Dynamic monitoring of open-surface water bodies in the middle reaches of the Tarim River based on the GEE cloud platform

Open-surface water bodies are a crucial component of global terrestrial ecosystems (Lu et al. 2023), providing essential water resources for both humans and natural systems. These surface water bodies, which include natural lakes, ponds, and rivers, as well as artificial reservoirs, ditches, and channels, serve as important water sources for terrestrial ecosystems, especially in arid inland areas (Song et al. 2024). They are key environmental factors for maintaining the stability of terrestrial ecosystems and supporting sustainable economic and social development (Yu et al. 2023). Furthermore, they provide the necessary resources for the survival of organisms and the development of human society (Jin et al. 2023; Kencanawati et al. 2023; Amindin et al. 2024; Tan et al. 2024). However, there has been a recent decline in global terrestrial water storage, particularly in arid and semi-arid regions where the natural water balance has been severely disturbed (Wang et al. 2018). The resulting changes in open-surface water bodies can lead to regional water shortages, unintended flooding, and other disasters (Ticehurst et al. 2014; Patel et al. 2015; Abbas 2023; Klein et al. 2024). Therefore, the rational regulation of water resources through riparian eco-hydraulic projects and the accurate, dynamic investigations of the spatial extent and time-series characteristics of open-surface water resources have attracted significant attention from scientific communities globally.

Remote sensing and GIS-based techniques utilizing Earth observation datasets have become popular tools. For many years, surface water monitoring has been a major focus of remote sensing tools and techniques globally (Abbas et al. 2023; Jiang et al. 2023). Two main methods are used for detecting open-surface water bodies from optical remote sensing images: sample-dependent machine learning methods requiring training data and sample-free indexing methods using mathematical formulas (Wang et al. 2018; Chen et al. 2024). Machine learning methods are emerging techniques characterized by strong noise resistance and high classification accuracy (Lin et al. 2022; Cui et al. 2024; Wang et al. 2024). However, these methods also have disadvantages, including strict training sample standards, high computational resource requirements, and high time consumption (Sun & Scanlon 2019).

Recent studies have leveraged the power of geospatial cloud computing (Gorelick et al. 2017) to integrate various indices for more effective water body monitoring. Zou et al. (2017) constructed a hybrid index rule set based on the MNDWI, the normalized difference vegetation index (NDVI), and the enhanced vegetation index (EVI) for automatic extraction of water bodies. This rule set does not require setting a threshold for the MNDWI and is more accurate than the existing MNDWI rule set (Zou et al. 2018a). Traditional indices, such as the normalized difference water index (NDWI), often produce significant noise due to the spectral similarity between open surface water and other landscapes like bare ground or urban areas, making differentiation challenging (Singh et al. 2015). Additionally, single indices may misclassify water bodies containing impurities, reducing extraction accuracy (Wang et al. 2023). To address these issues, we employed a mixed index rule set that integrates auxiliary data such as NDVI and EVI as constraints, which helps mitigate interference to some extent (Tamiminia et al. 2020). Furthermore, Jin et al. (2019) successfully addressed the issue of terrain shadows being misclassified by incorporating digital elevation models. This comprehensive approach using integrated indices has become a reliable strategy for monitoring open water bodies (Mayer et al. 2021). Compared with previous studies, this method represents a significant improvement, offering a more robust and adaptable solution for water body monitoring, particularly in complex landscapes where traditional methods may be less effective. Recent advancements in Google Earth Engine (GEE) have enhanced remote sensing capabilities for large-scale and long-term water body analysis. Pekel et al. (2016) utilized the GEE platform and water body extraction method to create a high-resolution global water dynamics dataset from 1984 to 2020, capturing various water characteristics. Zou et al. (2018b) analyzed U.S. open water surface trends using 200 TB of GEE-processed data.

Currently, extensive research has primarily focused on the impact of ecological water transfer on the restoration of severely degraded riparian vegetation along the Tarim River and the ecological water requirements of desert riparian forests. However, there has been limited study on how the regulation of water flow through these gates affects the distribution patterns of open-surface water bodies. To address this gap, this study focuses on the middle reaches of the Tarim River (MROTR) (between Yingbazha and Qiala) as the research area. Utilizing the GEE cloud platform and integrating Landsat-5/7/8 satellite imagery, this research employs a mixed index rule set for the automated extraction of water bodies to reveal the spatiotemporal dynamics of open-surface water area (SWA) over the past 30 years. The findings of this study can provide scientific support for water resource management in arid regions.

