Concepts in Spatial Data Science

Acknowledge

  • Dr. Elisabetta Pietrostefani &
  • Dr. Carmen Cabrera-Arnau

A course in Geographic Data Science

SCTIMST, Trivandrum

  • Sree Chitra Tirunal Institute for Medical Sciences & Technology, Trivandrum
  • An Institution of National Importance established by the Act of the Indian Parliament (Act No.52, 1980)

  • Dept. of Science & Technology

  • Three major focus areas

    • Bio-Medical Technology Wing
    • Super specialty Hospital
    • Public Health (AMCHSS)

  • Healthcare Technology development

  • inter-disciplinary initiatives

  • Running MPH program since 1997, PhD programs since 2003

Introduction

  • Public Health - Science vs. Advocacy
  • The need for participatory decision making in public health
  • The transparency of open data science approach
  • The beauty of computational reports, presentations, etc.

Work plan

Lectures

  • Essential concepts
  • Mainly to get the big picture
  • Enthusing interest rather than teaching
  • Welcome to the open data science initiative!

Lab work

  • Do it yourself
  • Get skilled in the process
  • Come out of your comfort zones and collaborate!
  • Use data for dialogue!

What information does GIS use?

  • Data that defines geographical features like roads, rivers

  • Soil types, land use, elevation

  • Demographics, socioeconomic attributes

  • Environmental, climate, air-quality

  • Annotations that label features and places

What is Spatial Data Science?

Spatial Data Science

  • Analyse and extract insights from geospatial data
  • Work with real-world data on a number of domains and problems
  • Acquire key data science skills and important tools to answer spatial questions

It is especially true in public health

GIS

Layers - Image - Data

GIS world vs. Real World

Skills for public health data science

Hard Skills - Programming Language - Transparency and Reproducibility - Version control

Soft Skills - Communication - Storytelling - Geospatial analytics acumen - Ethical skills

R software for Spatial Data Science (SDS)

Graphical User Interfaces (GUIs)

  • QGIS and GRASS has revolutionized Open source Spatial Information Systems (GIS).
  • However, the reproducibility aspect has many challenges

Command Line Interfaces (CLIs)

Command Line Interface (CLIs) of R software is a good way to bring in reproducible algorithms for GIS/SDS

The Spatial Data ‘Revolution’

Advanced Hardware: High-performance computer hardware and efficient algorithms allow us to process vast data sets quickly.

Scalable Software: Scalable solutions with the R environment help us to sift through the data deluge, and extract valuable insights from the noise.

Spatial Databases: The advent of spatial databases empowers us to store and manipulate manageable subsets within the vast ocean of spatial data.

SDS in Public Health

  • Data Science:“gathering data messaging it into a tractable form, making it tell its story and presenting that story to others”


Loukides (2011) What is Data Science?

Traditional datasets in healthcare

  • Collected for the purpose (carefully designed)
  • Detailed and informative (“rich profile and portraits of the country”)
  • High quality

Traditional health and allied sector data

  • Massive enterprises (very costly)
  • Coarse in resolution (to preserve privacy they need to be aggregated)
  • Slow - the more detailed, the less frequent they are available

Examples

  • Decennial census (census geographies)
  • Longitudinal surveys
  • Custom collected surveys, interviews etc.
  • Economic or well-being indicators

New Forms of spatial data

Tied into the geo-data revolution

  • Accidental : created for different purposes but available for analysis as a side effect

  • Very diverse in nature: resolution and quality but, potentially much more detailed in both space and time

Challenges (Arribas-Bel, 2014)

  • Bias
  • Technical barriers
  • Methodological “mismatch”

Part 2

(Geo)visualisation

By encoding information visually, they allow to present large amounts of numbers in a meaningful way.

A map for everyone

A real public health tool

Maps can fulfill several needs, looking very different depending on the end-goal.

MacEachren & Kraak (1997) identify three main dimensions:

  • Knowledge of what is being plotted
  • Target audience
  • Degree of interactivity

MacEachren & Kraak (1997)

DiBiase’s (1990) “Swoopy”

Translating numbers into a (visual) language that the human brain “speaks better”

Exploratory Visualization


“forces us to notice what we never expected to see” (Tukey 1977: vi)


  • Mostly for ourselves in the course of the research process.

  • Many, quick and dirty, and rather unattractive graphs.

Explanatory Visualization


“forces readers to see the information the designer wanted to convey” (Kosslyn 1994: 271)


  • Mostly for others after the research is completed.

  • Few, carefully crafted, and attractive graphs.

Choropleths

Thematic map in which values of a variable are encoded using a color gradient of some sort

  • Counterpart of the histogram

    Both allows us to gage the distribution of a variable

Part 3

Spatial Weights

For a statistical method to be explicitly spatial, it needs to contain some representation of the geography, or spatial context. One of the most common ways is through Spatial Weights Matrices

  • (Geo)Visualization: translating numbers into a (visual) language (colors) that the human brain can interpret.
  • Spatial Weights Matrices: translating geography into a (numerical) language that a computer can interpret.

Spatial Weight Matrices

Spatial Weights Matrices are building block for spatial analysis and statistics.

They are used to assign a weighted average or sum of neighbouring data values to an observation, or other point in space.

  • Relates to concepts of spatial ‘smoothing’ and interpolating data

  • They can be used to see how one’s characteristics or outcomes is correlated with their neighbours: e.g. education, criminality, disease risk factors,…

Core element in several spatial analysis techniques

  • Spatial autocorrelation
  • Spatial clustering/geo-demographics
  • Spatial regression

Spatial Weights

Spatial Weights represented by \[W\] N x N positive matrix that contains spatial relations that are translated into values

  • If you are not a neighbour, \(value = 0\)
  • If you are a neighbour, \(value<0\)

Website

https://drarunmitra.github.io/GIS4Epidemiology/