Background & Summary

Although lakes and reservoirs cover only about 2% of the Earth’s land surface, they represent an essential part of the planet’s bio- and hydrosphere: they store about 90% of the Earth’s unfrozen surface freshwaters1, are biodiversity hot-spots2,3, and contribute to human societies through diverse services and cultural values, including water and food supply, flood control, recreation, tourism, navigation, and spiritual values4,5,6,7. In addition, lakes and reservoirs play an important role in controlling global-scale hydrological and biogeochemical processes, in regulating regional climate and weather patterns, and in contributing to the global carbon budget8,9,10,11,12,13.

Despite the importance of lakes and reservoirs, our understanding of their functioning and their sensitivity to anthropogenic alterations at the global scale remains constrained by a historical lack of data regarding their hydro-environmental characteristics. Advances in remote sensing and computing over the past two decades have produced increasingly comprehensive and dynamic global maps of inland surface waters14,15,16,17. But these datasets, on their own, are limited in their ability to describe surface water characteristics and distinguish among lakes, rivers, floodplains, and other wetlands. To aid in this distinction, a global lake-only database of more than 1.4 million individual lake and reservoir shoreline polygons has been created through extensive manual processing of surface water maps, termed HydroLAKES18. However, although HydroLAKES provides basic information on the distribution and geometry of lakes, users in need of additional lake characteristics that shape their ecological and geomorphological functioning — such as climatic conditions, topography, or surrounding land cover — still have to derive or summarize this information from disparate sources.

In an era of global environmental pressures and climate change, researchers, conservation organizations, and decision-makers must adopt a large-scale perspective to understand, manage and conserve lakes and reservoirs3,19,20. Yet the inconsistency and inaccessibility of data sources across administrative units and lake basins stand in the way of taking such a perspective by precluding seamless analyses at regional to global scales. Even when data are available and sufficiently consistent, the compilation and harmonization of multiple data sources for a large region typically involves repetitive geospatial procedures, often necessitating the development of new algorithms or software customizations. This is not only a time-consuming process requiring specialized computational skills and resources, but the individual, non-standardized solutions also create results that are difficult to compare. For instance, to spatially correlate data that are processed at different resolutions, extents, or projections, the applied alignment procedures can result in potential artefacts and faulty interpolation. Therefore, baseline datasets providing consistent and homogeneous global coverage are needed to enable and support limnological research and water resource management in a timely manner across the world, particularly in remote areas where little monitoring is available.

Few initiatives to date have compiled and harmonized baseline data on the hydro-environmental attributes of lakes at large scales. As notable regional exceptions, the Lake-Catchment (LakeCat) dataset21 provides 135 environmental attributes for all 378,088 lakes in the conterminous U.S. at varying lake catchment scales; LAGOS-NE22 provides hydro-environmental attributes and in situ measurements of lake water quality for lakes across 17 states in the U.S. Northeast; and the forthcoming LakePulse database will provide attributes for lakes across Canada (https://lakepulse.ca). At the global scale, several databases provide hydro-environmental attributes for a small subset of lakes, or a limited selection of attributes for a larger number of lakes; examples include the International Lake Environment Committee Foundation World Lake Database (https://wldb.ilec.or.jp), LakeNet’s Global Lakes Database (http://www.worldlakes.org/lakes.asp), data compiled by the GloboLakes project (http://www.globolakes.ac.uk), or the Global Lake Database (GLDB)23. However, to our knowledge the only existing compilation of attributes for a near-complete global set of lakes (≥10 ha) is the Global Lake area, Climate, and Population dataset (GLCP)24. GLCP combines static lake polygons from HydroLAKES with calculated yearly lake surface area and basin-level temperature and precipitation metrics, as well as human population estimates for each of the 1.4 million individual lakes. GLCP is a valuable contribution to the global study of lake and reservoir dynamics as its data on lake surface fluctuations not only address important limitations in the component datasets of HydroLAKES (see Technical Validation) but also provide information on seasonal and long-term changes in water quantity which are key to understanding the global interplay between lakes, humans, and hydrology24. Nonetheless, GLCP shows some technical constraints associated with the spatial units used for the calculation of basin-level statistics (see Technical Validation for more details); and its restricted set of environmental variables limits its applicability when data on a broader range of lake characteristics are required (e.g., land cover, physiography, soil conditions). Also, GLCP derives its environmental variables for an encompassing lake surface and basin area. Yet for some environmental conditions it is important to distinguish whether the information is derived from the entire drainage area of the lake, from within the lake proper (i.e., across the lake surface area), or from within close vicinity around the lake’s shoreline, as each of these zones can provide unique contributions to lake ecosystem functioning and integrity3,21,25,26.

