Assessment of surface water dynamics through satellite mapping with Google Earth Engine and Sentinel-2 data in Manipur, India

Surface water refers to water bodies such as ponds, lakes, rivers, and reservoirs (Liao et al. 2014). It is a valuable natural resource that supports human well-being through various uses such as domestic water, irrigation, sanitation, dam power generation, navigation, and industrial operations (Vörösmarty et al. 2000; Lepcha et al. 2021). Water, a vital natural resource, undergoes transformations influenced by climate change and various environmental and social factors (Rokni et al. 2016). Surface water plays a crucial role in hydrological, energy, and carbon cycles (Papa et al. 2023). Waterbody changes affect natural resources, assets, and the environment. Its fluctuations can have various consequences, including the potential for flooding, water scarcity, and other environmental impacts. Therefore, detecting, mapping, quantifying, and monitoring surface water are vital for adequate water reserve supervision and disaster preparedness (Huang et al. 2018). This means supervising its changing aspects on time is crucial for informing policy and decision-making strategies.

Surface water undergoes spatial and temporal fluctuations due to aspects such as environmental change, variations in land use land cover (LULC), and ecological shifts worldwide (Karpatne et al. 2016; Nyatuame et al. 2020). Consequently, an academic investigation has extensively focused on surface water monitoring for many years (Teklu et al. 2015; Park et al. 2017; Yang et al. 2017; Li et al. 2021a; Saghafi et al. 2021; Tran et al. 2022; Jiang et al. 2023). The rise and fall in surface water can trigger disasters, including floods, water scarcity in arid regions, and outbreaks of waterborne diseases, resulting in human casualties (Nguyen et al. 2022). As a result, timely monitoring of surface water changing aspects and data dissemination are essential for disaster prevention, mitigation strategies, and decision-making (Tralli et al. 2005; Giardino et al. 2010; Liu et al. 2013; Zhang et al. 2018; Sogno et al. 2022).

Understanding the dynamic nature of surface water is vital due to its susceptibility to changes over time and across different regions, driven by factors such as climate variations and LULC shifts (Palmer et al. 2015; Lepcha et al. 2024). This holistic perspective highlights the intricate features of the Earth's surface, prompting extensive research efforts in monitoring surface water dynamics, including significant contributions from several researchers (Lake 2003; Klein et al. 2015; Teklu et al. 2015; Feng et al. 2016; Park et al. 2017; Wang et al. 2023). Moreover, changes in surface water can give rise to disasters, including floods, water shortages in arid areas, and occurrences of waterborne health, potentially harming human beings (Feyisa et al. 2014). Nowadays, remote sensing data with different spatio–temporal and spectral capabilities have become indispensable to observe dynamic information related to Earth's various resources, particularly surface water.

Various mathematical models for surface water extraction have been established using existing data, remote sensing, and geographic information system (GIS)-based methodologies applied to observational datasets. Moreover, remote sensing and GIS-based approaches used for Earth observational datasets are now among the most widely used strategies. Rapid data may be retrieved, continually monitored, and analyzed to demonstrate various land resources at the lowest possible expense. Using remote sensing technology and methodologies, surface water supervision has been the most engaged pursuit throughout history. Traditionally, surface water mapping relied on visual analysis of satellite images from different periods. However, this approach faced challenges in maintaining continuity across chronological data series, making it hard to outline the extent due to seasonal pattern variations accurately (Mohammadi et al. 2017). Subsequently, numerous automated procedures were presented to extract surface water to address the abovementioned issues. These approaches utilize satellite imagery that exhibits diverse spatial and spectral attributes. These attributes encompass perceptron models, neural networks with multilayer perceptron (Balázs et al. 2018), entropy-driven techniques (Horkaew & Puttinaovarat 2017), extraction of spatial and spectral sub-pixel information through regression models (Kaplan & Avdan 2017), band-ratio analysis (Xu 2006; Singh et al. 2015; Deng et al. 2020), the fusion of density slicing with Hue, Intensity, and Saturation (HIS) transformation (Frazier & Page 2000), decision tree algorithms (Acharya et al. 2016; Olthof 2017), image segmentation, and pattern recognition (Chen et al. 2014; Xie et al. 2016a, 2016b).

