Automatic calibration of groundwater models is a challenging task due to the complexity of the codes, parameters and workflow involved. Calibration of transient state models is even more complex due to changes in sensibility on the different stages of the aquifer requirement such as pumping and recovery.
Python has awesome packages for machine learning that can be coupled to groundwater models and perform automatic calibration of hydraulic parameters. This webinar covers the procedure to implement a neural network based on a set of parameters set with corresponding head values; the study case is on a transient pumping test model with an observation point located around 10 meters away from the pump well. The analysis predicts the calibration parameters from the spatially interpolated observed head values.
Content
Review of the sensibility analysis
Parameter / head set generation overview
Scaling and transformation for neural networks
Split data over the recovery phase of pumping
Construct MLP regressor
Evaluate the accuracy of the MLP regressor
Predict values and analyze the hydraulic response
Instructor
Saul Montoya M.Sc
Hydrogeologist - Numerical Modeler
Mr. Montoya is a Civil Engineer graduated from the Catholic University in Lima with postgraduate studies in Management and Engineering of Water Resources (WAREM Program) from Stuttgart University – Germany with mention in Groundwater Engineering and Hydroinformatics. Mr Montoya has a strong analytical capacity for the interpretation, conceptualization and modeling of the surface and underground water cycle and their interaction. He is in charge of numerical modeling for contaminant transport and remediation systems of contaminated sites. Inside his hydrological and hydrogeological investigations Mr. Montoya has developed an holistic comprehension of the water cycle, understanding and quantifying the main hydrological dynamic process of precipitation, runoff, evaporation and recharge to the groundwater system.
Event date
Friday, Feb 9 2024 6:00 p.m. Amsterdam Time