Here we introduce the LakeATLAS dataset27, which provides hydro-environmental information for all lakes and reservoirs globally with an area of at least 10 ha that are contained in HydroLAKES. LakeATLAS complements an existing global data repository of hydro-environmental sub-basin and river reach characteristics termed HydroATLAS28 (Fig. 1). The goal of the overarching HydroATLAS database is to provide a single, comprehensive, consistently organized and fully global digital data compendium at high spatial resolution, organized in three distinct hydrographic sub-datasets: BasinATLAS (containing ~1.0 million sub-basin polygons), RiverATLAS (containing ~8.5 million river reach lines), and LakeATLAS (containing ~1.4 million lake polygons). The hydro-environmental attributes of all three sub-datasets are compiled from publicly available data sources and are organized in six categories: hydrology; physiography; climate; land cover & use; soils & geology; and anthropogenic influences (Table 1). To match with the remainder of the HydroATLAS database, version 1.0 of LakeATLAS provides the same list of attributes as versions 1.0 of BasinATLAS and RiverATLAS comprising a total of 281 individual attributes that represent 56 different hydro-environmental variables (Table 2 and Fig. 2). The attributes of LakeATLAS are processed for three complementary spatial units: within the lake surface polygon, in a 3-km buffer around the lake, and/or within the entire upstream drainage area of the lake.

Fig. 1
figure 1

Conceptual design of HydroATLAS and relationship to underpinning HydroSHEDS database. HydroATLAS consists of three companion attribute datasets: BasinATLAS and RiverATLAS (fully described in Linke et al.28) as well as LakeATLAS (this paper). BasinATLAS provides sub-basin characteristics for hierarchically nested watersheds at twelve spatial scales. RiverATLAS contains the same attributes yet calculated for river and stream reaches rather than sub-basins. The geospatial units for both databases, i.e., sub-basin polygons and river reach line segments, respectively, have been derived from the global hydrographic database HydroSHEDS39 at a spatial resolution of 15 arc-seconds (~500 m at the equator). LakeATLAS is added here following the same overall format and structure.

Full size image
Table 1 Categories of hydro-environmental characteristics included in LakeATLAS and in the overarching HydroATLAS database (from Linke et al.28).
Full size table
Table 2 Hydro-environmental variables and attributes provided in version 1.0 of LakeATLAS (modified from Linke et al.28).
Full size table
Fig. 2
figure 2

Example attributes of LakeATLAS. Top panel shows average annual precipitation aggregated within the lake surface polygon. Bottom panel shows total population count summed within a 3-km vicinity buffer around each lake. Note that in areas of high lake density, such as in northern Canada, lake surfaces visually appear as contiguous lake regions rather than as distinct lake polygons.

Full size image

Table 3 provides an overview of the main statistical characteristics of each variable as derived from all lakes included in LakeATLAS. We find that lakes occur in all major climate zones, land cover types, biomes, and countries; their distribution spans elevations between −414 and 5,826 m a.s.l. and mean air temperatures between −25.9 and 31.9 °C; in average 29.0% of their upstream area is forested and 12.6% of their upstream area is protected; and we estimate that a total of 965 million people (more than 12% of the world’s population) live within 3 km of a lake.

Table 3 Statistical characteristics of selected hydro-environmental variables in LakeATLAS.
Full size table

A unique feature of HydroATLAS and its three companion datasets is the ability to spatially integrate the respective hydrographic features using standard Geographic Information System (GIS) technology. For example, each lake polygon is associated to the river network of RiverATLAS and the sub-basin network of BasinATLAS (via unique IDs), which allows to geolocate each lake within the drainage network. This facilitates topological assessments such as identifying and analyzing the sub-basins or river reaches that are upstream or downstream of a given lake. Using this functionality, additional attributes or advanced statistics can be calculated.

