After a first super succesful session in May 2023, from which we developed a package for radial analysis with the team of Rouen (UMR Idees), Port-Louis (Morbihan, France) will welcome our 2nd coding retreat from February 26th to March 1st.

The retreat is aimed as a team building moment where uni.lu geographers and associated researchers spend time together consolidating their code.

This is mostly an R spatial training, peppered with some Py spatial analytics.

Rationale: we do all have our “own” pieces of code on our personal machines, on a server and/ on our own Git repos, which we fine tune along various projects and need. Then we quickly discover other colleagues have done similar, sometimes better, would benefit from similar lines, or, when reproducing, that some parts do not lead to the exact same results for some reasons (parameters, cut-offs, pre-processing of data,…)

In view of (i) reproducibility, (ii) improving, and (iii) sharing our codes, the retreat is a moment to take stock, compare and develop further our scripts and data.

The program is super simple:

• Intense mornings: 8AM to 2PM Monday to Friday including a quick sandwich lunch,
• Some more relax time up to 4.30PM (a walk outside before sunset)
• And further reading/correcting in the evening 5PM-7PM before enjoying a creperie or a nice
• … starting anew the next day…

For the venue, we joint-venture with the Gîtes de Kerouzec (https://gitesdekerouzec.fr/) so we are housed together, and live for and eat for coding 24/7 (well…24/5) while enjoying a view, a beach and a small but fully serviced town where all our needs are fulfiled in under 5min walk! No time wasted!

Organizer: Geoffrey Caruso, University of Luxembourg.

Venue: 5 Place au Bois, 56290 Port-Louis FR

2 presentations by Yufei Wei and by Estelle Mennicken

When? 18th June 2021, 14:00-16:00

Where? Webex

The Quadtrees Hubs are to share opinions and discuss research in progress. The meetings of Quadtrees Hubs are open to anyone interested and somehow familiar with some quantitative techniques and willing to progress with these. Please contact Isabelle Pigeron-Piroth isabelle.piroth@uni.lu for more information and obtaining the webex link.

The Effects of Distance to City Centers and Population Size on the NO2 Concentrations in Europe

Yufei Wei (University of Luxembourg, DGEO)

Ground-level NO₂ surface concentrations measured by monitoring stations and tropospheric NO₂ columns from Sentinel-5P are the data sources of NO₂. We filter the data to get the annual mean NO₂ concentrations of European cities. We regress the data to find how urban population size influences the NO₂ concentrations, and how the NO₂ concentrations change within the cities.

The results show larger cities have higher levels of NO₂ concentrations. We also find distinct spatial patterns of the NO₂ concentrations within the cities. The results indicate monitoring stations and Sentinel-5P are reliable in describing and predicting the NO₂ concentrations of European cities.

European urban cores under pressure: quantifying the congestion of trips in and out from city centers as function of population size

Estelle Mennicken (LISER, UDM)

Traffic congestion has many negative social, environmental and economic consequences. We can cite among others the loss of time (hours of delay) inducing productivity and well-being losses, the excess fuel consumption, and the excess emitted CO₂. Understanding and quantifying this phenomenon at the scale of an entire continent is therefore a serious societal challenge. In particular, we focus on comparing the accessibility of city centers based on the geographical location of urban areas but also on the population size. Larger cities benefit from positive agglomeration effects but whether they proportionally show more radial traffic congestion is still an open question.

We simulate intra-urban trips between residential locations and city centers in 303 European cities and retrieve travel information around the clock during a typical weekday. We then compute several congestion indices to reveal the excess time people spend on roads driving during peak traffic time compared to a free-flow situation and establish a city ranking. Second, we observe how the total population size influences the indices. Finally, thanks to previously computed detour indices, we examine the relationship between the physical road network shape and the congestion.

