A Norwegian Approach to Downscaling

TL;DR

The paper addresses the need for reliable regional climate information to support adaptation by proposing a Norwegian downscaling framework that fuses empirical-statistical downscaling (ESD) with regional climate models (RCMs) through a climate-downscaling paradigm. It introduces a hybrid PP-MOS approach using common EOFs (PCA) for predictors, a nine-level evaluation scheme, and a purpose-built data-storage format to handle large multi-model ensembles from CMIP; it also documents pdf-based downscaling of seasonal statistics and extremes with explicit formulas such as $T_{2m} \sim N(\mu_T,\sigma_T^2)$ and $Pr(X > x) = f_w \exp(-x/\mu)$ and $x_\tau = \alpha \mu \ln(f_w \tau)$. The key contributions include an open-source R package 'esd', RifS metrics for regional societal information, and demonstration of scalable extraction of ensemble statistics (e.g., 95th percentile from 151 CMIP6 runs in seconds) along with a framework for robust climate-change adaptation. Overall, the work provides a reproducible, region-focused downscaling pipeline that integrates physics-based reasoning, statistical theory, and large multi-model ensembles to support climate services and decision-making.

Abstract

A comprehensive geoscientific downscaling model strategy is presented outlining an approach that has evolved over the last 20 years, together with an explanation for its development, its technical aspects, and evaluation scheme. This effort has resulted in an open-source and free R-based tool, 'esd', for the benefit of sharing and improving the reproducibility of the downscaling results. Furthermore, a set of new metrics was developed as an integral part of the downscaling approach which assesses model performance with an emphasis on regional information for society (RifS). These metrics involve novel ways of comparing model results with observational data and have been developed for downscaling large multi-model global climate model ensembles. This paper presents for the first time an overview of the comprehensive framework adopted by the Norwegian Meteorological Institute for downscaling aimed at supporting climate change adaptation. A literature search suggests that this comprehensive downscaling strategy and evaluation scheme are not widely used within the downscaling community. In addition, this strategy involves a new convention for storing large datasets of ensemble results that provides fast access to information and drastically saves data volume.

Figures (6)

• Figure 1: Illustration of the evolution of the ESD downscaling strategy, where the upper part illustrates how the downscaling strategy diverged from a more traditional approach adopted in 1998, and the lower part shows a crude timeline.
• Figure 2: Illustration of pdfs and an ideal climate change for daily temperature and wet-day precipitation amounts, where the grey curve represents an original climate and the red curve a new climate. The change in these pdfs are specified in terms of the parameters $\mu$ and $\sigma$.
• Figure 3: Example of of evaluation levels 1--2 for a PCA-based ESD test for December-February $\mu_T$ for 310 locations in the Nordic countries, with ERA5 1950-2020 as predictor for calibration. The upper left panel shows the weights of the leading PCA, upper right panel shows the predictor pattern, the EOFs of the ERA5 temperature weighted by the coefficients from the multiple regression (after aggregation to coarser grid to better match the GCMs). The lower left shows the results of a five-fold cross-validation, and the lower right presents the observed and predicted trends. The prediction of trend involved detrended data for calibration and original data as input to give the output.
• Figure 4: Example of evaluation levels 3--5 for downscaling results for December-February $\mu_T$ for Oslo, based on calibration with ERA5 reanalysis, common EOFs, and 3 leading PCA applied to 310 ECA&D time series of daily temperature from the Nordic countries (accounting for 99.7%). Here the downscaled results are based on 151 CMIP6 SSP370 runs with coverage 1951-2050. The trend statistics from downscaled ensemble is consistent with observed trends, but the downscaled interannual variability is somewhat reduced. The cross-validation correlation coefficients for the three PCs and the ensemble of common EOFs associated with each CMIP run were within the ranges 0.98--0.99, 0.73--0.91, and 0.49--0.77 for each respective modes.
• Figure 5: Example of evaluation levels 9 for downscaling results for December-February $\mu_T$ for Oslo. The left panel shows how well a normal distribution fits the daily winter temperature whereas the right panel shows a similar test for the parameter $\mu_t$ used for calibration. The fit to normal distribution is not perfect, but nevertheless a useful approximation. The largest divergence is founds near the tails, which is not surprising.