172
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
River Basin, Central Asia, the area is considered one of the most arid regions in the
world. Its water supply comes mainly from the mountain regions through stream networks, and the runoff is important for both agricultural activities and natural environments. However, surface runoff generation is a complex and dynamic process
in this watershed, especially in the context of spatial variability in mountain areas
with very few observations (Beskow et al. 2009). Challenge in hydrologic modeling
arises from the fact that the meteorological parameters often vary dramatically with
topographic change (Zhai et al. 1999).
As part of the Tarim River Basin, the core study area in this chapter, covering
1.8 × 10 3 km 2 , is a typically mountainous watershed in southern Tienshan. Its altitude
varies between 1100 and 5000 m above mean sea level, which results in a remarkable
spatial variation in precipitation and temperature. Precipitation can range from 30 to
1000 mm, depending on the complex climatic and topographic conditions created by
the distribution of glaciers on the central mountain tops. The headwater of the river
originating from glaciers runs through alpine meadows and narrow gorges and has
an annual discharge of 25–57 × 10 8 m 3 , recorded by the Dashankou hydrologic station at the mountain outlet (Sun et al. 2010).
From the perspective of hydrometeorology, runoff formation depends on the
interaction of climatic factors, specifically the ratio between heat and moisture.
Hydrologic models aim at mathematically representing the hydrologic system from
precipitation to stream flow. The complexity of the models varies with user requirements and data availability. These hydrologic models vary from simple statistical
techniques that use graphical methods for their solution to the first-principle physicsbased simulation models depicting the complex three-dimensional (3-D) nature of a
watershed (Chow et al. 1988). Due to the geographical complexity of this study area
and the lack of in situ observations, employing a complex 3-D simulation model is
highly unlikely. Hence, the concatenation of two artificial intelligence models (EOF
and ANN) driven by remote sensing data may overcome these barriers to predict
runoff over such a complex terrain.
8.3 MATERIALS AND METHODS
The flowchart for runoff simulation and prediction in this study (Figure 8.2) starts
with the collection and processing of TRMM/PR and MODIS/LST images to generate the essential precipitation and LST time series data for simulation. With the aid
of hydrologic data, such as the historical runoff data and the digital elevation model
(DEM), spatial analysis may be carried out in sequence with possible time lags in
an ANN model employing one hidden layer. When the criterion for stopping the
iteration is available, the ANN output ensures that the predicted runoff is as close as
possible to the observed runoff during the supervised training process. The model
can then be used for simulation and prediction.
8.3.1 Satellite data
Runoff cannot be measured directly with remote sensing images, but the remote sensing techniques can be used indirectly to support the measurements and predictions
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
River Basin, Central Asia, the area is considered one of the most arid regions in the
world. Its water supply comes mainly from the mountain regions through stream networks, and the runoff is important for both agricultural activities and natural environments. However, surface runoff generation is a complex and dynamic process
in this watershed, especially in the context of spatial variability in mountain areas
with very few observations (Beskow et al. 2009). Challenge in hydrologic modeling
arises from the fact that the meteorological parameters often vary dramatically with
topographic change (Zhai et al. 1999).
As part of the Tarim River Basin, the core study area in this chapter, covering
1.8 × 10 3 km 2 , is a typically mountainous watershed in southern Tienshan. Its altitude
varies between 1100 and 5000 m above mean sea level, which results in a remarkable
spatial variation in precipitation and temperature. Precipitation can range from 30 to
1000 mm, depending on the complex climatic and topographic conditions created by
the distribution of glaciers on the central mountain tops. The headwater of the river
originating from glaciers runs through alpine meadows and narrow gorges and has
an annual discharge of 25–57 × 10 8 m 3 , recorded by the Dashankou hydrologic station at the mountain outlet (Sun et al. 2010).
From the perspective of hydrometeorology, runoff formation depends on the
interaction of climatic factors, specifically the ratio between heat and moisture.
Hydrologic models aim at mathematically representing the hydrologic system from
precipitation to stream flow. The complexity of the models varies with user requirements and data availability. These hydrologic models vary from simple statistical
techniques that use graphical methods for their solution to the first-principle physicsbased simulation models depicting the complex three-dimensional (3-D) nature of a
watershed (Chow et al. 1988). Due to the geographical complexity of this study area
and the lack of in situ observations, employing a complex 3-D simulation model is
highly unlikely. Hence, the concatenation of two artificial intelligence models (EOF
and ANN) driven by remote sensing data may overcome these barriers to predict
runoff over such a complex terrain.
8.3 MATERIALS AND METHODS
The flowchart for runoff simulation and prediction in this study (Figure 8.2) starts
with the collection and processing of TRMM/PR and MODIS/LST images to generate the essential precipitation and LST time series data for simulation. With the aid
of hydrologic data, such as the historical runoff data and the digital elevation model
(DEM), spatial analysis may be carried out in sequence with possible time lags in
an ANN model employing one hidden layer. When the criterion for stopping the
iteration is available, the ANN output ensures that the predicted runoff is as close as
possible to the observed runoff during the supervised training process. The model
can then be used for simulation and prediction.
8.3.1 Satellite data
Runoff cannot be measured directly with remote sensing images, but the remote sensing techniques can be used indirectly to support the measurements and predictions