Automated Land Use and Land Cover (LULC) Classification for Hydrological Modelling and Physically-Based Inflow Forecasting

The problem considered in this work is how to produce highly accurate and consistent land-use/land-cover (LULC) maps significantly faster than current semi?automated methods for use by Manitoba Hydro. The goal is to improve the ability to produce maps quickly and efficiently as priority needs arise. This project will use an approach for automated LULC mapping from satellite images using deep learning methods pioneered by the applicants. By classifying each pixel in a satellite image into LULC categories using neural networks, rapid and accurate LULC maps can be successfully produced. These LULC maps can then be included in improving hydrological modelling and inflow forecasting as an additive layer to improve overall modelling processes. Secondly, this research will also develop a solution for higher resolution satellite data. Lastly, the final objective incorporates seasonal water levels into the LULC products that can contribute to the mapping of hydrological connectivity and disconnectivity.

Faculty Supervisor:

Christopher D Storie;Christopher Henry;Joni Storie

Student:

Rostyslav-Mykola Tsenov

Partner:

Manitoba Hydro

Discipline:

Computer science

Sector:

Energy

University:

University of Winnipeg