Evaluating the influence of meteorological drought on river flow in the transboundary Sunkoshi basin, Nepal

Drought variation has been exacerbated by ongoing climate change and anthropogenic activities (Sheffield & Wood 2008; Haile et al. 2020). Drought has been seriously affecting people and their livelihoods globally (Wilhite 2000). It has been recognized as one of the sensitive environmental disasters affecting natural ecosystems, agriculture, and hydrological systems for long-term approaches (Mishra & Singh 2010; Jiao et al. 2016). Generally, drought represents conditions of below-average rainfall (McKee et al. 1993). Nearly half of the world has suffered different types and severities of drought (Berner et al. 2011; Masud et al. 2015). Drought has been classified as meteorological, hydrological, agricultural, and socio-economic droughts (Wilhite & Glantz 1985), and some are flash droughts in nature (Rakkasagi et al. 2023). A meteorological drought arises from prolonged precipitation deficits in a region and is often considered the initial indicator of drought, while an agricultural drought is characterized by insufficient soil moisture to meet the needs of specific crops (He et al. 2016). Hydrological drought occurs when surface and subsurface water supplies become inadequate (Van Loon 2015), and socio-economic drought affects people individually and collectively, causing disruptions in various sectors (Chand & Biradar 2017). Over the past few decades, drought has become particularly prevalent in the tropics and subtropics, and its frequency and duration are expected to increase under global warming by the end of this century (Sheffield et al. 2012; Dai 2013; Trenberth et al. 2014). Consequently, drought studies are essential from multiple perspectives, and effective drought monitoring is necessary for the formulation of adaptation and mitigation strategies (Sivakumar et al. 2014). To assess and monitor drought, several drought indices have been derived in recent decades such as the standardized precipitation index (SPI) (McKee et al. 1993), the standardized precipitation evapotranspiration index (SPEI) (Vicente-Serrano et al. 2010), the Palmer drought severity index (Palmer 1965), the crop moisture index (Palmer 1968), the soil moisture drought index (Hollinger et al. 1993), and the crop-specific drought index (Meyer & Hubbard 1995). Besides these indices, indices of Penman, Thornthwaite (Penman 1948; Thornthwaite 1948, 1963) and Keetch (Keetch & Byram 1968) have been used in limited cases (Hayes 1996). Additionally, remote-sensing-based drought indices, such as the normalized difference vegetation index (Rouse et al. 1974), the temperature vegetation drought index (Sandholt et al. 2002), and the vegetation condition index (VCI) (Kogan 1990), are also used widely as a part of ecological drought (Baniya et al. 2019). Among these indices, SPI and SPEI are the most widely used drought indicators, considering precipitation and potential evapotranspiration (PET) (calculated from temperature data) to determine drought, providing a more realistic representation of water availability under global-warming conditions (McKee et al. 1993).

Climate variability has been responsible for inducing and exacerbating drought conditions in Nepal for several decades (Wang et al. 2013). Climate change has also influenced streamflow through glacier melting in mountain river systems (Baniya et al. 2024). The changing climate triggers the expansion of glacial lakes that increase the possibilities of outburst floods (Gupta et al. 2023). While there have been some studies in the past to understand the spatial and temporal evolution of drought across Nepal (Sigdel & Ikeda 2010; Wang et al. 2013; Kafle 2015; Dahal et al. 2016; Khatiwada & Pandey 2019; Sharma et al. 2020a; Hamal et al. 2021; Sharma et al. 2021a), these studies differ in their geographical focus, temporal periods, and use of different drought indices. Nepal experienced drought in 1972, 1977, 1982, and 1992 and also frequent dry spells since 2002 (i.e. from 2004 to 2006 in both dry and wet monsoons) (Joshi & Dongol 2018). The drought over the last 33 years showed moderate to extreme drought intensity in Nepal (Sigdel & Ikeda 2010). The drought often occurred in mid and far-western Nepal (Kafle 2015) especially during the winter and summer seasons (Wang et al. 2013). The worst widespread drought in central Nepal was recorded in 2004, 2005, 2006, and 2009 in summer and 2006, 2008, and 2009 in winter (Dahal et al. 2016). Tree rings also revealed an intensified spring drought (Panthi et al. 2017) and spring and summer drought occurrences in the Trans-Himalayan region in Nepal (Gaire et al. 2019). Based on satellite-derived VCI, severe drought was identified in 1982, 1984,1985, and 2000 (Baniya et al. 2019). However, all these studies converge in revealing an increase in the severity and frequency of drought in Nepal. Many of these studies have also established a connection between precipitation variability, large-scale land–atmosphere circulations, and drought events (Wang et al. 2013; Hamal et al. 2020a, 2020b). For instance, Sigdel & Ikeda (2010) linked the cause of summer droughts with El Nino and the occurrence of winter droughts with a positive Indian Ocean Dipole Moment Index. Nepal experiences around 80% of its total annual rainfall during the monsoon season (June–September) (Nayava 1980; Sharma et al. 2020b). The impact of drought on agriculture is evident from previous years when drought conditions and natural calamities resulted in a significant decrease in agricultural production (WECS 2011).

