Emmanuel Muteba Ngonga
Remote Sensing in the form of multispectral imaging from spaceborne platforms has the potential to improve the pace and precision with which breeders determine agronomic attributes of new soybean varieties. This is because multispectral images and the spectral indexes derived from them can highlight crop phenotypic characteristics that are not easily discernable with the naked eye. In this study, yield potential and maturity periods of three soybean varieties, namely Dina, SC-Safari, and SC-Spike, that were grown on three commercial farms in Chongwe and Lusaka districts in Zambia were monitored using Remote Sensing techniques. In order to optimise this process, Landsat-8, PlanetScope, and Sentinel-2 satellites were combined into a virtual satellite constellation using data augmentation techniques. This constellation was used to track the chlorophyll levels of the soybean field's biomass with high spatial, spectral and temporal resolutions. A Random Forest algorithm was used to classify pixels when masking clouds, cloud shadows, and dense haze from the satellite images. Normalised Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI) formed the basis for determining the photosynthetically active biomass levels of the canopies of these cultivars. Average values of NDVI and EVI per field were plotted against time using Gaussian process modelling and cubic spline interpolation to produce their time-series profiles for the 2016/2017, 2017/2018 and 2018/2019 farming seasons. It was observed that there is a significant positive linear correlation between the maximum average EVI and NDVI values of these profiles to the yield of their respective soybean varieties. Further analysis was done by using logarithm and power functions to model this trend. It became apparent from this analysis that
linear and logarithm functions explain this trend more accurately than power functions. Different regression equations were observed per variety for each of the functions used in these analyses. Extrapolation of the trendlines of these equations was used to rank the varieties according to their maximum yield potential. SC-Spike was found to have the highest maximum yield potential in terms of metric tons per hectare after the crops were threshed. SC-Safari was the second-highest yielding, and Dina was the least yielding. The maturity period prediction model measured the time it took for the EVI and NDVI values at germination to recur during senescence. It placed SC-Safari as the earliest maturing variety, followed by SC-Spike. Dina showed the most extended maturity period.
to improve the temporal resolution. The poster then goes to show each of these satellites. It then explains that chlorophyll reflects most of the near infrared radiation incident on it, we use the amount of near infrared radiation reflected by a plant to measure above ground biomass levels. This is done using the NDVI (Normalized Difference Vegetation Index) and EVI (Enhanced Vegetation
Index). It dsplays these equations in a colour box.
In a new block entitled Image "Acquisition and field isolation" the poster shows the process from top of atmosphere image to bottom of atmosphere correction to NDVI computation. It then shows the process of field clipping using QGIS (Quantum Geographic Information System)
In a new block entitled "Image analysis" the poster shows how we use machine learning to classify pixels and do away with unwanted cloud and cloud shadow pixels. It then shows the mathematical modelling of EVI and NDVI time-series profiles using Cubic spline interpolations and Gaussian Process Modelling. The following section shows the graphs of maximum EVI and NDVI against yield.
The "Discussion" block discusses the main sources of error in our analysis. These were misclassifications by our cloud and shadow masking algorithm. This could be improved by training our algorithm with many more scenarios to allow it to classify surface reflectances more accurately The conclusion block concludes that There is a significat linear, logarithmic and power correlation of EVI and NDVI with the resulting yield of a soybean variety. It recommends that using satellites to observe soybean varieties is feasible and can be introduced into the breeding
process as the spatial resolution of earth observation satellites increases.