Vegetation index based crop yield prediction model using convolution neural network - a case study of Kenya

Abstract

Predicting the crop yield is a vital food security strategy that can help a country take suitable measures and come up with policies that will help in crop production management. Such predictions will also support the farmers and industries involved in crop production for strategizing the logistics of their business or farming activities. Having sufficient production plans can improve food sufficiency and avoid situations of food emergencies. Climate change has had a huge impact on food production with variations in crop yields, creating uncertainty. Most of the studies on crop yield prediction have been done based solely on weather data, which is sometimes inaccurate due to scarcity of weather information especially in developing countries, where there is poor record keeping and insufficient resources to collect data. The use of vegetation indices derived from remote sensing data overcomes these challenges by providing data that is easily accessible and gives a comprehensive and multidimensional analysis. This study proposes a model that uses of vegetation index to predict crop yield using machine learning. Data from past crop yields in Kenya and vegetation greenness indices were the inputs applied to the algorithms. Various machine learning algorithms were applied and thereafter evaluated, so as to select the algorithm that gives better accuracy. To determine the accuracy level for the prediction model, the RMSE is calculated to compare actual and predicted values. The RMSE values obtained using convolution neural network for the three crops maize, rice and wheat were lower compared to those obtained using ridge regression, so it was selected as the optimal algorithm for the crop yield prediction model.