Kaarel Sikk (University of Luxembourg, C2DH and DGEO)

When? 21^st May 2021, 2 to 3 pm

Where? via Webex (request meeting link by registering to isabelle.piroth@uni.lu or geoffrey.caruso@uni.lu)

Point process modelling provides a framework for exploring systems that can be observed as a set of points. In this study, we studied archaeological settlement patterns with the purpose of isolating regions in landscapes that were suitable for habitation by different populations. We create point process models of the settlement systems of hunter-fisher-gatherer groups (Narva and Combed Ware Culture) and early agrarian communities (Corded Ware Culture) in Stone Age Estonia and compare the spatial structure.

We conceptualize settlement system formation as a point process and develop a first-order point process model representing the environmental suitability for habitation based on geomorphology, soil, and proximity to water. We use MaxEnt and the SDMTune machine learning framework for building the model, variable selection, and estimation. The model is applied to the two communities and the effects of the variables and the resulting spatial patterns compared.

The spatial comparison showed significant differences between the suitable environments for habitation between the two groups. While the hunter-fisher-gatherer population had an entirely shoreline-connected settlement system, the Corded Ware people inhabited the areas further away from water bodies. This resulted in significantly expanded potential space with differing spatial configurations for the incoming agrarian groups but the areas also had a certain overlap.

The results also indicated higher predictive power for hunter-fisher-gatherer sites, which might be caused by a higher variety of agrarian activities, different socio-economic organizations, or effects of the spatial structure of the landscape.

The aim of Quadtrees’ Hubs is to share and discuss research in progress. The hubs are open to anyone interested and somehow familiar with quantitative spatial analysis and modelling and willing to progress with these. Please contact Isabelle Pigeron-Piroth or Geoffrey Caruso for information or to obtain the linnk to the meeting.

On May 21^st, our sixth Quadtrees Hub will take place. The aim is to share and discuss research in progress. Quadtrees’hubs are open to anyone interested and somehow familiar with some quantitative techniques and willing to progress with these. Please contact Isabelle Pigeron-Piroth for information.

When? 21^st May 2021, 2 to 3 pm

Where? via Webex

2-3 pm : Marlène Boura  (University of Luxembourg, DGEO): Towards a spatially explicit urban CO₂ budget. Urban carbon emissions, dispersion and sequestration in Europe

Modelling a steady-state urban carbon balance for 802 European cities at a fine spatial resolution.

Anthropogenic CO2 emissions are downscaled spatially (down to 1 ha) and temporally (from annual to daily) based on the sector of activity, the land use category and the location. Sequestration of CO2 is estimated for different types of urban vegetation, following the IPCC guidelines at the same spatial and temporal resolutions.

For one typical day of each month, we simulate 2 steady-state situations for the CO2 molecules dispersion and capture. The absolute carbon balance and the relative carbon capture (as a percentage of effective anthropogenic emissions) are then computed. The data produced can be used to assess the spatial heterogeneity of the carbon balance within a specific urban area. It can also be used to assess how much of its own emissions an urban area can capture.

On March 10th (NEW DATE), our fifth Quadtrees Hub will take place. The aim is to share and discuss research in progress. Quadtrees’hubs are open to anyone interested and somehow familiar with some quantitative techniques and willing to progress with these. Please contact Isabelle Pigeron-Piroth for information.

HUB #5

When? 18^th of February 2020 CHANGED TO March 10th 14h-15h.

Where? Map Room (next to GIS room) 1st floor MSH, Belval.

14h-15h : Kerry Schiel  (Université du Luxembourg, GEO) : Portraying Urban Sprawl from Space: an update and geographical effects

We test the robustness to change in geographical grids and extents of the urban sprawl measure implemented a decade ago by Burchfield et al (2006) on the conterminous United States. They calculate urban sprawl as the percentage of open space surrounding residential cells using a fixed 1km2 grid.  They then used this index to determine which factors positively contribute to the development of sprawl.

Instead of a fixed grid, we use a moving 1km2 window, centered on each cell, which is more in line with standard GIS and landscape metrics applications and with the sprawl concept suggested. We also change the areal extent used within each metropolitan area for the calculation based on the development of residential cells in the period between 1976 and 1992.  This is to determine whether changing the definition of the urban fringe will affect the outcome of the sprawl index values.  Mostly recently, we calculated an updated measure of sprawl using data from 2016, to compare with that from 1992.  Through these methods, we analyze how our more spatially accurate calculations affect the sprawl index values, and hence, the understanding of the factors that contribute to sprawl.

