Temporal variation in glacier surface area and glacial lakes in glaciated river basins of Arunachal Pradesh

The cryosphere constitutes an integral aspect of the global climate system, exerting a significant impact on global change due to its interlinked influences (Kargel et al. 2014). The multifaceted and intricate phenomenon of climate change is best comprehended via substantial alterations occurring within the cryosphere (Kaushik et al. 2019). Glacier research is crucial for the systematic observation of the hydrological cycle and the evaluation of climate change impacts (Kaushik et al. 2019; Patel et al. 2019). Ice shelves and glacier fronts serve as crucial markers of glaciological processes, responsive to sensitivity and playing a critical role in the dynamics and overall mass equilibrium of ice sheets (Baumhoer et al. 2019). Glaciers retain essential information about the climate, historical climatic, and atmospheric disruptions. In the absence of direct communication, glaciers have been used as a proxy for measurements (Zhang et al. 2021). Changes in glaciers have a direct influence on the dimensions and occurrence frequency of lakes found in glaciated and high-altitude regions globally. The shifting dynamics of glacial and high-altitude lakes are at potential risk of glacial lake outburst floods (GLOFs), posing a significant threat to downstream communities (Worni et al. 2013; Peng et al. 2014). To effectively address this risk, it is recommended to prioritize the evaluation of lakes with significant flood volumes and the highest risk of self-destructive and dynamic failures (Rounce et al. 2017). By concentrating future efforts on these lakes, the focus can be placed on mitigating the most critical hazards. Because of their sensitivity to minute climate shifts, glaciers are acknowledged as crucial markers of climate change (Zhang et al. 2019).

Conventional ground-based methods are impractical for monitoring glaciers across extensive areas due to their remote and inaccessible locations in high-mountain terrain. Remote sensing technology is highly efficient due to its capability to offer detailed spatial and temporal data on various physical characteristics of the Earth's surface that can be effectively leveraged to map the extent of water bodies regionally or even globally and to closely track their dynamics at regular and frequent intervals (Huang et al. 2018) compared to traditional in situ measurement. In recent years, the extraction of more accurate glacier boundaries from satellite data has become a prevalent practice, particularly when assessing changes in glaciers (Paul et al. 2013). Leveraging satellite data to demarcate glacier limits is manifold. Many studies have been reported using these techniques for delineating clean ice glaciers (Bhambri et al. 2011; Mir et al. 2017; Alam & Bhardwaj 2020). Automated and semi-automatic methods exploit the distinctive spectral reflectance characteristics of ice and snow in the shortwave infrared (SWIR) and visible and near-infrared (VNIR) ranges to differentiate debris-free glacier margins from darker features like rock, soil, or vegetation (Zhang et al. 2019). The single band ratio approach incorporates the Red/SWIR ratio and the near-infrared (NIR)/SWIR ratio method, which detects glaciers efficiently in thin clouds and shadow zones; however, this method also tends to identify various water features, along with glaciers. On the other hand, the NIR/SWIR ratio precisely exempts most water features, is prone to overlooking locations with shadowing glaciers, and may misidentify shadowed vegetation as glaciers (Paul et al. 2013). To overcome these flaws in glacier mapping and to enhance the accuracy of glacier mapping by effectively differentiating glaciers located in shadowed areas and neighboring proglacial lakes, a new index called the automatic glacier extraction index (AGEI) has been devised by Zhang et al. (2019). This index incorporates a weighted average of the red and NIR bands, thereby improving the precision of delineating debris-free glaciers (Zhang et al. 2019). Several studies have proven that using various water indices from remotely sensed images to distinguish water bodies is effective in delineating surface water from its surroundings (Haibo et al. 2011; Rokni et al. 2014; Naik & Anuradha 2018). Multi-band techniques amalgamate different reflective bands to efficiently extract surface water. The normalized difference water index (NDWI) was specifically devised to extract water features from Landsat data, but NDWI cannot separate shallow water features from built-up land. A modified normalized difference water index (MNDWI) detects and enhances the water features for water regions within predominated built-up areas, vegetation, or soil features (Rokni et al. 2014; Mishra & Prasad 2015). The threshold method relies on differentiating water from other objects such as vegetation and bare soil based on the lower reflected radiance of water in the SWIR band. This method is crucial to determine whether pixels belong to a water class or not (Mishra & Prasad 2015; Huang et al. 2018). Pure water exhibits a unique radiometric response. Small ponds and other muddy and clogged waterbodies with suspended sediments behave like bare soils as a result of the rise in turbidity and the radiometric reactions (Lacaux et al. 2007; Gardelle et al. 2010; Bid & Siddique 2019). Images are then sorted using normalized difference turbidity index (NDTI) thresholds.

