Detecting Anthropogenic Climate Change Signals in Global-Scale Water Cycles in the Presence of Natural Decadal-Multidecadal Variability: A Study with Four Global Coupled Models

Natural climate variability can lead to erroneous estimation of longer-term climate trends from data. This problem is especially relevant to water cycle variables, such as precipitation, because they are often constructed from satellite measurements, which at best span only a few decades. Since the goal of many climate projects is to infer long-term trends from available datasets, it is important to know the impact that natural climate variability can have on trend estimates.

At CRCES, we are creating a method to estimate the probability of inferring a long-term climate trend with a given accuracy from a relatively short decadal-multidecadal climate time series. The method depends on statistical comparisons between the trend deduced from a given century-scale time series and several trends computed from shorter decadal-multidecadal time series segments taken from that century-scale times series. Since century-scale global precipitation data sets do not exist, in our research we use long-term (80-year) global- and annual-average time series of precipitation from four Coupled Model Intercomparison Project 2+ (CMIP2+) models: PCM, GFDL, CCCMA, and CSIRO. The coupled model runs from which these time series are generated have atmospheric carbon dioxide (CO2) increasing at a rate of 1 %/yr over the entire 80-year period.

After computing trend statistics for the precipitation output of the four CMIP2+ model simulations, and then averaging the results, we find that a 10-year interval of precipitation output has less than a 10 % probability of being within 10 % of the 80-year trend (see Figure 1a). On the other hand, a 60-year interval of precipitation output has about a 50 % chance of falling within 10 % of the 80-year trend. Before we perform the analysis, if we remove the climate variations in the data that occur on time scales of two to eight years, similar skill of trend detection is found (Figure 1b). On the other hand, when we remove the climate variations in the output that occur on the decadal-multidecadal time scales (eight to 80 years), trend detection visibly increases, as shown in Figure 1c. For example, the probability of determining the 80-year trend with a 10-year time series in Figure 1c rises to about 20 %, compared to Figure 1a. Furthermore, all trends in Figure 1c estimated from 30-year time series or longer have a 100 % chance of being within 10 % of the 80-year trend!

Figure 1: For global-average surface freshwater flux (SurfFWF) due to precipitation, the fraction of trends determined from a given time series length - i.e. 10, 20, 30, 40, 50, and 60 years - that fall within specified error ranges of the 80-year time series trend. The analysis has been performed for CMIP2+ model time series that are A) unfiltered, B) filtered for variations between two and eight years (Low Pass), and C) filtered for variations between eight and 80 years (High Pass).

As revealed in Figure 1, variability on decadal-multidecadal time scales can severely impede our ability to utilize relatively short time series to resolve long-term trends of precipitation in four CMIP2+ models. Since the Earth's climate system has been observed to contain detectable natural decadal-multidecadal climate variability (e.g., Metha and Lau, 1997; Metha, 1998; and Scott et al., 1999), then it can be inferred from this analysis of the coupled models that understanding and prediction of this variability must be improved in order to enhance detection and quantification of anthropogenic climate change impacts. Furthermore, it has been revealed that long-term monitoring of the climate system is crucial to climate studies.

Metha, V. M. and K.-M. Lau, 1997: Influence of solar irradiance on the Indian monsoon-ENSO relationship at decadal-multidecadal time scales. Geophys. Res. Lett., 24, 159-162.

Metha, V. M., 1998: Variability of the tropical ocean surface temperatures at decadal-multidecadal time scales. Part I: The Atlantic Ocean. J. Climate, 11, 2351-2375.

Power, S. and Coauthors, 1999: Decadal climate variability in Australia during the twentieth century. Int. J. Climatol., 19, 169-184.