Sentinel-2 is the second satellite in the European Commission and European Space Agency Copernicus program, providing global land surface multi-spectral photography with better spatio–temporal resolution. Sentinel-2 images demonstrate higher accuracy regarding water extraction through indices, particularly in delineating water bodies. The red, blue, green, and near-infrared (NIR) bands of the sensor have a 10 m spatial resolution, but the two shortwave infrared (SWIR) bands, which are often used for water index, have a 20 m resolution, somewhat reducing water extraction accuracy. The Normalized Difference Water Index (NDWI), which uses green and NIR bands with a 10 m ground resolution, showed significant misclassification errors, whereas the band sharpening approach achieved 10 m water precision but had a lengthy computation time. Li et al. (2021b) used four 10 m and two 20 m SWIR bands from Sentinel-2 for surface water mapping at 10 m resolution using Support Vector Machine (SVM). However, due to the small size of the training sample, the Multi-Spectral Water Index is challenging to apply universally.

Water extraction and monitoring through remote sensing leverage optical and radar data, primarily sourced from Earth observation satellites. Recent advancements focus on enhancing spatial, temporal, and spectral resolutions in data obtained from remote sensing of water bodies (Yan & Jinwei 2019). Optical imagery, a component of passive remote sensing, currently serves as the primary data source for extracting water body information due to distinctive low reflectivity in visible and NIR ranges and extremely low reflectivity in NIR and SWIR spectra. While diverse spatial resolution remote sensing data exists, fine resolution imagery such as Landsat and Sentinel has gained prominence for its increased accuracy.

Selecting optimal sensors is crucial for acquiring ground information, and often, a combination of two or more sources is preferred for water body extraction. Optical sensors offer eminent spatial and spectral resolution but are susceptible to cloud and shadow interference during water recognition. Synthetic aperture radar (SAR) images have some natural disruption, but they are still helpful for locations with shallow water and shadows. Combining textural features is one way to reduce the influence of noise, but it takes a lot of work and constant testing to find the best feature blend (Zhu et al. 2006). Furthermore, digital surface models and digital elevation models are important for water body estimates because they use terrain factors to reduce shadow interference and eliminate missing water bodies (Horkaew & Puttinaovarat 2017; Li et al. 2018).

In advancing the understanding of surface water dynamics, Yang et al. (2020) contributed insights by applying a rule-based superpixel approach on Sentinel-2 data, showcasing the integration of spectral index and superpixel techniques without training data, enhancing precision in surface water mapping. Halder et al. (2023) conducted a case study on flood hazard monitoring in Assam, India, using Google Earth Engine (GEE). They used Sentinel-2 data for landscape impact assessment, and Landsat 8 datasets for modified NDWI (MNDWI) within the GEE platform aligns with the scope of advancing methodologies in surface water dynamics. Dehkordi et al. (2022) examined spatiotemporal changes in Iran's surface water using Landsat data on GEE, highlighting the superiority of NIR-based water mapping rules in arid regions. Atay & Kaplan (2023) mapped Turkish inland water using a Multi-Band Water Index. Kandekar et al. (2021) analyzed the Jayakwadi dam in India, utilizing NDWI to monitor surface water dynamics within the scope of advancing methodologies. Li et al. (2022) contributed a review emphasizing the importance of higher spatial resolution data in analyzing surface water dynamics. Chen & Zhao (2022) demonstrated advances in surface water exploration by employing Sentinel-1 and Sentinel-2 data, achieving >96% overall accuracy. Singh (2022) utilized a SAR to analyze floods in Manipur, adding valuable insights to the scope of surface water dynamics. Sur et al. (2021) estimated Punjab's surface water dynamics using Sentinel-2, showcasing the effectiveness of the MNDWI.

Previous studies primarily focused on automated extraction and the utilization of single water indices, as discussed previously. In contrast, this study employs the delineation of surface water using multiple water indices in conjunction with the Otsu threshold and a Random Forest (RF) classifier for decision-making. Frequently, a synergistic approach involving both water indices and machine learning is adopted to achieve comprehensive and accurate results.

Various approaches have been exploited for extracting surface water such as density slicing in which an individual infrared (IR) band obtains a water map (Frazier & Page 2000). Alternatively, both supervised and unsupervised classification techniques (Ozesmi & Bauer 2002) generate LULC maps, from which the extent of water is obtained. Decision trees, utilizing multi-spectral bands, have also been employed for water delineation (Acharya et al. 2016), but the difficulty hinders their universal applicability in constructing robust classification rules.

