How to evaluate vulnerability in changing environmental conditions

Fig. E.26.
Schematic illustration of the
reduction of uncertainty in a
reservoir operation through
application of forecasting
techniques
535
With forecast! Uncertainty
Without forecast: Reduction
veld under good management. When converted to sedi- wrong, i.e. 1981/1982, 1987/1988 and 1990/1991, most of
ment yields, however, the factor difference becomes southern Africa scores more "wins" than "losses". The
2.5-7.5 times, and even> 7.5 times (Fig. E.24, bottom), impacts of short term climate perturbations such as the
clearly illustrating how a hazard, in this case stormflow EI Nino phenomenon have already been illustrated. This
and especially sediment yield, can be modified positively forecast analysis indicates that even at the current level
by good grazing management and/or rehabilitation of of seasonal forecast accuracy (around 62% if the 3 worst
overgrazed lands.
forecasts in 15 are omitted) these can potentially be
"translated" into an operational tool for water resources
managers which could prove statistically more accurate
Example of Vulnerability Modification through
than the current practice of forecasting based on his-
Seasonal Forecasting of Runoff
Vulnerability modification is a form of risk mitigation
which includes, inter alia, assessing the benefits of foretorical
expected, i.e. median, runoffs with wide uncertainty
bands, as shown in Fig. E.26.
casting streamflows for the rainy season ahead. Statisti- Are Certain Areas in South Africa Hydrologically
cally derived categorical seasonal rainfall forecasts four More Sensitive than Others to the Individual Forcing
months ahead are made for eight regions of South Africa
by the South African Weather Service, for three cat-
Variables of Climate Change?
egories, i.e. "above average", "near average" and "below Figure E.2? illustrates the relative sensitivities on mean
average" seasonal rainfalls. If seasonal rainfall forecasts annual runoff of dT (assumed to be a uniform increase
were a random process, such three-category forecasts of 2 °C over southern Africa) and M (changed through
would be correct 33% of the time. If seasonal categori- -10% to +10% of the present). In each case the other
cal rainfall forecasts are "translated" into seasonal run- two variables are held constant at present levels when
off forecasts, these could become very valuable reser- running the daily ACRU model. The hydrological sysvoir
operations and irrigation application planning tools tem is relatively insensitive to temperature changes that
for water resources managers. Seasonal categorical rain- affect evaporation and hence runoff. The increase of 2 °C
fall forecasts for the eight forecast regions in South Af- reduces mean annual runoff over most of summer
rica were downscaled to daily rainfall values using tech- rainfall areas in South Africa by only 5% (Fig. E.2], top).
niques described in Schulze et al. (1998b) for applica- However, in the south-west winter rainfall region, the
tion with the ACRU modelling system to over 1500 Qua- response to temperature becomes more dramatic, with
ternary catchments in South Africa. A simple benefit a 2 °C increase by itself producing a simulated reduc-
analysis of forecasting skill was undertaken, in which a tion in mean annual runoff in excess of 50%. The rea-
"win" was recorded if, for the historical seasonal rainsons for this are that under present climatic conditions
fall forecast, the simulated seasonal runoff was closer to evaporation losses there are relatively low from the moist
the runoff simulated with actual historical rainfall than soils in winter, but that with warming, faster drying soils
the median seasonal runoff, while a "loss" was recorded between rainfall events significantly reduce runoff. The
when median runoff was closer to the actual than the most significant sensitivity to climate change, however,
forecast runoff. "No difference" implies forecasted and remains that due to rainfall, with changes by one unit
median runoffs within 5% of one another. Figure E.25 manifesting themselves as runoff changes by a factor of
illustrates that, when excluding three seasons out of 15 2 to 5 (Fig. E.2?, bottom), with the sensitivity more domi-
for which the rainfall forecast accuracy proved 100% nant in the extreme south-west.