When? 21st of January 2019 14h-16h.

Where? Map Room (next to GIS room) 1st floor MSH, Belval.

14h-15h : Justin Delloye (LISER) : A mathematical approach of residential migration dynamics

The spatial distribution of a regional population, the residential pattern, is an important topic for public authorities because it is related to concerning issues such as access to housing or sustainable commuting. Its evolution is essentially driven by residential migrations, which form a complex dynamic system because of the feedbacks between residential migrations and residential patterns. To handle this complexity, researchers usually rely on numerical simulation tools such as agent-based models or cellular automata. However, these methods still suffer from a few shortcomings among which calibration and validation issues, as well as biases due to discrete time steps.

In this talk, I present a mathematical model of residential migrations that can complement current numerical approaches by addressing these calibration, validation and temporal issues. The model is presented with two applications. First, an individual model of residential migration is built, which reconciles previous stochastic representations of dynamic spatial interaction models by introducing an endogenous “decision rate” parameter. This parameter is then estimated using Belgian residential flows in 2005. Second, the individual model is aggregated to express the evolution of a residential pattern as a Markovian system. This aggregation procedure is then used to study the structural properties of a traditional core-periphery model of economic geography.

15h-16h :  Brano Glumac (LISER) :What are the housing needs and solutions for the households with frequent residential mobility patterns? A qualitative systematic review

The objective is to generate the transdisciplinary conceptual framework capable of sketching the attributes of flexible housing solutions and their benefits to households with frequent residential mobility patterns.

Residential mobility has an impact on available housing solutions and vice versa. Frequent residential change is potentially a useful marker for the household financial danger and lack of wealth accumulation via classical property tenures. Thus this systematic review supports the enlargement of housing solutions by generating and exploring a framework to support these residentially mobile households for whom lack of means for wealth generation may be identified. Therefore, a primary research question asks what are the housing needs and solutions reported in the literature that should suit households with frequent residential mobility patterns. Followed by a secondary question; what are the attributes of the identified flexible housing solutions?

Narrative and framework synthesis were used consecutively to generate and explore the conceptual framework of flexible housing. Both are qualitative synthesis based on configurative synthesis within a purposive sampling framework approach. This approach is the best suited to generate new emerging concepts. The first step of framework synthesis approach provides an answer to a broader question of the housing needs for the population with frequent mobility patterns. The second step identifies the attributes of the non-exhaustive list of the interventions. Searching for studies adopts a purposive sampling framework (author identified three groups of housing interventions or solutions for residents with high mobility pattern: technical, service-based and institutional) based on an existing body of literature. Studies that investigated international migration for asylum were excluded. Two reviewers assessed each study using quality criteria with particular attention to the consideration of bias. Data were extracted for analysis using a thematic analysis. Eight-hundred studies were included for this review.

Tuesday 17th of December, our third Quadtrees Hub will take place – to discuss ongoing research in quantitative methods, urban analytics and spatial data from both the Urban Development and Mobility Dpt of LISER and the Dpt of Geography and Spatial Planning at the University of Luxembourg. The aim is to share and discuss research in progress. Quadtrees’hubs are open to anyone interested and somehow familiar with some quantitative techniques and willing to progress with these. Please contact Isabelle Pigeron-Piroth for information.

HUB #3

When? 17th of December 2019 14h-16h.

Where? Map Room (next to GIS room) 1st floor MSH, Belval.

14h-15h : Marlène Boura : Landscape typology of urban forest ecosystem services across European urban areas

In this talk we will present a typology of more that 800 Functional Urban Areas (FUAs). Urban areas are clustered regarding spatial metrics associated to different urban forest ecosystem services (ES) potentials and threats.

Urban areas exhibit a large variety of patterns which may affect differently the potential of ES. ES are essentials to counteract the urban pressure on the environment and its impact on the well-being of its inhabitants. Yet, the effects of the relative spatial arrangement of vegetation, forests and water bodies, with respect to the artificial urban lands on potential ES are still not systematically analysed.

