| 23 Jun 2026
A 600-year gridded temperature dataset for East Asia based on Analogue Method
Abstract. Long-term gridded climate datasets are essential for for investigating the spatiotemporal variability and trends of regional climate. Developing reliable gridded reconstructions for the past few centuries requires preserving the spatial co-variability of climate variables while maintaining reconstruction efficiency. This study develops a gridded temperature dataset for East Asia (EA) spanning 1400–2000 CE, reconstructed using an improved Analogue Method (AM) based on climate proxy records and model simulations, at annual temporal resolution and 1°×1° spatial resolution. During the overlapping period of 1901–2000, the reconstructed mean temperature series is strongly correlated with instrumental observations (r=0.74, p<0.01). In addition, the leading empirical orthogonal function (EOF1) mode of the reconstruction is highly consistent with that derived from instrumental observations, indicating that the reconstruction captures the dominant mode of temperature variation over EA. The reconstruction further shows that thetemperature variations over the past 600 years can be divided into three phases: a cooling phase (1400–1510), a fluctuating cold phase (1511–1844), and a warming phase (1845–2000). The most rapid centennial-scale cooling and warming occurred during 1400–1500 (-0.31 °C/100 a) and 1900–2000 (0.58 °C/100 a), respectively. Spatially, temperature variability is strongest in the core region of the Siberian High. This dataset a valuable basis for understanding historical temperature variability and associated heat and cold extremes in EA, and for further examining long-term regional climate change. The dataset can open access on https://doi.org/10.5281/zenodo.18477496 (Yan et al, 2026).
Summary: The manuscript presents a reconstruction of air temperature over China during the past 600 years based on a pool of annually resolved proxy data, mostly dendroclimatological, and uses the analogue method with a pool of historical and past-millennium simulations from the CMIP6 simulation suite.
The analog meothod is slightly modified relative to the clasiscala analog metho, n which the distance between the target pattetn s and the analogue candidates is spatially weighted by the local correaltion between ecah proxy reecord and co-located correlations. This method has been previously published and applied to other reconstructions, so it is not the main contribution in this study.
The authors identify different temperature phases in China that broadly correspond to other northern hemisphere reconstructions published previously
Recommendation: I think that the study can eventually be published, but there are some important unresolved issues that the authors should address in a revised version. Other than that, the manuscript is clearly written, and the discussion and conclusion sections are informative.
Main points:
1) The reconstruction methodology is reasonable, but, as applied here, it has clear drawbacks, namely, it cannot produce uncertainty bounds. So the reconstruction shown in Figure 7 is just one time series without any uncertainties attached. This is not the state-of-the-art. Whereas I am aware that estimating uncertainties is to some extent subjective, as it may involve different sources of uncertainty, the authors should provide uncertainty bounds. For instance, the uncertainties may be related to the choice of analogue, the subsampling of the proxy network, the use of some model pools and not others, etc. The goal should not be to produce uncertainty ranges that take into account all possible sources of uncertainty - this would probably be too complex- but the method should be able to answer the question, for instance, of whether the temperature minimum in 1800 could be 0.4, or 0.5, or 0.6 instead of 0.3 K. The reader should have a sense of the degree of reliability the authors place on their reconstruction, given this methodology.
Perhaps one way to address this point is to compare the reconstructions with observations from the 20th-21st centuries. The authors show that the reconstruction underestimates trends and variability, so these biases and uncertainties should also be reflected in the 600-year reconstruction. The series of annual errors has a spectrum in the frequency space, so one way could be to extrapolate those amplitudes to the multidecadal and centennial timescales.
These are just suggestions, but this point must be somehow addressed.
2) That the method underestimates the variability, also in the instrumental period, is somewhat surprising, since the method selects just one analogue. Selecting several analogues and calculating the mean would indeed reduce variance, but selecting only the best analogue and assuming that the climate models are perfect should yield the observed variance, not an underestimated one. The reason is that the analogue method just resamples the simulated data.
The simulated data and the observations are all standardised to unit variance, but in the last step the analogues are re-scaled to the original variability of the (model or CRU? ) data. If the variance of the model data is smaller than that of the observations, the reconstructed variability is underestimated. This should be checked.
3) In addition to this underestimation in the observational period, an additional underestimation is expected, since most of the simulated data stem from historical (1850-present) simulations, and thus do not contain analogues that go beyond the observational period. How does the reconstruction change when only the past 1000 simulations are included in the analogue pool ?
So, points 2 and 3 together could lead to a too-flat reconstruction, as shown in Figure 7.
Particular points
3) Table S1 should specify the type of tree-ring data (ring-width, wood density,..early wood density, etc). The same applies to the ice-core data