Evaluation of an intelligent agro-climate decision support tool for small-scale farmers

Abstract

Food security in Africa and the rest of the globe has come under tremendous threat. This means that agriculture, being the main driving force behind many economies is under threat. Seventy percent of the food produced in Sub-Saharan Africa comes from the rainfed small-scale agriculture. This agricultural sector happens to be the most devastated by any disasters experienced in the agricultural sector such as floods, drought, and other extreme climatic conditions. Evidence show that Africa has accounted for the most global droughts. Droughts have also become increasingly common in recent years. For example, out of 106 global droughts observed between 2010 and 2019, eight occurred in Africa in 2019. Catastrophically, these droughts affected 66 countries and impacted 690.2 million people, 9.3 million of whom were in Africa. The dire case for Africa can be attributed to lack of timeous and relevant early warning systems which is a result of limited resources in the small-scale agricultural societies. Indigenous Knowledge is currently still the trusted prediction tool being used by small-scale farmers for their day-to-day operations and strategic agricultural decision support. Climate change and global warming have however rendered this knowledge unreliable and unpredictable. There is a plethora of scientific models, decision support tools and predictive methods that have been researched, all in the quest to find the solutions to the declining food production in the agricultural sector. Furthermore, much investment has been put behind research in the agricultural sector all over the world. The relationship between the Indigenous Knowledge (IK) and scientific methods has been substantially debated, but researchers have concluded that science and indigenous knowledge complement each other rather than compete. Limitations to both IK and modern scientific tools has led to this study which aimed to investigate an integration of Indigenous knowledge, mobile phone, and smart sensor technology with intelligence, in one system that can play a role in assistive decision support tool. In order to enhance and extend IK, the concept of installing smart sensors in the field with the view to measure humidity, Phosphorous, Potassium and Nitrogen concentration, and for detection of possible infestation of crops. For the theoretical framework, this study adopted and adapted the ITIKI framework as a foundation. ITIKI is a drought early warning system developed for small-scale farmers in Sub-Saharan Africa. The system uses input from observed indigenous knowledge indicators according to various farming activities and uses the fuzzy inference system for an assumption output. Weather forecast and historic climate data forms part of the system input. The necessary and relevant messages are conveyed to the small-scale farmers through mobile phone. The forecast provided by the ITIKI framework gives valuable information in making decisions on whether to plant, when to plant and even how and what to plant. To enhance the intelligence of ITIKI, this study investigated the application of machine learning algorithms to assist small-scale farmers in a crop selection process at the planting stage. The study explored machine learning classification models to find the best possible model which resulted in an Agro-climate Decision Support tool. This tool was developed to be able to assist farmers in making a crop selection decision prior to planting. This tool was developed using the blueprint for integrating the IK and scientific model. Machine learning (ML) was used to determine the model to be integrated with the IK for developing the intelligence of the system. Climatic data, together with edaphic data ran through different ML algorithms to determine the best algorithm. The best model was used to select the best crop to plant based on the edaphic and climatic data. The models were tested and evaluated using the Jupyter notebook’s different metrices. The system compares the results of the ML with the observed IK to determine the crop to be planted. In a case where the results prove contrary, a decision requires a crop scientist intervention. The evaluations proved that the latter was possible because the climatic and edaphic conditions could affect the accuracy of the ML model, and the confusion matrix results suggested different crops at times. In conclusion, the study developed a cropping decision support tool for integration of Indigenous knowledge, sensor technology data, machine learning and mobile phone technology in a form of the Intelligent Agro-climate decision support system to be used by small-scale farmers.