In the presence of the abovementioned limitations, the analysts have to rely on their individual ability to clean the data and compile the needed information. The lack of reliable metadata on river gauges and catchment boundaries affects the compilation and subsequent analysis of catchment-scale hydrometeorological variables. India's river basins have witnessed a rise in average temperature, a decrease in monsoon precipitation and a rise in droughts, along with a number of other hydroclimatological changes . In order to better understand and quantify such changes, the research community needs robust hydrometeorological datasets across all river basins of India. Reliable hydrographic data are a fundamental building block of such datasets. Increasingly available satellite and other high-resolution data products can be leveraged to build the needed datasets. Moreover, high-resolution river discharge measurements from the new Surface Water and Ocean Topography mission (SWOT, https://swot.jpl.nasa.gov/, last access: 1 February 2023) can be reconciled with historical information only when there are robust hydrographic data. Leveraging the state-of-the-art remote-sensing data to build the needed hydrographic datasets is the motivation behind this study.

The National Hydrography Dataset (NHD, https://www.usgs.gov/national-hydrography, last access: 1 February 2023), from the US Geological Survey (USGS) is an excellent example of a robust hydrography dataset. It is not feasible for one individual to build such a dataset for India's river basins; instead, this requires agency-level or community-wide effort. Based on the recent reports available from WRIS, it appears that WRIS is progressing towards creating such a dataset. In the interim, hydrologists and other analysts still need a reliable dataset on India's river gauging stations, catchment boundaries and other relevant data. By making use of the widely used Hydrological data and maps based on Shuttle Elevation Derivatives at multiple Scales (HydroSHEDS) dataset as the underlying template, this study leverages the valuable and abundant data resources available via HydroSHEDS. The specific goal of this study is to build a quality-controlled analysis-ready dataset from publicly available resources for use in hydrologic analyses while highlighting the major limitations of existing datasets.

The remainder of this paper is organized as follows: in Sect. , the data sources used in this study are discussed; in Sect. , the methodology used to differentiate the more reliable gauging stations from the less reliable stations is described; the final product from this study, called the Geospatial dataset for Hydrologic analyses in India (GHI), is described in Sect. ; in Sect. , a preliminary analysis of GHI's time series of data is discussed; Sect.  provides a discussion of the limitation of GHI and potential next steps; and the conclusions from this study are presented in Sect. .

Landmarks such as rivers, bridges and highways are often used as reference points by CWC to describe the location of its gauging stations. As publicly available maps from Google (Google Maps) and OpenStreetMap (OSM) contain such landmarks, they can be used to verify CWC's station description. The names of nearby towns and cities are also available from these sources and serve as reference landmarks. Such landmarks can be independently verified by users and are useful in the validation of station coordinates and other metadata. Google Maps products typically contain more information on cities and towns than OSM. However, OSM typically has more information than Google Maps on rivers and other waterbodies. An example graphic illustrating some of these differences between Google Maps and OSM is shown in Fig.  in Appendix . The publicly available QGIS software (https://qgis.org/en/site/, last access: 1 January 2022) has plug-ins for both Google Maps and OSM. QGIS was used throughout this study for GIS analyses, and the plug-ins for online maps were used wherever needed.

Long-term hydrometeorological products spanning the entire study domain, often cited in scientific literature as reasonable products for use in hydrologic analyses, and currently being maintained and/or developed were chosen for this study. The gridded precipitation dataset from the Indian Meteorological Department (IMD) has become the benchmark precipitation dataset and has been used by a number of studies e.g.,. evaluated ERA5 products (ERA5 is the predecessor of ERA5-Land) and found that ERA5 is superior to other reanalysis products analyzed in their study and is suitable for hydrologic assessments over India. used ET from ERA5-Land and the Global Land Evaporation Amsterdam Model (GLEAM), among other products, for water resources assessment in the Godavari and Krishna river basins. This study specifically uses data from IMD, ERA5-Land and GLEAM. Sample precipitation and ET data for water year (WY) 2020 are shown in Fig.  in Appendix .