The LakeATLAS dataset is expected to create novel opportunities for a mix of theoretical and applied limnological studies, enabling multi-variable statistical assessments and model-based analyses, and to stimulate large-scale limnological research in otherwise data-poor or remote regions. For example, LakeATLAS can facilitate large-scale assessments of anthropogenic pressures on lakes29, support systematic lake classification efforts30,31, and reveal biases in monitoring and conservation networks32,33, tasks previously unachievable at this scale in the absence of consistent global data. Globally comprehensive data on lake characteristics can also be combined with datasets from in situ monitoring networks like the Global Lake Ecological Observatory Network (GLEON)34, long-term ecological research networks35, other data compilation efforts (e.g., ref. 36), or information derived from remote sensing imagery (e.g., refs. 15,17) to yield novel insights and refined estimates of the magnitude and distribution of lake processes worldwide. Finally, by connecting individual lakes to the river and stream network and their associated drainage areas, the overarching HydroATLAS database enables advanced mechanistic macroscale analyses of inland waters whereby the fluxes of materials and organisms among the elements of the waterscape can be explicitly modeled3.

It should be noted that every effort is made throughout this article to distinguish ‘natural lakes’ from ‘human-made reservoirs’, yet the general term ‘lake’ is used in all instances where this distinction is considered non-critical. Also, the expressions ‘catchment’ and ‘watershed’ are used interchangeably in this article to describe the lake drainage area.

Methods

Lake polygons and morphometric attributes

The geospatial foundation of LakeATLAS is formed by the HydroLAKES dataset (https://www.hydrosheds.org/hydrolakes) which provides lake polygons and basic morphometric attributes (e.g., area, depth, volume) for all lakes globally with a surface area of at least 10 ha, comprising a total of 1,427,688 individual lakes18. HydroLAKES contains both freshwater and saline lakes, including the Caspian Sea, as well as human-made reservoirs and regulated lakes. HydroLAKES polygons and morphometric attributes were used in their original form for LakeATLAS. This section and the following (Spatial association of lake polygons with drainage network) include a brief description of the original production process implemented by Messager et al.18 — for more details, see the original publication18 and the associated Technical Documentation37. HydroLAKES was built with the aim to be as comprehensive and consistent as possible at a global scale. It was created by compiling, correcting, and unifying several near-global and regional datasets (see ref. 37 for more details) — foremost the Shuttle Radar Topography Mission (SRTM) Water Body Data (SWBD)14 for regions from 56°S to 60°N, and CanVec38 for lakes in Canada (62% of all lakes in the database). To ensure spatial consistency among the different data sources, map generalization methods were applied to the original polygons and some polygon outlines were smoothed. The resulting map scale is estimated to be between 1:100,000 and 1:250,000 for most lakes globally, with some polygons having a coarser scale of 1:1 million (mostly in Russia north of 60°N).

Each lake polygon in HydroLAKES is associated with a suite of geometric attributes, including lake area, shoreline length, average depth, and volume. Lake area and shoreline length were directly calculated based on the geometry of each lake polygon. Average lake depth and volume were estimated with a geostatistical model based on surrounding land surface topography for all lakes with a surface area under 500 km2, and from values reported in the literature for the other 347 lakes over 500 km2 in size18. Lakes are also distinguished by type (natural lake, artificial reservoir, or natural lake with human-made regulation structure), although ‘natural lake’ was assigned by default unless another type could be confirmed.

Spatial association of lake polygons with drainage network

HydroLAKES was spatially associated with hydrographic baseline information from the HydroSHEDS raster database39 (Fig. 1). This allows for linking each lake to the drainage network by its pour point (the most downstream pixel that drains the lake) and delineating its upstream watershed area. HydroSHEDS consists of a hydrologically conditioned digital elevation model and a corresponding drainage direction map from which auxiliary layers were derived, including flow accumulations, flow distances, river orders, watershed boundaries, and stream networks39. HydroSHEDS was initially derived from elevation data of the Shuttle Radar Topography Mission (SRTM)14,40 at a pixel resolution of 3 arc-seconds (~90 m at the equator) and was subsequently upscaled to a resolution of 15 arc-seconds (~500 m at the equator). More information on HydroSHEDS is provided at https://www.hydrosheds.org.

HydroLAKES also utilizes river discharge estimates from the global WaterGAP model (version 2.2 as of 2014)41, a well-documented and validated integrated water balance model42,43,44. Each 15 arc-second pixel in HydroSHEDS was associated with an estimated long-term (1971–2000) average naturalized discharge value derived from WaterGAP’s 0.5-degree resolution runoff and discharge layers through geospatial downscaling45. See the Technical Validation section for more details and a quality assessment of the resulting discharge estimates.

