Holocene Dynamics of Vegetation Cover and Their Driving Mechanisms in Asian Drylands

0. INTRODUCTION

The Asian drylands, encompassing the northern East Asian monsoon region (NMA), the westerlies-dominated arid central Asia (ACA) and arid west Asia (AWA), are ecologically fragile areas and among the most sensitive regions to global change. These regions are significant dust sources of the Northern Hemisphere (e.g., Uno et al., 2009), playing a vital role in global climate change and marine biogeochemical cycles. Moreover, they served as crucial corridors for prehistoric human migration and transcontinental cultural exchanges, largely overlapping with the terrestrial Silk Road. Vegetation plays a crucial role in maintaining the ecological balance and ensuring the sustainable development of arid ecosystems (Berdugo et al., 2020). Against the backdrop of global warming, understanding the processes and mechanisms of vegetation cover changes and their responses to climate change is essential for achieving scientific vegetation management.

The Holocene is characterized by an abundance of geological archives, which provide critical foundations for climate and environmental reconstructions. Extensive precipitation reconstructions conducted in the Asian drylands have revealed distinct regional patterns. In the AWA (e.g., Chen et al., 2024; Ding et al., 2024) and the NMA (Li et al., 2020a; Chen et al., 2015), precipitation was higher during the Early to Middle Holocene, followed by rapid drying trend in the Late Holocene. In contrast, the ACA experienced a gradual increase in precipitation during the Holocene, reaching in the wettest conditions during the Late Holocene (Chen et al., 2022; Li et al., 2020b). This has led to a tripole spatial pattern of Holocene precipitation variations in the Asian drylands on the sub-orbital timescale, characterized by positive-negative-positive pattern across different subregions.

Previous studies have reconstructed vegetation cover in localized areas, yet it remains unclear how vegetation cover responded at a continental scale under the backdrop of significant spatial heterogeneity in Holocene precipitation variations (Tian et al., 2023; Sun et al., 2022). The Random Forest (RF) model has proven effective in construct the nonlinear relationship between pollen assemblages and vegetation cover, making it a widely used tool to reconstruct the past vegetation cover changes (Liu et al., 2024). This study based on the 5 379 modern pollen data and 42 published Holocene fossil pollen records, reconstructs for the first time the Holocene vegetation cover changes across the Asian drylands. Furthermore, it analyzes the processes by which vegetation cover responded to climate changes.

1. HOLOCENE VEGETATION COVER RECONSTRUCTION IN THE ASIAN DRYLANDS

For reconstruct past vegetation cover changes, we collected 42 fossil pollen records from NMA (13), ACA (11) and AWA (18), each spanning at least 10 000 years during the Holocene (Table S1). A total of 5 379 modern pollen samples with vegetation cover data were selected from 1 000 km around the fossil pollen records (Figure 1a). These samples are sourced from a new dataset we compiled using the Eurasian Modern Pollen Database (Davis et al., 2020) and the Asian Modern Pollen Dataset (Cao et al., 2022), which included 16 016 sampling sites and 245 terrestrial pollen taxa (Figure 1a). The modern vegetation data from the MOD44B version 6.1 vegetation continuous fields yearly product, with a spatial resolution of 250 m (DiMiceli et al., 2017). In ArcGIS 10.8 software, the vegetation covers for the modern pollen samples were extracted and calculated.

Figure  1.  Distributions of modern and fossil pollen data and paleoenvironment changes in the Asian drylands. (a) Map showing the spatial distribution of modern and fossil pollen sites (circle) were selected for this study. (b)–(d) The synthesized reconstruction results of pollen-based vegetation cover in AWA, ACA, and NMA, calculated at 200-year intervals using the Z-score normalization method. (e) Hydroclimate reconstruction based on average chain length (ACL) and (C₂₈ + C₃₀)/(C₂₂ + C₂₄) ratio of n-alkanoic acids from Almalou Peatland in the AWA (Chen et al., 2024). (f) The synthesized moisture of 27 moisture records from the ACA (Chen et al., 2022). (g) Pollen-based reconstruction of East Asian summer monsoon precipitation from Lake Gonghai in the NMA (Chen et al., 2015).