A more effective approach involves using water indices calculated from multiple bands to discern contrasts between water and non-water bodies. Numerous indices, such as the Tasseled Cap Wetness Index (Crist 1985), Normalized Difference Vegetation Index (NDVI), and others, have been developed for this purpose. Notably, the NDWI (Mcfeeters 1996) and MNDWI (Xu 2006) have gained prominence. The MNDWI, replacing the NIR band with SWIR, proves more stable and reliable due to SWIR's reduced sensitivity to optical constituents within the water. However, caution is advised in snow-covered regions, where the normalized difference snow index might be more suitable.

The critical issue of thresholding in water index applications is addressed through various approaches. While a threshold of 0 is commonly applied, studies suggest that adjusting the threshold value enhances extraction results (Ji et al. 2009). Automation challenges arise when thresholding time series or single images covering multiple water bodies, prompting the development of automated methods. Li & Sheng (2012) proposed an adaptive thresholding scheme, later applied to MNDWI by Allen & Pavelsky (2015). Feyisa et al. (2014) pioneered the Automated Water Extraction Index (AWEI), offering a single-threshold solution for water detection from time series Landsat imagery, while Fisher et al. (2016) presented the WI2015 index, demonstrating comparable strength to existing water indices.

In Manipur, despite ample monsoon rainfall, water scarcity persists for agriculture, drinking, and domestic use due to inadequate management, ultimately affecting the rivers and ecological balance of the region. Most districts rely on surface water and groundwater, covering only 1.65% of the land area. Surface sources such as wetlands, lakes, ponds, and rivers are insufficient for sustainable development. Deforestation and shifting cultivation have led to the depletion of springs in hilly areas, further worsening the situation. These water sources are vital for economic and social development and environmental integrity. Some regions experience persistent water shortages, even during the monsoon, while valley districts face recurrent floods, causing damage. Improved water resource management is urgently needed for sustainable development and ecological preservation in Manipur.

The uniqueness of this study is to analyze the surface water dynamics in Manipur, especially during pre- and post-monsoon season extending from 2016 to 2021, using the Sentinel-2–2 data with the state-of-the-art approach in the GEE platform. Notably, no similar studies have been conducted in the region before. The study holds promise for providing fresh perspectives on surface water dynamics, seasonal fluctuations, and their significance for water resource management in Manipur. Furthermore, it may present inventive solutions and policy suggestions for addressing water scarcity challenges efficiently.

Tackling the growing scarcity of water resources intensified by climate change and human activities requires immediate attention. A crucial aspect of effective water conservation involves the thoughtful and sustainable management of water sources, necessitating consistent monitoring, mapping, and evaluation of water bodies. Conventional methods for updating national water maps often present challenges, but recent advancements in remote sensing provide viable solutions. In this study, we utilized remote sensing data integrated into GEE to develop an application for mapping the inland water bodies of the Northeast Indian state of Manipur. Based on the above discussion, the primary aims of this study are (1) to extract the surface water spread-out area of the selected districts of Manipur. (2) To analyze seasonal surface water changes during 2016 and 2021 in selected districts of Manipur. To achieve this objective, we explored different surface water extraction indices within GEE, capitalizing on Sentinel-2 satellite imagery and applying it comprehensively across the research area.

Sentinel-2, part of the Copernicus Land Monitoring program, has higher spatial resolution multi-spectral imaging capabilities designed for monitoring various aspects of the Earth's surface, including vegetation, soil, water bodies, and coastal areas. It comprises two main satellites, Sentinel- 2A and 2B, offering a rapid revisiting cycle of about 5 days. Sentinel-2 has a multi-spectrometer featuring 13 spectral bands covering the visible, NIR, and SWIR ranges. To extract surface water automatically, cloud-free images from the database are selected based on a thresholding technique outlined in the product algorithm (Richter et al. 2012). This procedure ensures the inclusion of high-quality data less than 10% cloud. An extensive 5-year database to average seasonal surface water dynamics was used. The specifications of the data used are provided in Table 2.

GEE is an open cloud platform for processing and analyzing geoscience data (Gorelick et al. 2017). This platform primarily stores satellite and Earth surveillance data while offering ample processing power for data transfer and processing, in contrast to conventional image processing software and Environment for Visualizing Images (ENVI) and Earth Resource Data Analysis System (ERDAS). GEE excels at rapid batch image processing, delivering analysis results without the need for image data downloads. The platform supports both an online JavaScript API and an offline Python API, facilitating the swift creation of web services via Google Cloud by utilizing these APIs.

