Raw model data have biases which make them inappropriate as direct inputs to environmental models. Hence, statistical processing is needed to reduce this bias and generate Ready to use data that can be directly used in climate change impact modelling studies.
A climate model is an approximate representation of the real world climate drivers. This is due to incomplete understanding of climate physics and/or simplifications required for computational purpose. This inevitably introduces discrepancies (“errors”) when model results are compared to observations. Some of these errors are called bias because they affect basic statistical properties of climate models simulations (e.g., mean, variance). For example, some climate models tend to be systematically “warm” or “cold” by a couple of degrees but still capture the seasonal and interannual variability of temperature. Other climate models show average conditions that are “too dry” or “too wet” when compared with precipitation observations.
Climate change impact assessments often rely on environmental models that are calibrated with present day climate observations. However as mentioned, Raw data don’t have a good fit with present day climate observations and are thus not suitable as direct input for impact models. Hence, Raw climate model data also need to be calibrated with observations to reduce bias. The resulting Ready to use data can then be directly used in environmental models to assess climate change impacts.
A picture is worth a thousand words, so we compared Raw (biased) and Ready to use (calibrated) model data with observations for Sao Paulo in 2 graphics:
- The mean seasonal cycle of surface temperature averaged on the period 1980-2010.
- The annual trend of summer days, from 1951 to 2100
Mean seasonal cycle of the surface temperature, averaged on the period 1980-2010, in Sao Paulo (Brazil).
Figure A1. Raw model data (grey crosses) show a warm bias and don't match observations for most of them (black crosses).
Figure A2. Ready to use model data (orange crosses) have their bias reduced and better match observations (black cross)
Annual trend of Summer days in Sao Paulo (Brazil)
Figure B. Raw model data (upper) are “shifted” above observations over the historical period (grey envelope) because models tend to have a warm bias. Ready to use data (bottom) have a better fit with observations over the historical period (grey envelope) and a more realistic decreased spread over the future (red and blue envelopes).
Statistical bias-adjustment methods bridge the gap between Raw and Ready to use model data. Ready to use data are more consistent with historical observations and produce more robust future climate projections. Although these methods have their own limitations (see our related article), they are required to make climate model data usable in environmental models. We co-develop one of these methods with research organizations and use additional processing techniques to create Ready to use data that better fit researchers and adaptation practitioners requirements.