Several studies show that glaciers in the Indian Himalayas are experiencing rapid decline at an alarming rate, contributing to an overall loss in the frozen component of the Himalayan cryosphere indicative of climate change consequences (Kulkarni et al. 2005, 2007; Gurung et al. 2011; Kaushik et al. 2019). Because of the rocky and inaccessible alpine environment, Himalayan glaciers are typically challenging for field evaluation (Mir et al. 2017), making frequent monitoring and data gathering using field methods challenging (Mir et al. 2014). Arunachal Pradesh in the Eastern Himalayas is the region with the least glacier coverage in the Himalayas but it has an immense impact on the runoff regime and hydrology of the drainage region, which manifests the necessity of obtaining a data hub exclusively for watershed management of the region. Studies have used satellite images to create glacier and glacier lake inventories for various sections of the Himalayan region (Worni et al. 2013; Bhambri et al. 2018); however, Arunachal Pradesh, Eastern Himalayas, lacks an updated glacier, glacial lake inventory, and classification type (Mal et al. 2020). This study aims to employ a remote sensing method to gather glacier extent and glacier lake inventory in Arunachal Pradesh and analyze changes over a decade to fill the study gap.

The Mago River Basin, located in the Tawang district of Arunachal Pradesh, is marked by narrow undulating features extending from 92° 0′ 28.5″ E to 92° 0′ 28.5″ E longitude and 27° 53′ 17.5″ N to 27° 31′ 16.5″ N latitude at 2,438 to 6,443 m.a.s.l., with a drainage basin area of 841 km² originating from the Gorichen glacier in India. The Subansiri River Basin is the largest tributary of Brahmaputra River, extending between the latitude of 28° 55′ 25.1868″ N to 27° 25′ 58.6416″ N and longitude of 91° 33′ 31.9968″ E to 94° 48′ 46.5372″ E, with a drainage basin area of 32,640 km² between an elevation range extending from 99 to 6,692 m.a.s.l. originating near Mount Porom in Tibet, Himalayas. The Dibang River Basin, extending between a latitude of 29° 22′ 26.1876″ N to 28° 5′ 45.078″ N and a longitude of 95° 14′ 25.044″ E to 96° 38′ 49.0236″ E, courses along the Mishmi hill toward the south, with a drainage basin of 13,933 km² and elevation from 178 to 5,421 m.a.s.l.

For automated/semi-automated glacier mapping approaches, Landsat data offer enhanced accuracy, efficiency, and repeatability (Zhang et al. 2019). Because of the huge swath width (185 km), moderate spatial resolution (30 m), and extensive temporal record, Landsat data are commonly recognized as extremely valuable for glacier mapping within the wide array of remote sensing datasets available (Bolch et al. 2010). The most recent Landsat data (5 Thermal Mapper (TM) and 8 Operational Land Imager (OLI)) for least snow and cloud coverage were acquired from the United States Geological Survey (USGS) website (https://glovis.usgs.gov/) and processed for atmospheric correction and scanline correction resulting from any scanline error in the image. The Randolph Glacier Inventory (RGI) is a comprehensive collection of glacier outlines on a global scale. This dataset supplies a singular delineation for each glacier and is created in collaboration with the Global Land Ice Measurements from Space (GLIMS) initiative. The Landsat images obtained for the study area were mosaiced and clipped within the areal extension for each River Basin. Randolph Glacier Inventory version 6.0 released on July 28, 2017 (RGI v 6.0) for Southeast Asia was obtained from the GLIMS website (https://www.glims.org/RGI/rgi60_dl.html). Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) was acquired from the NASA website (http://search.earthdata.nasa.gov/) for the study area. The data are detailed in Table 1.

To extract the glacier area accurately in the study area, different remote sensing methods were applied and compared with each other through high-resolution Google Earth images and RGI v6.0 within the Mago River Basin.