We propose a typology, based on the intra-urban structure of cities and the associated ES potentials coming from the urban forest. More particularly, we investigate the share of different land uses and the distance between human settlements, forests and the other vegetated lands as well as their relative spatial distribution within urban settlements. We then use spatial metrics as proxies for urban ES associated with urban forests – e.g., micro and macro climate regulation, air pollution removal, rainwater runoff, mental and physical health. The typology is created using an unsupervised machine learning approach (clustering) with standardized spatial metrics as input data.

15h-16h : Kaarel Sikk : Model exploration with OpenMole software

In the talk we discuss model exploration techniques and present them using dedicated OpenMole software. The talk focuses on sensitivity analysis and calibration of spatial Agent-based simulation models. Some other model exploration techniques are described. 

To do so we are using an ABM simulation of hunter-gatherer central place foraging model. The model was developed based on traditional aspatial analytical model coming from the domain of human behavioural ecology. Spatial ABM implementation added opportunity of testing the model with heterogeneous environments. With presented examples I am testing models robustness to environment heterogeneity generating random environments with different spatial autocorrelations. 

For robustness testing we present a input space exploration experiment with Sobol sampling over the parameter space. We are also exploring ideas designing an origin search experiment by extending traditional model calibration techniques. 

First of all we load the mtcars data set. Then we want to see what is contained within the data set and also view what type of data each variable is.

The raw code is also available from my github, see:

https://github.com/granger89/SCALEITUP/blob/master/graphs%20in%20ggplot2

library(ggplot2)
cars <- mtcars
cars
str(cars)

We can see there is no categorical variable. We are going to create one which we will use for plotting we are going to create two; one for the make of car and another for its weight class.

cars$car_type <- rownames(cars)

Next we use summary to help us when deciding upon the thresholds.

summary(cars$wt)
cars$wt_class <- NA 
indx <- cars$wt<2.5 
cars[indx, "wt_class"] <- "light"
indx <- cars$wt>=2.5 & cars$wt<3.4
cars[indx, "wt_class"] <- "medium"
indx <- cars$wt>=3.4
cars[indx, "wt_class"] <- "heavy"

We now have the data frame in the correct format we want for plotting

str(cars)

The following code gives us a simple plot of two variables mpg (x-axis) and hp (y-axis) using the dataset cars.

ggplot(cars) +
geom_point(aes(mpg, hp))

We now introduce a third variable. We can change the point shapes using a categorical variable. For this we use the weight class variable we created earlier.

ggplot(cars, aes(x=mpg, y=hp, group=wt_class))+
geom_point(aes(shape=wt_class), size=2, alpha=1)

Now we have the graph we next steps involved editing the axes, legend, title and background so that we get a cleaner graph.

Editing the axes and background

Firstly we will import new fonts.

library(gridExtra)
library(extrafont)
library(ggthemes) # Load
fonts()
windowsFonts("Arial" = windowsFont("Arial"))
windowsFonts("Times New Roman" = windowsFont("Times New Roman"))
windowsFonts("Goudy Stout" = windowsFont("Goudy Stout"))

Give the plot a white background and change the fonts

ggplot(cars, aes(x=mpg, y=hp, group=wt_class))+
geom_point(aes(shape=wt_class), size=2, alpha=1)+
theme_bw()+
theme(panel.background = element_rect(fill = "white", colour = "white"))+
theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
theme(panel.grid=element_line(color = "grey80"))

Change the x and y axis names and also remove spacing around the origin. We can also change the limits of the x and y axes.

ggplot(cars, aes(x=mpg, y=hp, group=wt_class))+
geom_point(aes(shape=wt_class), size=2, alpha=1)+
theme_bw()+
theme(panel.background = element_rect(fill = "white", colour = "white"))+
theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
theme(panel.grid=element_line(color = "grey80")) +
labs(x = "Miles per Gallon (mpg)", y = "Horse Power (HP)" )+
labs(title = "Cars power, weight and fuel economy")+
scale_x_continuous(limits = c(0,35), expand = c(0, 0)) +
scale_y_continuous(limits = c(0,350), expand = c(0, 0))

Fixing the legend

One thing I would like to fix is the legend as it does not look that pleasant. I would like the first letters to be capitals and ‘wt_class’ does not explain what the values show.

scale_shape_manual(values=c(1, 2, 3), name = "Weight Group", breaks=c("heavy","medium","light"), labels=c("Heavy >3.4 Tons", "Medium 2.5-3.4 Tons", "Light <2.5 Tons"))

To explain this code, the values are the type of shapes used to represent the different classes. See the link for what the different values mean.

