PIXELS AREN’T PLACES

Rachel Franklin (@rsfrankl)

Center for Geographic Analysis | Harvard University

2026-05-20

Disclaimer

I’m not here on behalf of the anti-pixel lobby

This is a talk about how we leverage the magic of satellite imagery for social science research.

Main argument

Satellites are a game changer but require careful thinking about how we believe place matters and this has implications for both theory and data.*



What I’m going to talk about

  • Geography
  • Some aspects of measuring place
  • How pixels can help
  • A quick example

But first

Some ways we are interested in “place”


Why is this place more developed, happier, healthier than that one?


Which places are most vulnerable to effects of economic crisis or climate variability?


Why do people move from this place to that one?


How is this place changing over time?


How do place conditions affect outcomes…for people, businesses, policies?


What hazards are individuals exposed to as they move through these places?

Spatial data helps us do better at understanding and measuring place

Traditional data (often also spatial!)

  • Foundational role in social science research
  • Reliable for socio-demographic characteristics
  • Trade-off between location and individual information
  • Rigid conceptualization of place (which is often just fine for us!)
  • Only periodic updates

Novel forms of spatial data

  • Frequent updates, maybe even real time
  • More fine-grained, exact location
  • Potentially greater insight into individual movement, behaviors, preferences
  • Often spatially and socio-economically selective
  • In many cases more applicable for placing individual movement, behaviors, preferences within a context, than actually measuring that context

A exception: satellite imagery

Geography

There are lots of ways social scientists use geography


Different applications demand different conceptualizations

Place


Has political, economic, and/or psychological valence. Often has a name.

Could be formal and administratively defined, like a statistical unit, or could be perceptual, like your neighborhood.

Boundaries matter.

As researchers, we often don’t control the definition of a place

  • Geographies are “off the shelf”
  • Our challenge is data. What are places like?
    • Traditional data give only an incomplete picture
    • A gap emerges for novel types of data to fill

Pixel


The unit upon which satellite images (and photos) are based. Resolution and boundaries are determined by a machine in space. Boundaries are arbitrary and meaningless.

Pixel size, sensor, and post-processing together will determine what we “see” for a particular location.

We are unlikely to go out and look for other data for that geography.

Context


Spatial influence on some outcome of interest (like education or health). Could be a place!

(probably isn’t a pixel)

Context is possibly the toughest: how do we know the spatial zone of influence and how do we get data that matches that geography?

Lots of other geographical concepts too

  • Location: Attaching information to a point, line, or polygon
    • Information could come from pixels…or not…
  • Trajectory: Movement through space
    • We often want to estimate exposures, influences, or contexts for routes or movement
    • Pixel could make sense…or not…
  • View-shed: What can we see from here?
    • Inherently pixel-based

Some aspects of measuring place

What are the spatial mechanisms or relationships we’re hypothesizing?

Are there undefined or fuzzy boundaries?

Not always: administrative versus perceptual

However, place can be ego-centric and uncertain

However, place can be ego-centric and uncertain

So far, so good


Research should be grounded in a conceptualization of place that suits our purposes.


Then we need data for these geographies.

What do pixels have to do with it?

Pixels aren’t places but they are pretty cool



  • Leaving aside the breadth of metrics and information they can provide
  • Uniform spatial building blocks
  • Flexible aggregation power
  • In short: we can make places from pixels

And sometimes a pixel is just a pixel

Getting the big picture (literally)

What was the temperature at a particular point?

What about heat exposure along a route?

Getting from pixel to place

We can aggregate pixels to known place geographies

And hypothesized local contexts (this is great!)

A quick example




Project website: https://imago.ac.uk

Predicting neighborhood cloudiness

  • Clouds affect everything from surface temperatures to mental well-being to (probably) local spending
  • Traditional data are not capable of providing cloudiness data
  • Satellites are!

Predicting neighborhood cloudiness



  • However, makes no sense to look at cloudiness at the pixel level
    • PIXELS AREN’T PLACES
  • The magic is not only the data (probability of cloudiness) but the power of the pixel

Southern Orkney

Presenting Sun Probability Framework (SPF) for the UK

Parting thoughts I

  • We talk a lot about the potential of satellite imagery (and other spatial data)
  • Conversation often revolves around what new information, or data, imagery can provide
  • Or other advances in frequency, coverage, etc.
  • All of this is true

Parting thoughts II

  • An under-appreciated aspect of imagery, though, is its chameleon-like ability to aggregate to geographies we use in the social sciences
  • With a few caveats:
    • We need to understand what sorts of places we’re working with
    • We typically don’t need the data in pixels
    • Instead, what is needed is pixel-based data for places*

Parting thoughts III

  • Different data or more data doesn’t obviate the need for a solid conceptual foundation
  • Geography and space have an important role to place in social science research
    • Spatial inequalities, exposure, vulnerability—all of these have strong spatial components
  • BUT this means the onus is on us to think carefully about spatial relationships and mechanisms.
    • What kind of place are we talking about?
    • At what scale do we think causes or inputs are operating?
  • Good news is that, increasingly, we can locate the data we need for the places we’re working with.

The End.


rachel_franklin@cga.harvard.edu


https://www.rachelfranklin.org


Thank you!!