In 2000, the government initiated comprehensive management and protection measures for the Tarim River Basin, constructing water conveyance dikes and ecological gates along the mainstream, completed in 2009. The objective was to protect natural vegetation outside the dikes by promoting lateral overflow of the river (Qian et al. 2024). Along the MROTR, from Yingbazha to Qiala, 34 ecological gates were installed on the water conveyance dikes, with 11 gates designed for flows greater than 20 m³/s. We divided the Yingbazha to Qiala section into four subregions: S1 (Yingbazha to Shazihe), S2 (Shazihe to Wusiman), S3 (Wusiman to Arqike), and S4 (Arqike to Qiala).

• (1) Landsat series remote sensing data (https://landsatlook.usgs.gov/): Landsat 5, 7, and 8 surface reflectance Tier 1 images from 1990 to 2022 for the MROTR were selected. These images underwent atmospheric correction using the LEDAPS and LaSRC algorithms and were preprocessed to remove clouds, snow, etc., using the CFMASK algorithm. These products were accessed and processed on the GEE cloud platform, where batch operations were employed to calculate the annual NDVI and FVC from 1990 to 2022.

• (2) Global surface water data (JRC-GSW dataset https://global-surface-water.appspot.com/): Data from 1990 to 2020 were collected, including annual water history, maximum water extent, transitions, change intensity, and seasonality.

• (3) Sentinel-2 MSI remote sensing data (https://scihub.copernicus.eu/): Sentinel-2 MSI is a high-resolution multispectral imaging satellite with a revisit cycle of 10 days. The spatial resolution of its remote sensing data is 10 m, which is greater than that of the Landsat series satellites, and the imaging intervals are relatively short. High-resolution images with cloud cover of less than 10% were obtained through filtering operations on the GEE cloud platform, and these images were used for accuracy validation of the water body extraction results. Detailed information on all the data is provided in Table 1.

• (4) Hydrological data: Annual runoff data from 1990 to 2022 were collected at Yinbazha and Qiala hydrological stations. Annual runoff data from 2000 to 2022 were collected at Wusiman hydrological station and from 2005 to 2022 at Arqike monitoring section. Monthly data on ecological and agricultural water transfers at each ecological lock of the Tarim River main stream from 2013 to 2022 were also collected.

• (5) NASA's digital elevation model (DEM) data (https://search.earthdata.nasa.gov/).

• (6) Land use data for the Tarim River Basin from 1990 to 2020 were obtained from the Resource and Environment Science Data Center of the Chinese Academy of Sciences (https://www.resdc.cn/).

Based on the periods of dike construction and ecological sluice implementation along the Tarim River, three distinct phases can be identified: the no ecological gate (NEG) phase from 1990 to 2000, the ecological gate construction (EGC) phase from 2000 to 2010, and the ecological gate control water (EGCW) phase from 2010 to 2022.

The annual SWA from 1990 to 2022 was categorized into four levels: SWA < 200 km², 200 km² < SWA < 300 km², 300 km² < SWA < 400 km², and SWA > 400 km². We defined SWA I as dry years, SWA II as normal years, SWA III as wet years, and SWA IV as extremely wet years, as detailed in Table 3.

We analyzed the SWA for years with different water inflow levels. For SWA level II, the years 1993, 2001, and 2014 were selected for analysis. For SWA level III, the years 1996, 2005, 2013, 1998, 2006, and 2019 were analyzed. For SWA level IV, the years 1994 and 2022 were selected for analysis.

After 2010, with the completion of the Tarim River embankments and EGC, there was a reduction in ineffective water spillage within the 8 km range from the riverbank. This facilitated long-distance lateral water conveyance, increasing the SWA within the 23–35 km range during average water years and extending the conveyance distance up to 40 km during wet years, thereby increasing the SWA within the 10–40 km range.

This section will assess changes in the contribution of SWA in each subregion to the overall SWA, and analyze the correlation between river loss and SWA in these areas, evaluating the impact of ecological gate-controlled water on water distribution and river water resource allocation across different regions. An analysis of the correlation and contribution of the SWA and runoff loss in each subregion to the SWA of the MROTR revealed that the contribution of the SWA of the S1 region decreased from 15.79% during the NEG phase to 10.49% during the EGCW phase. The contribution of the SWA of the S2 region significantly increased from 29.86% during the NEG phase to 52.81% during the EGCW phase. In the S3 region, the contribution of the SWA decreased from 46.35% during the NEG phase to 24.41% during the EGCW phase. The S4 region's contribution of SWA to the MROTR first significantly increased from 8.01% during the NEG phase to 23.88% during the EGC phase and then decreased to 12.29% during the EGCW phase. Among the subregions, the SWA of S2 had the highest correlation with that of S3. The SWAs of the S2 and S3 regions were significantly impacted by ecological water diversion projects.