The study domain comprises only those river basins of India where daily streamflow data are publicly available (Fig. ). For ease of analysis and to be consistent with WRIS-GIS, the study domain was separated into composite river basins. Such composite basins serve as a regional spatial reference. Every gauging station within GHI is tagged with the parent composite basin information. Starting with PF-12 watersheds from HydroSHEDS, composite basins were manually delineated using the QGIS software. It was ensured that the boundaries of these basins followed topographic divides, and the resulting basins were consistent with those from WRIS-GIS to the extent possible. The study domain was grouped into 15 composite hydrologic basins (Fig. ). The names of these composite basins are generally consistent with those from WRIS-GIS, but they are more consistent with those from CWC-19. Most composite basins are contiguous and are shown in gray shading in Fig. . Basins that are not contiguous are shaded; these include the East-Flowing Rivers North Basin (EFR North, dark blue) and the East-Flowing Rivers South Basin (EFR South, light blue).

The catchment area of these GHI composite basins is compared with equivalent basins from other sources in Table . The categorization of basins is not the same across sources; hence, a new category (West-Flowing Rivers, or WFR, North and South) was created to facilitate comparison. For ease of comparison, the different sources are compared against WRIS-GIS. The discrepancies between the different sources are typically less than 5 %. GHI-estimated catchment areas are much closer to CWC-19 than to WRIS-GIS. Discrepancies between the various estimates are attributed to differences in the underlying topographic data's spatial resolution, land vs. ocean demarcation, used within the data and GIS projection system and/or coordinate system used.

Within GIS analyses, it is customary to choose an appropriate coordinate projection system that determines how three-dimensional landscapes can be projected onto two dimensions for analysis and visualization. Estimation of area on the surface of the Earth, such as catchment area, is affected by the choice of the projection system or coordinate system and is based on a number of factors (e.g., https://pro.arcgis.com/en/pro-app/latest/help/mapping/properties/coordinate-systems-and-projections.htm, last access: 1 September 2022). While the basin boundaries from WRIS-GIS are in the Lambert conformal conic projection (https://pro.arcgis.com/en/pro-app/latest/help/mapping/properties/lambert-conformal-conic.htm, last access: 1 September 2022), it is not known whether this is the best or recommended projection system for India's river basins. Data sources such as MERIT estimate catchment areas using the World Geodetic System 1984 (WGS84) coordinate reference system (https://pro.arcgis.com/en/pro-app/latest/help/mapping/properties/specify-a-coordinate-system.htm, last access: 1 September 2022), without using a specific projection system. Gridded meteorological data often use the WGS84 coordinate reference system as well. For the sake of convenience, this study also uses the WGS84 coordinate reference system.

The second geospatial layer in GHI contains metadata on river gauging stations. Station-specific metadata – latitude, longitude, site name and river name – were used to verify the location of each CWC station within Google Maps and/or OSM. CWC's stations are typically named after nearby towns, cities or other landmarks. A reference landmark was identified for each station whenever possible, and the coordinates of the landmark are included within GHI. Such landmarks provide a definitive reference location which can be reverified by other users and/or repositioned if needed. The distance between CWC's station location and corresponding GHI landmark as well as the direction of the landmark relative to the CWC station are also noted.

An example of the abovementioned site verification and landmark identification is presented in Fig.  for the Neeleeswaram station on Periyar River in the WFR South Basin. The CWC location is shown as a red circle. The town of Neeleeswaram is about 4 km to the west of the CWC location and is chosen as the landmark (black square) for this station. The CWC locations do not fall on the pixel-based river network from HydroSHEDS nor MERIT due to the approximate nature of these networks. For feasibility of catchment area delineation and other subsequent analyses, the original station location needs to be relocated onto these river networks such that the relocated station is in the middle of a HydroSHEDS (or MERIT) pixel. The relocated station is shown by the blue circle on the river network from HydroSHEDS (Fig. a) and by the green circle on the river network from MERIT (Fig. b).