The Sentinel-2 data from 2016 to 2021 underwent de-clouding and radiometric calibration processes. Cloud removal employs two methods: image-based cloud sorting and a pixel-based GEE algorithm. The former method opts for imageries with less than 30% cloud cover, while the latter calculates per-band percentile values from the pixels with the lowest cloud scores. Thus, the second approach prefers scenes with minimal cloud cover in areas with more scenes than the specified maximum limit. However, a drawback of this method is the persistence of visible residues around removed cloud patches, potentially affecting water extraction processes.

The surface water extraction involved employing key algorithms, including the NDWI, MNDWI, NDVI, and the AWEI. NDWI, focusing on the green and NIR bands, and MNDWI, emphasizing the green and SWIR bands, played crucial roles in capturing spectral distinctions between water and non-water features. AWEI specifically addressed shadow effects, enhancing overall water extraction accuracy. Additionally, supervised classification utilized machine learning for pixel categorization, with a set of known land cover types training the algorithm. The application of RF regression, a robust technique, facilitated efficient surface water mapping, especially adept at handling large datasets. These combined methodologies ensured a comprehensive and precise mapping of surface water extent in the study area.

The Earth's surface presents a diverse array of aquatic and terrestrial landforms, each characterized by unique complexities. Surface water features, including rivers, lakes, reservoirs, and seas, exhibit distinct shapes and spectral profiles. Conversely, non-water elements such as shadows, roads, and buildings may share spectral and spatial similarities, introducing potential confusion with actual water bodies. Consequently, achieving precise identification of water bodies remains imperative despite these challenges.

Optical image-based surface water extraction takes advantage of the lower reflectance of water, particularly in IR channels, compared to other LULC types. Various techniques tailored for this purpose have emerged in optical remote sensing imagery. A practical methodology involves computing water indices from multiple bands to discern contrasts between water and non-water bodies. In this study, we employed several methods – such as NDVI, MNDWI, AWEI, supervised classification, and RF regression – to map the extent of surface water.

The predominant approach for quantitatively analyzing remote sensing image data is supervised classification. This method entails identifying spectrally similar areas in an image using known target ‘training’ sites and applying their spectral signatures to unknown target areas. It necessitates using specific, homogeneous training data and implementing techniques or decision rules.

In order to instruct the computer to identify comparable patterns within imagery accurately, it necessitates high-quality training data. This constitutes only a fraction of the entire image or region to be classified. In this research, training samples were sourced from a notably high-resolution Google Earth Pro dataset, ranging from 15 cm to 15 m, to curate diverse classes for the training dataset. Approximately 100 sample points were gathered for each class to facilitate algorithm training. Additionally, field visits were conducted to validate the precision of the chosen training dataset.

The goal is to provide a quantitative description of how each thematic (LULC) class in the image appears. The purpose is to select numerous regions throughout the image for each LULC category. The type of training data used can substantially impact the classification outcomes. The first step must thus be determining the classes of interest, which necessitates careful consideration of the study's goal.

RF is a combined procedure developed based on a decision tree classifier. Using the bagging resampling method, multiple samples are removed from the original dataset, and for each bagging, the sample creates decision trees, and as the final output, it estimates the mean estimate of all decision trees. They overfit their training set when trees are full-grown and likely to learn extremely uneven patterns; it has significant variance but a small bias. RF is a method of averaging numerous deep decision trees, and the same training was done on different parts of the dataset, aiming to reduce variance. In the RF algorithm, every selected decision tree of the samples randomly generates the tree nodes.

The accuracy of the surface water mapping using the water indices was assessed over designated periods in different districts of Manipur. The PA, OA, and K coefficients calculated during the selected period from 2016 to 2021 for both pre-monsoon (PrM) and post-monsoon (PoM) seasons are listed in Table 3. The results suggest higher accuracy in mapping surface water extent over different districts of Manipur. The OA was found greater than 97% (minimum at Thoubal), PA with greater than 91% (minimum observed at Thoubal during pre-monsoon), UA with minimum of 93%, and above (minimum observed at Bishnupur during post-monsoon), and the minimum Kappa value with 0.96 at Thoubal (pre-monsoon) and Bishnupur (post-monsoon).

