Using data assimilation to advance REACHES reconstructed temperature index data for east Asia.A new 0.5o x 0.5o REACHES-LME reanalysis temperature data set for East Asia since the 14th centurey

To understand and reconstruct paleoclimate, historical records are commonly used. In this study, we utilize the REACHES (Reconstructed East Asian Climate Historical Encoded Series) data, derived from Chinese historical documents, to derive new 0.5o x 0.5o latitude/longitude reanalyais temperature data since the mid-14th centuries. The REACHES reconstructed temperature index (four-point ordinal scale from -2 extreme cold to 1 warm) data covers more than 1,400 sites in the east China. But it has a significant flaw in a large number of missing values most likely reflecting normal weather (index value 0) and so produced a very biased and skewed distribution. To enhance the prediction for the missing data and improve REACHES data quality, we apply simple kriging to impute the missing data and set the mean of the underlying spatial process to zero (normal weather) to adjust for the missing patterns. To improve climate reconstruction accuracy in China, we propose a data assimilation approach by combining the REACHES reconstructed temperature index data with Last Millennium Ensemble (LME) reanalysis data. We propose a nonstationary time series model for the LME data and apply regularized maximum likelihood with a fused lasso penalty for parameter estimation. We treat the resulting distribution as the prior for historical temperatures, which are then updated to acquire refined temperatures based on the REACHES data using the Kalman filter and smoother. Overall, our approaches, which combine historical climate records, climate models, and statistical techniques, provide insights into past climate variations and enhance the accuracy of historic temperature estimation in east Asia.