High-throughput monitoring of crop and soil traits using airborne hyperspectral imagery

Sheng Wang, University of Illinois at Urbana-Champaign

Usage Details

Kaiyu Guan, Bin Peng, Chongya Jiang, Wang Zhou, Yi Yang, Yizhi Huang, Jingwen Zhang, Ziqi Qin, Ziyi Li, Zewei Ma, Kaiyuan Li, Chenhui Zhang, Sheng Wang, Qu Zhou, Xianrui Zhong, Han Qiu, Sichen Liu, Weiye Mei, Xiaocui Wu

Airborne  hyperspectral  techniques  can  efficiently  collect  high  spectral  (35  nanometer)  and  spatial  (0.5 meter)  resolution  of  surface  reflectance  with  designed  revisit  times,  which  makes  them  more  appealing  than ground  or  satellite  sensing.  With  such  merit s  of  high  spectral  and  spatial  resolutions,  airborne  hyperspectral data  have  high  capability  to  accurately  estimate  crop  and  soil  traits,  when  coupled  with  deep  learning  and radiative  transfer  modeling  approaches.  These  unprecedented  datasets  of  traits  can significantly  facilitate highthroughput  field  phenotyping  and  sustainable  agriculture.  To  make  the  above  work  possible,  highperformance  computing  (HPC)  is  highly  needed,  as  airborne  hyperspectral  data  is  large,  and  data  processing and  modeling are computationally  intensive. Any  scalable use of  airborne hyperspectral data is  impossible without HPC.

This  project  aims  to  apply  our  novel,  reliable,  and  comprehensive  framework,  which  integrates  machine learning,  radiative  transfer  modeling,  and  supercomputing,  to  predict  diverse  crop  and  soil  traits  accurately from  airborne  hyperspectral  imagery.  Our  developed  automated  processing  pipeline  requires  large computational  resources  to  complete  radiometric  calibration  of  imagery,  orthorectification,  atmospheric radiative  transfer  correction,  and  canopy  radiative  transfer  modeling. To perform these tasks, this proposal  requests computational  resources  to  facilitate  large-volume  airborne  hyperspectral  data processing,  machine  learning,  and  radiative  transfer  modeling.  Specifically,  we  aim  to  predict  49  crop and soil traits for UIUC’s  experimental fields in  Champaign County  from 2019  to  2021. The tangible outcomes  generated  by  work  described  in  this  proposal  will  permit  comprehensive  monitoring  of  crop growth  and  soil health  which ultimately promotes sustainable precision  agriculture.

Geometric  and  radiometric  processing  of  largevolume  airborne  hyperspectral  data  cubes,  and  using hyperspectral  data  to  quantify  crop  and  soil  traits  through  machine  learning  and  radiative  transfer modeling  is  computationally  demanding.  The  Blue  Waters  Supe rcomputer  facility  offers  an  ideal platform  for  our  large  processing  demands,  highfrequency  I/O,  and  output  postprocessing  and visualization.  Interactions  with  the  system  staff  and  scientists  in  NCSA  will  bring  strong  technical support  for our project.