Data Fusion Of Remote Sensing Products And Machine Learning Feature Engineering Strategies For Near-Real-Time Runoff Forecasting

Hydropower Generation In Ecuador Is Vital For Its Economic Development And For Covering The Energy Deficit Of The Country. One Of The Most Important Plants (Hydropower Generation) Is The Minas-San Francisco (Msf). However, Hydropower Generation Is Not Optimal Due To The Lack Of Hydrological Forecasting Tools, I.E., Msf Operators Cannot Anticipate The Incoming Discharge And Thus Optimize The Energy Generation Schedule, Producing Significant Economic Losses. There Are No Operational Rain Gauge Networks In The Associated 4000 Km2 Basin, So The Only Data Source Are Satellite Precipitation; And While There Are Multiple Products Available, All Of Them Have Problems Because The Basin Goes From Sea Level To More Than 4000M. Here, We Propose To Develop An Hourly Runoff Forecasting System Fed By A Fusion Of Multiple Satellite Products Employing State-Of-The-Art Machine Learning (Ml) Techniques. Therefore, The Main Challenges Of The Project Are (I) The Spatial Exploitation Of Satellite Products Through Data Fusion, And (Ii) The Input Data Optimization Using Ml Feature Engineering Strategies; And Both Of Them For Improving The Foresting With Several Lead Times. Runoff Forecasts Will Provide Hydropower Operators With The Tools For Optimizing The Energy Generation Schedule And Planning Maintenance Activities In A Secure Way. This Will Be Possible Since The Project Contemplates An Operational Near-Real-Time Automatized Scheme Of Data Acquisition And Processing, Running Of The Forecasting Models, And Delivery Of Forecasts To Hydropower Operators For Its Immediate Use. The Participation Of The Operators Will Provide A Quality Feedback As To Tailor The Products To Their Day-To-Day Needs.