Applications of remote sensing in sugarcane agriculture at Umfolozi, South Africa.
The aim of this study was to evaluate potential applications of remote sensing technology in sugarcane agriculture, using the Umfolozi Mill Supply Area as a case study. Several objectives included the evaluation of remotely sensed satellite information for the following applications: mapping of sugarcane areas, identifying sugarcane characteristics including phenology, cultivar and yield, monitoring the sugarcane inventory throughout the milling season and yield prediction. Four Landsat 7 ETM+ (Enhanced Thematic Mapper Plus) images were obtained for the 2001-2002 season. Mapping of sugarcane areas was conducted by .means of unsupervised hierarchical classifications, on three relatively cloud free, Tasseled Cap transformed images. The Brightness, Greenness and Wetness bands for each Tasseled Cap transformed image were combined into a single image for this classification. The investigation into relationships between satellite spectral reflectances and phenology, cultivar and yield involved the cosine of the solar zenith angle (COST) method for atmospheric correction of all four Landsat 7 ETM+ images. Detailed agronomic records and field boundary information, for a selection of sugarcane fields, were used to extract the at-satellite reflectances on a pixel basis . These values were stored in a relational database for analysis. Monitoring of the sugarcane inventory throughout the milling season was conducted by means of unsupervised classifications on the Brightness, Greenness and Wetness bands for each of the four time-step Tasseled Cap transformed images. Accurate field boundary information for all sugarcane fields was used to mask out non-sugarcane areas. The remaining sugarcane areas in each time-step image were then classified by means of unsupervised classification techniques to ascertain the relative proportions of the different land covers, namely: harvested immature and mature sugarcane by visual interpretation of the classification results. The yield forecasting approach utilized a time-step approach in which Vegetation Indices (VIs) were accumulated over different periods or time frames and compared with annual production. VIs were derived from both the National Oceanic and Atmospheric Administration (NOAA) and Landsat 7 ETM+ sensors. Different periods or times were used for each sensor. The results for the mapping of sugarcane areas showed that the mapping accuracies for the large scale grower fields was higher than for the small-scale growers. In both instances, the level of accuracy was below that of the recommended sugar industry mapping standard, namely 1% of the true area. Despite the low mapping accuracies, much benefit could be realized from the map product in terms of identifying new areas of sugarcane expansion. These would require detailed accurate mapping. The results for monitoring of the sugarcane inventory throughout showed that remote sensing, in conjunction with detailed field information, was able to accurately measure the areas harvested in each time-step image. These results may have highly beneficial applications in sugarcane supply management and monitoring. The results for time-step approach to yield forecasting yielded poor results in general. The Landsat derived VIs showed limited potential; however, the data were only available for one season, making it difficult to quantify the impact of climatic conditions on these results. All results for the time-step approach using NOAA data yielded negative results. The results for the investigation into relationships between satellite spectral reflectances and phenology, cultivar and yield showed that that different phenological stages of sugarcane growth were identifiable from Landsat 7 ETM+ at-satellite reflectances. The sugarcane yields and cultivar types were not correlated with the at-satellite reflectances. These results combined with the sugarcane area monitoring may provide valuable information in the management and monitoring of sugarcane supply.