The risk of frequency of drought and its effect on rainfed agriculture and socio-economic status is severe in response to climate change in the Koshi River basin in Nepal (Zhu et al. 2020). Water scarcity has been a great challenge in the middle mountains of Nepal (Baniya et al. 2019). Drying-up of rivers, insufficient water available for irrigation, and long queues at public water sources reveal that water is a scarce resource in the Koshi basin (Merz et al. 2003). After 2000, the temporal and spatial trends of drought have increased at annual and seasonal scales, especially in the hilly region of the Koshi basin (Dahal et al. 2021). The river basin receives approximately 900–1,200 mm annual precipitation, and the annual mean PET is approximately 1,500–2,000 mm, which indicates severe drought in the basin (Chen et al. 2013). The basin shows increasing weather extremes (Shrestha et al. 2017), and seasonal precipitation is unevenly distributed (Agarwal et al. 2014). The daily minimum and maximum temperatures and consecutive dry days are increasing in the basin (Shrestha et al. 2017). The monthly average maximum and minimum temperatures range from 27.0 to 32.0 °C and from 0.0 to 5.0 °C, respectively (DHM 2023). Approximately 70% of the population in the river basin relies on rainfed agriculture for their livelihood; therefore, the uneven precipitation distribution and increasing temperatures cause water stress in the basin (Dixit et al. 2009). The Sun Koshi basin, a sub-basin of the Koshi River, has also been identified as vulnerable to both frequent floods and droughts (Khanal et al. 2014). Much of the previous literature has used SPI and SPEI indices on national and basin scales to study drought. However, the connection between drought and stream flow (river discharge) in the basin has not yet been well studied. In this context, this study aims to analyze summer drought characteristics using SPI and SPEI at three-month and six-month scales and link them with river discharge during the period of 1985–2015. The drought studies in Nepal have been crucial to understanding the occurrence and severity of drought events. The increased intensity and frequency of drought do not only create water shortage but also severely impact on river discharge. Thus, this study provides a new dimension of understanding the relationship between drought and stream flow patterns in the river system of Nepal.

While the basin experiences South Asian monsoons in summer and westerlies in winter (Hamal et al. 2020a, 2020b; Sharma et al. 2020a), uneven precipitation (Sharma et al. 2021b) affects water availability. Temperature is higher in low-elevation areas, declining toward the north. Understanding the impact of drought on stream flow is essential for effective water resource management amidst changing climates. The large areas above 5,000 m asl were covered by the glacier in the upper basin (Khadka et al. 2023). In the Sunkoshi basin, we have selected nine meteorological stations based on their time-series data availability (Table 1).

This analysis is based on data collected from nine rainfall stations (station nos. 1202, 1219, 1102, 1103, 1016, 1009, 1024, 1027, and 1036) and four hydrological stations (station nos. 630, 652, 665, and 680) spanning the years 1997 to 2015. These six hydrological stations have been divided into 14 sub-basins using the HEC–HMS (Hydrological Engineering Center Hydrologic Modeling System) model, as shown in Figure 1.

