| 
   
 
The ISF'06 ANNEX Neural Network Forecasting Competition
 
Prof Rob Wilby, Drs Bob Abrahart, 
Asaad Shamseldin, Linda See, Christian Dawson and Sven Crone would like to 
welcome you to the ISF06 ANNEXG forecasting competition . The intention of ISF06 
ANNEXG is to initiate a project to evaluate the effectiveness of artificial 
neural networks in rainfall-runoff modelling/flow forecasting.  
Our intention is to provide a 
standard rainfall-runoff data set (available below) that will be used by all 
participants to create a number of neural network models. We then hope to 
integrate the model results to examine the skill of individual members, as well 
as the ensemble forecast(s) for the catchment. 
All the data you send back to us 
will be anonymous. In other words, when the model statistics are calculated and 
published, there will be no indication as to who has produced the most accurate 
model (although parameters will be published so you should be able to work out 
which model(s) is your own!). 
Following on from the 2001 / 2002 
experiments (see below) a new data set is provided which is 
coded at a 6 hourly time interval. The data sets and rules can be downloaded 
directly from this page below. Model results will be evaluated using statistics 
calculated by the HydroTest web site at:
www.hydrotest.org.uk. 
  
Data Files and 'Rules'
The 'rules' are available as a Word 
document from
here
 . 
All the data are presented in Excel 
format and can be download as a single zipped file from
here
 .  
Alternatively, you can download the individual files below. 
The calibration data is split into two files;
Train1.xls
 and
Train2.xls
 . 
The test data is available as
Test.xls
 . 
A replication website is located at
http://www-staff.lboro.ac.uk/~cocwd/Annexg/annexg.htm  
  
Timing
Please submit an abstract to the 
ISF06 conference website by the deadline of  28 February 2006
(passed) including a 
reference to the ISF06 ANNEXG in the abstract. 
Please return all results to Christian Dawson by 28 April 2006
extended to: 30 November 2006! 
  
ANNEXG 2001 / 2002
The first ANNEXG experiments were 
undertaken during 2001 / 2002. This exercise involved the dissemination of a 
benchmark catchment data set to seventeen neurohydrologists world wide. Each was 
given the freedom to develop up to two ANN models for t+1 and t+3 days ahead 
forecasting in an unknown catchment. An additional motivation for this exercise 
was to investigate the potential of ensemble forecasting to improve forecast 
accuracy and, taking this work further, using ensembles to provide confidence in 
modelling performance. 
The results were presented at the 
8th BHS National Hydrology Symposium at the University of Birmingham, UK (8 - 11 
September 2002). You can download a copy of the paper from
here. 
You can access the original 
benchmark data set as a zip file
here. 
You can view the 'rules' that the 
participants had to follow from
here. 
  
Contact
If you have any comments / 
questions / queries please feel free to contact Christian Dawson at :
C.W.Dawson1@lboro.ac.uk 
 |