BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.earthmonitor.org//global-workshop-2026//NEFXXX
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-TW9RF7@pretalx.earthmonitor.org
DTSTART;TZID=Europe/Amsterdam:20261007T150000
DTEND;TZID=Europe/Amsterdam:20261007T151500
DESCRIPTION:Foundation models for Earth systems have advanced rapidly for w
 eather and climate prediction\, but remain largely confined to physical va
 riables\, omitting the human systems that drive emissions\, shape land use
 \, build infrastructure\, and mediate vulnerability. We argue that this ga
 p is fundamentally a data problem: the information exists but is fragmente
 d across incompatible grids\, projections\, temporal frequencies\, and for
 mats. We present two complementary contributions that address this challen
 ge.\nFirst\, WorldTensor is a harmonised global dataset that aligns over 7
 50 environmental and socioeconomic variable families onto a common 0.25° 
 latitude–longitude grid and annual temporal framework. It integrates cli
 mate\,  emissions\, land use\, satellite vegetation indices\, gridded popu
 lation and GDP products\, power plant registries\, and natural hazard and 
 conflict catalogues into a single ML-ready NetCDF corpus. Constructing Wor
 ldTensor required solving nontrivial harmonisation problems including regr
 idding across heterogeneous native resolutions\, rasterising point and vec
 tor datasets into spatially meaningful fields\, and reconciling temporal c
 overages spanning daily observations to sparse multiyear socioeconomic sna
 pshots. The dataset and processing code will be released under open licens
 es.\nSecond\, TerraNova is a foundation model designed to learn from World
 Tensor's multimodal structure. It combines coordinate-based spatial encodi
 ng\, learned country-level embeddings\, Fourier temporal encoding\, and a 
 hypernetwork decoder to jointly predict climate\, land surface\, socioecon
 omic\, and infrastructure variables in a unified multi-task framework. Ear
 ly results demonstrate successful learning across multiple heterogeneous E
 arth system tasks simultaneously\, validating that foundation models can l
 earn shared representations across the coupled human–Earth system.\nToge
 ther\, WorldTensor and TerraNova provide an open\, end-to-end pipeline fro
 m harmonised planetary data to multimodal foundation model training\, supp
 orting applications in climate impact assessment\, cross-domain pattern di
 scovery\, and evidence-based environmental policy.
DTSTAMP:20260624T070956Z
LOCATION:Rooms 12+14
SUMMARY:WorldTensor and TerraNova: Open Data and Foundation Models for the 
 Coupled Human–Earth System - Carlos Rodriguez-Pardo
URL:https://pretalx.earthmonitor.org/global-workshop-2026/talk/TW9RF7/
END:VEVENT
END:VCALENDAR