The document introduces remote sensing and machine learning, emphasizing how satellites capture data related to Earth's surface and atmosphere through electromagnetic radiation. It discusses the importance of vegetation indices and spectral signatures, along with the challenges in machine learning training data for land cover classification. The Radiant Earth Foundation aims to provide open access to geospatial data and enhance global collaboration in tackling critical challenges through its platform.

1. An Intro to
Remote Sensing and
Machine Learning
HAMED ALEMOHAMMAD
LEAD GEOSPATIAL DATA SCIENTIST, RADIANT EARTH FOUNDATION
IMED, 2018, Vienna, Austria

2. Remote Sensing
Measurement of a quantity associated with an object by a
device not in direct contact with the object

3. Satellite Remote Sensing
Satellites carry instruments or
sensors which measure
electromagnetic radiation
coming from the earth-
atmosphere system.
3

4. Measuring Earth Surface and
Atmospheric Properties
 The intensity of reflected and emitted
radiation to space is influenced by the
surface and atmospheric conditions.
 Thus, satellite measurements contain
information about the surface and
atmospheric conditions.

5. Electromagnetic Radiation
Earth-Ocean-Land-Atmosphere System:
• Reflects solar radiation back to space
• Emits Infrared and Microwave
radiation to space

6. Interaction with Vegetation
Example: Healthy, green vegetation absorbs Blue and Red
wavelengths and reflects Green and Infrared.
Since we cannot see infrared radiation,
we see healthy vegetation as green.

7. Spectral Signatures in Imagery
Remotely sensed imagery acquires information in different wavelengths,
representing different parts of the Electromagnetic Spectrum.

8. Vegetation Indices

9. Solar Induced Fluorescence (SIF)
 Energy absorbed by plant through its chlorophyll
used for gross primary production (GPP)
lost as heat
re-emitted (SIF: byproduct)
 SIF responds to stressors (water, light, T).
Babani, F., et al. 2005

10. Except for Indonesia all tropical regions exhibit some
seasonal cycle due to light/water limitations

11. 11

12. Microwave | Thermal | Infrared |Visible|
Visible-NIR
Vegetation Index
Solar Induced
Fluorescence
IR Thermal
Radar
Backscatters
Passive-Microwave
Optical Depth NDVI, EVI, …
Photosynthesis
Canopy Temp.
& Evapotranspiration
Top-Canopy
Biomass
Canopy-column
Water Content

13. Mean Annual Soil Moisture

14. Mean Annual Precipitation

15. Nitrogen Dioxide from Sentinel-5P Satellite
credit: ESA

16. Nitrogen Dioxide Mapping
credit: Google

17. Satellite vs Sensors

19. Spatial Resolution
Actual size of each pixel of the image

20. Spatial Resolution vs Extent
Generally, the higher the
spatial resolution the less
area is covered by a
single image.

22. The European Copernicus Initiative

23. Atmospheric Transparency
Average cloudiness (2002 - 2015)
NASA Earth Observatory

25. Radar Measurements across
Pivotal Agricultural Systems
Google Earth

26. Credit: Jörgen Eriksson
Artificial Intelligence (AI) is about
bringing together computers and
humans in ways that enhance
human life.

27. Intelligence Augmentation (IA):
Computation and data used to create services that augment
human intelligence and creativity.
 Search engine
 Natural language translation
Intelligent Infrastructure (II):
A web of computation, data and physical entities that makes
human environments more supportive, interesting and safe.
 Starting to appear in domains such as transportation, medicine,
commerce and finance.
Credit: Michael Jordan, Professor at UC Berkeley

28. Computer
Computer
Data
Program
Output
Data
Program
Output
Traditional Programming
Machine Learning

30. source: COGNUB

31. Caution

32. Random Forest

33. Neural Networks

34. Deep Learning
SegNet architecture

35. Rural schools in Liberia
Courtesy of Zhuangfang Yi
Development Seed

36. credit: Space-Net

37. Road Tracer
Credit: MIT CSAIL

38. Crop Classification
Credit:
Rose M. Rustowicz

39. Training Data Challenges
 Capturing the wide range of possible outcomes both in
space and time;
 Accuracy;
 GeoDiversity
 Accessibility;
 Inter-Operability;
 ML-Readiness;

40. Open source machine learning commons
for Earth Observations.
Promoting creation of open libraries of labeled images and
algorithms to advance ML for global development, and
democratize ML applications for EO data.
Developers can join the collaborative initiative and
contribute their tools and knowledge on Github.
Imagery training data will be created as STAC compliant
and in COG format.

41. • The Problem: Need for an open,
dynamic, global, and comprehensive
LC map
Open Training Library for Land Cover
Classification:
• Using Deep Learning for labeling
imagery
• Crowdsourcing and citizen
science to verify / correct the
labels
Sponsored by:
Open Source
10 m resolution
Global
ML Centric
• Solution: Training labeled image library
for land cover classification

42. Radiant Earth Foundation:
Vision & Mission
 Open Geospatial Data for Positive Global Impact
 Connecting people globally to Earth Imagery, geospatial data, tools and
knowledge to meet the world’s most critical challenges

43. What we do
Provide Open Access to
Earth Imagery & Tools
Provide Education on
Geospatial Data & Tools
Provide Neutral Leadership
to Enhance Industry-Wide
Collaboration

44. Attributes of the Platform
AGILE
Experiment with data,
visualization, and collaborate in
a cross-domain multidisciplinary
ecosystem.
OPEN
Work with open
imagery, data sets and
technology standards.
NEUTRAL
Discover both government &
commercial imagery, and
collaborate with tech-and non-
technical users at the intersection
of global development & remote
sensing.
COLLABORATIVE
Learn and share ideas to
improve collaboration across
domains.
FEDERATED
Find and work with diverse
imagery data sets covering the
globe with a federated
catalogue.

45. Available Open Imagery
Datasource Temporal Coverage Temporal Revisit Spatial Resolution
Sentinel 2-A/B 2015 - present 5 days 10 m
Landsat 4/5/7/8 1982 - present 16 days 30 m
MODIS 2000 - present 8 day composite from daily 250 m
ISERV 2012 - 2015 Specific operation times 3.5 m

46. Platform Features
 Supporting any imagery type:
 Satellite
 Drone
 Airborne
 Uploading pipelines:
 Local
 Dropbox
 Amazon Web Services (AWS) S3 Bucket
 Planet API Connection
 Radiant Earth Foundation API

47. Platform Interfaces
app.radiant.earth doc.radiant.earth

48. Radiant APIs
Raster APIs
api.radiant.earth/platform
Imagery from Drones, Aerial,
Balloons, Satellites
Projects
Area of Interests
Annotations
Band math algos in Labs
Sharing via OGC (e.g. WMTS, etc)
Teams, Organizations
Data APIs
api.radiant.earth/{endpoint}
Weather forecasts / weather
Air Quality / air-quality
Population / population
Crop Suitability / crop-suitability
Satellites / satellites

49. Platform Demonstration

51. Get in touch Follow Us
740 15th St NW, Suite 900
Washington DC 20005
+ 1. 202.596.3603
hello@radiant.earth
www.radiant.earth | app.radiant.earth | help.radiant.earth | demos.radiant.earth
@OurRadiantEarth
https://www.facebook.com/OurRadiantEarth
Q & A