5 | CONCLUSION
Based on water table measurement in 16 months, the water table was
increasing and directly proportional to rainfall data. It can be
concluded that in the arid region, for shallow groundwater aquifers
water table fluctuation is highly influenced by rainfall.
The best-fitted-semi-variogram is the exponential model type, which has
the lowest RMSE value. Within this given period of water table
measurements, the first nine months of data (dry season, November 2018
– July 2019) has a weak spatial dependence (>75%), and
during wet season or the second period (wet season) data from August
2019 – February 2020, has strong spatial dependence (<25%).
Based on the cross-validation analysis of the Kriging map using data of
weak and strong spatial dependences, the following conclusions can be
drawn:
- A negative mean error has occurred in data with weak spatial
dependence, which means that the predicted value is lower than the
actual measurement. A positive mean error has occurred in data with
strong spatial dependence, which means that the predicted value is
higher than the actual measurement.
- Most of the root-mean-square standardized error is close to 1, which
indicates the errors are following a normal distribution. It is also
confirmed by the QQ plot that shows the prediction error for both weak
spatial dependence (November 2018 – July 2019) and strong spatial
dependence (August 2019 – February 2020) are close to normal
distribution.
- The RMSE for weak spatial dependence data is higher than the RMSE
value for strong spatial dependence data, which means that the weak
spatial dependence has a higher variance compared to the strong
spatial dependence. A similar result occurs for average standard
error, where the data with weak spatial dependence has a higher value
compared to strong spatial dependence.
- Most correlation coefficients for wells data with strong spatial
dependence are close to one, which means that the data from August
2019 – February 2020 has a positive and high correlation (0.76 –
0.97). On the other hand, most of the wells data with weak spatial
dependence have a relatively low correlation of coefficient (0.37 –
0.98). It indicates that during high rainfall season, the water table
rises in all wells and provides high correlations, while during the
dry season, the water table drops at each well individually, and the
correlation at some wells are poor and can reach a negative value.
- The graph of actual measurement and prediction shows that in the weak
spatial dependence, the points are far from the line of 1:1 compared
to strong spatial dependence. Therefore, the accuracy of the Kriging
map using data of strong spatial dependence (August 2019 – February
2020) is more reliable compared to the case of weak spatial dependence
(November 2018 – July 2019).
Based on the results of the water table map, the water table is
increasing from August 2019 to February 2020. The groundwater is flowing
from East to West direction and South-East to North-West direction,
which is almost similar to the stream direction on the surface. This
result corresponds to the streams elevations where it is assumed that
the groundwater is flowing from the higher stream elevation (Southern
stream) to the lower stream elevation (Northern stream). On the other
hand, a Kriging map using weak spatial dependence data (November 2018 –
July 2019) shows that the flow direction is going outside the farm in
random directions. It happens because the water tables in the wells do
not change in time due to there is no feeding from the rainfall, and
therefore, there is no specified direction on the flow.