SCCS Colloquium - Sep 11, 2019
|Date:||September 11, 2019|
|Time:||15:00 - 16:00|
Vyshakh Unnikrishnan: Implementation of a Deep Learning Based Model for Rainfall-Runoff Modelling
This is a Master's thesis submission talk. Vyshakh is advised by Ivana Jovanovic.
This thesis explores the idea of using a data driven model to predict the runoff from rainfall and meteorological data using the Long Short Term Memory (LSTM) network, which is a variant of a recurrent neural network. An LSTM network with better temporal memory was used to account for the effects of long term dependencies in the data, for example the effect of snowfall on runoff. Data from different meteorological stations which measures precipitation, runoff, air temperature, air pressure, global radiation, solar radiation, and sunshine duration are used as input to the model. The model was trained on input data for the Regen region in Bavaria, from the past 15 years and then predictions were made on the test data available for the latest 2 years. An exhaustive hyper-parameter search was conducted to identify the best model settings to produce the best performing model. Using this approach we were able to achieve a Nash Sutcliffe efficiency (NSE) of 0.96 for the period of high runoff events in the test data which was unseen by the trained model. This indicates that the trained model was able to closely predict the high runoff values for the discharge station at Marienthal, Regensburg, Germany. Results also show that we could obtain a good performing model that was trained on data that excluded the runoff values of the interested station as input. It shows that the model could predict runoff values from other meteorological data alone.
Keywords: rainfall runoff modelling, lstm, rnn, deep learning
We are looking for a second talk! Please use the Colloquium registration form.