LESOTHO
GPOW 5.0 Riparian Factor 0.5
6.1 INTRODUCTION
6.1.1 Climate Model Data and Observations
The historical rainfall data used in this study were obtained from the South African surface water resources assessment study (WR2005) of Middleton and Bailey (2008) and consist of monthly totals for each of the quaternary catchments for the whole of South Africa, Lesotho and Swaziland. The rainfall data from the WR2005 study are the best available historical data series in the country, although they are not perfect because the density and coverage of the gauging network are not adequate in many regions. The observed historical data set used in this study
137 spans the period from 1920 to 2005 and has been accepted to represent the long term climate patterns of the region (Middleton and Bailey, 2008).
The downscaled climate data sets comprise daily maximum and minimum temperatures and rainfall at the quinary catchment scale. Within the South African hydrological and water management system, there are at least three quinary points within each quaternary catchment.
The GCM climate simulation data was scaled down using an empirical statistical procedure (Hewitson and Crane, 2006). The model simulations used here cover three climate scenario periods: baseline (1961-2000), near-future (2046-2065) and far-future (2081-2100). The quinary scale baseline daily rainfall data were interpolated to the quaternary catchment scale using the inverse distance squared method and then summed to monthly values for comparison with the historical WR2005 data. The climate model data do not represent real historical sequences of rainfall and each model depends on the initial boundary conditions. It is therefore not possible to compare the climate model time series data with each other, nor with the historical WR2005 data. The comparisons and bias correction methods (Chapter 4) are therefore mainly based on calendar month statistical properties (means, standard deviations and skewness).
The GCM-simulated baseline annual rainfalls for the four sub-basins are substantially biased relative to the observed rainfall. The bias varies significantly in both magnitude and direction.
Table 6.1 shows that while some GCMs are biased by as much as 68% others severely under-simulate rainfall, by up to -35%. The annual rainfall bias of the GCMs seems to increase with increasing rainfall amounts. While the GCMs show a generally negative bias in the wettest region (D21B), the opposite is true in the driest part (D23H). However, rainfall simulations by CCCMA and IPSL show a consistent negative bias across all sub-basins, whereas GISS consistently show the highest positive bias in the four sub-basins.
Figure 6.1 provides more detailed comparisons of the calendar month mean rainfall values and the deviations from the observations for the baseline simulations of the nine GCMs. One immediate observation is that the percentage deviation varies in both time and space. There is higher deviation during the dry winter period (May to September), as well as in the drier sub- basins, with the high summer rainfall amounts being generally under-estimated and the lower winter rainfalls being over-estimated. The deviation values range between approximately 250%
(in June) to about -50% (in January). Figure 6.1 suggests that there is generally a negative deviation during the rainy season and that the range of uncertainty appears to be greater in the drier area (D23H) than in the wetter parts of the basin (D21B and D22A). However, the general
138 differences in percentage bias between the nine GCMs appear to be quite consistent across the four example sub-basins.
Table 6.1 Percentage bias (PBIAS) of the annual rainfall simulated for the baseline scenario by the climate models relative to the observed.
Catchment D21B D22A D23C D23H
Historical
rainfall (mm) 1013 679 633 517
CCCMA -35.8 -23.9 -19.52 -2.0
CNRM -13.2 2.3 11.24 35.3
CSIRO -15.5 -11.3 10.53 22.8
GFDL -13.6 -7.0 8.62 21.3
GISS 4.5 23.9 31.64 68.6
IPSL -31.9 -20.0 -17.54 -9.7
MIUB -24.7 -11.8 -3.58 18.6
MPI -23.1 -9.1 -1.54 19.5
MRI -27.7 -15.3 -6.60 17.4
Figure 6.1 Percentage deviation of the simulated baseline monthly rainfall from the observed rainfall.
139 6.1.2 The Bias Correction Method
In the preliminary analysis of the GCM baseline rainfall data, several calendar month statistical measures were used to compare the characteristics of the climate models with the observed WR2005 rainfall records. The analyses indicate that the GCM baseline data differ significantly against the observations, in terms of the various statistical indicators used (see Table 6.1 and Figure 6.1). These discrepancies indicate that the outputs of the future climate scenarios cannot be directly applied for any impact assessment without some form of bias correction if sensible results are to be attained from a model previously established using climate forcing data based on historical data.
In the current study, a bias correction method introduced by Hughes et al. (2014b) is used to adjust the statistically downscaled precipitation data sets (see Chapter 4). The bias correction method is based on the use of the calendar month means and standard deviations and relies to a certain extent on the frequency distributions of the rainfall data to be near-Normal.
Preliminary analyses of all of the rainfall data sets suggested that a square root transformation would produce the closest approximation to Normal distributions in all cases. There are some situations where the requirement for low skewness values after transformation is still not met in the dry winter months, however, this was not considered to be a critical problem as the rainfall values are almost always very small and have little influence on the hydrological modelling results. In the bias correction method the future rainfall estimates are expressed in terms of standard deviates of the baseline scenario monthly distributions and the standard deviates are scaled with the monthly distributions of the historical rainfall data. The bias correction method was presented in detail in Chapter 4.
The application of the bias correction removes bias in the monthly rainfall means and standard deviations between the historical and the GCM simulated baseline data. At the same time, the bias correction method is able to preserve the differences between the GCM baseline and future scenario predictions.