This study utilized daily observed precipitation and temperature data from nine meteorological stations across Nepal for the period 1985–2015 collected from the Department of Hydrology and Meteorology, Government of Nepal. The details regarding the stations, including their unique identifier, name, district, geographic coordinates (longitude and latitude), and elevation, are presented in Table 1. For the analysis of streamflow, daily discharge data for 14 sub-basins within the Sunkoshi River basin was estimated. This estimation employed data from four hydrological stations and was performed using the Hydrologic Modeling System (HEC–HMS) model version 4.10 (Adhikari et al. 2023; Baniya et al. 2024) downloaded from https://www.hec.usace.army.mil/software/hec-hms/downloads.aspx. It is important to acknowledge the influence of the transboundary area of China on the Sunkoshi basin's flow. However, due to a lack of precipitation and temperature data in this region, a kriging interpolation method was implemented using Golden Surfer software to fill these missing values (Surfer 2023) downloaded from https://www.goldensoftware.com/. Furthermore, this research also used the ClimPACT2 software, an R package (WMO & ET-SCI 2016) available at https://github.com/ARCCSS-extremes/climpact2/, to calculate drought indices. Specifically, the SPI and SPEI were derived and missing data were addressed Data points were excluded if five or more consecutive years of data were missing (Alexander & Herold 2016).

Both SPI and SPEI have been widely adopted due to their simplicity, data availability, and usefulness in assessing different types of droughts, from short-term soil moisture deficits to long-term water resource depletions (Mishra & Singh 2010). SPEI, on the other hand, considers both precipitation and PET (ET₀) to calculate drought conditions, which involves the climatic water balance on different time-scales (Vicente-Serrano et al. 2010).

This departure of rainfall and discharge represents the difference between the actual R and Q and the average R and Q. Actual R refers to the amount of rainfall that occurred during a specific period, while average R signifies the typical amount of rainfall over an extended historical record for the same period.

This research investigates the relationship between meteorological drought and streamflow in a small transboundary Nepalese basin. Specifically, a hydrological model calibration was developed from the gauged basin using parameters for model calibration and validation. The Clarke method was applied to estimate the time of concentration from basin delineation (subbasin characteristics data) using the Kirpich equation. The storage coefficient was calculated as twice the time of concentration, and the baseflow was estimated using the recession method, with a coefficient ranging from 0.98 to 1, an initial value of 0.01 m³/s per km², and a ratio to peak of 0.1. The calibrated results showed a root mean square error (RMSE) of 0.39, a Nash–Sutcliffe efficiency of 0.84, and a coefficient of determination of 0.86. After calibration and validation in four basins, the SCS (Soil Conservation Service) curve number (CN) method was applied. For this method, the percentage impervious area was estimated from land-use data (built-up area), the initial loss was assumed to be between 5 and 10 mm, and the SCS CN was set between 70 and 79. The CN itself is estimated based on land use, soil group, and an assumed initial abstraction range of 5–10 mm according to land-use data. Evaluating the impact of drought on continuous river flow values is essential for our study. Therefore, we employed the HEC–HMS model to delineate watershed boundaries, input historical precipitation data, identify meteorological losses, and generate the flow of the Sunkoshi basin outlet at hydrological station no. 680, as recommended by USACE (2020) for streamflow computation. HEC–HMS offers various discharges to simulate the generation of surface runoff from precipitation using the SCS CN method (Chow 1964). Based on studies in similar geographic locations, the Snyder and SCS CN models were identified as potentially more accurate for this basin outlet. Therefore, we applied the SCS CN method to determine the peak flow parameters for estimating missing flow values in our extensive study of basin outlets. The channel-routing method was employed to simulate the flow of water through the stream channels and conveyance system (Yuan et al. 2019). The model was calibrated using curve numbers, routing coefficients, and storage coefficients to match the observed streamflow data. After calibration, the model was validated using independent streamflow data from 2015 at hydrological station no. 655 to assess its performance and accuracy. The anticipated precipitation data of 2015 was used as input for the desired simulation by the HEC–HMS model, which computes the hydrographs and streamflow at the basin outlet. The regional discharge of the sub-basin was estimated in HEC–HMS using catchment-area methods (USACE 2020). The determined monthly flow equations are given in Table 2.

In Table 2, the monthly regional discharge was assumed to be Y in m³/s and X is the catchment area in km², and a and b are the coefficients specific to each month using the power equation (Y = a × Xb) from January to December. These equations can be used to estimate discharge within the other small catchments of the Sunkoshi River basin, given a specific catchment area and assuming similar hydrological conditions of the regions. However, it is important to note that these are empirical relationships and may not be accurate under conditions where rainfall, land use, or other hydrological factors are significantly different.