Currently, monitoring the dynamic changes in SWA using Landsat imagery and index threshold methods is the primary approach for studying open-surface water bodies (Pekel et al. 2016; Zou et al. 2018b; Wang et al. 2020). Although previous studies have proposed various indices to highlight water features (McFeeters 1996; Xu 2006; Wang et al. 2015; Fisher et al. 2016), their performance in mixed pixels in certain regions is not optimal, particularly for water extraction in arid environments. Therefore, some studies suggest establishing hybrid index rules between water indices and vegetation indices to address the errors in mixed pixels. Zou et al. (2017) proposed using a set of hybrid index rules to extract surface water bodies (using the rule of MNDWI > NDVI or MNDWI > EVI, and EVI < 0.1), which reduces the error of mixed pixels between water and other elements; this method has achieved promising results and has been widely applied by many scholars (Zou et al. 2018b; Zhou et al. 2019; Wang et al. 2020). Studies have shown that the simple and efficient hybrid index rule method for extracting water bodies is suitable for large-scale open-surface water body detection, as demonstrated in our MROTR study (Figures 4 and 5), this method effectively distinguishes between permanent and seasonal water bodies.

Accurate extraction and calculation of the SWA are crucial for ensuring the reliability of research results on the spatiotemporal characteristics of open-surface water bodies (Deng et al. 2020; Wang et al. 2020). In this study, all available Landsat images and hybrid index algorithms were utilized to calculate the annual water body frequency, defining water bodies with 25% < WIF ≤ 75% as seasonal and those with WIF > 75% as permanent (Zou et al. 2018b). The JRC-GSW dataset shows significant data gaps in specific years (before 1999), leading to anomalously reduced water areas in years such as 1997. The permanent water area in the JRC-GSW falls within the frequency thresholds of >65 and >75% (Xiao et al. 2024). Therefore, the SWA calculated in this study is smaller than that determined from the JRC-GSW data. This study validated the surface water extraction results of the 2019 Landsat time series data using high-resolution Sentinel-2 MSI imagery from 2019 to 2022 against the JRC dataset (Pekel et al. 2016). A confusion matrix was constructed to quantify the accuracy of water body extraction, with 670 samples correctly extracted, achieving an extraction accuracy of 93.71%. The classification results showed a kappa coefficient of 0.87 upon consistency testing (Table 2). Upon comparison, both the JRC dataset and the water bodies extracted in this study effectively depicted lakes and rivers. However, the details and noise in the results extracted in this study are more pronounced in smaller water bodies than in the water boundaries in the JRC data. The algorithm used in this study enabled more precise and comprehensive detection of changes in different types of open-surface water bodies, identifying more seasonal and small water bodies.

The primary purpose of regulating irrigation with ecological gates along the Tarim River basin is to increase the water flow into the lower reaches of the river, thereby improving water supply conditions to protect and restore the natural vegetation within the control area. However, the available ecological water volume required to ensure the protection and restoration of natural vegetation varies with different incoming water volumes. Therefore, water transfer through ecological gates, a unique method of water resource allocation and ecological restoration along the Tarim River, requires an understanding of the impacts of eco-hydraulic projects on open-surface water bodies based on the movement and spatial distribution of water resources in ecological control areas. From 1990 to 2022, the spatial and temporal patterns of the open surface water (SWA) in the MROTR differed across three phases: the NEG phase, the EGC phase, and the EGCW phase. During the NEG phase, the annual SWA increased from 79.36 to 285.01 km², with a change rate of 20.5 km²/year. In the EGC phase, the annual SWA decreased from 256.55 to 87.73 km² in 2009, at a rate of −16.88 km²/year. During the EGCW phase, the annual SWA showed a significant increasing trend, with a change rate of 28.8 km²/year. These trends aligned with the annual runoff consumption data from the Yingbazha and Qiala hydrological stations from 1990 to 2022. In the NEG phase, the average annual runoff consumption in the MROTR exceeded 20 × 10⁸ m³, with a considerable portion supplementing the P. euphratica forests outside the river channel through natural overflow. In the EGC phase, the river water could not overflow, reducing the average annual runoff consumption in the MROTR to 15 × 10⁸ m³, showing a linear decline. In the EGCW phase, the orderly regulation of water through ecological gates for both agricultural and ecological purposes increased the average annual runoff consumption to 24 × 10⁸ m³.

