Abstract :
Abstract – The study was carried out to develop a weather
based model on yield of cotton using Neural Network Fitting
Tool for various sizes of data set. There is significance effect
of various climatological data on yield of cotton. The
climatological data for the hot weather season are collected
for the period 1981- 2006 and correlated with yield of cotton
in Vallabh Vidyanagar using neural network fitting. To form
the dataset for model development and Evaluation, three
alternatives are considered. In Alternative 1, 70% of data of
Maximum and minimum temperatures, sunshine hours,
relative humidity and wind velocity are correlated with yield
data and 30% data are used for Evaluation of the model. In
Alternative 2, 60% of above mentioned data are used to
develop the model and 40% data are used to validate the
model. In Alternative 3, 80% data are correlated with yield
and 20% data are utilized for Evaluation of the model. Then,
all the model are re-trained until the best coefficient of
correlation is obtained and this corresponding model is
considered as the best model and this is further used to
validate the remaining dataset. This whole procedure is
repeated for three different Alternatives. The results shows
that the best model is obtained for alternative 1, in which R
for training, testing and validation for model development
are 0.92, 0.66 and 0.86 respectively and R for Evaluation is
0.72. The overall R value is 0.5 for this alternative. It can be
concluded from the overall study that considering all the five
climatological parameters, with use of 70% data for model
generation and 30% data for Evaluation, the correlation is
achieved as the best. The use of Neural Network fitting is
quite helpful in studying the effect of distribution of dataset
for model development.