The intensity of drought during the summer months of June, July, and August (JJA) exhibited varying degrees of severity across different monitoring stations within the Sun Koshi basin. The SPI and SPEI were applied for nine meteorological stations to derive summer drought mainly for JJA in the basin from 1985 to 2015. The Dhulikhel and Panchkhal stations displayed the most pronounced levels of overall drought and extreme drought in 2010 and 2015, respectively (Table 1 and Figures 2(b) and 3(b)). The higher negative SPI values show that the year 2010 was a drought year identified in all the stations except Okhaldhunga (1,206) and Salleri (1,219). In Okhaldhunga and Salleri, the year 2015 was a drought year where three-month SPI values were negative. The years 2010 and 2015 were the most drought years in the majority of the stations. The SPEI also showed that the years 2010 and 2015 were the drought years. In 2010, drought was observed in Chautara (1,009), Sermathang (1,016), Dhulikhel (1,024), Panchkhal (1,036), Charikot (1,102), and Jiri (1,103). However, Jiri (1,103), Okhaldhunga (1,206), and Salleri (1,219) experienced a higher magnitude of drought in 2015. During the last three decades, the drought (SPI and SPEI) fluctuated and increased in the majority of the observed locations in the basin, which has affected the discharge in the monsoon season in the basin (Supplementary Tables S1 and S2). Both drought indices (SPI and SPEI) have a positive degree of correlation, mainly the highest in Panchkhal (1,036) (r = 0.96) and the lower positive r = 0.72 in Salleri (1,219) (Supplementary Table S3). Most of the basin areas experienced different magnitudes of drought during 2010 and 2015 (Table 3).

Drought and regional flow are interconnected phenomena influenced by local and regional conditions that reduce precipitation, leading to decreased water inflow into streams and rivers. Drought diminishes water availability in rivers, affecting regional flow patterns, both monthly and annually. When regional low flow falls below the long-term average (monthly or annual), drought conditions prevail. The severity of drought and its impact on regional flow are contingent upon the water availability of specific geographic regions and river basins.

The evaluation of drought, based on sub-basin-wise departures of rainfall and discharge from the estimated average values during the drought year 2010, revealed that the maximum discharge deviation from the average rainfall value was −0.412 at hydrological station 652. However, in the drought year 2015, the maximum discharge deviation from the average rainfall value was −0.197 at hydrological station 630. All sub-basin-wise departure values of rainfall at hydrological station no. 680 are presented (Table 4).

