Geology Reference
In-Depth Information
Fig. 5.47 Overall Model
utility case study: solar
radiation modelling
5.6 Discussions and Conclusions
Due to success of the GT in selection required data sets for modelling and their data
length, the study has applied the GT to daily data from the Brue catchment. This
application lead to a new approach to estimate daily solar radiation from meteoro-
logical data sets with the Gamma Test in combination with nonlinear modelling
techniques. The study successfully demonstrated the informative capability of the
Gamma Test in the selection of relevant variables in the construction of non-linear
models for daily (global) irradiation estimations. The study has used six variables for
estimating the daily solar radiation (four highly relevant inputs like daily extrater-
restrial radiation, daily mean air temperature; daily maximum air temperature and
daily precipitation; two relatively less signi
cant inputs like daily minimum air
temperature and daily wind velocity). The quantity of data required to construct a
reliable model was determined using the M-Test, which has identi
ed M = 770 as the
suf
cient data scenario. Abrahart and See [ 3 ] have argued the need to have consistent
measures of merit and trust in hydrological modelling requires. The study has
identi
ed that proper controlled experiments are inevitable for more authenticity and
before acceptance of any model. Good examples of such controlled experiments
could
find in hydrological modelling [ 4 ] and for sediment transfer in [ 5 ].
To check the reliability of GT modelling, the study has performed extensive
modelling experiments with LLR and ANN on all input combinations. The modelling
with LLR has shown that the modelling results from input combinations like
[110101] than [111111] are comparable with better numerical values associated with
[110101] model. But the ANN modelled results were perfectly matched with that of
GT
finding associating better statistics with the ANN model. The mismatch in the
finding with different models shows the importance of controlled experiments as that
mentioned in [ 3 ]. The explicit method of testing [ 3 ] for the presence of non-linear
relationships in each input combinations gave different results for simpler model
(LLR didn
'
t match with GT results) and complex model (ANN matched with GT
finding) in this case study. This contradiction strengthens the question raised by [ 3 ].
The question of which tool would be more appropriate in a practical issue that equates
to picking the
ed the GT
methodology is more reliable and less time consuming than trial and error, has an
right tool for the right job
. However, the case study identi
 
Search WWH ::




Custom Search