BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.earthmonitor.org//E8HUES
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-E8HUES@pretalx.earthmonitor.org
DTSTART;TZID=Europe/Amsterdam:20261007T182500
DTEND;TZID=Europe/Amsterdam:20261007T183000
DESCRIPTION:Large-scale and highly accurate wheat yield prediction is of gr
 eat importance for \nensuring food security\, supporting agricultural poli
 cymaking\, and guiding grain \nallocation. In recent years\, the rapid dev
 elopment of remote sensing technologies and \ndeep learning algorithms has
  provided powerful tools for large-scale crop yield \nprediction. However\
 , crop yield is jointly influenced by multiple environmental factors\, \ns
 uch as climate\, soil\, and topography. Existing studies often adopt simpl
 e feature \nconcatenation or fixed-weight fusion strategies\, lacking adap
 tive modeling of relative\nmodality importance\, which limits further impr
 ovement in prediction accuracy. To \naddress this issue\, this study propo
 ses a Transformer-based multi-modal adaptive Gated \nFusion model (TMMGF).
  The model employs Transformers to model dynamic time \nseries of remote s
 ensing spectral data and climate variables\, applies multilayer \nperceptr
 ons (MLP) to handle static environmental factors including soil and topogr
 aphy. \nMultiple modalities are then integrated through a gated fusion mec
 hanism to achieve\nadaptive weighted fusion. This study was conducted acro
 ss the conterminous United \nStates\, based on county-level winter wheat y
 ield records from 2008 to 2023. The \nTMMGF was systematically compared wi
 th an LSTM-based multimodal adaptive \nGated Fusion model (MMGF)\, Transfo
 rmer single-modal remote sensing model\, \nTransformer single-modal climat
 e model\, MLP single-modal soil model\, and MLP \nsingle-modal topography 
 model. The results show that TMMGF achieves the best \nperformance\, with 
 an average R² of 0.813\, RMSE of 0.571 t/ha\, and MAPE of 14.49% \nin 10-
 fold cross-validation\, significantly outperforming the baseline models. I
 n \nparticular\, compared with the LSTM-based multimodal model MMGF (R² =
  0.796\, \nRMSE = 0.598 t/ha\, MAPE = 15.11%)\, TMMGF shows clear advantag
 es in both \naccuracy and stability. This study demonstrates that a Transf
 ormer-based adaptive \nmultimodal fusion framework can effectively integra
 te heterogeneous data sources and \nprovides a promising technical pathway
  for high-accuracy large-scale wheat yield \nprediction.
DTSTAMP:20260624T083907Z
LOCATION:Aula Magna
SUMMARY:Transformer-Based Adaptive Multimodal Fusion Model for Remote  Sens
 ing Large-scale Winter Wheat Yield Prediction - Haoran Meng
URL:https://pretalx.earthmonitor.org/global-workshop-2026/talk/E8HUES/
END:VEVENT
END:VCALENDAR