Geoscience Reference
In-Depth Information
the USA, the USDA STATSGO database defines soil characteristics for the whole of the country, but at
a scale of 1 km
2
(USDA SCS, 1992). In Europe, the European Soils Database (ESDB) is also available
in vector form or as a raster 1 km
2
grid.
This is also true of the other GIS data types. In general, another model is required to interpret the
GIS data in a form that can be used in a hydrological model. An example is the type of
pedotransfer
function
models suggested by Rawls and Brakensiek (1989). Regression analysis is used to provide
relationships between soil texture and soil hydraulic parameters. Soil texture is a common characteristic
reported in soil surveys and a soil classification can normally be associated with a texture. A pedotransfer
function can then be used to derive values for parameters such as porosity and hydraulic conductivity
(see Box 5.5).
However, values derived in this way should be interpreted with care. We have already noted that
measured soil hydraulic parameters may be highly variable in space, even within a single soil unit,
and that the effective parameter values required for different models may be model structure and scale
dependent (Section 1.8). Thus, GIS-derived values of parameters may be associated with a considerable
degree of uncertainty which is often ignored. Having said that, a number of studies have reported success
in rainfall-runoff modelling based on GIS data (see Section 6.6).
Within this GIS framework, rainfall-runoff modelling may be just one component of a larger catchment
management or decision support system (DSS). Examples are the WATERSHEDSS package developed
by the USDA (Chaubey
et al.
, 1999) and the UK NERC-ESRC Land-use Programme (NELUP) DSS
(Dunn
et al.
, 1996). Few GIS packages, however, allow flexible modelling structures to be built directly
onto the spatial framework of the GIS. One exception is PC-RASTER which has a built-in programming
language that allows model structures, together with input and output data, to be readily incorporated into
the GIS. Other general programming languages, such as MATLAB and PV-WAVE, can also be used with
large raster spatial databases for modelling (see, for example, Clapp
et al.
, 1992; Romanowicz, 1997).
3.9 Remote-sensing Data
Another source of distributed data for hydrological modelling comes from remote sensing. A full review
of the types of data that can be provided by remote sensing is beyond the scope of this topic but excellent
coverage is given in Section 5 of the
Encyclopaedia of Hydrological Sciences
(Anderson, 2005). Remote
sensing may be used in the estimation of input data (including topography, rainfall and evapotranspiration
rates), state variables (including soil moisture, snow cover, snow water equivalent and areas of flood
inundation) and model parameter values (mostly derived through the classification of soil and vegetation
types from remote sensing). In fact, topography or land cover distribution derived from remote sensing
may contribute to the database stored in a GIS. Many of the same problems apply as for GIS data:
remote sensing does not generally give information that is directly hydrologically relevant; a model is
required to interpret the remote-sensing signal into a form that is hydrologically useful (as discussed
earlier, for example, for the case of radar-derived estimates of rainfall). It is not often recognised that
the interpretation model can be a significant source of uncertainty in the resulting images supplied to the
user. Corrections for atmospheric effects for satellite sensors, for example, involve empirical coefficients
or parameters that are not known precisely and that may be time variable. These uncertainties are not
often quantified and the user has little option but to assume that they are small. However, remote sensing
will increasingly be an important source of spatial information and future hydrological models will make
increasing use of different types of imaging in model calibration, evaluation of predictions and data
assimilation to improve short-term forecasting.
The main uses of remote sensing in hydrological modelling to date have been in the estimation of
precipitation (primarily using ground-based radar), land cover types and vegetation parameters, soil
moisture and snow cover. Snow cover mapping from satellite images is now used operationally in the
