BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.earthmonitor.org//global-workshop-2026//XR9NXF
BEGIN:VTIMEZONE
TZID:Europe/Madrid
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-E9DW8S@pretalx.earthmonitor.org
DTSTART;TZID=Europe/Madrid:20261007T140000
DTEND;TZID=Europe/Madrid:20261007T141500
DESCRIPTION:Forests\, as a land use\, are central to climate mitigation\, b
 iodiversity conservation\, and international policy reporting\, including 
 the Sustainable Development Goals\, the FAO Global Forest Resources Assess
 ment (FRA)\, and the EU Regulation on Deforestation-free Products (EUDR). 
 Yet\, global monitoring based on Earth Observation (EO) primarily captures
  land cover\, while policy frameworks often require land use distinctions.
 \nHere\, we present FLUM 2024 (Forest and other Land Use Map)\, the first 
 global\, wall-to-wall land use map at 10 m resolution aligned with FAO FRA
  definitions. FLUM distinguishes 18 land use classes—including naturally
  regenerating versus planted forests\, stocked versus temporarily unstocke
 d forests\, forest versus other tree-covered land uses—leveraging the la
 rgest expert-annotated dataset to date for this purpose. The deep learning
  model is trained on over 350\,000 reference samples collected through the
  FAO FRA Remote Sensing Survey\, annotated by more than 800 experts across
  120+ countries\, and on time series of satellite embeddings. FLUM achieve
 s an overall accuracy of 71.7 ± 0.7% across 18 classes\, increasing to 85
 .9 ± 0.6% for the four main FRA RSS classes and 93.4 ± 0.2% for forest/n
 on-forest discrimination. The product includes spatially explicit uncertai
 nty estimates that reflect annotation complexity.\nBeyond the base map\, w
 e elaborate on the critical importance of reference data quality based on 
 local expert knowledge\, currently the main constraint on the highest achi
 evable model performance\, and on synergetic integration of expert knowled
 ge and artificial intelligence. For example\, by identifying high-uncertai
 nty regions\, trained deep learning models enable targeted quality control
  and adaptive sampling in expert surveys\, improving both data reliability
  and model performance over time. We further show how FLUM can be deployed
  and fine-tuned on open platforms\, including FAO SEPAL and Google Earth E
 ngine\, supporting transparent\, scalable\, and country-driven monitoring.
 \nAs the first global EO product trained and evaluated directly on FRA-com
 pliant data\, FLUM represents a critical step toward operational\, policy-
 relevant land use monitoring.
DTSTAMP:20261023T012346Z
LOCATION:Room 18
SUMMARY:FLUM 2024: a global 10 m Forest and other Land Uses Map aligned wit
 h FAO definitions - Stassin Timothée
URL:https://pretalx.earthmonitor.org/global-workshop-2026/talk/E9DW8S/
END:VEVENT
END:VCALENDAR