This study has investigated drought patterns based on SPI and SPEI drought indices in the Sun-Koshi River basin using ClimPACT2 software. It subsequently examines and evaluates the responses of downstream hydrological station discharge values. The spatial evaluation of drought and streamflow patterns was conducted using available precipitation and discharge data spanning 18 years (1997–2015), with river flow data available only for the years 1985–1996. This analysis is based on data collected from nine rainfall stations (station nos. 1202, 1219, 1102, 1103, 1016, 1009, 1024, 1027, and 1036) and four sub-basin hydrological stations (station nos. 630, 652, 665, and 680) spanning the discharge from 1997 to 2015. These indices play a critical role in evaluating drought conditions and analyzing prevailing climate trends in Nepal (Sigdel & Ikeda 2010; Wang et al. 2013; Kafle 2015; Dahal et al. 2016; Khatiwada & Pandey 2019). This evaluation explores the increasing severity of drought and its far-reaching consequences, particularly in critical sectors such as agriculture and water resources. Both the SPI and SPEI are valuable because they are normalized using univariate probability distribution (McKee et al. 1993; Vicente-Serrano et al. 2010). Both the SPI and SPEI variations showed that the basin experienced severe drought during the years 2010 and 2015. Spatially, more than 90% of the basin area experienced the drought (Table 1; Figures 2 and 3). This analysis shows the importance of these indices as major tools for effectively quantifying drought, which is consistent with previous research studies. While the SPI primarily uses precipitation data, the SPEI incorporates a combination of precipitation and PET, improving its accuracy in assessing drought conditions. In the context of the Sun Koshi River basin, both indices are synergistically important, as they collectively contribute to a nuanced understanding of the dynamics of drought. Spatially, the Dhulikhel and Panchkhal stations observed extreme drought severity. The basin received lower annual precipitation (900–1,200 mm) and higher mean annual evapotranspiration (1,500–2,000 mm) in what is a relatively drought-prone region, mainly in the hilly regions of the basin (Dahal et al. 2021). Due to erratic precipitation patterns in frequency and magnitude, notorious flash droughts are also possible in Nepal. Likewise, more than 80% of flash droughts were experienced in the Indus and Cauvery basins of India during monsoon and non-monsoon seasons (Rakkasagi et al. 2023). Drought is a major concern in Nepal, with significant impacts on key sectors, particularly agriculture and water resources. Drought has been observed in many places of Nepal during both dry and wet monsoons (Sigdel & Ikeda 2010; Joshi & Dongol 2018). Climate change, mainly precipitation and temperature, has been responsible for exacerbating the drought condition in Nepal (Wang et al. 2013). The Sunkoshi River basin is one of the drought-prone regions in Nepal where the monthly drought magnitude obtained from SPI and SPEI is observed to be higher (Supplementary Tables S5 and S6). Drought has also influenced river discharge. During 2010 and 2015, the discharge is lower, and the departure of discharge percentage is higher in the basin (Figures 2 and 3). Furthermore, this research conducts an exhaustive investigation, with a specific focus on the years 2010 and 2015, using a hydrological model (HEC–HMS). This comparative analysis involves assessing both drought occurrences and basin discharge in the Sunkoshi River basin. The monthly regional flow and sub-basin-wise discharge were estimated using the HEC–HMS model (Supplementary Tables S4 and S6), which showed that the discharge of the river was highly affected by lower regional flow during drought periods. By combining observed data with simulated discharge results, the discharge response rate of the basin's SCS CN is influenced by the choice of a loss model (HEC–HMS), including factors such as infiltration, storage, interception, evaporation, and precipitation received within the basin. Accurate prediction of the drought's impact on river flow requires the reciprocal use of rainfall and discharge percentages. In our study, the analysis of discharge percentages from the four sub-basin divisions (station nos: 630, 652, 665, and 680) with areas of 4,920, 10,000, 2,680, and 17,600 km², respectively, at the hydrological stations yielded better results in assessing drought-prone effects in smaller basin areas (Table 4). This suggests that selecting a smaller area is advisable for optimal drought-impact analysis; the study provides valuable insights into the dynamic interplay between these factors. Additionally, the integration of geographical coordinates of hydrological stations into GIS (Geographic Information System) mapping significantly enhances the scope and accuracy of the assessment, enabling a comprehensive evaluation of the profound impact of drought on the complex dynamics of streamflow across both the basin outlets and sub-basin areas of the Sunkoshi River. The study also has significant implications for agricultural practices and water resource management in the Sunkoshi transboundary river basin. The identification of drought-prone areas, particularly during the critical summer months, can inform targeted interventions such as improved irrigation systems, water conservation measures, and early warning systems. Additionally, understanding the spatial variability of drought impacts can help optimize water allocation strategies and prioritize water-stressed regions. By assessing the impact of drought on different sub-basins, policymakers can develop more effective water resource management plans, ensuring equitable distribution and sustainable utilization.

A careful examination of drought magnitude during the summer months (JJA) reveals a diverse range of results across different stations in the Sun Koshi basin. Of the nine stations under investigation, the Dhulikhel and Panchkhal stations are particularly noteworthy, as they exhibit the highest total and extreme drought scores. It is important to note that the absence of river flow data for the years 1985–1996 necessitates a modified approach. The relationship between the percentage of rainfall and river flow showed that the discharge percentage is lower during the drought time (Table 4), which is an indication of the impact of drought on the stream river flow in the basin. Besides the rainfall, river damming and regional flow also contribute to river discharge in the context of the Himalaya River system in Nepal (Adhikari et al. 2023; Baniya et al. 2023). The departure of discharge percentage was also found to be higher during flooding years in all the stations. Despite this limitation, the assessment of drought and streamflow patterns is conducted with precision, using available rainfall and discharge data from the years 2010–2015. This comprehensive study emphasizes the need for robust strategies and proactive measures to effectively address the multifaceted challenges posed by drought in Nepal. The implications of this study extend beyond the academic realm, urging policymakers to incorporate these findings into the development of effective strategies.

