Geoscience Reference
In-Depth Information
Fig. 2. Scatter plots of the observed and predicted sediment discharges by the corre-
sponding ANNs and MLR models.
5. Conclusion
This study demonstrated that ANN is capable of modeling the monthly sed-
iment discharge with fairly good accuracy when proper variables that rep-
resent the driving forces and their lag effect on sediment discharge are used
as inputs. The network using both climate variables and water discharge as
inputs can provide best simulation; whereas the network with only climate
variables as inputs could be used for climate change impact assessment.
Compared to the multiple regression models, ANNs can produce better fits
to the observed sediment discharge and provide more reasonable results at
the extreme points.
The previous researches using ANNs to model water and sediment dis-
charge employed the independent variable at previous time steps as inputs
13
or as the only type of input to the network.
5
,
9
The ANNs established in
this research with only climate variables as inputs have the potential of
being used to fill the missing data in sediment time series and to predict
the influence of climatic change on sediment discharge.
Acknowledgments
This project is funded by National Basic Research Program of China
(Project No. 2003CB415105-6) and National University of Singapore