To link lake polygons to HydroSHEDS and delineate their upstream drainage area, the most downstream pixel that drains each lake was identified as its pour point18. This pour point (or lake outlet) pixel is typically near the lake’s shoreline but can also occur near the center of a lake polygon for terminal lakes in endorheic basins (which have no outlet). To create a single pour point for each lake, the cell accumulation values (i.e., the number of upstream pixels as provided by HydroSHEDS, representing a proxy for watershed area) were analyzed within each lake at a resolution of 15 arc-seconds. Note that in this step, all 15 arc-second pixels that fully or partially (≥1%) overlapped with a lake polygon were considered as candidate pour point pixels. Then, the pixel with the maximum accumulation value per lake was identified as the lake’s pour point. Where multiple pixels existed with equal maximum cell accumulations for the same lake, the pixel with the highest modeled long-term mean annual discharge value (from WaterGAP) was selected. If there were still multiple pixels, a random choice was made among them. This random choice was necessary for only 3.8% of lakes, all small ones (≤6 km2 in size) and mostly with low discharge (i.e., 94% of affected lakes presented a mean annual flow at the outlet of less than 0.1 m3 s−1). This ultimately resulted in one pour point pixel per lake. Finally, the precise coordinates of the pour point location were calculated as the centroid of the intersection between the lake polygon and the pour point pixel, assuring that all pour points are located inside their corresponding lake polygon.

After creating the lake pour points, the upstream watershed area from HydroSHEDS (corrected for latitudinal pixel distortions due to the applied geographic projection) as well as the downscaled discharge estimates were extracted at the location of each lake pour point and assigned to the lake polygon. An estimate of the average residence time for each lake was calculated as the ratio between lake volume and discharge18.

Acquisition and selection of hydro-environmentally relevant attribute data

LakeATLAS relies on the same sources of hydro-environmental data as RiverATLAS and BasinATLAS28 (Fig. 1 and Table 2). These input data were acquired either from free and publicly available sources, or from collaborators who provided their data for this project. All data sources were originally assessed regarding their suitability for the entire HydroATLAS project using the following selection criteria:

  1. a)

    completeness of global coverage (allowing only for minor spatial gaps, such as small remote islands, or omission of non-critical areas, such as Greenland or deserts);

  2. b)

    consistency in data quality (i.e., no regional or local biases);

  3. c)

    sufficiency of the native resolution, precision, and accuracy (i.e., resolution of the source data should be in the range of 15 arc-seconds, or finer); and

  4. d)

    permission to use and distribute derivatives under a free license.

If multiple datasets were available for the same attribute, priority was given to the most widely recognized and/or best resolution and/or most recent dataset on a case-by-case basis28. It should be noted, however, that the selection of an attribute dataset does not imply any kind of endorsement or warranty of its quality or superiority over other data. In addition, we acknowledge that new and improved sources of data have been published since the creation of RiverATLAS and BasinATLAS (e.g., an updated WordClim climatological database46, an updated SoilGrids pedological database47, and new/updated global land cover products48,49), yet we opted to keep all data sources originally included in HydroATLAS to warrant consistency and the capacity to link and compare attributes across the three datasets. Data sources will be updated simultaneously for the three datasets in future versions of HydroATLAS.

Preprocessing of attribute data

Before extracting their attribute information into LakeATLAS format, the original attribute datasets were preprocessed into a standardized grid format with the same geometric specifications as the HydroSHEDS 15 arc-second resolution grids. Standardization was applied to datasets in both raster (i.e., representing data as a grid) and vector (i.e., representing data as points, lines, or polygons) formats. The aim of this step was to ensure full spatial congruency between (preprocessed) attribute data and HydroSHEDS, thereby avoiding misalignments in the subsequent conversion processes. All original datasets had already been standardized for the extraction of attributes at the scale of sub-basins (for BasinATLAS) and river reaches (for RiverATLAS), so the same preprocessed grids were used for extracting attributes for LakeATLAS. We provide a brief description of the general workflow below; a full description is available in Linke et al.28. Additional details for each individual attribute, including on format and resolution, are provided in the Technical Documentation accompanying the LakeATLAS dataset.

The production of standardized attribute grids involved: re-projecting original data sources to a geographic coordinate system with the horizontal datum of the World Geodetic System 1984 (GCS_WGS_1984); converting to grids those datasets that were originally in vector format; aggregating or resampling grids to 15 arc-seconds if needed; cropping or interpolating grids to address mismatches with HydroSHEDS along the coastlines (i.e., layers showing some pixel values in the ocean or lacking pixels on land compared to HydroSHEDS); and filling voids to replace small areas lacking data with spatially interpolated values (though large areas encompassing entire regions or biomes without data were kept unaltered, e.g., if they covered all of Greenland or large deserts regions).

