Step 9: Correct Possible Biases
You need to do this step if your objective is to develop modelled time series for an impact model assessment.
Climate models are not intended to produce an exact value of climate variables on a specific day or in a particular sequence, so climate model simulation data will never be exactly the same as observed data.
Additionally, all models have inadequacies due to such factors as resolution, differing internal dynamics, and parameterizations. Therefore, different climate models can respond differently to the same inputs. This is an important reason to use a multi-model approach rather than the output of a single climate model.
Bias correction ‘adjusts’ climate model time series so that the current/historical climate more correctly matches the observed climate.
There are many techniques to adjust the model data so that it is closer statistically to the observed data, ranging from simple bias corrections to more sophisticated approaches that adjust the distributions of fields generated by the model to match the observed distribution. In all these cases, high-quality observational data is needed (see step 2). In addition, regardless of the technique used, it is essential that the climate change signal as projected by the models is altered as little as possible.
You should check to see if the climate change signal is affected by the correction and, if it is, you should find out why. For example, if your initial projected change indicates an increase in rainfall but your bias-corrected data suggests a decrease in rainfall, then you need to check whether this is because of incorrect calculation or because of the real results of bias correction application. Some studies indicate that bias correction could have an impact on the climate change signal and hence introduce another layer of uncertainty in the modelling chain.
The choice of method must be matched to the intended application, taking into account constraints of time, resources, human capacity, and supporting infrastructure. The techniques must also preserve the projected climate change signal and the internal variability.
Two popular bias correction techniques to make climate scenarios for use in impact models are the delta method and quantile-quantile bias-correction method.
This method applies the projected changes in mean climate, as simulated by a climate model, to observed climate data. This may be in the form of an additive or multiplicative factor, depending on the variable. Some modifications of this method incorporate projected changes in daily climate variability using the quantile-quantile bias-correction method.
To use this method:
- Adjust the baseline observations by applying the ‘scaling factor’, constructed in step 7, namely the difference (or ratio) between the period-average of modelled future and the corresponding model’s historical period.
- Differences are usually applied for temperature changes (e.g. 2040–2069 minus 1961–1990).
- Ratios are commonly used for precipitation change (e.g. 2040–2069 divided by 1961–1990), though differences may be preferred in some cases.
- When this procedure is completed across some or all the model grid boxes, a pattern of differences or ratios known as a ‘change field’ is produced.
Quantile-quantile bias-correction method
This method computes the observed percentiles and the simulated current climate percentiles. For each percentile class, compute the mean value for each dataset. Then compute a factor to adjust the model data mean value to match the observed.
To use this method:
- Compare the statistical properties of the observational data and the historical period of the regional climate model output.
- Adjust the statistical properties of the regional climate model to match the observed dataset.
- Adjust the whole time series of regional projections with the correction.
CASE STUDY EXAMPLES
 For example: Hagemann S, Chen C, Haerter JO, Heinke J, Gerten D, Piani C. 2011. Impact of statistical bias correction on the projected hydrological changes obtained from three GCMs and two hydrology models. Journal of Hydrometeorology, 12, 556–578.
 For example: Santer BD, Wigley TML, Schlesinger ME, Mitchell JFB. 1990. Developing climate scenarios from equilibrium GCM results. Max-Planck-Institut für Meteorologie Rep. 47, 29 pp
 For example: Bennett JC, Ling FLN, Post DA et al. 2012. High-resolution projections of surface water availability for Tasmania, Australia. Hydrology and Earth System Sciences, 16, 1287-1303.