The Landsat 8 OLI images, post-processing, featured a range of threshold values for various methods. These images were transformed into polygons, and glacier maps for the year 2017 were created. The generated glacier maps were imported into Google Earth with the closest available Landsat image acquisition date to ensure accuracy. A meticulous assessment was conducted to determine optimal threshold values for all the methods. The resulting glacier maps were collectively imported into Google Earth for a comprehensive comparative analysis to verify the glacier extent, as well as identifying and distinguishing water features and shadowed glacier areas. This comparison enabled us to identify the most effective remote sensing method.

Following the method selection, the glaciers for Mago Basin, Subansiri Basin, and Dibang Basin were generated using the chosen approach and manually delineated down to their snouts. The delineated glaciers underwent a validation process by cross-referencing them with the Randolph Glacier Inventory version 6.0 (RGI v6.0) for accuracy and consistency.

Generation of water bodies was carried out using NDWI, MNDWI, and NDTI indices along with the slope of the region. The indices used in the present study have the potential to demarcate the waterbodies individually. However, these individual indices could not efficiently bifurcate waterbodies from hill shadows, built-up areas, and sometimes glaciers. The present study utilizes the thresholding method to classify water bodies in the region. Thresholding or reclassification is a method to create a binary image in terms of foreground and background. To find out the threshold value, multiple sample points were created and pixel values for waterbodies were extracted. The extracted pixel value for water was then used for thresholding the indices to extract the area that has a high probability of water. In this study, the water spread area is considered as foreground, and non-water pixels are considered as background values. The resulting individual reclassified layers were then intersected with each other to estimate the final waterbodies. However, the present study is only concentrated on high-altitude lakes; therefore, waterbodies above 3,000 m.a.s.l. were considered. The Eastern Himalayan region has high topographical variations, and these variations create noises in the resulting waterbody layer due to hill shadows. To avoid these hill shadows, a percent slope was incorporated by considering that the waterbodies were present in relatively flatter areas. The size of the mapping of waterbodies is limited by the spatial resolution of the dataset (30 m in this case). The assumption was made that at least three continuous pixels both in the x and y directions are required to identify any object efficiently in an optical dataset. Therefore, in the present study, a 3 × 3 window size was considered while mapping the waterbodies. The entire process was performed in the ERDAS Imagine model builder tool.

Here, omission and commission error are the classification errors used for calculating the producer's and user's accuracy, respectively. Omission errors evaluate the pixels that are left out to classify in that particular class, whereas commission errors evaluate the pixels that are incorrectly classified.

Due to the restricted availability of suitable cloud- and snow-free sceneries, mapping glacier lakes using satellite imagery is a difficult undertaking in the Himalayas. It is critical to have reliable data on glacier surface area and to monitor changes regularly (Worni et al. 2013; Nie et al. 2017). The years for temporal change analysis were selected differently for the three basins based on image availability: 1988 and 2017 for Mago Basin; 2013 and 2022 for Subansiri Basin; and 1995 and 2021 for Dibang Basin. The Landsat data available were acceptable based on governing factors, i.e., cloud cover and snow cover.

The temporal analysis reveals distinct patterns in the number of glaciers falling within the ‘very small’ category (<0.5 km²). Both the Mago Basin and the Subansiri Basin displayed an increase in the number of these small glaciers. In the Mago Basin, the total area of such glaciers rose from 6.64 km² (19 glaciers) in 1988 to 6.41 km² (24 glaciers). A similar trend was observed in the Subansiri Basin, where the area increased from 5.17 km² (18 glaciers) in 2013 to 5.49 km² (21 glaciers). This phenomenon is primarily attributed to the deglaciation process, resulting in the larger glaciers melting and transforming into smaller ones. Conversely, the Dibang Basin exhibited a decrease in the number of ‘very small’ glaciers between 1995, when 32 glaciers covered 8.787 km², and 2021, when there was only 25 glaciers with a total area of 6.40 km². This decline is attributed to the basin's lower elevation, which is more susceptible to deglaciation, causing the melting and eventual disappearance of these small-sized glaciers.

