Enter Note Done
Go to previous page in this tab
Session
  • Presentation | GC43R: Satellite Solutions: Agricultural Monitoring Through Remote Sensing I Poster
  • Poster
  • Share Icon
    Bookmark Icon
  • GC43R-1030: In-Season Crop Growth Forecasting Using Satellite Imagery and Context-Aware Deep Learning
    • Schedule
    Notes
  • Board 1030‚ Hall EFG (Poster Hall)
    NOLA CC
    Set Timezone

Generic 'disconnected' Message
Author(s):
Ramana Heggadal Math, Texas A&M AgriLife Research (First Author, Presenting Author)
Ashutosh Tiwari, Texas A&M AgriLife Research and Extension Center at Corpus Christi
Lei Zhao, Texas A&M AgriLife Research and Extension Center at Corpus Christi
Benjamin Ghansah, Texas A&M University Corpus Christi
Scott Jose L Landivar, Texas A&M AgriLife Research Center, Corpus Christi
Juan Landivar-Bowles, Texas A&M AgriLife Research
Mahendra Bhandari, Texas A&M AgriLife, Corpus Christi


Despite advances in precision agriculture, many farmers still lack early and reliable crop growth forecasts. Traditional simulation models require extensive field data and often struggle with noisy or missing observations. Yet, accurate monitoring is essential for making timely decisions under uncertain weather and farming conditions. We propose a scalable, data-driven framework that uses satellite-derived NDVI (Normalized Difference Vegetation Index) as a measure of crop health. Trained on multi-year PlanetScope imagery from cotton fields in Texas, our model mimics how humans predict future growth from incomplete patterns. This allows early forecasting and rapid updates as new data comes in. At its core is a sequence-to-sequence model with autoregressive teacher forcing. It predicts NDVI step by step, using each output as input for the next. To improve accuracy across time and space, the model uses a normalized timeline and group-based crop behavior statistics for added context. A key innovation is its ability to fine-tune in real time (5 day intervals). This helps align NDVI patterns across years and locations, a long-standing challenge. We also add randomness during training to improve stability across growth stages. This framework enables adaptive crop monitoring, supporting smarter decisions in farm management, logistics, yield estimation, and planning.



Scientific Discipline
Suggested Itineraries
Neighborhood
Type
Main Session
Discussion