During 2016, it was observed that the actual areal coverage of water bodies increased during post-monsoon with reference to pre-monsoon season except in Imphal West district, which decreased by 4.66 km². On the other hand, a substantial increase was observed in Bishnupur and Kakching districts by 4.61 and 8.39 km². Similarly, during 2021, a considerable increase in water bodies was observed at Bishnupur and Imphal West district by 6.49 and 13.69 km² from pre-monsoon to post-monsoon season. The remaining stations have shown a small decrease, with a maximum of 1.65 km² at Churachandpur. Table 5 shows the quantitative surface water available in different districts of Manipur during pre- and post-monsoon during 2016 and 2021.

The change in water bodies from 2016 to 2021 during post-monsoon season revealed that substantial gain and loss were observed in Bishnupur, Imphal West, and Kakching districts. Bishnupur gained 19 km² from and lost 30 km² to non-water bodies; Imphal West and Kakching gained 18 and 13 km² from and lost 26 and 30 km² to non-water bodies. Relatively, only the Thoubal district has lost its water bodies by 19 km². It is worth mentioning that the overall change in water bodies during post-monsoon season from 2016 to 2021 revealed that the loss in areal coverage of water bodies is relatively higher than gain, which shows a trend towards the water scarcity scenarios in all districts of Manipur. Overall, the change in the remaining district was very small, as shown in Table 6.

During the year 2016, from pre- to post-monsoon season, a resultant gain in water bodies was (relatively higher than loss) observed at Bishnupur, Kakching, and Thoubal districts by 11, 8.5 and 6 km² while only Imphal West lost about 5 km² to non-water bodies. In general, the change in the remaining district was very small and negligible, as shown in Table 7.

During the pre-monsoon season from 2016 to 2021, it was revealed that substantial loss in water bodies was observed in Bishnupur, Churachandpur, Imphal West, Kakching, and Ukhrul districts. They lose about 11.4, 2.77, 25.98, 8.74, and 1.47 km² to non-water bodies. Overall, it is observed that the areal coverage of surface water bodies in all districts of Manipur is losing to non-water bodies during pre-monsoon season from 2016 to 2021, showing the way towards the water scarcity scenarios in all districts of Manipur as shown in Table 8.

Numerous studies have applied varied methodologies for surface water extraction, showcasing notable outcomes. Rapinel et al. (2019) employed a support vector machine and an RF approach in the Couesnon floodplain, France, favoring the former for superior performance. Chen & Zhao (2022) devised an algorithm based on topological relationships and a digital surface model local search algorithm, achieving heightened accuracy in delineating surface water across the plain and mountainous terrains along the Yangtze River, China. Mayer et al. (2021) utilized a deep learning approach to map surface water over Cambodia and Southeast Asia, generating results comparable to conventional methods. Goffi et al. (2020) mapped flooded areas in northern Italy through soft fuzzy integration, reporting consistently greater and robust mapping accuracy. Sur et al. (2021) employed an MNDWI with the Otsu method, yielding promising results in western India. In our investigation, the analysis demonstrated that, under consistent conditions, the RF classifier reliably extracted surface water extent over the Otsu threshold approach.

Traditionally, global-scale studies have been the focus of surface water mapping. However, there is now a growing need for higher-resolution mapping at the regional level (Sur et al. 2021) to better comprehend inundation patterns and formulate effective management strategies (Laonamsai et al. 2023). GEE offers the computational power and analytical capabilities necessary to process Sentinel-2 datasets and generates a detailed final product with an impressive resolution of 10 m. MNDWI, NDVI, Normalized Difference Build-Up Index (NDBI), AWEI-sh, and Bare Soil Index (BSI) algorithms were used on the GEE platform to extract surface water using an RF classifier for pre- and post-monsoon seasons in this study. The backscatter coefficient's median value was used to synthesize Sentinel-2–2 images of the study region during the pre- and post-monsoon in 2016 and 2021. The RF method was then used to generate water extraction results.

Verifications were done from Joint Research Centre datasets from 16 March 1984 and 31 December 2020 that are freely available in the GEE platform to check the proposed effectiveness of the framework. The proposed method has overall accuracy and a relatively high kappa coefficient. Meanwhile, related to the state-of-the-art method, the Otsu threshold approach in terms of qualitative and quantitative evaluations, the proposed water extraction framework employing the RF method can provide better performance. Argenti et al. (2013) emphasized the challenges of setting initial and automated thresholds. Speedy and less computationally intensive threshold segmentation methods pose difficulty due to dependence on various parameters such as wavelength, incidence angle, and dielectric characteristics.

