When and where higher-resolution climate data improve impact model performance

TL;DR

The paper investigates how the spatial resolution of climate forcing data influences climate impact-model performance across sectors using the ISIMIP3a framework and CHELSA-W5E5 v1 data at four resolutions. It finds that the largest performance gains occur when moving from coarse (≈60 km) to intermediate (≈10 km) resolutions, with temperature-driven outcomes and topographically complex regions benefiting most, while precipitation-driven and low-relief systems show variable responses. The results underscore the need for careful benchmarking of resolution choices, improvements in downscaling and process representation, and ensuring high-resolution validation data to translate added detail into skill. The study provides practical guidance for data providers and modelers on where and how to invest in higher-resolution climate data to maximize impact and policy relevance.

Abstract

Climate impact assessments increasingly rely on high-resolution climate and forcing datasets, under the premise that finer detail enhances both the accuracy and policy relevance of projections. Yet systematic evaluations of when and where higher resolution actually improves impact model outcomes remain limited, and it is unclear whether increasing spatial resolution consistently enhances performance across sectors, regions, and forcing variables. Here we show that gains in climate input accuracy and impact model performance are largest when moving from coarse (60 km) to intermediate (10 km) resolution, while further refinement to 3 km and 1 km yields more modest and inconsistent benefits. Using cross-sectoral simulations from the Inter-Sectoral Impact Model Intercomparison Project, we find that higher resolution substantially improves model skill in temperature-sensitive impact models and topographically complex regions, whereas precipitation-driven and low-relief systems show weaker and less systematic improvements. For temperature, both climate inputs and model outputs improve most strongly at the 60 km to 10 km transition, with diminishing gains at finer scales; for precipitation, some models even exhibit reduced performance beyond 10 km. These results highlight that optimal resolution depends on sectoral and regional context, and point to the need for improving model process representation and downscaling techniques so that added spatial detail translates into meaningful skill gains. For data providers, this implies prioritizing resolutions that maximize improvements where they matter most, while for modelling groups and users it underscores the need for explicit benchmarking of resolution choices in climate impact assessments.

Figures (5)

• Figure 1: Overview of participating ISIMIP sectors and models, and location/geographic extent of the performed simulations.
• Figure 2: Evaluation of CHELSA-W5E5 v1 climate forcing data at multiple spatial resolutions over the European Alps. Panel (a) shows CHELSA-W5E5 v1 temperature and precipitation at four spatial resolutions. The first column displays absolute values at 1800$^{\prime\prime}$ resolution; subsequent columns show differences relative to the 1800$^{\prime\prime}$ baseline. Mountainous regions, such as the European Alps, are apparent in the first column of (a) as areas with lower temperatures (bluer shading) and higher precipitation (darker blue shading). Panels (b–c) show evaluation against GHCN-D station observations, with dot size indicating the number of daily measurements available per station. The first column presents absolute error metrics for 1800$^{\prime\prime}$ resolution; the remaining columns show relative changes in error compared to this baseline. Panel (b) corresponds to temperature, and panel (c) to precipitation. For KGE in the relative-change panels of (b) and (c), the color scale shows changes in the absolute KGE value (i.e. movement toward or away from the ideal value of 1).
• Figure 3: Comparison of NRMSE (interquartile range–normalized) and KGE across spatial resolutions for all participating ISIMIP sectors and models using sector-specific variables and sector-specific temporal scales (daily, monthly, or annual) for the evaluation (see Table \ref{['tab:model_overview_part2']}). (a) NRMSE and (b) KGE results from impact model simulations, each evaluated against observational reference data using sector-specific variables. For NRMSE, cell colors indicate the percentage difference in error relative to the coarse 1800$^{\prime\prime}$ ($\sim$60 km) resolution (green indicates improvement), while absolute error values are displayed within each cell. For KGE, cell colors indicate the change in KGE relative to the coarse 1800$^{\prime\prime}$ resolution, with negative values indicating worse performance and positive values indicating improved performance. Colorbar limits are clipped to the 95th percentile; triangular shaped colorbar indicates values exceeding the scale. Panel (c) shows the mean topographic complexity at each model’s evaluation locations. We show the Terrain Ruggedness Index (TRI), which quantifies surface heterogeneity, with higher values (blue on colormap) indicating more complex terrain around the evaluation sites. As the labour sector model was evaluated globally, no TRI was computed.
• Figure 4: Example of spatial results for the tree species Larix decidua simulated with the TreeMig model from the forest sector. Simulated basal area from TreeMig is compared to observational data from the national forest inventory (LFI) across the four spatial resolutions (1800$^{\prime\prime}$, 300$^{\prime\prime}$, 90$^{\prime\prime}$, and 30$^{\prime\prime}$). Results show spatial variation, with notable improvements in simulation accuracy at higher resolutions (see NRMSE, MAE and KGE metrics in the lower left corner) - particularly in the eastern mountainous regions. White in the first row indicates species absence; in the second row white indicates no difference between simulated and observed basal area. Grey in both rows indicates missing LFI data.
• Figure 5: Per-sector comparison of percentage changes in NRMSE for model performance and climate accuracy across spatial resolutions at all model evaluation locations where climate accuracy reference data is available. Panel (a) uses GHCNd temperature station data, and panel (b) uses GHCNd precipitation station data for climate accuracy assessment. In both panels, the subplots from left to right correspond to resolution changes from 1800$^{\prime\prime}$ (approximately 60 km) to 300$^{\prime\prime}$ (approximately 10 km), from 300$^{\prime\prime}$ to 90$^{\prime\prime}$ (approximately 3 km), and from 90$^{\prime\prime}$ to 30$^{\prime\prime}$ (approximately 1 km). Each data point represents a model-variable evaluation pair with co-located GHCNd or local station data.For spatial models, one point is shown per location with a co-located GHCNd station, which is why there can be multiple points for the same model–variable combination. The plots are divided into four quadrants: the green lower-left quadrant represents simultaneous improvement in both climate accuracy and model performance; the blue upper-left quadrant indicates improved climate accuracy but degraded model performance; the yellow lower-right quadrant indicates worse climate accuracy but better model performance; and the red upper-right quadrant indicates a decline in both climate accuracy and model performance.