This study investigates the influence of meteorological drought on river flow in the Sunkoshi River basin, a transboundary region. However, a key limitation lies in the lack of complete rainfall data across the basin, particularly in areas with complex terrain. To address this, the study employs Surfer software to interpolate missing data points. While interpolation provides estimates, it may not fully capture the true spatial variability of rainfall, especially in the Sunkoshi basin's mountainous regions with sparse rain-gauges. This can lead to uncertainties in assessing the severity and extent of meteorological droughts, potentially impacting the accuracy of the correlation between drought indices (SPI and SPEI) and observed variations in river flow across the basin's subdivisions. Future studies could incorporate additional data sources, such as satellite imagery, or explore alternative interpolation techniques that account for the Sunkoshi basin's unique spatial characteristics. This would enhance the analysis by providing a more robust foundation for linking meteorological drought to river flow variations within the basin. In the future, incorporating socio-economic factors and stakeholder perspectives into the analysis can provide a more comprehensive understanding of drought impacts and adaptation strategies. Developing advanced hydrological models and integrating remote-sensing techniques can enhance the accuracy of drought monitoring and prediction.

This study focuses on assessing the temporal and spatial pattern of meteorological drought and its influence on river discharge in the Sunkoshi River basin. Severe drought episodes were notably observed in 2010 and 2015. Spatially, over 99% of the basin area experienced drought, varying from moderate to extreme magnitudes. The years 2010 and 2015 were marked by lower rainfall and higher deviations in discharge when discrepancies emerged between rainfall and discharge percentages across stations 652, 680, 630, and 665, respectively. Over the past three decades, drought has displayed fluctuating but predominantly increasing trends across most observed stations within the basin, impacting discharge, particularly during the monsoon season. The estimated annual rainfall and basin outlet discharge were 1,800, 1,907 mm, and 1,899, 1,086 m³/s in 2010 and 2015, respectively. During drought periods, the stations indicated significantly reduced discharge percentages, indicating a marked departure from normal conditions across the basin. Analysis of regional flow contributions and monthly discharge within different sub-basins indicated that the drought-affected regions identified by SPI and SPEI experienced reduced contributions to the overall river discharge in the basin. The complexity of representing discharge percentages in response to drought was highlighted in the basin's subdivisions, notably 4,037 km² (Dudhkoshi) and 2,599 km² (Tamakoshi at Sunkoshi Duvan). Evaluation of the impact of drought response to the discharge in the four sub-basins, with areas of 4,920, 10,000, 2,680, and 17,600 km², revealed flow departures of −0.405%, −0.412%, −0.412%, and −0.382% in 2010, and 0.208%, 0.064%, 0.064%, and −0.009% in 2015, respectively. The maximum flow departure was recorded at 0.208% in 2015, while the minimum flow departure was −0.382% in 2010. It found that the utilization of SPI and SPEI in conjunction with discharge data enabled an exploration of the drought's influence on river flow patterns. The performance of station data and HEC–HMS with the SCS CN model demonstrated a strong influence of drought on discharge. In the smaller basins, this performs more precise and accurate results, ultimately enhancing our understanding of drought dynamics and its diverse impacts.

This first and corresponding author are supported by the Collaborative Research Program of the Alliance of International Science Organization (ANSO) (ANSOCR- KP-2021-09), CAS Interdisciplinary Innovation Team (xbzg-zdsys-202104) and President's International Fellowship Initiative (PIFI) visiting scientist grant for the Chinese Academy of Science (CAS) international talent (2024VEA0001, 2023VCC0001). The authors would also like to express their gratitude to the Department of Hydrology and Meteorology (DHM), Government of Nepal, for their data support. The authors are also grateful to the Institute of Science and Technology (IoST), Tribhuvan University (TU), and the College of Applied Sciences (CAS), Nepal.

A.T.R. conceptualized the process, contributed in research design, rendered support in data analysis, wrote the original draft, reviewed and edited the article; B.B. conceptualized the process, contributed in research design, rendered support in data analysis, wrote the original draft, reviewed and edited the article; C.S. rendered support in data analysis and wrote the original draft; T.Q. contributed in research supports, rendered support in data analysis, reviewed and edited the article; L.H. reviewed and edited the support; C.D. reviewed and edited the article; K.N. reviewed and edited the article; S.M. rendered support in analysis, reviewed and edited the article; A.R.P. rendered support in data analysis, reviewed and edited the article.

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

The authors declare there is no conflict.