Steinbeis experts help improve the agro-ecosystem in Africa
Artificial intelligence (AI) is used today in many areas. The Horb-based Steinbeis Transfer Center for Industrial Digitization is now even involved in one area that is literally a down-to-earth field: agriculture. As part of a pioneering project called Agricultural and Rainforest Development in Africa, it has been providing expert support to international non-profit food and agricultural research organizations, with a focus on AI topics. Most of the research institutions taking part in the project are involved on a voluntary basis in cross-border environmental protection and nature conservation in Africa. This relies on partnerships with organizations based in the respective countries. As part of the project, the Steinbeis experts developed data models, AI algorithms, and analysis models to be used in vegetation management with the aim of regularly assessing the status of greenery in agricultural areas in Africa.
To conduct the project, the Steinbeis experts merged artificial intelligence, data analysis, and data from the agricultural internet of things (IoT) with the expertise of experienced agricultural scientists. This was in order to overcome difficulties in the agricultural sector with digital technology. Their work has resulted in a solution that facilitates quicker and more informed agricultural decisions. It has also spawned an opportunity to improve the entire agro-ecosystem in Africa.
The main objective of the project was to determine growth rates and crop changes in areas where precipitations fall under 80% of climatological normal values. The team also wanted to assess resulting changes in land use. To do this, the Steinbeis experts focused on data validation, since smart solutions require coherent data and data models. It was therefore important to harmonize data – i.e. spot inconsistent data, such as so-called outliers, which have to be cleaned up without being forced to determine clear rules for the process in advance. Due to other differences in factors such as time periods and data formats, it was also important to standardize data models and formats.
The developed solution identifies high-risk cycles or corridors most likely to have the greatest impact on vegetation. The scores and KPIs calculated by the solution provide effective support in prioritizing and determining possible areas in which to take action. Steinbeis project manager Hans-Dieter Wehle describes the development as a first and important step on the path to smart agriculture and an optimized food supply chain.
It all starts with the data
To assess the situation, use was made of different sources of public information: spatio-temporal data provided by NASA satellites and the European Center for Medium-Range Weather Forecasts (ECMWF). MODIS (Moderate Resolution Imaging Spectrometer) on board NASA’s Aqua satellite provided the NDVI (Normalized Difference Vegetation Index) data needed for calculations, underpinned every one or two days by Earth surface observations from NASA’s Terra satellite.
MODIS creates images of the Earth in 36 different spectral bands (wavelength intervals) offering spatial resolutions of 250, 500, or 1,000 meters. The images represent an area measuring 1,200 x 1,200 km in the form of 4,800 rows and columns in 16-bit signed integers. For each pixel, the best value is selected from a 16-day period in order to minimize errors resulting from cloud cover or the viewing angle, and in order to maximize NDVI – a standardized measurement of greenery that is calculated as a ratio, based on the difference between near-infrared reflectance (NIR) and red spectral bands and the sum of those bands. This makes it possible to plot vegetation, since plant leaves reflect almost all incidental near-infrared light but absorb red in chlorophyll. The ECMWF data includes forecasts and accumulations of events such as snowfall, wind, and solar radiation, which play an essential role when it comes to context-based analysis and visualization of the results.
The Steinbeis experts used a variety of AI tools and techniques for their analysis. For example, they employed data mining observation techniques such as association analysis to search for relationships between data in order to identify a rule of inference. They also used cluster analysis to create groups of data that had greater similarities than other groups of data.
The results produced by analyzing the data
Climate data were analyzed for the last ten years, showing a noticeable rise in temperatures as well as clusters of extreme weather and climate incidents. Such changes are not favorable for the continuous development of greenery, and in many cases they go hand in hand with the destruction of arable land, which in turn leads to logging of rainforest areas in order to open up new arable land.
Although only minimal differences can be perceived in actual land use year on year, the data relating to cultivated farming land does reflect year-on-year fluctuations in the proportion of farmed land in relation to overall areas of vegetation – with a steady rise in barren areas.
The described solution makes it possible for different stakeholders to automate data transfer between different systems, establishing a transparent, connected ecosystem that delivers benefit to up- and downstream organizations – such as
- Food and drinks manufacturers, which can introduce integrated supply chains based on improved harvesting schedules and more reliable volume forecasts
- Financial lenders to agriculture, which can monitor the performance of yields compared to potential
- Insurance agencies, which can use validated information to gauge risk more accurately and thus offer growers more logical premiums
- Governments, which can improve strategies for reducing dependence on food supplies by providing producers and authorities with access to shared instruments