Through these preprocessing steps, all attribute datasets were standardized to the following target grid specifications: a global extent of 180°W to 180°E in longitude and 84°N to 56°S in latitude; a cell size of 15 arc-seconds; a global projection defined by GCS_WGS_1984; and a land-ocean distribution of pixels following the land mask of HydroSHEDS.

Calculation of statistics within lakes and in their vicinity (‘local’ statistics)

After preprocessing the data sources of hydro-environmental attributes into standardized grids, their values were assigned to individual lakes using one, or more, of three different ‘local’ extraction options (depending on the nature of the attribute): values were extracted either at the lake’s representative pour point; or as a spatial aggregation across the entire surface area of the lake (i.e., the area within its shoreline boundary); or as a spatial aggregation across the area in the lake’s close vicinity (i.e., in a 3-km buffer around the lake). The three different extraction zones were implemented as described below, and the resulting spatial statistics were joined as new attribute columns to the vector layer of HydroLAKES. Calculations were performed using the ‘Zonal Statistics’ tool of ESRI’s ArcGIS 10.4 software package50 embedded in customized batch scripts. The zonal statistics tool produces spatial summary statistics, including mean, majority, sum, maximum, and minimum, by performing calculations on cells from a value grid (i.e., the preprocessed hydro-environmental attribute grid) within the unique spatial units of a zone grid. These zones are defined by cells with the same value, i.e., the unique lake identifier codes (IDs).

In preparation for the zonal statistics calculations, three zone grids of lake IDs were produced: (i) a grid showing each lake’s ID only in the grid cell that represents the lake’s pour point; (ii) a grid showing each lake’s ID in all grid cells that represent the lake’s surface polygon; and (iii) a grid showing each lake’s ID in all grid cells within a 3-km vicinity buffer around the lake (Fig. 3). The first two zone grids (i and ii) were produced because two possible zoning options were applied to derive aggregated attribute statistics within lakes depending on the nature of the attribute variable: one zoning option represents lake conditions at the pour point and the other represents lake conditions across the lake’s entire surface area. This distinction was made because some attributes are well suited to be calculated as the average or sum across the entire extent of the lake, such as precipitation or air temperature, whereas for other attributes, such as discharge, using the entire lake polygon as the zone does not deliver a meaningful metric. In this case, a single, clearly defined grid cell at the outlet of the lake creates a better representation of the lake’s overall condition (i.e., by capturing the discharge at the lake’s outlet).

Fig. 3
figure 3

Different spatial aggregation units used for the extraction of lake attributes in LakeATLAS, and their relationship with the flow network. Panel (a) shows the flow directions of every pixel from which the river network (red lines) and drainage areas are derived. Panel (b) shows three lakes within the flow network. Other panels show the spatial extraction or aggregation zones of: (c) grid cells that represent lake pour points; (d) grid cells that represent areas ‘inside’ lakes (i.e., across the lake surface polygon); (e) grid cells that represent the ‘vicinity’ around lakes (i.e., in a buffer zone); and (f) upstream drainage areas associated with the lakes. Note that in panel (f) the drainage area of the lower lake also encompasses the two nested drainage areas of the upper lakes.

Full size image

For the vicinity calculations (iii), a 3-km buffer (~6 grid cells buffer distance) was chosen as a compromise between technical and functional considerations: a smaller buffer would approach the limits of the inherent mapping accuracy of some of the underpinning source data, whereas a larger buffer was deemed increasingly disassociated from the lake proper and thus less representative of its immediate surrounding characteristics. For example, in regions of high lake density in Canada, the distance between neighboring lakes is often less than 5 km and thus a 3-km buffer around one lake already includes areas that are closer to — and potentially more associated with — the neighboring lake. Furthermore, no single distance is optimal to relate characteristics surrounding lakes to in situ patterns and processes, and the appropriate buffer dimension may vary by lake size, type, and region as well as by the processes under study51. For example, Soranno et al.51 found moderately strong relationships between lake nutrients and land use within 1-km and 1.5-km buffers around lakes for 346 northern temperate lakes, ranging in size between 0.2 km2 and 75 km2, across the state of Michigan, USA. For a larger lake, i.e., Lake Taihu in China (2,250 km²), water quality in a selected bay mainly reflected land use within 2-km and 4-km buffers compared to wider areas of influence, but the land use factors that impacted water quality differed within the two buffer zones52. Our chosen 3-km buffer therefore also serves as a compromise to accommodate the wide range of lake sizes in LakeATLAS. Future iterations of LakeATLAS may include additional buffer sizes.

