This dataset holds the high resolution (0.5 x 0.5 deg; 8 vertical levels) monthly means of 5 atmospheric variables (air temperature, pressure, water vapour pressure, vertical velocity and horizontal wind speed) over the Amazon Basin for the period 1972 to 2009 (version 1.0). This data is public and citable (DOI: 10.5285/2dfce039-cd71-43b3-bed4-98978e78f1bb).
It was constructed using the predictive capabilities of Time-Delayed Neural Networks (TDNN) method. Thirty years of monthly averages of current climate data (1971-2000) of the NCEP/NCAR reanalysis dataset were used to train the TDNNs, which were then validated on the next 10 years (2001-2010). Once validated, the downscaling model was fed with the higher resolution CRU TS3.1 data and SRTM-1km elevation data (thereby obtaining the higher resolution dataset).
No news update for this record
|Previously used record identifiers:||
Public data: access to these data is available to both registered and non-registered users.
Use of these data is covered by the following licence: http://creativecommons.org/licenses/by/3.0/. When using these data you must cite them correctly using the citation given on the CEDA Data Catalogue record.
Compressed data files were submitted to the BADC in August 2013. Data files were extracted and archived at the BADC. Compressed files were also renamed to follow the BADC file name convention. The metadata was submitted as a text file to the BADC and archived in the doc directory. A DOI was issued for version 1.0.
As described in the dataset description above and in the metadata file in the archive. Paper describing the dataset is being prepared for publication.
The files are ASCII however these are structured in the ArcINFO format with the file extension .asc (i.e. the files can be read immediately in ArcGIS. Please see details of the File Format and units under "Docs".
|Title||Time-Delayed Neural Networks (TDNN) deployed on Fundacion Entropika Computers|
|Abstract||This computation involved: Time-Delayed Neural Networks (TDNN) deployed on Fundacion Entropika Computers. Time Delayed Neural Networks are Multi-Layer Perceptrons that keep the previous states of the input layer in memory. Such networks have shown to be efficient in the prediction of time series, the hypothesis being that a climate variable depends or relates to the previous state of that same variable (i.e. that climate variables are temporally dependent). The exact architecture of the network (i.e. number of hidden layers, number of nodes in the hidden layers, adaptative learning function, transfer function and number of previous input states kept in memory) has to be arbitrarily determined by testing the performance of various structures.|