Water and food security are at risk in many places in the world: now and most likely even more in the future, having large economic and humanitarian consequences. Risk managers and decision-makers, such as water management authorities and humanitarian-aid agencies/NGOs, can prevent harmful consequences more efficiently if information is available on-time on (1) the impact on the system, economy or society, and also (2) the probabilities for a failure in the system. EO information has proven to be extremely useful for (1). For looking into the future, considering the uncertainties, novel machine learning techniques are becoming available.
The proposed development is incorporated into an existing solution for providing Drought and Early Warning Systems (DEWS), called InfoSequia. InfoSequia is a modular and flexible toolbox for the operational assessment of drought patterns and drought severity. Currently, the InfoSequia toolbox provides a comprehensive picture of current drought status, based mainly on EO data, through its InfoSequia-MONITOR module. The proposed additional module, called InfoSequia-4CAST, is a major extension of current InfoSequia capabilities, responding to needs that have been assessed in several previous experiences.
InfoSequia-4CAST provides the user with timely, future outlooks of drought impacts on crop yield and water supply. These forecasts are provided on the seasonal scale, i.e. 3-6 months ahead. Seasonal outlooks are computed by a novel state-of-the-art Machine Learning technique. This technique has already been tested for applications related to crop production forecasting and agricultural drought risk financing. The FFTrees algorithm uses predictor datasets (in this case, a range of climate variability indices alongside other climatic and vegetative indices) to generate FFTs predicting a binary outcome – crop yields or water supply-demand balance above or below a given threshold (failure: yes/no).
The activity includes intensive collaboration with stakeholders in Spain, Colombia and Mozambique, in order to establish user requirements, inform system design, and achieve pilot implementation of the system in the second project year. Generic machine learning procedures for training the required FFTs will be developed, and configured for these pilot areas. An intuitive user interface is developed for disseminating the output information to the end users. In addition to development of the forecasting functionality, InfoSequia-MONITOR will be upgraded by integrating state-of-the art ESA satellite data and creating multi-sensor blended drought indices.