Mapping surface water with Sentinel-2 in the GEE platform encountered issues with clouds and shadows, particularly in high-altitude regions. Cloud masking algorithms proved effective against clouds while removing mountain shadows using the AWEI-sh Index, and Digital Elevation Model (DEM) enhanced water continuity. However, selecting optimal textural elements for water extraction remained challenging, and mitigating shadow effects was complex. Overcoming water continuity issues in large-scale imagery was crucial. Zhang et al. (2018) highlighted the GEE platform's role in improving efficiency for large-scale and time series surface water dynamics studies. They proposed using level-set segmentation with Sentinel-1 data, reducing shadow and noise interference, leading to enhanced accuracy in water feature extraction, particularly in hilly regions.

About 90% of Manipur is hilly regions, so mountain shadows were challenging to remove directly from water extraction findings. Therefore, the DEM was utilized to construct a slope model, which was beneficial in eliminating mountain shadows. After initial water extraction calculations, the mean slope was computed, and the shadows were eliminated for slopes exceeding 15 degrees, as described by Li & Sheng (2012) and Jiang et al. (2021). Moreover, merging the open and closed procedures of the mathematical morphology model enhances the accuracy of small water bodies.

The analysis has identified a concerning trend in most districts of Manipur where water bodies are consistently shrinking (potential droughts in the study area for 2019 and 2021), whether during the pre-monsoon or monsoon periods from 2016 to 2021. This ongoing reduction in water bodies contributes to water stress scenarios across Manipur. Consequently, there is an urgent need for enhanced water resource management to promote sustainable development and preserve the regional ecological balance.

This study employs advanced remote sensing techniques to analyze surface water dynamics in Manipur, providing crucial insights into water resources management, sustainable development, and environmental conservation. By harnessing technologies such as GEE and Sentinel-2 datasets, the research sets a precedent for integrating technological solutions into regional resource management, enhancing accuracy and efficiency.

In water resources planning, the study delivers a high-resolution mapping of surface water dynamics in Manipur, offering vital insights into water availability and distribution. Identifying diminishing water bodies underscores the impact of climate change, emphasizing the necessity for adaptive strategies to mitigate water stress. Key findings include determining maximum and minimum surface water areas in various districts during pre-monsoon and post-monsoon periods, facilitating more efficient resource allocation by decision-makers. The observed decline in water bodies poses a direct threat, emphasizing the urgency of enhanced water resource management for sustainable use and regional well-being. Significantly, the negligible water loss in the Churachandpur district indicates a stable water ecosystem, guiding conservation efforts. The effectiveness of the RF classifier method in surface water extraction holds practical implications for land use and development decisions, fostering sustainability.

Recognizing the concerning trend of diminishing water bodies underscores the need for proactive measures to uphold ecosystem health and biodiversity. The study's results extend beyond scientific inquiry, providing tangible implications for addressing water challenges in Manipur. This information can inform policies, empower communities, and guide regional efforts towards a sustainable and resilient future.

Based on the study, the main conclusions are listed as follows:

• (1)

  Among the selected districts of Manipur, the maximum surface water area found at Bishnupur is 123.62 km², and the minimum surface water area was observed at Tengnoupal as 0.24 km² during pre-monsoon for the study period from 2016 to 2021.

• (2)

  For post-monsoon from 2016 to 2021, the maximum surface water area of 100.03 km² was observed at Bishnupur, whereas a minimum surface water area of 0.21 km² for Pherzwal district.

• (3)

  The findings of the change analysis revealed that the maximum water gain was 6% in the Bishnupur district during pre- and post-monsoon for the year 2016, and the minimum water gain was 9.5 × 10⁻³% in the Tengnoupal district for post-monsoon during 2016 and 2021.

• (4)

  The findings of the change analysis revealed that the maximum water loss was 7% in the Bishnupur district for pre-monsoon during 2016 and 2021.

• (5)

  Almost negligible water loss was found in the Churachandpur district during pre- and post-monsoon for 2016.

• (6)

  Compared to the Otsu thresholding method, the RF classifier method is profoundly satisfactory for surface water extraction.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.