It's insanely tough to get high-res data to build super-resolution models.

However, now there's a new (free) dataset we can use.

It contains over 17,657 high-res/low-res image pairs.

Here's everything you need to know: Satellite image super-resolution is the process of increasing the resolution of satellite images.

Typically, this can be done to take the resolution of Sentinel-2 from 10m to 2.5m (or even 1m).

High-resolution satellite images can be insanely expensive to buy.

So here's a list of free datasets you can access.

These datasets can be used to build foundation models, super-resolution models, or for segmentation. The most commonly used free multi-spectral satellite images are from:

• Sentinel 2
• Landsat-8

However, Sentinel 2 has ~10m resolution (for RGB), while Landsat-8 is ~30m (for RGB).

But what free high-resolution datasets exist to train foundation models on?

The most important geospatial concept?

Spatial Autocorrelation.

It can make the difference between spurious and accurate results.

Here’s what you need to know about it: In short, spatial autocorrelation is a measure of how similar or dissimilar things are based on their locations.

To be a good geospatial data scientist, you need good tools.

My tool of choice is R.

Here are the packages I can't live without:

#rstats 𝟭. 𝗣𝗮𝗰𝗸𝗮𝗴𝗲𝘀 𝘁𝗼 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝗴𝗲𝗼𝘀𝗽𝗮𝘁𝗶𝗮𝗹 𝗱𝗮𝘁𝗮

• 𝗿𝗴𝗲𝗲𝗱𝗶𝗺: provides the easiest way to download data directly from Google Earth Engine as GeoTiffs.

• 𝗿𝗴𝗲𝗲: an R wrapper for Google Earth Engine, so you can run GEE commands in R.

We've all heard about corruption of national governments.

However, quantifying corruption of 𝘭𝘰𝘤𝘢𝘭 governments is incredibly tough.

A new dataset provides data on local government corruption for 1,473 regions in 178 countries from 1995-2022: A new paper on local government corruption was just published in @scientificdata by @LamarCrombach and Jeroen Smits:

Most countries don't provide granular data on economic activity.

Luckily, there are a range of geospatial datasets we can use as proxies.

Here's a list of the best (free) resources for estimating sub-national economic activity: I've previously discussed geospatial datasets that provide GDP estimates at a granular level.

Instead, in this post I'm providing proxies for economic activity.

If you're interested in the granular GDP estimates you can check this post out:

https://x.com/yohaniddawela/status/1805565620928356365

Inequality isn't just about income and wealth.

It's also related to environmental factors.

Here's how poorer people are more impacted by heat-related deaths: A new study from @natcities has found significant environmental inequalities.

Vulnerable residents in major European cities were found to receive inadequate ‘green cooling’ compared to wealthier residents.

One of the best use cases for geospatial data is examining environmental factors.

Here's a list of my favourite geospatial environmental datasets: 1. CO2 Emissions

@EU_ScienceHub's EDGAR database provides global geospatial data on:

• CO2 emissions
• GHG emissions
• pollutants data

Link: edgar.jrc.ec.europa.eu/emissions_data…

Mapping informal settlements is incredibly tough.

Now we can use (free) remote sensing data to map these places and deliver aid.

Here's the breakdown: One of the biggest issues of using remote sensing data to map informal settlements is its COST.

It can cost a lot of $$ to purchase very high-resolution images, that provide enough detail to identify informal settlements.

NASA has created a new foundation model for geospatial data.

It can provide granular weather forecasts, and air pollution forecasts.

Here's what you need to know about it: The foundation model—called Aurora—was announced this month.

But before going into the details of what it can do, what the heck is a foundation model?

The best way to accurately predict where deforestation will occur?

By identifying 'Ghost Roads'.

200 people spent 7,000 hours identifying them.

Here's what you need to know about it: A recent study in @nature has examined the identification of 'ghost roads' as a leading indicator for deforestation.

The research was conducted by

Want to know how to downscale nightlights data from ~500m resolution to ~30m resolution?

Here are the exact steps and codebase in R:

#rstats Instead of using AI, we can upscale nightlights data using:

• daytime satellite data (Landsat-8 or Sentinel-2); and
• @openstreetmap data on roads.

Free geospatial datasets from Meta.

PLUS: the history of R, how to download historical OpenStreetMap data without code, and more.

Here's everything I wrote about this week: The history of #rstats.

From the University of Auckland, to one of the most popular statistical programming languages:

https://x.com/yohaniddawela/status/1787440188358017420

Meta is known for Facebook, WhatsApp and Instagram.

But did you know they provide a range of free geospatial datasets for researchers?

These include granular measures of household wealth, population, and network access.

Here's what you need to know about it: 1. Meta provides granular estimates of household wealth for low and middle income countries.

Read more about it here:

https://x.com/yohaniddawela/status/1786359573139329406

Meta is using Facebook and mobile phone data to produce super-granular household wealth estimates.

Here’s what you need to know about the Relative Wealth Index: The Relative Wealth Index (RWI) is a geospatial measure, which shows wealth disparities within countries.

Main features:

• It's open source
• It measures 'asset-based' wealth
• It covers 135 countries in the world
• Data is provided at 2.4km resolution

Did you know we can access 𝙙𝙖𝙞𝙡𝙮 geospatial climate forecasts up until 1 Jan 2100?

It covers forecasts for:
• temperatures
• precipitation
• snowfall

Plus it's completely free.

Here's what you need to know about it: Daily geospatial climate forecasts are part of the CMIP initiative.

CMIP is a framework for research centres across the world to develop standardised climate forecasts.

Measuring agricultural GDP at a pixel level is notoriously challenging.

It requires precise information on crop type and crop yield.

Now, @esa has launched a (free) dataset that provides this information.

Here's what you need to know about it: @esa Last year ESA's WorldCereal project launched a dataset that provides data:

• at 10m resolution
• on farmland
• on seasonal maps of maize and cereals
• on where irrigation is used during different seasons
• on annual maps for where crops are grown temporarily

Let's unpack:

The most efficient way to process nightlights data?

Using Google Earth Engine.

But using it with R can be tough.

So here's a detailed guide on how to do it:

#rstats By the end of this post, you'll know the steps required to create the following graph of monthly luminosity data in R using Google Earth Engine.

GEE is a fantastic resource—the computations are done on the cloud, so you won't be limited by your computer's hardware.

High-resolution satellite images can be insanely expensive to buy.

So here's a list of free datasets you can access.

These datasets can be used to build foundation models, super-resolution models, or for segmentation. The most commonly used free multi-spectral satellite images are from:

• Sentinel 2
• Landsat-8

However, Sentinel 2 has ~10m resolution (for RGB), while Landsat-8 is ~30m (for RGB).

But what free high-resolution datasets exist to train foundation models on?

My recommended R packages for geospatial analyses.

PLUS: how to get €50k (free) to launch a geospatial business, recommended geospatial courses, top websites for free geospatial data, and more.

Here's everything I wrote about this week: An introduction to U-Nets: one of the most important models in the geospatial space

https://x.com/yohaniddawela/status/1775128062234886649

Geospatial data science is all about having access to good data.

Here are a list of my favourite free geospatial resources: 1. Aiddata

Aiddata has an excellent list of geospatial datasets.

It's all available as a CSV.

So you don't even need to know GIS to access their data.

link: geo.aiddata.org