The analog-method as statistical upscaling tool for meteorological field reconstructions over Northern Europe since 1850

Abstract

In this study, the analog-method is introduced, evaluated and applied as a nonlinear statistical upscaling tool to reconstruct multivariate atmospheric fields from long historical station records. The difficulty of currently available high resolution meteorological fields is their shortness mostly limited to the last up to 60 years. In contrast, a growing number of homogenized daily pressure and monthly mean temperature observations allow to study regional climate variations over Northern Europe back to 1850. The aim of this study is hence to develop a method which combines the advantages of numerically downscaled atmospheric fields by a regional climate model (RCM) for the last 50 years with statistical upscaling of long homogenized pressure and temperature observations since 1850.

This study shows that this aim can be achieved by the analog-method. Based on a set of long historical station records, the method searches for the highest pattern similarity between every time step in the past and its closest "analogue" stored in an archive of observations of the last 50 years. In this archive, corresponding fields to the station records are available from the RCM which has been forced by reanalysis data at its boundaries. Based on the maximized similarity of the analogous pairs, the method assumes that also the corresponding atmospheric states should be very similar. Atmospheric fields since 1850 are then reconstructed by taking the RCM fields from its closest analog in the last 50 years.

The first part of the thesis focuses on testing different settings of the analog-method to evaluate the robustness of the field reconstruction on daily to monthly scale. This involves changes in the number and spatial distribution of predictor stations, the size of the analog-pool, the stationary of the calibration period, the density of suitable neighbouring analogs, frequency distributions and the introduction of persistence into the reconstruction of daily variables. As a measure of skill, temporal correlations, mean bias and variance are estimated between the reconstruction and the RCM simulation. The validation confirms a good reconstruction skill with realistic statistical properties for variables with a strong physical link to pressure and temperature.

In the second part, the reconstruction is extended to the period 1850-2009. Low-frequency variations and long-term trends of annual storminess are compared to the novel 20th century reanalysis (20CR) since 1871 and different pressure-based storm indices over Northern Europe. The analog-reconstruction shows a highly realistic reproduction of the storm climate since 1850 compared to storm indices derived from homogenized pressure observations. In contrast, large inconsistencies in 20CR prior to 1920-40 lead to spurious long-term trends i.e. over open sea areas. The reconstruction confirms that no long-term trend exists in annual storminess while the observed NE-shift of storm tracks since the 1970s is unprecedented since 1850. Large decadal variations and a return to average conditions in the last decade indicate no systematic change so far.

The third part of the thesis provides a thorough validation of the temperature reconstruction's ability to reproduce spatial coherency of the trend patterns, low-frequency variations and trends of seasonal temperatures since 1850. While seasonal correlations are high with very realistic long-term trends, the evaluation suggests also a dependency on deficiencies inherited from the used RCM. A larger analog-pool and spectrally nudged RCM fields should allow here further improvements.

A brief overview on applications in chapter 5 confirms that the reconstructed atmospheric forcing fields allow a realistic simulation of the Baltic Sea ecosystem since 1850. While the evolution of eutrophication is clearly dominated by anthropogenic nutrient discharge, a reasonable reproduction of salinity, oxygen and water temperatures indicates realistic meteorological forcing conditions provided by the reconstruction. This is confirmed by a high agreement of simulated and observed sea-ice extents and a short evaluation of reconstructed vs. observed weather conditions connected to a strong inflow event into the Baltic Sea.

This thesis concludes that analog-upscaling provides an optimal tool to reconstruct physically consistent high resolution meteorological fields over Northern Europe based on sparse station records linked to the output of a RCM. Further improvements with the current settings should be mainly achieved by more realistic RCM fields e.g. by using spectral nudging. The results of this study motivate further applications of analog-upscaling in the context of paleoclimate reconstructions e.g. by directly searching for analogs in the RCM space of paleoclimate simulations.