1. Introduction
In recent years, with population growth, economic development, and industrial progress, the emission rate of greenhouse gases has intensified, impacting the carbon cycling processes within terrestrial ecosystems [
1]. Climate warming has become an undeniable reality [2, 3]. Wetland ecosystems are a vital component of terrestrial ecosystems, playing a crucial role in regulating runoff, mitigating floods, and improving water quality. They are of paramount importance for maintaining ecosystem stability and preserving biodiversity [
4,
5,
6]. Since 1900 AD, influenced by hydro-meteorological natural factors and human activities, global wetland area has decreased by 50%, as a result, monitoring wetland dynamics and exploring the causes of wetland dynamic changes have become essential aspects of wetland research [7, 8].
Remote sensing-based wetland land cover mapping has emerged as an effective means of wetland dynamic monitoring. Studies combine field observation data, high-resolution remote sensing satellite data, visual interpretation, and pixel-based and object-based features to achieve land cover dynamic monitoring [
9]. In the domain of wetland remote sensing mapping, for instance, Mao et al. [
10]. constructed a national sample library of various wetland categories, coupled with object-oriented methods, to achieve refined mapping of 30-m resolution wetland types in China. Gong et al. [
11] utilized Landsat data and automated classification methods to achieve fine classification mapping of 30-m resolution wetlands in China for 1990 and 2020. However, single-category land cover datasets cannot explore the transitions between different land cover types and the potential ecological and environmental issues they may raise [
12]. In the field of multi-category land cover mapping, Chen et al. [
13] used global 30-m Landsat TM5/ETM+/OLI multispectral images and HJ-1 multispectral images, employing the Pixel-Object-Knowledge (POK) classification method, to achieve land cover mapping for the years 2000, 2010, and 2020. Due to its high classification accuracy and resolution, Globe Land 30 (GLC-30m) has been applied in land cover resource assessment and evaluation of land cover products [
13]. For instance, Zhang et al. [
14] assessed the roughness of underlying surfaces in the Yunnan Datuan Mountain wind farm using the 2010 GLC-30m data, thereby enabling wind energy resource assessment. However, it has limitations in terms of its longer time cycle, which makes it unsuitable for monitoring rapidly changing wetlands. Zhang et al. [
15] utilized Landsat TM/ETM+/OLI satellite data from 1984 to 2020, employing the random forest method on the Google Earth Engine platform to produce a global 30-m refined land surface cover product (GLC_FCS30) for 1985–2020.
While the GLC_FCS30 has a higher average overall accuracy compared to GLC-30m, it also provides more diverse land cover types and has a shortened cycle of 5 years [
15]. With 29 land surface cover types, this product is often used in the analysis of spatial and temporal change characteristics in regions with diversified land surface vegetation cover. For instance, Wang et al. [
16] analyzed the reasons for mutual conversion between secondary land cover types of farmlands and woodland based on the 2000–2020 GLC_FCS30 dataset and socioeconomic data from Myanmar. However, there is still a need to address continuous time-series land dynamic evolution analysis. To rapidly and accurately analyze the response of land cover to climate change, the European Space Agency has produced the global land cover dataset (ESA_CCI) with a time scale of 1992–2020 [
17]. Due to its advantages of long-time scale, diverse land cover types, and continuous updating, ESA_CCI is widely used in the production of refined land cover type products and long time-series dynamic change monitoring in the field of remote sensing mapping [
18]. The China Land Cover Database (CLCD) developed by Yang et al. [
19] and the CLC_FCS30 training dataset [
15] are both extracted from ESA_CCI. However, the spatial resolution of ESA_CCI is 300 m, and uncertainties in land cover types from coarse resolution data may hinder our understanding of the dynamics of time-series land cover [
18]. Based on all available Landsat data on the Google Earth Engine (GEE) platform, Yang et al. [
19] utilized the 1980–2020 China Land Use Data Set (CLUDs), selected time-series invariant sample points, combined with long time-series NDVI and Google Earth to filter invariant sample points, producing CLCD. The CLCD product can detect more surface water and impermeable areas and is more suitable for fine-scale environmental and pavement process simulation research [
19]. However, China’s mountainous terrain, complex topography, diverse climatic zoning, varied vegetation cover, urban development conditions, and other complexities lead to varying accuracy of the land cover datasets in different regions [
20]. In summary, establishing a regional high-resolution and long time-series annual coverage product set is an effective approach to achieve more accurate land resource assessment and precise spatiotemporal change analysis. This is crucial for the response of terrestrial ecosystems to climate change.
