~0.10
Area-mean intermodel spread in dlnP remains nearly unchanged with BCSD
Model Development and Evaluation
Uncertainty in Model Projections of Precipitation Change
Downscaling improves present-day bias but does not reduce intermodel spread in projected rainfall change because projection spread is tied to climatological-structure dependence.
Intermodel Spread
Bias Correction
Model Screening Constraint
16-20%
Spread reduction from simple model screening across downscaling products
GRL (2019)
Constraining climate model projections of regional precipitation change
Paper Citation
Zhang, B., and B. J. Soden, 2019: Constraining Climate Model Projections of Regional Precipitation Change. Geophysical Research Letters, 46, 10,522-10,531. https://doi.org/10.1029/2019GL083926
Scientific Logic
- Question: Does statistical downscaling reduce intermodel spread in future precipitation projections?
- Method: Comparison of downscaled land rainfall across models, relating projected change spread to baseline climatological rainfall structure.
- Mechanism: Projection spread is tied to physically meaningful climatology-change relationships, not only removable mean-state bias.
- Main Findings: Bias correction improves historical fit but does not materially narrow future spread, so uncertainty reduction requires process constraints beyond statistical correction.
Scientific Question
Do statistical downscaling and bias correction reduce uncertainty in future regional precipitation projections, and if not, can present-climate model performance be used to constrain that spread?
Approach
- Analyze CMIP5 land precipitation with BCSD and SD_noBC products (plus BCCA/LOCA over CONUS).
- Compare historical climatology (1986-2005) with late-century RCP4.5 projections (2080-2099).
- Use fractional change dlnP and intermodel spread sigma(dlnP) to evaluate uncertainty reduction.
Key Findings
- Bias correction materially improves historical rainfall fields.
- Future projection uncertainty remains large after downscaling due to structural climatology-change coupling.
- Uncertainty reduction needs process constraints, not statistical correction alone.
Figure from the Study
Intermodel Dependence and Spread
