BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//speaker//FLUVJK
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pydata-amsterdam2026-YVEPAV@pretalx.com
DTSTART;TZID=CET:20260911T141000
DTEND;TZID=CET:20260911T144000
DESCRIPTION:In the Netherlands\, groundwater level time-series stored acros
 s partially overlapping databases need to be harmonised before they can be
  used in hydrogeological models for studying and predicting Dutch hydrolog
 y. Manual harmonisation does not scale: the search space contains millions
  of time-series\, and a false match can propagate into model inputs. In th
 is setting\, missing a match is often less harmful than accepting the wron
 g one.\n\nThat makes this a certainty-first matching problem rather than a
  task of retrieving the closest match. Spatial proximity is not sufficient
  in dense monitoring networks\, and a single similarity score breaks down 
 under common artifacts such as vertical shifts and gaps introduced or remo
 ved during quality control. More generic approaches\, including spectral r
 epresentations and learned embeddings\, can also fail when they compress a
 way details that remain critical in the time-series.\n\nWe compare complem
 entary matching scores based on metadata fingerprinting\, Dynamic Time War
 ping\, and point-wise value comparison. We found that no method dominates 
 across archetypes: each succeeds on some data pathologies and fails on oth
 ers. We show how these  failure modes can in turn be used to design a more
  reliable matching workflow.\n\nAttendees working on time-series matching\
 , record linkage\, or other large-scale data harmonisation problems applie
 d to real-world data shaped by natural variability\, human error\, and imp
 erfect observation histories\, will leave with a practical framework for d
 eciding when to trust different matching scores and how to combine them in
 to an interpretable\, scalable workflow that favors certainty over forced 
 decisions.
DTSTAMP:20260710T150510Z
LOCATION:Unconference
SUMMARY:When one score is not enough: matching real-world groundwater time 
 series at scale - Luisa Orozco
URL:https://pretalx.com/pydata-amsterdam2026/talk/YVEPAV/
END:VEVENT
END:VCALENDAR