181
Modeling Stream Flow Changes
likely present could disturb the training process of the ANN model. Output of this
modeling work is satisfied, mainly because the training process covers the turning
point where a sudden drop occurred (e.g., covers the period before and after 2002).
If the training process is carried out using the data sets before 2003, then the factors
causing the sudden drop cannot be captured by the ANN model; thus, the quantitative relation achieved in the ANN modeling work may be deemed unreliable.
8.6 CONCLUSIONS
The applications of remote sensing images and the EOF analysis in this study provide
more spatially representative data to verify some scientific hypotheses on the effects
of climate change. By using the EOF analysis, the behavior of the spatial patterns of
monthly precipitation and LST was smoothly extracted. A group of input variables
was derived from identified spatial patterns via the EOF analysis to support the ANN
modeling analysis. This success led to the smooth prediction of stream flow changes
during the study period, which would otherwise be limited by the proper handling
of tremendous amount of input data in the ANN modeling analysis. Results indicate that the approach integrating the ANN model into spatial statistical output in
association with the EOF analysis shows promise in rainfall–runoff modeling for an
ungauged, glacier-fed basin environment.
To improve the prediction accuracy, the hydrologically and hydraulically relevant
variables (e.g., soil moisture, vegetation, land cover, and water stage) and basin characteristics (e.g., topography and surface roughness) can also be included in future
modeling work. Some of these variables may be acquired by using special remote
sensing technologies. Based on these additional data sources, the proposed approach
may be reinforced, signified, and magnified to properly handle a variety of ungauged
watersheds with spatial variation of rainfall, the heterogeneity of watershed characteristics, and their complicated impacts on runoff at different temporal and spatial
scales. Methods of making use of the spatiotemporal data that can be more efficient to aid in different modeling platforms than the ANN model require additional
research in the future.
ACKNOWLEDGMENTS
This work was financially supported by the Natural Science Foundation of China
(grant nos. 40701025 and 40801040), the German Academic Exchange Service
(DAAD), and the Robert Bosch Foundation on Sustainable Partners (grant no.
32.5.8003.0063.0/MA01).
REFERENCES
Abdulla, F., Eshtawi, T., and Assaf, H. (2009). Assessment of the impact of potential climate
change on the water balance of a semiarid watershed. Water Resources Management,
23(10), 2051–2068.
Beskow, S., de Mello, C. R., Coelho, G., da Silva, A. M., and Viola, M. R. (2009). Surface runoff in a watershed estimated by dynamic and distributed modeling. Revista Brasileira
De Ciencia Do Solo, 33(1), 169–178.
Modeling Stream Flow Changes
likely present could disturb the training process of the ANN model. Output of this
modeling work is satisfied, mainly because the training process covers the turning
point where a sudden drop occurred (e.g., covers the period before and after 2002).
If the training process is carried out using the data sets before 2003, then the factors
causing the sudden drop cannot be captured by the ANN model; thus, the quantitative relation achieved in the ANN modeling work may be deemed unreliable.
8.6 CONCLUSIONS
The applications of remote sensing images and the EOF analysis in this study provide
more spatially representative data to verify some scientific hypotheses on the effects
of climate change. By using the EOF analysis, the behavior of the spatial patterns of
monthly precipitation and LST was smoothly extracted. A group of input variables
was derived from identified spatial patterns via the EOF analysis to support the ANN
modeling analysis. This success led to the smooth prediction of stream flow changes
during the study period, which would otherwise be limited by the proper handling
of tremendous amount of input data in the ANN modeling analysis. Results indicate that the approach integrating the ANN model into spatial statistical output in
association with the EOF analysis shows promise in rainfall–runoff modeling for an
ungauged, glacier-fed basin environment.
To improve the prediction accuracy, the hydrologically and hydraulically relevant
variables (e.g., soil moisture, vegetation, land cover, and water stage) and basin characteristics (e.g., topography and surface roughness) can also be included in future
modeling work. Some of these variables may be acquired by using special remote
sensing technologies. Based on these additional data sources, the proposed approach
may be reinforced, signified, and magnified to properly handle a variety of ungauged
watersheds with spatial variation of rainfall, the heterogeneity of watershed characteristics, and their complicated impacts on runoff at different temporal and spatial
scales. Methods of making use of the spatiotemporal data that can be more efficient to aid in different modeling platforms than the ANN model require additional
research in the future.
ACKNOWLEDGMENTS
This work was financially supported by the Natural Science Foundation of China
(grant nos. 40701025 and 40801040), the German Academic Exchange Service
(DAAD), and the Robert Bosch Foundation on Sustainable Partners (grant no.
32.5.8003.0063.0/MA01).
REFERENCES
Abdulla, F., Eshtawi, T., and Assaf, H. (2009). Assessment of the impact of potential climate
change on the water balance of a semiarid watershed. Water Resources Management,
23(10), 2051–2068.
Beskow, S., de Mello, C. R., Coelho, G., da Silva, A. M., and Viola, M. R. (2009). Surface runoff in a watershed estimated by dynamic and distributed modeling. Revista Brasileira
De Ciencia Do Solo, 33(1), 169–178.