19 Detection of Land Cover Change
423
bration periods should be comparable. Otherwise the ability of the model to reflect
the hydrological impacts of land use change is hidden by its sensitivity to different
hydro-meteorological conditions.
The second problem concerns the general possibilities to detect the hydrological
effects of land use changes. Uncertainties in determination of the areal extent and
location of these changes and also the inaccuracy of measured meteorological and
hydrological data limit our capability to detect their sometimes relatively small
effects. Nandakumar and Mein (1997) estimated in a case study the uncertainty of
model predictions which results from the random characteristics of model parameters and climatic data. Under consideration of this level of background 'noise'
in some cases extreme changes of land use (between 20 and 65% clear cutting of a
forest) would be necessary to estimate for example detectable flow increases at the
90% prediction level.
Under the aspects of uncertainty the hydrological impacts of land use changes
are mostly discussed for two types of changes which can be identified relatively
easily and which are connected with specific changes in rainfall- runoff processes
and water balance: urbanisation and deforestation.
Remote sensing reduces the uncertainty of estimations of land cover changes
significantly. To estimate the hydrological effects ofland cover changes by remote
sensing a very careful classification of the satellite scenes or air photography is
needed. This necessity can be explained by an example which was given by Allewijn and Baker (1993). These authors estimated the relative differences of the
classification results of two Landsat TM scenes for 15 catchments between 26.0
and 184.1 km 2 in the Vechte River Basin in the Netherlands. For agricultural
crops, forest types, nature areas, urbanized areas, water and bare soil the similarity
between the two classifications varies between 40% to more than 80% if they were
compared on catchment scale base although the time difference of the two Landsat
scenes was only two years (from 1984 to 1986). The authors explain these differences with random and systematic errors caused by classification but also by "real
world differences" in satellite recording dates, harvesting and crop rotation. If
these data are used in hydrological modelling uncertainties in areal extent and
location of land use units may result in uncertainties in areal evaporation and runoff estimation. Especially distributed models demand a high accuracy in the estimation of the location of land cover changes as their hydrological effects may be
increased or decreased by the other aspects (e.g. soil, aspect etc.).
If land use changes within a catchment are identified by remote sensing two possibilities exist to characterise their hydrological effects:
trends in runoff can be estimated by time series analysis,
- the hydrological effects of these changes can be simulated by application of a
hydrological model where its parameters are changed.
If time series analysis is used to identify the effects of land use changes on runoff
also possible changes in the time series of meteorological variables (esp. precipitation) should be considered. The effects of land use changes could be hidden by
changes in these time series. By utilization of a hydrological model a physical
explanation of its parameters in relationship to land use characteristics is neces-
423
bration periods should be comparable. Otherwise the ability of the model to reflect
the hydrological impacts of land use change is hidden by its sensitivity to different
hydro-meteorological conditions.
The second problem concerns the general possibilities to detect the hydrological
effects of land use changes. Uncertainties in determination of the areal extent and
location of these changes and also the inaccuracy of measured meteorological and
hydrological data limit our capability to detect their sometimes relatively small
effects. Nandakumar and Mein (1997) estimated in a case study the uncertainty of
model predictions which results from the random characteristics of model parameters and climatic data. Under consideration of this level of background 'noise'
in some cases extreme changes of land use (between 20 and 65% clear cutting of a
forest) would be necessary to estimate for example detectable flow increases at the
90% prediction level.
Under the aspects of uncertainty the hydrological impacts of land use changes
are mostly discussed for two types of changes which can be identified relatively
easily and which are connected with specific changes in rainfall- runoff processes
and water balance: urbanisation and deforestation.
Remote sensing reduces the uncertainty of estimations of land cover changes
significantly. To estimate the hydrological effects ofland cover changes by remote
sensing a very careful classification of the satellite scenes or air photography is
needed. This necessity can be explained by an example which was given by Allewijn and Baker (1993). These authors estimated the relative differences of the
classification results of two Landsat TM scenes for 15 catchments between 26.0
and 184.1 km 2 in the Vechte River Basin in the Netherlands. For agricultural
crops, forest types, nature areas, urbanized areas, water and bare soil the similarity
between the two classifications varies between 40% to more than 80% if they were
compared on catchment scale base although the time difference of the two Landsat
scenes was only two years (from 1984 to 1986). The authors explain these differences with random and systematic errors caused by classification but also by "real
world differences" in satellite recording dates, harvesting and crop rotation. If
these data are used in hydrological modelling uncertainties in areal extent and
location of land use units may result in uncertainties in areal evaporation and runoff estimation. Especially distributed models demand a high accuracy in the estimation of the location of land cover changes as their hydrological effects may be
increased or decreased by the other aspects (e.g. soil, aspect etc.).
If land use changes within a catchment are identified by remote sensing two possibilities exist to characterise their hydrological effects:
trends in runoff can be estimated by time series analysis,
- the hydrological effects of these changes can be simulated by application of a
hydrological model where its parameters are changed.
If time series analysis is used to identify the effects of land use changes on runoff
also possible changes in the time series of meteorological variables (esp. precipitation) should be considered. The effects of land use changes could be hidden by
changes in these time series. By utilization of a hydrological model a physical
explanation of its parameters in relationship to land use characteristics is neces-