Spying on the winter wheat crop - generating objective planted area and crop production estimates using MODIS imagery

Spying on the winter wheat crop - generating objective planted area and crop production estimates using MODIS imagery

With world commodity markets becoming more competitive and the deregulation of the wheat industry in
Australia during the nineties, advanced knowledge of likely production and its geographical distribution has
become highly sought-after information. During the past 5 years, the Queensland Department of Primary
Industries & Fisheries (DPI&F) has generated shire/state and national yield (t/ha) forecasts for wheat and
sorghum crops on a monthly basis throughout the crop-growing season with appreciable success. However, to
achieve an accurate near real-time production forecast, a real-time estimate of the crop area planted is required.
Generating objective estimates of planted area will allow near real-time crop production estimates, which can
then be used in updating supply chain information at the regional, state and national levels. While there are
alternative methods (e.g. subjective opinions, surveys, censuses, etc.) to derive the required information, the use
of remote sensing (RS) offers more objectivity, timeliness, repeatability and accuracy. Furthermore, the use of
multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) imagery (spanning an entire
cropping season) is novel, and has been rarely used in determining crop area planted in targeted agricultural
systems. In this paper, we provided a brief background of regional commodity forecasting in Queensland, and
have reported some preliminary results on the use of digital image processing techniques to determine crop area
planted. More specifically, different multivariate approaches to analysing remote sensing data [i.e. Harmonic
Analysis of Time Series (HANTS) and Principal Component Analysis (PCA)] were compared in determining
winter crop area planted from MODIS imagery for a specific case study in the Darling Downs region,
Queensland. The methodology was validated for the 2003 and 2004 seasons at a shire level by contrasting
aggregated shire total area planted with surveyed ABARE estimates. Finally, the ability of these methods to
discriminate area planted for wheat, barley and chickpea at the shire level was determined. Preliminary results
showed a significant potential to capture total crop area planted at a regional level and a good overall capability
(>95% correct classification) in discriminating between these winter crops.