The ‘small’ glacier category (0.5–1 km²) exhibited varying trends in glacier retreat within the Mago Basin and the Subansiri Basin. In the Mago Basin, there was a notable reduction from a total of 10.1 km² (14 glaciers) in 1988 to 5.72 km² (8 glaciers) in 2017. Similarly, the Subansiri Basin experienced a retreat from 10.89 km² (16 glaciers) in 2013 to 9.82 km² (15 glaciers) in 2021. These changes are primarily attributed to the process of larger glaciers melting and transforming into smaller sizes, particularly those falling below 0.5 km². Conversely, the Dibang Basin exhibited an upward trend, with the total area covered by ‘small’ glaciers increasing from 11.78 km² (16 glaciers) in 1995 to 13.95 km² (20 glaciers) in 2021. This trend is associated with the melting of larger glaciers into smaller ones, contributing to this expansion in the ‘small’ glacier category.

The medium-sized glaciers (1–3 km²) experienced a consistent retreat across all three basins. In the Mago Basin, the total coverage of such glaciers decreased from 25.85 km² (14 glaciers) in 1988 to 20.23 km² (11 glaciers) in 2017. Likewise, the Subansiri Basin saw a reduction from 21.32 km² (12 glaciers) in 2013 to 19.67 km² (11 glaciers) in 2022. The Dibang Basin displayed a similar trend, with the coverage of 1–3 km² of glaciers falling from 29.38 km² (19 glaciers) in 1995 to 20.47 km² (12 glaciers) in 2021. This widespread retreat is indicative of the glacier reduction trend across the basins.

Glaciers of large size (3–5 km²) exhibited a retreating pattern in both the Dibang Basin and the Subansiri Basin. The Dibang Basin, with 28.22 km² of glacier coverage (seven glaciers) in 1995, experienced an area decrease to 19.48 km² (five glaciers) in 2021. Similarly, the Subansiri Basin, with 9.68 km² of glacier coverage (three glaciers) in 2013, witnessed a reduction to 6.28 km² (two glaciers) in 2022. This decline is attributed to the melting of glaciers exceeding 5 km², leading to their transformation into smaller glaciers. The Mago Basin displayed an opposite trend, characterized by an increase in glacier size ranging from 3 to 5 km². The basin's glacier coverage expanded from 20.27 km² (five glaciers) in 1988 to 25.3 km² (six glaciers) in 2017. This growth is primarily attributed to the conversion of larger glaciers into smaller ones within the basin.

Large-size glaciers >5 km² diminished in Dibang Basin and Mago Basin such that the Dibang Basin had a coverage of 10.70 km² (two glaciers) in 1995 and Mago with 17.13 km² coverage (three glaciers) in 1988. Subansiri Basin experienced a retreat in the area coverage of large glaciers from 28 km² (three glaciers) in 2013 to 26.60 km² (three glaciers) in 2022.

Small glaciers (less than 0.5 km²) exhibited dominance and a clear upward trend. This trend was characterized by an increase in the number of small glaciers, while the number of medium, large, and very large glaciers experienced a relative decline in the context of our study within Arunachal Pradesh. This observation aligns with a previous study conducted by Borah et al. (2022), which also noted glacier thinning and a subsequent transformation of larger glaciers into smaller ones.

Glacial lakes are produced by or fed by glacier meltwater (Cao et al. 2016). When a glacier retreats or melts, it leaves behind a plateau lake called a glacial lake, which serves as a water resource. Song & Sheng (2016) categorized high-altitude lakes into categories: glacier-fed lakes, non-glacier-fed lakes, and upstream lakes. In this study, we differentiated the extracted high-altitude lakes into glacier lakes which are fed by glaciers and non-glacier lakes which do not originate from any glacier source. The study observed that few noises were present in the output after extraction due to shadows and clouds, which were removed manually.

Glacial lakes are a valuable supply of fresh water and an integral part of the alpine environment, but they also catalyze numerous glacier risks (Richardson & Reynolds 2000). The average lake expansion rate of 889.08 m²/year was seen in the Mago Basin between 1988 and 2017. The average lake expansion rate of 154.45 m²/year was observed for Subansiri Basin during the observation period (2013–2022), and 137.93 m²/year for Dibang Basin during the observation period (1995–2021). These lakes may pose a risk for GLOF in the basin's drainage regions and further study on the GLOF susceptibility and hazard classification of these lakes can be performed.

