Western Canada spring wheat yield estimation using weekly AVHRR NDVI composite data
Loading...
Date
1995
Authors
Hochheim, Klaus P.,
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis evaluates tbe use of weekly NOAA (National Oceanic and Atmospheric Administration) AVHRR (Advanced Very High Resolution Radiometer) maximum value NDVI (Normalized Difference Vegetation Index) composite data to predict spring wheat yields for Western Canada for 1991 and 1992. Since 1987 the Canada Centre for Remote Sensing (CCRS) in cooperation with the Manitoba Remote Sensing Centre (MRSC) has produced maximum value NDVI composites. Based on four years of NDVI data (1987-1990) a method was developed to generate potential yield estimates of spring wheat well before harvest on a broad regional scale over Crop Reporting Units (CRUs) that are both physically and spectrally distinct. NDVI spring wheat yield estimates for 1991 were found to be within 1-5% of official yield estimates for Western Canada. Yield estimates for 1992 were approximately 10% below official yield estimates. Underestimation of wheat yields in 1992, despite favorable growing conditions, were attributed to the possible effects of extensive cloud cover during the 1992 season and possibly due to the attenuation of NDVIs resulting from residual stratospheric aerosols of the Mount Pinatubo eruption of July 1991. The underestimation of 1992 yields using AVHRR data illustrates the need for the implementation of atmospheric correction algorithms to improve the integrity of the weekly NDVI composite data. While evaluating the use of the NDVI composites for yield estimation a number of data quality issues arose. First, the current crop mask distibuted by CCRS was found to have serious deficiencies. A new crop mask was developed to include only intensely cropped areas resulting in significant changes in early season and late season NDVIs. Second, residual cloud contamination in weekly NDVI composite data was found to be significant. It is shown that cloud masking alone is insufficient to remove residual cloud contamination and that linear interpolation of data points is often required on a CRU level. Third, the prelaunch calibration coefficients used to generate the weekly NDVI composites were inappropriate especially for NOAA 11 AVHRR data (1989-92). New time-dependent piecewise linear coefficients recently adopted by CCRS were found to be much more appropriate. It is shown that the improved calibration can be implemented quickly and effectively with minimal disruption to operational users of the NDVI data.