After 2010, the construction of water conveyance dikes and ecological gates in the Tarim River reduced ineffective overflow within 8 km of the riverbank, enabling long-distance lateral water transport. During normal water years, this increased the SWA within 23–35 km of the gate, while in high water years, the water transport distance increased to 40 km, with SWA extending 10–40 km from the gate. In these ecologically sensitive areas far from the river, water diversion through ecological gates created overflow conditions conducive to seedling growth (Ling et al. 2016). Previous studies, including Ling et al. (2016), Jiao et al. (2022), and Zhu et al. (2022), have shown that ecological water diversion supplements groundwater supplies and improves the growth conditions for vegetation in arid areas. In the MROTR, the construction of ecological water conveyance dikes reduced water usage and increased water transport efficiency. From 2000 to 2016, the cultivated area along the Tarim River increased by 4.12%, and water volume from 2011 to 2016 increased by 11.8% compared with 2000 to 2010 (Ling et al. 2019). Although water usage increased from 2000 to 2016, the proportion of water usage decreased by 33.4% after the construction of ecological water conveyance dikes (Ling et al. 2020). By improving the water supply and overflow conditions along the riverbanks, the construction of these dikes and ecological gates achieved suitable groundwater depths for vegetation growth (Ling et al. 2020). According to the river loss results of each subregion, the YBZ-WSM section accounted for 70.53% of the total river loss in the MROTR from 2000 to 2010, decreasing to 62% in the EGC and EGCW stages, indicating a reduction in water resource allocation in the S1 and S2 regions after the construction of eco-hydraulic projects. The WSM-QL section accounted for 29.47% of the total river loss in the MROTR from 2000 to 2010, increasing to 37.23% in the stages of completed water transfer embankment construction and ecological gate control. These data indicate that the construction of water transfer embankments and ecological gates is beneficial for reducing ineffective water loss, improving water resource utilization, and increasing water transfer to downstream river channels. The ecological water supply and overflow conditions for riparian forests should be improved constructing water conveyance dikes and ecological sluice gates on the banks of the Tarim River to produce a groundwater depth suitable for vegetation growth.

Riparian ecosystems are particularly sensitive to hydrological changes caused by ecological water management and related engineering projects. An earlier study (Ling et al. 2015) noted that the Tarim River's two to three annual floods, lasting 15–20 days, help promote seed germination and seedling growth in desert riparian forests. Therefore, real-time groundwater depth data on both sides of the river could be used to maintain groundwater depth at levels favorable for the natural vegetation's growth in the study area, thereby adjusting the flood discharge regime of ecological gates. The latest research (Li et al. 2024) proposed an ecological rotation irrigation model tailored to the ecological gate control zones and vegetation gradient patterns and evaluated the spatial match between the layout of ecological gates and ecological water requirements. However, there have been few studies in the Tarim River Basin that use more precise water extraction standards combined with the power of geospatial cloud computing.

Moreover, compared with previous studies, this research emphasizes the importance of using precise water extraction standards and the power of geospatial cloud computing in the context of ecological water management projects. These technologies offer greater accuracy in understanding the impact of ecological gates and water diversion dikes on open-surface water bodies. Although this study did not analyze the impact of SWA on the ecological environment by integrating ecological indicators and vegetation gradient patterns, existing research indicates that ecological water diversion positively influences vegetation restoration in the Tarim River Basin. Therefore, the findings of this study provide essential scientific support for ecological water resource allocation in arid regions and serve as a valuable reference for water resource management in other similar arid basins.

The mixed index method employed in this study demonstrated excellent performance in extracting open-surface water bodies in the MROTR, achieving an overall accuracy of 93.71% and a Kappa coefficient of 0.87. These results indicate that the mixed index method is highly reliable and practical for monitoring open-surface water bodies in arid regions. Between 1990 and 2022, there was a general increase in the open SWA in the MROTR, particularly after 2009. This trend suggests that ecological gates and water diversion dikes have played a significant role in improving water resource distribution and utilization efficiency, especially in the S2 and S4 regions, where ineffective overflow was substantially reduced, leading to enhanced water transfer efficiency. The study reveals the spatiotemporal distribution patterns of water resources in the MROTR, which is crucial for optimizing water resource allocation and improving utilization efficiency.

However, the spatial and temporal limitations of this study, coupled with the potential errors in water body extraction under extreme arid or complex terrain conditions, highlight the need for future research. Expanding the temporal and spatial scope of the dataset and optimizing algorithms to address varying topographic and climatic conditions are necessary. Additionally, integrating ecological indicators and climate models would provide a deeper understanding of the long-term impacts of ecological gates on water resources and ecosystem restoration, thereby facilitating the development of more effective water resource management and ecological restoration strategies.

This work was financially supported by grants from the Major Science and Technology Projects of Xinjiang Uygur Autonomous Region [grant number 2023A02002-1].

J.W.: Conceptualization, Methodology, Data curation, Writing – original draft. F.G.: Writing – review and editing, Supervision, Funding acquisition. B.H.: Writing – review and editing, Supervision. K.L.: Writing – review and editing. H.X.: Writing – review and editing, Supervision, Funding acquisition.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.