# Z-socre of Central Asian loess dust accumulation rate

## Description

In this study, we collected chronological data of CA loess profiles according to the following criteria: (1) The profile must be composed entirely of typical aeolian loess without any secondary modification by humans or gravity. (2) The profile has independent OSL chronological data, and a small number of 14C ages must fall within the OSL age range. (3) Interval of the long-time scale profile (duration > 50,000 years) is less than 15,000 years, whereas the short-time scale profile is less than 10,000 years. If there is no age control point beyond the time range, we eliminate the unreliable age data or consider it has sedimentary hiatus. 43 loess profiles were selected to synthesize the DAR curves in CA. Among the 43 CA loess profiles, 37 of these profiles are with OSL chronological data, whereas 6 profiles are with 14C chronological data. We directly used the published data interpolated by Bayesian age-depth model, However, for profiles (n=32) without this model age,, we use the rbacon v2.3.9.1 package in R to produce the depth-age model for each profile. We establish the Bayesian age-depth model with resolution of 2 cm. We set “thick=10”, whereas acc.mean use the recommended value of the program (If the data overlaps with the age-depth model less than 90%, we will appropriately change the program parameters to ensure that the Bayesian age data is highly reliable.). We calculated the first-order derivative of the age-depth curve of each loess profile to get the DAR(m/kyr). Z-Score converts two or more sets of data into unitless Z-Score scores using (x-μ)/σ, which unifies data standards and improves data comparability. Therefore, we used the Z-score method to normalize the DAR data of each loess profile. The normalized formula is DARnormalized=(DAR-DARmean)/DARstandard deviation. According to studies during the past decade, Song et al. (2021) divided loess distribution in CA into three subregions Here, we used 0.5-kyr interval to synthesize the DARnormalized (DARn) data into three curves respective to the three subregions. Due to human activities influence DAR, we removed the upper 1kyr of these three curves.

## Files

## Steps to reproduce

In this study, we collected chronological data of CA loess profiles according to the following criteria: (1) The profile must be composed entirely of typical aeolian loess without any secondary modification by humans or gravity. (2) The profile has independent OSL chronological data, and a small number of 14C ages must fall within the OSL age range. (3) Interval of the long-time scale profile (duration > 50,000 years) is less than 15,000 years, whereas the short-time scale profile is less than 10,000 years. If there is no age control point beyond the time range, we eliminate the unreliable age data or consider it has sedimentary hiatus. 43 loess profiles were selected to synthesize the DAR curves in CA. Among the 43 CA loess profiles, 37 of these profiles are with OSL chronological data, whereas 6 profiles are with 14C chronological data. We directly used the published data interpolated by Bayesian age-depth model, However, for profiles (n=32) without this model age,, we use the rbacon v2.3.9.1 package in R to produce the depth-age model for each profile. We establish the Bayesian age-depth model with resolution of 2 cm. We set “thick=10”, whereas acc.mean use the recommended value of the program (If the data overlaps with the age-depth model less than 90%, we will appropriately change the program parameters to ensure that the Bayesian age data is highly reliable.). We calculated the first-order derivative of the age-depth curve of each loess profile to get the DAR(m/kyr). Z-Score converts two or more sets of data into unitless Z-Score scores using (x-μ)/σ, which unifies data standards and improves data comparability. Therefore, we used the Z-score method to normalize the DAR data of each loess profile. The normalized formula is DARnormalized=(DAR-DARmean)/DARstandard deviation. According to studies during the past decade, Song et al. (2021) divided loess distribution in CA into three subregions Here, we used 0.5-kyr interval to synthesize the DARnormalized (DARn) data into three curves respective to the three subregions. Due to human activities influence DAR, we removed the upper 1kyr of these three curves.