全球变化知识资源中心

Nature Climate Change 6, 508–513 (2016); published online 7 March 2016; corrected after print 1 February 2017.

In our Letter we analysed changes in total annual precipitation (PRCPTOT) and extreme precipitation (Rx1day) in the dry and wet regions of the world as defined using precipitation amounts. The estimates of change have been partly biased due to the choice of reference period used for identifying wet and dry regions and normalizing the time series¹. Here we present our analyses corrected for these biases. While the corrections affect the exact quantification of precipitation changes, the biases are comparable to the uncertainties already discussed in our study, and our conclusions remain unchanged. The corrected results confirm that observations and CMIP5 models show robust increases in precipitation totals and extremes over the past 60 years in the world's dry regions, and in precipitation extremes in the wet regions, while there are only small changes in precipitation totals in the wet regions. Climate projections for the rest of the twenty-first century show that for precipitation over land these observed tendencies are expected to continue into the future, although there are substantial uncertainties about the magnitude of change, in particular in the dry regions.

In our original study, we used the sample mean of the first 30 years of the investigation period (that is, 1951–1980) to normalize the local time series of either precipitation totals or precipitation extremes, and to identify the dry and wet regions. This procedure, however, may artificially increase the spatial mean of the precipitation totals or precipitation extremes outside the climatological reference period because of variability over time in the sample means. Comparable biases related to normalization have previously been discussed for the analysis of temperature time series^{2, 3}. In addition, although 30 years is the recommended time length to calculate climatologies⁴, it may be too short to derive robust estimates of the spatial distribution of precipitation extremes. As a result, the calculated changes of the dry and wet regions' spatial aggregations may be affected by the 'regression to the mean' phenomenon^{5, 6}. This means that selecting from the dry part of the spatial distribution may result in a higher probability of wetter conditions outside the reference period, and similarly a higher probability of dry conditions when analysing the wet selection.

To avoid these biases, we use the entire 60 year period 1951–2010 as reference to identify dry and wet regions and to normalize the data. The corrected estimates of change over the 1951–2010 period are shown in Figs 1 and 2 of this Addendum. The increases in the dry regions are somewhat reduced in magnitude, but statistically significant (p ≤ 0.05) increases of 1–2% per decade still occur in both precipitation indices in the dry regions and in the Rx1day index in the wet regions, as reported in our original study. Changes in PRCPTOT in the wet regions remain comparatively small.