The IPCC’s Sixth Assessment Report highlights that global climate warming has intensified the frequency and intensity of extreme climate events, with significant impacts on agriculture, water resources, and ecological security [
21]. Among the most severe meteorological and hydrological disasters, extreme drought events stand out. Drought, characterized by prolonged water deficiency, reflects the comprehensive influence of all climatic factors on the ecosystem [
22]. In recent years, regional drought events have become more frequent, such as the autumn drought in southwestern China in 2003, spring drought in Yunnan in 2005, and summer drought in Sichuan and Chongqing in 2006 [
23,
24,
25]. Land cover serves as the carrier of regional drought response. Regional land cover changes exhibit a strong correlation with the response level to drought. For instance, Li et al. [
26] analyzed the response of vegetation cover to drought based on land cover in the northern slope of the Tianshan Mountains, finding a positive correlation between vegetation cover and SPEI from 2001 to 2015. Mokhtar et al. [
27] investigated the impact of shifting surface evaporation and land cover alterations in southwest China in response to climatic factors. Their study revealed a significant correlation between, evapotranspiration, precipitation, and changes in vegetation ecosystems, indicating feedback mechanisms influenced by climatic variability [
27].
Currently, drought events and severity are evaluated using drought indices. Commonly used drought indices include the Meteorological Drought Composite Index (MCI), Effective Drought Index (EDI), Palmer Drought Severity Index (PDSI), Standardized Precipitation Index (SPI), and Standardized Precipitation Evapotranspiration Index (SPEI) [
28,
29,
30,
31]. Due to different spatiotemporal scales of research, the applicability of different drought indices varies. For example, Yang et al. [
32] suggested that the PDSI index is best suited for China, while Feng et al. [
33] argued that MCI is more suitable for drought studies in northeastern China. SPEI, considering the impact of both precipitation and evapotranspiration, can characterize the long-term and broad-spatial-scale drought change trends of research subjects. In recent years, it has gained widespread application [34, 35]. Chen et al. [
36] and used SPEI to study China’s drought change characteristics from 1961 to 2012, finding an escalating trend of drought conditions in China. Wang et al. [
37] used SPEI to study the changing drought characteristics of different time scales in the Huang-Huai-Hai Plain of China from 1901 to 2015, identifying a clear wetting trend in the region.
The Qaidam Basin is located in the northwestern part of the Qinghai-Tibet Plateau, surrounded by the Kunlun Mountains, Qilian Mountains, and Altun Mountains. The basin contains marshes, water bodies, and glacier wetland resources, playing essential roles in water conservation, biodiversity protection, sand control, carbon sequestration, and regional climate regulation. Therefore, changes in wetlands within the Tibet Plateau have significant impacts on people’s livelihoods, economic activities, and ecological protection [
38]. In the last few decades, the dynamic changes in wetlands within the basin have intensified due to climate change and human activities [
39]. Therefore, studying the dynamic evolution characteristics of wetlands and their response to wet/dry climatic conditions in the basin holds great significance [
40]. Currently, there is limited research specifically focusing on the characteristics of climate change and drought and their effects on wetlands in the Qaidam Basin. Previous studies suggest that SPEI is particularly applicable in arid and semi-arid regions in northwest China [
41]. Furthermore, Bai et al. [
42] examined the temporal and spatial evolution characteristics of wet/dry conditions of SPEI and their effect on Lake Dalinor in the semi-arid Mongolian Plateau. Their analysis demonstrated the significant impact of wet/dry climatic conditions on the lake area and water level [
42]. Hence, in this study, we first used Landsat TM/ETM/OLI L2 surface emissivity images from June to August in 1990–2020. Employing the Google Earth Engine platform and the random forest method, we extracted wetland land cover types. Subsequently, based on meteorological data, we calculated monthly SPEI indices. By analyzing the dynamic characteristics of SPEI at different time scales, including duration, severity, intensity, peak, and the response degree of wetland dynamic changes to different time-scale SPEI, we explored the impact of drought on wetlands. The aim is to provide significant support for wetland protection within the basin and the sustainable development of the ecological environment in the Qaidam Basin. This study focuses on the following questions: (1) The dynamic evolution characteristics of marshes, lakes, and glacier wetlands in the basin from 1990 to 2020. (2) The characteristics of drought changes at different time scales within the basin. (3) The response degree of wetland dynamic changes in the basin to drought changes at different time scales.
Author Contributions
Conceptualization, A.F., W.Y., methodology, A.F., K.A.; formal analysis, A.F., W.Y., and K.A.; investigation, X.Y.; data curation, A.F., X.Y., and J.S.; writing—original draft preparation, A.F., Y.Z., and K.A.; writing—review and editing, K.A., B.B., A.A., and W.C.; visualization, A.F.; supervision, W.Y., and K.A.; funding acquisition, B.B., A.A., and W.Y. All authors have read and agreed to the published version of the manuscript.