Geoscience Reference
In-Depth Information
MODELING SEDIMENT DISCHARGE WITH ARTIFICIAL
NEURAL NETWORK: AN EXAMPLE OF THE
LONGCHUANG RIVER IN THE UPPER YANGTZE
YUN-MEI ZHU
∗,†
,X.X.LU
∗
,YUEZHOU
†
and YOUAN GUO
§
∗
Department of Geography, National University of Singapore, 119260, Singapore
†
Department of Environmental Science, Kunming University of Science
and Technology, China
§
Yunnan Hydrological and Water Resources Bureau, Yunnan, China
Artificial neural network (ANN) was used to model the monthly sediment dis-
charge in the Longchuang River in the Upper Yangtze, China. The variables
including the averaged rainfall and temperature, rainfall intensity and water
discharge were used. The results suggest that ANN is capable of modeling the
monthly sediment discharge with fairly good accuracy when proper variables
and their lag effect on sediment discharge are used as inputs. Compared to the
multiple regression models, ANN produced a better fit to the observed sediment
discharge and provided more reasonable results for the extreme points.
1. Introduction
Artificial Neural Network (ANN) is based on the concepts derived from
the researches on the nature of the human brain.
1
Its distinct advantages
make it a competitive tool in hydrological modeling.
2
,
3
The current appli-
cation of ANN in hydrology mainly focuses on the river flow modeling and
prediction,
4
,
5
but much less on sediment discharge. The researches con-
ducted by Abrahart and White,
6
Jain,
7
Tayfur,
8
and Kisi
9
may be deemed
as pathfinder experiments in this area. They have demonstrated the capa-
bility of ANN in sediment concentration or discharge modeling. However,
these researches usually took water and sediment discharge at previous
time steps as inputs, which may increase the accuracy of the simulation
but failed to explain the physical relations between the sediment and its
control variables. This research attempted to relate the sediment discharge
with the original driving forces such as rainfall, temperature and rainfall
intensity, aiming to establish an ANN model that can be used to explore the
relationships between the causal variables and the sediment responses. The
advantage of the ANN over the multiple linear regression (MLR) models
was also evaluated by comparing their performances.
117
