BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.earthmonitor.org//global-workshop-2026//XXRLBZ
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-global-workshop-2026-H7T3W3@pretalx.earthmonitor.org
DTSTART;TZID=Europe/Amsterdam:20261008T184500
DTEND;TZID=Europe/Amsterdam:20261008T190000
DESCRIPTION:EO foundational models transform satellite images from a space-
 time grid of raw values into high-dimensional latent spaces called embeddi
 ngs. These embeddings encode relationships between pixel values and the co
 rresponding biophysical characteristics. Seasonal crop phenology (plant li
 fe cycle events)\, urban patterns\, and forest canopy texture are each rep
 resented in different combinations of embedding dimensions. Researchers us
 e these embeddings to train lightweight\, downstream models for specific t
 asks\, such as LULC (land use and land cover) classification\, biomass est
 imation\, or deforestation detection. These tasks require only a fraction 
 of the computational power and labelled data.\nThe trend is to build massi
 ve\, global-scale foundational EO models (such as TESSERA or AlphaEarth). 
 Nevertheless\, there is a strong case for developing dedicated regional fo
 undational models. Global foundation models inherently seek universal stat
 istical patterns\, pushing representations toward generalised\, highly sim
 plified categories. A regional foundational model avoids this homogenizati
 on by optimising representations for local landscapes. By pre-training a f
 oundation model on regional Earth observation data cubes\, the latent spac
 e represents those specific regions. This prevents the model\nfrom importi
 ng spatial biases learned from entirely different continents\, resulting i
 n much higher-quality embeddings for local downstream tasks.\nThis present
 ation will show how to build regional EO foundational models using an easy
 -to-use API associated with the R/Python package SITS. Users can merge var
 ious sources\, such as optical\, radar\, topographic\, and climate data. T
 he resulting EO embeddings will be better suited to regional applications 
 than global products.
DTSTAMP:20260624T065624Z
LOCATION:Room 18
SUMMARY:Regional Earth Observation Foundational Models: Improving  Represen
 tation of Domain-Specific Patterns - Gilberto Camara\, Felipe Carlos\, Rol
 f Simões\, Alexandre Assunção\, Felipe Souza
URL:https://pretalx.earthmonitor.org/global-workshop-2026/talk/H7T3W3/
END:VEVENT
END:VCALENDAR