Glaciers are natural freshwater reservoirs as well as a dynamic element of energy and mass interchange on Earth. It is a significant element of the water balance and contributes to the global water cycle (Rai et al. 2017); therefore, obtaining precise information about glacier surface areas and monitoring their changes regularly is of utmost importance. However, conducting field measurements for observing glacier variations is incredibly challenging and often unfeasible at consistent intervals, particularly in rugged terrains like the Himalayan region. This study introduced glacier inventory within the study area by comparing and selecting the best glacier delineation method among various remote sensing methods, namely, NDSI, single BRs (NIR/SWIR, Red/SWIR), and AGEI, for assessing glacier extent using Landsat series images. Meticulous determination of the optimal threshold value for each method and comparison of their results with the Google Earth images was performed to identify the most effective remote sensing technique for our study area. The comparative analysis established that the AGEI method with a 0.5 weighted coefficient was the best approach for mapping glaciers. Subsequently, the glacier map was derived for Mago Basin, Subansiri Basin, and Dibang Basin using AGEI and manually delineated up to the snouts of the glaciers. The final validation was conducted by comparing it with RGI v6.0, revealing a strong agreement between the two datasets.

It is critical to have reliable data on glacier surface area and to monitor changes regularly (Worni et al. 2013; Nie et al. 2017). When employing low revisit images like Landsat 8-OLI, it can be challenging to choose cloud-free images (Zhang et al. 2021). Due to the limited availability of cloud- and snow-free scenes, mapping glacier lakes using satellite imagery is a difficult undertaking in the Himalayas. Landsat imagery resources are scarce within the region encompassing our study area due to the presence of cloud cover (>10%), along with substantial snow cover, which compounded the complexity and limited the research task in 2017, 2021, and 2022. Hence, the subsequent latest years were selected for Mago Basin, Dibang Basin, and Subansiri Basin, respectively.

This study showed that the AGEI method with a weighted coefficient (α = 0.5) and water bodies extraction indices (NDWI, MNDWI, and NDTI) proficiently extracted glacier boundary and high-altitude lakes, which were differentiated into glacier-fed lakes, in the glaciated River Basins of Arunachal Pradesh. Further analysis concluded an inference that glaciers are retreating, and glacier lakes are increasing in number and area due to the global scenario of climate change.

The Mago Basin, Subansiri Basin, and Dibang Basin observed glacier retreat at the rate of 1.08, 1.01, and 1.42% per year, respectively, over the study period. The Dibang Basin experienced the most retreat (178–5,421 m.a.s.l.) whereas the Subansiri Basin (2,304–6,409 m.a.s.l.) experienced the least retreat. Most glaciers span across elevations 5,000 and 6,000 m.a.s.l. (Zhang et al. 2021), and the least glacier coverage was observed in elevations above 6,000 m only in the Subansiri Basin. This indicates that the deglaciation rate is increasing inversely to the elevation range, such that lower elevation is experiencing maximum deglaciation. Comprehending temporal analysis of glacier size, it can be inferred that the larger-sized glaciers retreated to form medium-sized glaciers and smaller-sized glaciers. The small-sized glaciers (<0.5 km²) are dominant due to their location at lower elevations (4,000–4,500 m), which experienced maximum deglaciation and will be diminished in future years at a higher rate due to the rising global warming scenario.

The formation of glacial lakes is due to variations in topography, elevation, meteorology, climate change, rate of glacier change, and the presence of ice areas with gentle slopes (Mir et al. 2014). The glacier lakes inventory showed that the largest glacier lake area coverage (3.19 km²) in 2021 was in the Dibang Basin. The Dibang Basin consists of the highest number of glacier lakes due to its relatively lower elevation compared to the other two basins and higher deglaciation factor leading to glacier melt feeding glacial lake formed by topographical structures holding the feed. The Subansiri Basin with the highest elevation is composed of the least numbers of glacial lakes due to direct drainage of glacier melt to the streams. Retreating glaciers feed glacier lakes, leading to increasing tendencies quantitatively both in number and area extent at the rate of 3.31, 0.48, and 0.68% per year for the Mago Basin, Subansiri Basin, and Dibang Basin, respectively.

In Dibang Basin, glacier lakes and area coverage experienced an overall increase during the study period, but two small glacier lake (<0.1 km²) diminished and one large glacier lake (1.02 km²) notably retreated in 1995 to form a medium-sized glacier lake in 2021 due to the reduced feeding from the diminishing glacier.