All zone grids were created at the 15 arc-second resolution to ensure proper alignment with the preprocessed attribute grids. Special attention was needed during the conversion of lakes from vector format into zonal grids to accommodate cases where multiple lake pour points or parts of multiple lake polygons occupied the same grid cell; or where a lake only occupied a partial cell area (including small lakes that do not cover even a single cell). To account for these issues and ensure that every lake was represented on each of the three zone grids, the following processing and prioritization rules were applied (for uncertainties related to these steps, see the Technical Validation section):

  1. i)

    ‘Pour point’ zone grid. If multiple lake pour points coincided in a single 15 arc-second grid cell (n = 7,567 affected lakes, representing 0.5% of all records), the respective pour point cell received the ID of only one representative lake. All coinciding lakes were then redefined to be associated with the same lake ID as the shared pour point cell. This ensures that the same attribute value will later be extracted to all coinciding lakes in the zonal statistics procedure. Note that in each of these cases, the ID of the smallest coinciding lake (based on its total polygon area) was assigned to the pour point cell in order to prioritize the preservation of small lakes in the zone grids of the following steps.

  2. ii)

    ‘Inside’ zone grid. First, all pour point cells and their assigned lake IDs were kept as defined in the previous step. Additional lake cells were then defined by converting the lake polygons to grid format. To increase precision, the initial conversion was conducted at a spatial resolution of 3 arc-seconds (i.e., five times finer than the target resolution of 15 arc-seconds) by creating a lake ID for any 3 arc-second pixel that was at least half (50%) covered by a single lake’s polygon. The 25 sub-pixels forming a 15 arc-second grid cell were then aggregated by assigning the majority ID to the 15 arc-second cell. In cases of ties in the majority, the ID of the smallest lake in the grid cell (based on its total polygon area) was used in order to increase the likelihood that small lakes are preserved in the grid format (whereas larger lakes are more likely to be represented by alternative grid cells). For lakes that were not uniquely preserved in this process due to exceptional geometrical constellations (n = 4 lakes; e.g., where two small lake polygons occupied the same 15 arc-second cell), they were redefined to be represented by the lake ID at their pour point location, ensuring that in the resulting zone grid every lake was at least represented by one grid cell.

    It should be noted that this conversion process, by design, defines every 15 arc-second grid cell to be a ‘lake cell’ if at least 2% of its area overlap with a lake surface polygon (i.e., 50% of at least one of its 25 sub-pixels). Grid cells that represent the shoreline of a lake are therefore assigned to be part of the lake and are included in the ‘inside’ zone calculations. This choice of slightly overrepresenting the lake surface area was made intentionally to accommodate for minor spatial uncertainties, and because a larger number of cells reduces the influence of outliers in the derivation of summary statistics. However, this expanded interpretation of a lake’s shoreline should not be confused with the specific definitions of the functional pelagic or littoral zones of a lake. In particular, we consider the lake’s polygon boundary to designate the actual land shoreline rather than the end of the pelagic zone.

  3. iii)

    ‘Vicinity’ zone grid. To define the vicinity of lakes, first a hull polygon was generated by extending each lake’s surface polygon with a ‘geodesic’ buffer going three kilometres beyond its shoreline. The resulting hull polygons were transformed into a 15 arc-second grid by converting all grid cells whose center fell within any lake hull polygon. Next, every 15 arc-second grid cell that was covered entirely by (25) lake sub-pixels at 3 arc-second resolution (see step ii) was removed from the hull grid, ensuring that the resulting buffer grid represented only the land surrounding lakes, including partial shoreline and pour point cells but excluding the actual lake surfaces. Finally, each cell within the remaining buffer grid was assigned the ID of the lake it is nearest to using the result of step (ii) as source grid for the proximity calculations. Associating each cell with a single (nearest) lake, rather than with all lakes within a distance of 3 km, prevents double-counting in global statistics (e.g., when summarizing the total population in all lake buffers for a country). However, some statistics, such as the dominant surrounding land cover, may be affected by the proximity-based partitioning of overlapping lake buffers, so users should exercise caution in the interpretation of results. Like before, lakes that were not preserved in this process due to exceptional geometrical constellations (n = 4) were represented by the lake ID at their pour point location, ensuring that in the resulting zone grid every lake was at least represented by one cell.

It should be noted that the ‘geodesic’ algorithm used to produce the buffers for LakeATLAS accounted for the Earth’s actual geoid shape by computing distances away from lakes on a curved surface, rather than using straight-line (Euclidean) distances on a flat surface. This approach was required to correct for the latitudinal distortion in the size and shape of pixels in the GCS_WGS_1984 geographic coordinate system (the coordinate system of the preprocessed grids).

