Multilayered Ensemble Learning for short-term forecasting in agro-climatology

Abstract

Agriculture is one of the areas whose activities depend heavily on weather forecasts. Indeed, in order to optimize their production, farmers must be able to anticipate climate conditions favorable or not to their activities by deploying the appropriate action plans. For this purpose, they consult the data daily from various suppliers of weather forecasts. However, the reliability of the forecasts of each supplier is variable according to the period, the climate or the geographical area. Farmers, therefore, have to arbitrate between suppliers daily. This paper proposes a new set of learning architecture that significantly improves the accuracy of weather short-term forecasts for the next 1-12h in order to assist farmers in decision-making.