Artificial Intelligence on Water Resources

Online
The registration deadline has expired.

IHE Delft Alumni & Partners Online Seminars are designed as a means of providing lifelong learning to the Institute Alumni community and Partners. Since the topics will be of interest to a broader audience, they are open to all interested persons. The Online Seminars are like interactive lectures transmitted over the web, with a duration of one hour including time for discussion. The seminar is organized in collaboration with IHE Delft partner TheWaterChannel.

In November we have the pleasure to organize two seminars on 'Artificial Intelligence on Water Resources. Hydroinformatics applications in Machine Learning, trends and alumni experiences' to be held on 6 and 13 November. Please find below the information the first.

Description

In recent years the increase of machine learning applications to water resources have allowed us to propose new solutions to complex problems. Alumni from the Hydroinformatics program have explored new areas that in many cases have led to implementations at different places in the world, and have shown to be able to compete with ongoing traditional solutions. For this seminar we will make an overview of some of the most recent ideas of applications of machine learning in Hydroinformatics. These presentations will be divided into two sessions that will cover forecasting problems. An introduction in both sessions to a variety of machine learning basic concepts will be given to introduce the topics, limitations and a friendly way to see the theory. 

Introduction 1st seminar | 6 November 2020, 13:00 to 14:00 CET
How to understand the problem of forecasting and how machine learning has been applied to increase accuracy, understand uncertainty and provide extended lead times.

Best practices

By alumna Teo Kai Wen. To cater to the imminent threats of changing weather patterns and increase urbanisation, Singapore’s national water agency, PUB, aims to improve its flood forecasting system. In light of the recent advancement in deep learning, many studies have cited promising results of deep learning model (ConvLSTM) on radar rainfall nowcasting. Thus, KaiWen’s research work is dedicated to pilot study of the application of deep learning models (ConvLSTM and its variants) in the area of radar rainfall nowcasting and flood forecasting, for the case of Singapore. The deep learning model adopts a hybrid combination of convolutional neural network which is commonly used in computer vision tasks such as face recognition and image classification; and the recurrent neural network which is typically used in language translation. The deep learning models were trained using past observed radar images from Meteorological Service Singapore to forecast future water levels at 5 locations in Bedok Catchment of Singapore. The results of this research would be useful for tropical countries with dynamic weather patterns and small flashy catchments like Singapore. 

By alumnus Jose Valles. Flood Early Warning System (EWS) has been implemented in the Grande de San Miguel catchment in El Salvador with local inhabitants, governmental sector, and the National Hydro-meteorological Service since 1998. Nevertheless, a rainfall-runoff forecast model has not been implemented in this catchment to predict and anticipate to flood conditions. In Jose Valles’s research work at the Ministry of Environment and Natural Resources of El Salvador (MARN), he proposed the implementation of Multilayer Perceptron Artificial Neural Network (MLP-ANN) model as operational flood forecast model in Grande de San Miguel catchment to address the short-term flood forecast. The MLP-ANN is defined by Solomatine and Wagener (2011) as a device that consists of several layers of mutually interconnected neurons, which transform the inputs using a multiparameter nonlinear transformation, so the resulting model is capable of approximate complex input-output relationships. The MLP-ANN models were trained and validated using past observed rainfall and discharge values from MARN to predict future flood conditions with a forecast lead time of 12 hours. The trained MLP-ANN forecast model uses real time observations from the catchment and provides a real time flood guidance for stakeholders and decision makers in El Salvador. The forecast model output can be seen in the following web address: http://srt.snet.gob.sv/apps/public/pronosticorgsm 


23 views Visits

Like

Webinar
Released on: Alumni in Water, Energy and Climate
Please log in to see or add a comment

Suggested events

Sat.,

12

Dec

Holland Alumni Lecture (Composing Bachelor and Master Thesis into Journal Article)