Once computed, the statistics derived for the lake’s ‘pour point’, for ‘inside’ the lake (i.e., across the lake’s surface polygon) and/or within its ‘vicinity’ (i.e., within the lake’s 3-km buffer zone) were appended to all lake polygons via each lake’s unique ID (or its redefined ID in exceptional cases as outlined above). The specific spatial zones and statistics (e.g., mean, majority, sum) that were applied to extract each individual attribute are reported in the Technical Documentation of LakeATLAS.

It is important to note that the suitability or meaningfulness of a variable and its zone may differ between lake polygons and their buffers, as well as between lakes. Therefore, the interpretation and use of the provided information remain a user’s choice. For instance, the majority of the area within the 3-km buffer around a small headwater lake may extend beyond its upstream drainage area and may thus reflect the pressures exerted on the lake less closely than the 3-km buffer of a larger downstream lake which can be a more representative descriptor of the immediate zone of influence and thus the lake’s condition.

Calculation of ‘upstream’ statistics

Statistics within lakes and in their vicinity allow for a characterization of ‘local’ hydro-environmental conditions, such as the population living in proximity to a lake shoreline or the mean annual temperature across the surface of a lake. Due to the hydrologic connectivity of the river network, however, many characteristics are better suited to an upstream perspective where the entire contributing drainage area is taken into account. For example, if an application wanted to model the water quality of a lake, this would depend both on the conditions along its shorelines (e.g., urban cover) and on conditions that originate in the entire contributing drainage area that is connected to this lake, i.e., the lake’s watershed. The latter conditions can be described with upstream statistics, such as the average population density or the total glacier, snow, or forest extent across the entire upstream watershed area that contributes to the lake. In fact, it is the very nature of freshwater systems that they depend both on local conditions and on the conditions of the entire upstream drainage area which can include parts that are far away.

To allow for the duality of both local and upstream perspectives, LakeATLAS offers pre-calculated upstream statistics for many of its characteristics. Upstream perspectives are not provided for attributes where these calculations are not meaningful, such as for ‘minimum elevation’ (as the local elevation of a lake’s water surface is identical to the lowest elevation of the entire upstream watershed), or for local attributes with an already inherent upstream perspective, such as river discharge. All upstream watershed statistics in LakeATLAS were extracted at the pour point location of each lake.

Upstream values were calculated with the standard ‘Flow Accumulation’ tool of ESRI’s ArcGIS 10.4 software package50 to accumulate all upstream pixel values of an attribute grid along the drainage direction map of HydroSHEDS. In order to produce upstream averages, a correction was performed to account for the latitudinal distortion in pixel sizes due to the applied geographic projection: each pixel value was first multiplied by its individual pixel area and the accumulated sum of multiplied values was then divided by the accumulated sum of pixel areas to derive an area weighted average for the watershed. In a similar way, the upstream extent of an attribute (in percent coverage), such as percent forest cover, was calculated by dividing the total area of the attribute in the upstream watershed by the total watershed area, using latitude-corrected pixel areas.

Data Records

All hydro-environmental attributes available in version 1.0 of the LakeATLAS dataset, as well as their sources, are listed in Table 2. Most attributes with a time component (i.e., based on time series data) are provided as long-term annual averages in the attribute table of LakeATLAS, while some also include a monthly climatology (i.e., long-term monthly averages).

Each attribute offered in LakeATLAS is identified by a unique 10-character column name. This name is composed of a 3-digit key describing the name of the attribute, a 2-digit key describing the unit of the attribute, a 1-digit spatial aggregation key, and a 2-digit dimension key (plus two underscores). For example, the variable tmp_dc_lyr depicts air temperature (tmp) in degrees Celsius (dc) inside the lake polygon (l) as an annual average (yr), whereas the variable aet_mm_v06 depicts actual evapotranspiration (aet) in mm (mm) within a 3-km vicinity around the lake (v) as an average for the month of June (06), and the variable glc_pc_u16 depicts the global land cover extent (glc) in percent spatial coverage (pc) within the entire watershed upstream of the lake pour point (u) for class 16 (16) which represents ‘cultivated and managed areas’. All attributes refer to spatial averages across their respective aggregation unit unless explicitly stated otherwise.