The third geospatial layer in GHI contains station-specific catchment boundaries and river networks. Catchment boundaries were derived using the PF-12 watersheds from HydroSHEDS. Using information on the upstream watersheds associated with each PF-12 watershed, all of the PF-12 watersheds upstream of the relocated station were recursively identified. The polygons corresponding to the most downstream PF-12 watershed and all of identified upstream watersheds were merged (or “dissolved” in GIS jargon) to create a catchment boundary topographically consistent with HydroSHEDS. At this juncture, the delineated boundary includes the most downstream PF-12 watershed in its entirety. However, the portion downstream of the relocated station is not needed. Using pixel-specific flow direction information from HydroSHEDS, only the portion contributing to flow at the relocated location was extracted and the remainder was discarded. The boundary delineation procedure just described is conceptually similar to standard procedures available within GIS software. The specific procedure used here ensures that the final delineated boundary is consistent with HydroSHEDS' underlying topographic data and is also accurate to the extent feasible. The station-specific river network was identified by extracting (or “clipping” in GIS jargon) the HydroSHEDS river network present within the upstream catchment boundary.

A schematic illustrating the abovementioned catchment boundary delineation procedure is presented in Fig.  for the Watrak Dam station in the Sabarmati River basin. The original CWC location (red circle), the relocated GHI station (blue circle) and the GHI river network (blue lines) are overlaid on Google Maps information (Fig. a). The gray polygons are the HydroSHEDS PF-12 watersheds upstream of the Watrak Dam station. The portion downstream of the station, not contributing to flow at the station, was discarded, and the final catchment boundary was delineated.

The catchment area associated with each station was obtained by estimating the area enclosed by the delineated HydroSHEDS-based catchment boundary. An additional estimate of catchment area was obtained using the MERIT 90 m raster layer on upstream catchment area. CWC's stations were relocated onto the MERIT river network. The catchment area corresponding to the station was estimated as the raster value at the 90 m pixel containing the relocated location. The HydroSHEDS and MERIT relocated station locations as well as the estimated upstream catchment areas are included within the metadata available through GHI.

The quality control process used to assess the reliability of station-wise metadata and hydrography data involves several steps (as outlined in Fig. ) and is described in the following text. There are 10 quality check (QCs) and 1 data check (DC), and each station is assigned a “P” (for pass) or an “F” (for fail) corresponding to each of these 11 checks. If a station passes a specific check, then it is flagged as P for the particular check. Otherwise, the station is flagged as F for the particular check. A station is placed in Group 3 if it fails to meet any one of the 10 QCs. Otherwise, stations are placed in Group 1 or Group 2 based on the availability of streamflow data (i.e., based the status of the only DC). Thus, metadata associated with stations in Group 3 should be considered the least reliable. Stations in Group 1 and Group 2 are equally reliable when it comes to metadata, but Group 1 has reliable daily streamflow data available.

QC1 is on the availability and reliability of station coordinates – latitude and longitude. Stations with missing or spurious coordinates are placed in Group 3. Latitude and longitude from CWC are presented in “DDMMSS” format. The numerical value of minutes and seconds under such a format should span the values from 0 to 60. When values violate such bounds, the coordinates are considered spurious. QC2 ensures that CWC's description of the station is broadly consistent with its coordinates. If the river basin specified by CWC did not match the composite basin associated with the station's coordinates, then it was placed in Group 3.