Next ‘name’ is the title of the legend.
‘breaks’ are the different classes. Once we know the values we can then put them in whatever order we want. (“heavy”,”medium”,”light”) or (“light”,”medium”,”heavy”) . That is the order they appear in the legend.
Next we use labels to replace the raw data values with whatever value we want.
http://www.sthda.com/english/wiki/ggplot2-point-shapes

ggplot(cars, aes(x=mpg, y=hp, group=wt_class))+
   geom_point(aes(shape=wt_class), size=2, alpha=1)+
   theme_bw()+
   theme(panel.background = element_rect(fill = "white", colour = "white"))+
   theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
   theme(panel.grid=element_line(color = "grey80")) +
   labs(x = "Miles per Gallon (mpg)", y = "Horse Power (HP)" )+
   labs(title = "Cars power, weight and fuel economy")+
   scale_x_continuous(limits = c(0,35), expand = c(0, 0)) +
   scale_y_continuous(limits = c(0,350), expand = c(0, 0))+
   scale_shape_manual(values=c(1, 2, 3), name = "Weight Group",
                      breaks=c("heavy","medium","light"),
                      labels=c("Heavy >3.4 Tons", "Medium 2.5-3.4 Tons", "Light <2.5 Tons"))

Finally we can position the legend. Currently it is squeezing the size of the plot area. We are likeyl to have some empty space where we can place it. The position varies on the x and y axes from 0 to 1. So how about we place it at 0.2 on the x-axis and 0.2 on the y-axis. I’ve also given it a border and made the background white.

ggplot(cars, aes(x=mpg, y=hp, group=wt_class))+
   geom_point(aes(shape=wt_class), size=2, alpha=1)+
   theme_bw()+
   theme(panel.background = element_rect(fill = "white", colour = "white"))+
   theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
   theme(panel.grid=element_line(color = "grey80")) +
   labs(x = "Miles per Gallon (mpg)", y = "Horse Power (HP)" )+
   labs(title = "Cars power, weight and fuel economy")+
   scale_x_continuous(limits = c(0,35), expand = c(0, 0)) +
   scale_y_continuous(limits = c(0,350), expand = c(0, 0))+
   scale_shape_manual(values=c(1, 2, 3), name = "Weight Group",
                      breaks=c("heavy","medium","light"),
                      labels=c("Heavy >3.4 Tons", "Medium 2.5-3.4 Tons", "Light <2.5 Tons"))+
   theme(legend.position=c(0.2,0.2))+
   theme(legend.text.align = 0)+
   theme(legend.title=element_text(size=10))+
   theme(legend.text=element_text(size=10))+
   theme(legend.background = element_rect(colour = 'black', fill = 'white', size = 1, linetype='solid'))

Vary size by value of variable

Finally lets introduce a fourth variable. Instead of all points having a fixed size let us vary the size based on the number of gears.

ggplot(cars, aes(x=mpg, y=hp, group=wt_class, size=gear))+
   geom_point(aes(shape=wt_class), alpha=1)+
   theme_bw()+
   theme(panel.background = element_rect(fill = "white", colour = "white"))+
   theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
   theme(panel.grid=element_line(color = "grey80")) +
   labs(x = "Miles per Gallon (mpg)", y = "Horse Power (HP)" )+
   labs(title = "Cars power, weight and fuel economy")+
   scale_x_continuous(limits = c(0,35), expand = c(0, 0)) +
   scale_y_continuous(limits = c(0,350), expand = c(0, 0))+
   scale_shape_manual(values=c(1, 2, 3), name = "Weight Group",
                      breaks=c("heavy","medium","light"),
                      labels=c("Heavy >3.4 Tons", "Medium 2.5-3.4 Tons", "Light <2.5 Tons"))+
   theme(legend.position=c(0.2,0.3))+
   theme(legend.text.align = 0)+
   theme(legend.title=element_text(size=10))+
   theme(legend.text=element_text(size=10))+
   theme(legend.background = element_rect(colour = 'black', fill = 'white', size = 1, linetype='solid'))+
   scale_size_continuous(range = c(3,5),  
                         breaks= c(3,4,5), name="Number of \nGears")

