The modular design allows to work with the components independently to slice and plugin, update and improve and/or centralise and make generic as required.
AKILIMO hosts a modular data analytics system where different analytical procedures create components that could be improved, updated, expanded and/or made generic to serve different purposes. At the desired resolution, the prediction engine sources field observations and geospatial data and runs feature selection to identify variables defining crop growth in the target area. Target area delineation for research and/or scaling agronomic solutions is one of the components developed at an earlier stage of the research.
Using geospatial data, georeferenced observations and crop masks, target areas to conduct field research and/or to scale agronomic solutions are conducted with application of clustering algorithm and maximum-entropy approach, Maxent. GIS and remote sensing data for this and subsequent use are sourced using in-house developed and automated scripts.
Agronomic data gathered from field trials are processed using a linear mixed-effects model to extract maximally the structured variability in the data. Based on the nutrient supply, crop uptake and yield relationships and with the application reverse crop model, AKILIMO developed a fairly accurate system to estimate indigenous soil nutrient supply. Machine learning techniques are further used to train models to characterise soil nutrients as a response to soil properties which is later used within the Spatio-temporal yield gap analysis. An economic optimizer implemented in AKILIMO generates a profit surface and searches for the point of a high return on investment considering the fertiliser types, prices, crop yield estimates and required nutrient input.