QC3 ensures that CWC's description of the station is verifiable using either Google Maps or OSM. CWC's description includes the name of the station, the name of the main river (and sometimes a tributary) on which the station is located. Within a general vicinity of 50 km around the station coordinates, a visual search was performed for the namesake river body and a namesake village, town or other landmark (such as a bridge or a highway). The visual search was performed first using Google Maps and then, if needed, OSM. An approximate match was also considered acceptable when searching for names of rivers and places. Sometimes, CWC's stations were nowhere near a waterbody, and sometimes one station's location exactly coincided with the location of another station with a different name. There were also instances in which a station was present at a river confluence and it was not evident if the station was intended to be downstream of the confluence or upstream of the confluence (on one of the river branches). One such example is presented in Fig.  in Appendix . All such ambiguous situations resulted in the station being placed in Group 3.

QC4 is related to QC3 but is an independent check on CWC's station names. QC4 ensures that a reference landmark matching the station name can be found, regardless of whether a river body is present in the vicinity or not. If a reliable landmark was not found, the station was placed in Group 3. QC5 ensures that the identified reference landmark is not far from the original CWC station. Some of CWC's stations began operation more than 50 years ago. It is possible that population centers (and their names) have experienced changes during this time and do not always reflect those shown within present-day Google Maps or OSM. A distance of 20 km was selected as a reasonable threshold based on an examination of the typical distances between stations and their landmarks. A majority of the stations meet this criterion and were within 5 km of the corresponding GHI landmark (Fig. a). If a station was more than 20 km away from its landmark, it was placed in Group 3.

During the catchment delineation process, stations were relocated onto the pixel-based river networks. QC6 ensures that the relocated station is in the proximity of the original CWC location. Typically, the relocated stations are only a few pixels away from the original location. However, the relocation distance was much larger on some occasions. Given that some of the larger rivers can have channels spanning multiple HydroSHEDS pixels, a distance of 5 km (approximately 10 HydroSHEDS pixels) was assumed to be a reasonable relocation distance. A majority of the stations meet this criterion and were relocated less than 1 km (Fig. b). If a station was relocated more than 5 km, it was placed in Group 3.

QC7, QC8 and QC9 pertain to the availability and reliability of catchment area estimates from CWC. Catchment area was not available for certain CWC stations, and QC7 ensures that such stations are placed in Group 3. As discussed in Sect. , CWC's catchment area estimates are not reliable, and they could sometimes have more than a 50 % discrepancy. The discrepancy between GHI's catchment area estimates and those from CWC and MERIT are shown in Fig. . Based on this chart and an examination of other discrepancies, a discrepancy of 80 % between GHI and CWC (GHI relative to CWC) was considered an appropriate threshold. As MERIT-based catchment area estimates are considered equally reliable with respect to those from GHI, a smaller discrepancy of 10 % between GHI and MERIT (GHI relative to MERIT) was considered an appropriate threshold. QC8 and QC9 ensure that the catchment area estimates from GHI stay below these thresholds.

QC10 is on the adequacy of the delineated HydroSHEDS-based river network upstream of the gauging station. While the HydroSHEDS river network typically resembles the actual river network based on Google Maps or OSM, it can be grossly inadequate, particularly in the presence of braided river channels, distributaries or other complex flow regimes. The reasoning behind the use of QC10 is to qualitatively assess the adequacy of the delineated network by visually comparing it with actual river flow paths from Google Maps and OSM. A number of stations in the river deltas of the Mahanadi and Cauvery basins and a number of stations on smaller tributaries did not pass this check. One such example is presented in Fig.  in Appendix .

The final check is on the adequacy of daily streamflow data. As discussed earlier, for many of CWC's stations in Peninsular India, daily streamflow data are available via WRIS. However, there are often many missing records. It should be noted that a flow value of zero was not considered to be missing data in this study. Missing values from WRIS are blanks in the downloaded data spreadsheets. When available, daily data were aggregated to monthly and annual data. Only months with a maximum of 5 missing days and years with a maximum of 60 missing days were used. This would mean that data would be available for at least 80 % of the days in a typical month and for 80 % of the days in a year. Finally, only those stations with a minimum of 5 years of annual streamflow data were selected. As water resources assessments and other hydrologic studies often desire sufficiently long time series, a minimum threshold of 5 years was used in this study. However, these 5 years need not be sequential.

