Indicator Delta Scaling (IDS): Rethinking bias adjustment for climate risk indicators

Abstract. Translating global climate projections into local risk assessments relies heavily on Bias Adjustment (BA). Yet, standard BA methods often distort historical observational baselines and modeled trends, suffer from significant inconsistency across different regions, variables and processed datasets, while imposing heavy computational costs. Driven by the need for credible, consistent and computationally efficient climate information, we propose Indicator Delta Scaling (IDS), a unified framework that generalizes the traditional concept of Change Factors for direct application to any type of climate indicator. IDS relies on three core principles: securing accurate historical baselines by anchoring climate indicator estimates strictly to observations; generating bias-adjusted future indicators by directly scaling these baselines with raw climate model-based deltas; and shifting any required adjustment from the underlying input data to the indicator computation itself. Beyond its conceptual and physical transparency, a systematic evaluation across a multi-model ensemble and a wide spectrum of climate indicators shows that IDS consistently outperforms state-of-the-art methods (QM, QDM, and ISIMIP3b) by reducing their error in adjusted future projections by 23% on average across indicators and climate models. At the same time, IDS provides notably superior consistency, shrinking error variability to one-third of that observed in standard methods. This performance gap stems primarily from the failure of traditional approaches to eliminate historical biases, leaving behind roughly one-third of the initial raw model error alongside severe spatial outliers, whereas IDS always ensures zero historical errors. Furthermore, while conventional methods introduce unpredictable artifacts, including artificial reversals in model trends and compounding errors, the skill and uncertainty of IDS remain strictly tied to climate model performance. Ultimately, by avoiding the BA of the daily variable distribution, IDS delivers massive computational speedups ranging from x1.2 to x12, depending on benchmark. Overall, IDS offers a transparent and computationally efficient framework that preserves observational baselines and model projected changes, while improving indicator level performance and consistency across the evaluated applications, making it a promising approach for large-scale climate indicator production and climate risk applications.