Mr Doug Hamilton1, Mr Nicholas Berryman1
1CSBP Fertilisers, Perth, Australia
Biography:
Doug joined the CSBP Fertilisers business in early 2007 as a trainee sales manager before moving across to the AgTech services team. He has been a key member in developing and delivering several innovative projects, including DecipherAg and CSBP Detect. Doug has a degree in Agricultural Science from the University of Western Australia. While working full-time at CSBP, he completed a Post Graduate Diploma in Geographic Information Science through Curtin University. Before employment at CSBP he held a role for 2 years as a Farming Systems Development Office with the Department of Primary Industries and Regional Development.
Abstract:
This project used publicly available satellite imagery datasets from the MODIS sensors (high temporal resolution, low spatial resolution) and Sentinel-2 sensors (moderate temporal resolution, moderate spatial resolution). A common smoothing method was used to fuse the MODIS and Sentinel-2 images based on relative differences to create a daily revisit moderate spatial resolution image timeseries dataset. On a per pixel basis, a double logistical curve was fitted to describe both the growth and senescence phase of the crop. A machine learning model was trained to calculate the baseline level of reflectance per pixel. The baseline and summer weeds were removed from the crop growth curve to determine an emergence date for each using the curve’s first derivative. From a dataset of 154 point specific emergence date observations, the image fusion product was able to estimate the emergence date with a mean absolute error of 6 days (±2 weeks at a 97% rate). Scalable and accurate emergence dates can help build more accurate models to predict crop phenology and yield potential as well as inform growers and agronomists of delayed or failed crop emergence.