The compiled annual time series was checked against statistics presented by CWC-YB for select stations to ensure consistency between the two sets of estimates. A self-consistency check was also performed to ensure that there were no spuriously large values. After a visual examination of annual time series of streamflow across all available stations, a value of 2 times the 95th percentile was deemed to be a reasonable threshold. Two stations – Pathagudem in Godavari Basin and Sadalga in Krishna Basin, had annual values that exceeded the abovementioned threshold for WY 2017. Both annual and monthly data for WY 2017 for these two stations were discarded.

The number of stations falling within each group are tabulated in Table . Out of the 645 stations analyzed here, 213 are in Group 1, 259 are in Group 2 and the remaining 173 are in Group 3. Thus 472 of 645 (or about 73 %) of the stations are in Group 1 and Group 2. Except for the Pennar Basin, all river basins have at least one station falling in Group 3. All of the basins have at least a few stations in Group 1. Figure  shows the spatial distribution of stations across groups. Figure d shows a cluster of stations in the Mahanadi and Cauvery deltas. These stations are associated with river networks that are too complex for GHI to delineate. Stations in the WFR South Basin correspond to catchment areas that are typically much smaller than those in other basins. This is due to the presence of the Western Ghats (or mountains) along the coast, resulting in many small coastal watersheds along the western coast of India. Several such catchments of the WFR South Basin have catchment area discrepancies larger than the thresholds defined in the quality control process (QC8 and QC9 in Fig. ). Such stations typically have catchment areas of a few hundred square kilometers or less. As discussed in Sect. , the 500 m resolution of HydroSHEDS data is probably too coarse to adequately resolve the boundaries and river networks of such catchments.

Station-wise monthly and annual time series of data of select hydrometeorological variables are compiled for only those stations falling into Group 1 or Group 2. The variables compiled include observed streamflow; observation-based precipitation from IMD; model-based precipitation, ET and runoff from ERA; and ET from GLEAM.

In order to facilitate comparison across these variables, they are expressed in the same units of volume. As various reports from CWC (e.g., CWC-19 and CWC-YB) present summary statistics on streamflow (and other hydrometeorological variables) in units of “MCM” (or 106 m3), this unit was chosen to facilitate cross-checking. For the purposes of discussion and graphical display, sometimes the unit of billion cubic meters (“BCM”, equivalent to km3) was also used. Observed daily streamflow, available in units of cubic meters per second (m3s-1) was aggregated to cumulative monthly and annual volumes (MCM per month and MCM per year, respectively). Gridded data on precipitation, modeled ET and runoff, available in units of depth per unit area per month (e.g., millimeters per month), were also aggregated to cumulative monthly and annual volumes. The process of aggregating grid-based products to a catchment scale involves identifying the spatial overlap between the grids and the catchment. Such relationships were identified using a GIS analysis. Grid-specific fractional areas were used to aggregate the gridded data to each catchment. A schematic illustrating the process of aggregation is given in Fig. .

Instead of the calendar year (January–December), a water year (WY) is often used in hydrological analyses. Consistent with CWC-19 and other CWC publications, the water year in this study is defined as the period starting on 1 June and ending on 31 May of the following calendar year. For example, WY 2020 spans the period 1 June 2020 through 31 May 2021.

The time span of GHI's time series of data is WY 1950 to 2020 (71 water years), whenever data are available. Precipitation data from IMD and all data from ERA span the 71 WYs, whereas GLEAM data span WY 1980 to 2020 (41 WYs). The availability of streamflow data from WRIS is station dependent, and no station spans the entirety of the above 71 WYs. The compiled streamflow observations for stations in Group 1 span WY 1965 to 2017, with individual station record lengths varying from 6 WYs to 53 WYs (median length of 34 WYs). There are 175 stations within GHI that have at least 20 WYs of streamflow data.