Geology Reference
In-Depth Information
Table 7.8 Correlation coef
cient of different inputs with pan evaporation data
Correlation coef
cients of inputs with pan evaporation (E)
Inputs
W
T
RH
Ed
E
Coef
cients
0.59739
−
0.734759
0.851086
0.86153
1.0
7.5.4 Data Analysis in Evaporation Modeling with Data
Splitting and Cross Correlation Approaches
The study has applied traditional cross-correlation and data splitting approaches to
see whether there is any deviation from the
finding explained in the previous
section. The cross correlation coefficient values of evaporation data with other
inputs are shown in Table
7.8
. It is shown that the highest correlation is associated
with daily values of saturation vapor pressure de
cit and the least with daily relative
humidity, as found in the earlier cases with GT and entropy theory.
We split the available data and made training data sets with lengths ranging from
250 to 4,000 and these were used for modeling with LLR model, the remaining data
points of 4,019 being used as a validation set. The results of the data splitting
approach are shown in Fig.
7.27
.
As the training data increases, the modeling error during the training and testing
phase appears to be decreasing. The root mean squared error associated with the
validation phase was lower in quantity than that of the training phase, as antici-
pated. The RMSE curve during the training phase has shown a sudden decreasing
trend at 1,500 data points but there was no such indication on the validation curve.
The validation curve had a sudden decrease in error when modeling was performed
