Sugarcane yield prediction at farm scale using remote sensing and artificial neural network

Accurate predictions of regional yield for sugarcane agro-industries are of vital importance for formulating harvesting, milling and forward selling decisions, while at a block scale they provide growers with an understanding of both in-crop variability and total production (Robson et al. 2016). Lack of a reliable model and inadequate image processing equipment analysers lead the sugarcane agro-industries toward yield prediction through manual sampling (destructive), which is not only associated with sampling problems but also requires spending money and labour. Moreover, vegetation models require agronomic and meteorological data, which are not easily available due to their wide spatial distribution. Many plant indices based on canopy spectral reflectance have shown the ability to estimate crop physiological properties accurately; including plant, biomass and crop yield (Tucker et al. 1979; Zhao et al. 2003). Current rates of sugar production in Iran could not meet the demands, so that the unsatisfied demand shall be supplied by imports (ISFS 2017). By increasing the farm yield and achieving higher sugar purity, one can attenuate the need for imports. Numerous studies have been performed on finding relationships between remotely sensed vegetation indices and yield across sugarcane farms, most of which have been performed on regional scale. Bégué et al. (2010) used SPOT4 and SPOT5 images and assumed an exponential relationship between maximum NDVI and sugar yield, ending up with a R2 value of 0.78. Lofton et al. 2012 Predicted sugarcane yield based on NDVI and stipulated that the best time for estimating sugarcane yield is the interval between 601 and 750 Growing Degree-Days. Fernandes et al. (2017) used NDVI time series of MODIS images. The result showed that R2 was 0.63 for the Stacking method, anticipating the crop forecast by three months before the harvest. The present research is aimed at identifying the appropriate date for acquiring best satellite images to achieve maximum correlation between vegetation indices and end of season yield at farm scale. The research further looks for the optimal model and the best vegetation index for predicting irrigated sugarcane yield.