DownLoad: Full-Size Img PowerPoint

The relationship between pollen and vegetation is not linear due to variations in pollen productivity, deposition, and transport, as well as differences in the representation of pollen taxa within high-dimensional pollen data (Xu et al., 2016). The RF model is a machine learning method based on decision tree algorithm, which performs well in handling high-dimensional data and nonlinear problems (Breiman, 2001). We built the RF model between modern pollen percentages and vegetation covers and applied it to reconstruct past vegetation covers of fossil pollen percentages. To evaluate the performance of the RF model, we randomly selected 20% of modern pollen samples as out-of-bag data for testing. The decision tree was set to 500. The prediction was repeated 1 000 times by setting different 'seed' parameters. The RF model with the best coefficient of determination (R²) and root mean square error (RMSE) was applied to the prediction of fossil pollen data (Table S2). The RF models were carried out using R language (version 4.4.1) with the randomForest package (version 4.7–12; Breiman et al., 2022).

We constructed three RF models with robust performance for the NMA, ACA, and AWA, achieving R² of 0.60, 0.66, and 0.62, respectively, with corresponding RMSE of 13.89, 16.15, and 15.87. These RF models were subsequently applied to fossil pollen data to reconstruct past vegetation cover. The reconstructed results were subjected to Z-score standardization and integration at 200-year intervals to ensure consistency and comparability across data. The integration results reveal a general increase in vegetation cover across Asian drylands prior to the Early Holocene (~8 ka; Figures 1b–1d). During the Middle Holocene (~8–4 ka), vegetation cover in both NMA and AWA reached its peak. In the Late Holocene (~4–0 ka), vegetation cover in NMA and AWA exhibited a declining trend (Figures 1b, 1d). In contrast, vegetation cover in the ACA showed a continuous increase throughout the entire Holocene (Figure 1c).

2. HOLOCENE VEGETATION COVER DYNAMICS IN RESPONSE TO PRECIPITATION VARIATION

Further comparison of vegetation cover and precipitation records across NMA, ACA, and AWA reveals a significant correlation between vegetation cover dynamics and precipitation variation during the Holocene (Figures 1b–1g). In the ACA, vegetation cover exhibited a marked increase corresponding to the rise in precipitation transported by westerlies (Chen et al., 2022; Routson et al., 2019; Figures 1c, 1f). However, over the past few centuries, a marked decline in vegetation cover has been observed in the ACA region, which is likely driven by human-induced land cover changes. Similarly, in the NMA, a significant expansion of vegetation cover in response to enhanced summer monsoon precipitation during the Early to Middle Holocene (Chen et al., 2015; Figures 1d, 1g). In contrast, during the Late Holocene, a weakening of monsoon intensity led to a rapid decline in vegetation cover in the NMA (Figures 1d, 1g). This decline was further exacerbated by intensified human activities, which likely accelerated terrestrial environmental degradation and contributed to the continued loss of vegetation cover (Cao et al., 2022).

In the AWA, the peak precipitation during the Middle Holocene coincided to the highest vegetation cover, whereas a decline in precipitation during the Late Holocene corresponded to a reduction in vegetation cover (Figures 1b, 1e). Interestingly, despite relatively humid conditions, vegetation cover was at its lowest during the Early Holocene (Figures 1b, 1e). This apparent decoupling between precipitation and vegetation cover may be attributed to the seasonal distribution of precipitation. During the Early Holocene, precipitation in the AWA was mainly concentrated in the winter, with minimal or absent spring rainfall (Stevens et al., 2001). Consequently, despite higher total precipitation levels, the limited availability of water during the growing season (spring and early summer) likely restricted vegetation development, which may explain the asynchrony between precipitation and vegetation cover during the Early Holocene. The δ¹⁸O record from Lake Zeribar in Iran indicate a significant increase in spring precipitation after ~7 ka, which shortened summer droughts and promoted the development of forests and the expansion of vegetation into local arid areas (Stevens et al., 2001). Around ~6 ka, the precipitation regime transitioned to a modern-like pattern (Stevens et al., 2001), after which changes in vegetation cover and precipitation became more closely aligned.

This exploratory study reveals a strong correlation between vegetation cover and precipitation variations in the Asian drylands during the Holocene. Specifically, the increase or decrease in regional precipitation directly influences the expansion or degradation of vegetation cover. In the context of ongoing global warming, the Asian drylands face an escalating risk of desertification and extreme climate events, which are projected to disrupt the spatiotemporal patterns of regional precipitation. Hence, regions experiencing declining precipitation require focused attention through the implementation of targeted intervention, including water resource management, soil conservation, and artificial vegetation restoration, to mitigate the risks of ecological degradation.