Geoscience Reference
In-Depth Information
perspective. This is also a problem of regionalisation since the land surface parameters at the atmospheric
grid scales cannot be inferred from either direct measurement or calibration since most of the land surface
is effectively ungauged. There is now a global network of several hundred observation sites for land
surface fluxes, also including carbon dioxide fluxes (FLUXNET, see Appendix A) but these are local
scale measurements that reflect fluxes in the effective fetch area around each site. So some regionalisation
methodology is still required to provide parameters at large scales.
There is a relatively small number of common methods for regionalisation. These can be differentiated
between those that make estimates at the catchment scale, and those that try to incorporate knowledge
from geographical information systems about the distributed nature of catchment characteristics. The
catchment scale methods generally use regression against catchment characteristics or define donor
catchments or pooling groups using similarity measures based on summary measures of catchment
characteristics (Vogel, 2005). The distributedmethods generally relatemodel parameter values to overlays
of soil, land use and topographic characteristics.
10.5 PUB as a Learning Process
The regionalisation problem would not be so difficult if it was possible to find a “similar” catchment to
the ungauged site of interest in the same physioclimatic region. Unfortunately, even catchments that look
superficially similar often exhibit rather different hydrological response characteristics. Catchments vary
in their characteristics in so many ways (in area, in rainfall characteristics, in geology, in topography
and geomorphological history, in soil characteristics, in vegetation and land management, in degree of
urbanisation, ...) that close similarity should not be expected (especially since the sample of gauged
catchments is generally very limited). However, it is worth noting that where the ungauged site of interest
is within a catchment with one or more existing gauges, simply scaling the observations at those gauges
will be difficult to beat, unless there are good physical reasons to suspect that this will not be successful
because of a striking change in geology, urbanisation or strongly different rainfall inputs. This is the first
approach recommended by the UK
Flood Estimation Handbook
(IH, 1999).
This is one argument for using a distributed parameters approach to regionalisation since the pattern of
parameters can then reflect the spatial patterns of characteristics and processes within a catchment, at least
partially. There remains a problem that we do not know enough about the nature of the processes at any
point, particularly the subsurface processes, to be confident about whether the responses are being properly
represented. This is because the small scale variability in hydrological responses is sufficiently complex
that, even given the same rainfall regime, soil type, vegetation type, upslope contributing area, slope angle
and aspect, it is difficult to be sure that the balance of surface and subsurface flow processes will be the
same. This might because of fracturing in the bedrock, highly variable soil depths, or modifications by
people in ditching and installing drainage. This is the problemof “uniqueness of place” (Beven, 2000). It is
evident in the fairly common experience that some catchments appear to be “outliers” in regionalisation
methods (e.g. Wagener and Wheater, 2006). It implies that we should expect local predictions to be
uncertain, even if distributed parameters are used, because the use of generalised parameter values are
expected to give only approximate results locally.
In extreme cases, of course, the generalised parameter values might give predictions that are (locally)
quite wrong in some of the predicted catchments (e.g. Lamb and Kay, 2004). Where this is made known,
however, action can be taken to try to improve the local model representations (Beven, 2007; see also
Chapter 12). The philosophy that has developed over the course of the PUB decade has been to treat the
ungauged basin problem as a learning process in which initial estimates of how a catchment responds
are gradually improved, and uncertainty contrained, as new information becomes available (Sivapalan,
2003; Buytaert and Beven, 2009; Wagener and Montanari, 2011). Associated with this philosophy has
been a change from an approach centred on model parameters to one based more on trying to estimate the