These fluctuations will impact the pace of lake expansion and the likelihood of outburst events in each of the sub-regions in the future (Wilson et al. 2018). According to the study, the Eastern Himalayas' cryosphere is adapting to climate change via retreating glaciers and changing glacial lakes. Other studies have noted a similar pattern in the Eastern Himalayas (Shrestha & Joshi 2009; Yao et al. 2012; Che et al. 2014; Ye et al. 2017; Shukla et al. 2018; Debnath et al. 2019; Alam & Bhardwaj 2020; Mal et al. 2020; Borah et al. 2022). Drawing upon the study analysis, it is inferred that the changes in climatic conditions are the key factor responsible for the decrease in glacier coverage observed in the study area.

This study illustrates a comparative analysis of various remote sensing, namely NDSI, single BR (NIR/SWIR, Red/SWIR), and AGEI methods to delineate glaciers by employing Landsat image. The newly developed AGEI method, which excelled in the accuracy assessment, was used to delineate the glacier boundary within the study region to improve the accuracy of mapping glaciers at the end of the melt season. The Mago Basin, Dibang Basin, and Subansiri Basin of Arunachal Pradesh in the Eastern Himalayas were selected for the study. Weighted coefficient α ∈ [0,1] as 0.5 was assigned to obtain the optimal threshold for mapping glaciers by optimizing multiple thresholds between 1.5 and 2.5 to automatically obtain the glacier pixels and thereby the glacier map of the basins. The Mago Basin, Subansiri Basin, and Dibang Basin observed glacier retreat at the rate of 1.08, 1.01, and 1.42% per year, respectively, over the study period. High-altitude lakes were extracted using NDWI, MNDWI, and NDTI, which were further differentiated into glacier-fed lakes as glacial lakes. Water indices pixel value ranged from −1 to 1; the higher range of positive values observed in water indices indicate water pixel class, and negative values indicate non-water land cover types. The negative NDTI indicates clear water, whereas the positive NDTI indicates highly turbid water. The thresholding or reclassification method was applied to create a binary image in terms of foreground (water pixels) and background (non-water pixels) which compared each pixel of an image to an optimum threshold value and pixels with higher water area probability were extracted. Retreating glaciers are feeding glacier lakes, leading to ascending tendencies quantitatively both in number and area extent was observed at the rate of 3.31, 0.48, and 0.68% per year for the Mago Basin, Subansiri Basin, and Dibang Basin, respectively. Glacier recession, largely attributed to the consequences of warming global climate, has led to the formation and sustenance of glacial lakes, thereby contributing to heightened risk of GLOFs. Simultaneously, the diminishing or recession of glaciers has led to the eventual disappearance of glacial lakes, impacting water source availability in the region. The outcomes of this current study are anticipated to provide valuable insights for governmental agencies, policymakers, and officials, enabling them to make informed decisions and implement essential measures for the sustainable management of water resources within the region.

The authors gratefully acknowledge the help, encouragement, and financial support provided by the Climate Change Programme (CCP), Strategic Programmes, Large Initiatives and Coordinated Action Enabler (SPLICE), Department of Science and Technology, Govt. of India under National Mission on Sustaining Himalayan Ecosystem (NMSHE) through Grant No. DST/CCP/MRDP/184/2019.

The authors gratefully acknowledge the help, encouragement, and financial support provided by the Climate Change Programme (CCP), Strategic Programmes, Large Initiatives and Coordinated Action Enabler (SPLICE), Department of Science and Technology, Govt. of India under National Mission on Sustaining Himalayan Ecosystem (NMSHE) through Grant No. DST/CCP/MRDP/184/2019.

All data used in this study are openly available in the public domain.

ArcMap and ArcGIS Pro are used under license which needs to be procured from ESRI. ERDAS Imagine is used under license which needs to be procured from Hexagon Geospatial.

Rimum Murtem rendered support in data acquisition, data preparation, methodology, and original draft preparation. Sameer Mandal rendered support in data acquisition, data preparation, and methodology. V. Nunchhani rendered support in data acquisition, data preparation, and methodology. Megozeno rendered support in data acquisition, data preparation, and methodology. Arnab Bandyopadhyay rendered support in supervision, manuscript editing, and communicating, and as a grant recipient. Aditi Bhadra rendered support in conceptualization, supervision, and visualization.

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

The authors declare there is no conflict.