are qualitatively different and justify the adoption of more detailed data and analysis.

Moreover, accessibility has a shining futur, made of more flexibility and adaptation, in order to handle uncertainty: “Smart public transport must be adaptive!” says Itzhak BENENSON.
Parking is a relevant issue in such perspective: up to 30% of traffic in city is cruising for parking, which is highly uncertain. Adaptive parking price: parking is inherently heterogeneous in space, as is readiness to pay. Agent-based modeling, fed with detailed data, help us investigating these difficult issues.

Park-Agent is a data-driven ABM that uses and independant variable as a validation milestone: distribution of park car distance to destination. In collaboration with a private company, they created the parknav.com application. Then, the key problem of information sharing has to be felt with as, being informed, people adapt their behavior.

As a third example, Itzhak BENENSON addressed the question of pedestrian modeling and crowd simulation. The ABM model developed integrates very detailed motion, collision avoidance and decision making algorithms and is able to explore cars/pedestrians interactions in urban environments.

Smart city demands is pushing us towards new generation of models that:

• work at the resolution of objects and are spatially explicit ;
• are built as a bunch of application rather than one model for all ;
• exploit new database tools – implement new algorithms (parallel computing, cloud, HPC,…) ;
• are competitive and properly marketed (web-based, serious games).

All this will work on the solid basis of:

• Big data collection and analysis ;
• Field and lab experiments ;
• Calibration and validation ;
• Complex systems theory.

Smart city turns urban modeling into an industry! The question is: are we ready for this?