Label specific points

Also I would like to label the outliers at the upper and lower end. For this we use ggrepel

library(ggrepel)

ggplot(cars, aes(x=mpg, y=hp, group=wt_class, size=gear))+
   geom_point(aes(shape=wt_class), alpha=1)+
   theme_bw()+
   theme(panel.background = element_rect(fill = "white", colour = "white"))+
   theme(axis.text=element_text(size=11, family="Arial"),axis.title=element_text(size=12,face="bold", family="Arial"))+
   theme(panel.grid=element_line(color = "grey80")) +
   labs(x = "Miles per Gallon (mpg)", y = "Horse Power (HP)" )+
   labs(title = "Cars power, weight and fuel economy")+
   scale_x_continuous(limits = c(0,35), expand = c(0, 0)) +
   scale_y_continuous(limits = c(0,350), expand = c(0, 0))+
   scale_shape_manual(values=c(1, 2, 3), name = "Weight Group",
                      breaks=c("heavy","medium","light"),
                      labels=c("Heavy >3.4 Tons", "Medium 2.5-3.4 Tons", "Light <2.5 Tons"))+   theme(legend.position=c(0.2,0.3))+   theme(legend.text.align = 0)+   theme(legend.title=element_text(size=10))+   theme(legend.text=element_text(size=10))+   theme(legend.background = element_rect(colour = 'black', fill = 'white', size = 1, linetype='solid'))+   scale_size_continuous(range = c(3,5),                           breaks= c(3,4,5), name="Number of \nGears")+   labs(size="Number of Gears")+   geom_text_repel(data = subset(cars, mpg > 25 | mpg <16), size=3,aes(label = car_type))

For other customisable options please refer to the ggplot2 documentation which is excellent at explaining how to add further options.

https://cran.r-project.org/web/packages/ggplot2/ggplot2.pdf

HUB #1 will have some focus on cities and air pollution with 2 inputs, by Yufei Wei and Hichem Omrani. See below.

When? 22nd of Oct 2019 14h00-16h00. Where? Map Room (next to GIS room) 1st floor MSH, Belval.

Yufei Wei : Scaling of urban heat island and nitrogen dioxide with urban population: a meta-analysis.

In this research, at the beginning a qualitative synthesis is performed to collect literature introducing the relations of urban heat island intensity and nitrogen dioxide concentration with urban population. We then find and validate the linearity of urban heat island intensity and nitrogen dioxide concentration with urban population size by ANOVA test and linear regression, based on the selected literature from qualitative synthesis.

Hichem Omrani : Spatio-temporal Data on the air pollutant nitrogen dioxide derived from Sentinel satellite

Do you see the cube or the hexagon? Both maybe? (and the Q of Quadtrees*, obviously!)

Sometimes we geographers have a flat x, y understanding of space. A useful simplification for sure…. sometimes. But space comes with many z dimensions: physical height, time, socio-economic, or environmental dimensions. A quadtree is a manner to compress a 2D (x,y) based information (z usually being a color code on a map) within a nested structure.

Measuring and modelling in geography is in all case a mental projection and a kind of scientific compression of what is happening. They are insufficient and do not pretend to bring up truth, but they are helpful simplifications and scientific constructs.

…all this wrapped up in a sleek logo!

 * Note that the name, Quadtrees, chosen for our virtual grouping of quantitative geographers in Lux, was suggested by Kate Jones within a brainstorming effort with Geoffrey Caruso who struggled to find a catchy and sensical name.