Mending and extending observational temperature series by linear and nonlinear regression

<p>While time series of meteorological measurements from land-based weather stations still represent one of the basic types of data employed in climate research, it not uncommon for these records to be incomplete, interrupted by periods of missing or otherwise compromised values. Such gaps typically need to be filled before a subsequent analysis can be performed, and records from other nearby measuring sites are frequently used for this purpose. In this presentation, results of central European daily temperatures estimation from other concurrent measurements by various statistical methods are showcased, with a particular emphasis on assessing potential benefits of application of nonlinear regression techniques. Using multi-decadal daily temperature series originating from a dense network of weather stations covering the territory of the Czech Republic, we show that while nonlinear regression does not always outperform its linear counterpart, it can substantially improve accuracy of temperature estimates for some target locations. The gain is shown to be especially prominent for sites exhibiting atypical behavior compared to their local geographic neighborhood, such as isolated mountain-based stations. In addition to regression-based restoration of compromised segments in the temperature records, use of this methodology for extending the temperature records beyond their original period of measurements is also discussed, as well as its potential for homogeneity testing.</p>

Comparison of homogenization methods for daily temperature series against an observation-based benchmark dataset

Homogenization of daily temperature series is a fundamental step for climatological analyses. In the last decades, several methods have been developed, presenting different statistical and procedural approaches. In this study, four homogenization methods (together with two variants) have been tested and compared. This has been performed constructing a benchmark dataset, where segments of homogeneous series are replaced with simultaneous measurements from neighboring homogeneous series. This generates inhomogeneous series (the test set) whose homogeneous version (the benchmark set) is known. Two benchmark datasets are created. The first one is based on series from the Czech Republic and has a high quality, high station density, and a large number of reference series. The second one uses stations from all Europe and presents more challenges, such as missing segments, low station density, and scarcity of reference series. The comparison has been performed with pre-defined metrics which check the statistical distance between the homogenized versions and the benchmark. Almost all homogenization methods perform well on the near-ideal benchmark (maximum relative root mean square error (rRMSE): 1.01), while on the European dataset, the homogenization methods diverge and the rRMSE increases up to 1.87. Analyses of the percentages of non-adjusted inhomogeneous data (up to 39%) and substantial differences in the trends among the homogenized versions helped identifying diverging procedural characteristics of the methods. These results add new elements to the debate about homogenization methods for daily values and motivate the use of realistic and challenging datasets in evaluating their robustness and flexibility.

Homogenization of Daily Temperature Data

This paper develops a method for homogenizing daily temperature series. While daily temperatures are statistically more complex than annual or monthly temperatures, techniques and computational methods have been accumulating that can now model and analyze all salient statistical characteristics of daily temperature series. The goal here is to combine these techniques in an efficient manner for multiple changepoint identification in daily series; computational speed is critical as a century of daily data has over 36 500 data points. The method developed here takes into account 1) metadata, 2) reference series, 3) seasonal cycles, and 4) autocorrelation. Autocorrelation is especially important: ignoring it can degrade changepoint techniques, and sample autocorrelations of day-to-day temperature anomalies are often as large as 0.7. While daily homogenization is not conducted as commonly as monthly or annual homogenization, daily analyses provide greater detection precision as they are roughly 30 times as long as monthly records. For example, it is relatively easy to detect two changepoints less than two years apart with daily data, but virtually impossible to flag these in corresponding annually averaged data. The developed methods are shown to work in simulation studies and applied in the analysis of 46 years of daily temperatures from South Haven, Michigan.

SPLIDHOM: A Method for Homogenization of Daily Temperature Observations

One major concern of climate change is the possible rise of temperature extreme events, in terms of occurrence and intensity. To study this phenomenon, reliable daily series are required, for instance to compute daily-based indices: high-order quantiles, annual extrema, number of days exceeding thresholds, and so on. Because observed series are likely to be affected by changes in the measurement conditions, adapted homogenization procedures are required. Although a very large number of procedures have been proposed for adjustment of observed series at a monthly time scale, few have been proposed for adjustment of daily temperature series. This article proposes a new adjustment method for temperature series at a daily time scale. This method, called spline daily homogenization (SPLIDHOM), relies on an indirect nonlinear regression method. Estimation of the regression functions is performed by cubic smoothing splines. This method is able to adjust the mean of the series as well as high-order quantiles and moments of the series. When using well-correlated series, SPLIDHOM improves the results of two widely used methods, as a result of an optimal selection of the smoothing parameter. Applications to the Toulouse, France, temperature series are shown as a real example.