Increasingly, scientists are using tree growth–climate relationships quantified from tree-ring data to spatially project climate change impacts on forests, from local to global scales. Local ring-width measurements are often combined through statistical aggregation to represent larger regions. Aggregation maximizes the common climate signal in tree-ring time-series while dampening local factors such that climate emerges as a strong predictor of variability in ring-widths, particularly at broad spatial scales. We ask whether spatial aggregation of ring-width data influences estimates of tree growth–climate sensitivity and biases projections of future forest response to climate change. We aggregated plot-level annual ring-width time-series of Douglas-fir in the southwestern United States, collected by the USFS Forest Inventory Analysis (FIA) program, and corresponding climate data from small (e.g., 40 km grain) to large (e.g., 600 km grain) spatial scales. We then modeled aggregated annual ring-widths as a function of aggregated annual warm season temperature and cool season precipitation, two important drivers of tree growth in the region identified by earlier research.
As aggregation scale increased, so did the proportion of variation in ring-widths explained by climate (coefficient of determination). This was not accompanied by an increase in the magnitude of climate sensitivity (regression slopes), which remained stable, on average, across aggregation scales. Rather, increasing aggregation scale resulted in a strong reduction in the variability of regression slope estimates. The fine scale heterogeneity in climate sensitivity was partially explained by climatic conditions (normals) that vary geographically, with greater tree growth–climate sensitivity at warmer and drier locations. While we did not find an effect of aggregation on the average change in ring-width predicted under future climate scenarios, the decreased variability of climate sensitivity estimates at large spatial scales resulted in predictions that underestimate the variability of future tree growth response to climate. Detecting such landscape-level heterogeneity in climate sensitivity may be particularly important for understanding and managing forest resilience to climate change – i.e., identifying and actively managing climate refugia. Hierarchical models, which can draw inference from local tree growth and nest effects from multiple spatial scales, may offer a more robust alternative to widely used spatial data aggregation practices.