Full explanations and details on the syntax of the column names, the used abbreviations, and other specifications pertaining to each attribute and its associated data source are provided in the Technical Documentation that is part of the LakeATLAS dataset27 (also available at https://www.hydrosheds.org/hydroatlas). In particular, the Technical Documentation includes a browsable catalog that provides an information sheet and overview map for every available variable.

Data format and distribution 

All derived hydro-environmental attributes are provided in attribute tables associated with either the lake polygon layer or the pour point layer of HydroLAKES. LakeATLAS data are publicly available for download at https://www.hydrosheds.org/hydroatlas and as a static copy at the figshare data repository27. The LakeATLAS data layers are offered in ESRI Geodatabase and Shapefile formats. The data are projected using a geographic coordinate system with the horizontal datum of the World Geodetic System 1984 (GCS_WGS_1984). The attribute table can also be accessed as a stand-alone file in dBASE format which is included in the Shapefile format. All data are distributed with an accompanying Technical Documentation.

Technical Validation

Like the BasinATLAS and RiverATLAS datasets (Fig. 1), the data compendium of LakeATLAS does not create new data from scratch but rather re-formats existing source data. Unless specified otherwise, all source data are used “as is”, i.e., without modification except for preprocessing (described in the Methods section and by Linke et al.28) and the accumulation of values along the drainage direction map of HydroSHEDS. Validation of the quality of original datasets remains with the source publications or documentations as cited in LakeATLAS.

HydroLAKES

The reliability of LakeATLAS is largely driven by the quality and limitations of its geospatial foundation (i.e., lake polygons from HydroLAKES), which is itself dependent on its component datasets (see Lake polygons and morphometric attributes in Methods section). HydroLAKES has two main limitations with respect to mapping lake surface areas: the incomplete representation of small lakes; and its inability to represent temporal fluctuations in lake extents.

Given the characteristics of the underpinning source data, lakes with a length below 600 m but an area of 10 to ~25 ha (0.10 to 0.25 km2) are susceptible to be incompletely portrayed in HydroLAKES for most areas worldwide, except in Canada. While Canadian lake polygons were provided by the high-resolution digital Canadian hydrographic dataset CanVec38, most other lake polygons, in particular those below 60°N, were generated from the SWBD dataset (see Methods). Lake polygons from SWBD were created with orthorectified imagery of radar intensities at 1 arc-second resolution (~30 m at the equator) that was processed with semi-automated extraction protocols in combination with manual supervision and rule-based editing14. The technicians used land cover water masks as well as existing maps and charts from 1:50,000 to 1:1 million scales as guidance and confirmation of the presence or absence of water. In this dataset, a discontinuity affects the representation of lakes below 25 ha because the minimum size threshold used by technicians for digitizing a waterbody was a length of 600 m (and a width of 90 m)14. The largest lake missing due to this constraint is theoretically a round lake of 570 m in diameter spanning ~25 ha, and the proportion of omitted lakes increases with decreasing lake area. This discontinuity is apparent when comparing the size distribution of lake polygons from HydroLAKES to polygons from the reference dataset in the contiguous United States, the U.S. National Hydrography Dataset Plus (NHDPlus) medium resolution53 (version 2.1; scale 1:100,000; Fig. 4). In Canada, no discontinuity in lake abundance exists below 25 ha because the Canadian reference dataset (CanVec) portrays lakes down to 0.01 km2, an order of magnitude smaller than the surface area cut-off of 10 ha set for HydroLAKES. In Siberia, HydroLAKES polygons above 60°N latitude, i.e., in areas where SWBD is not available, were generated from the MODerate resolution Imaging Spectro-radiometer (MODIS) MOD44W Collection 5 water mask, which provides a global coverage of waterbodies at 250-m resolution54. As explained by Carroll et al.55, it is likely that a considerable proportion of surface waterbodies between 10 and 25 ha (≤4 pixels) were not detected in this region due to the coarse pixel size of the MODIS instrument (each pixel is ~6.25 ha in area) and were thus also missed in HydroLAKES.

Fig. 4
figure 4

Comparison of the number of lakes represented in HydroLAKES and in the National Hydrography Dataset Plus (NHDPlus) for the contiguous United States by size class. The histogram shows the size distribution of lake polygons (left-hand axis) included in HydroLAKES (dark grey bars) compared to the NHDPlus medium resolution (light grey bars). The black line shows the ratio in the number of lake polygons for each size class between the two datasets (right-hand axis; i.e., the number of polygons in HydroLAKES divided by the number of polygons in NHDPlus, for each size class). While HydroLAKES includes lakes down to 0.10 km2, it comprehensively represents lake prevalence in the United States down to 0.25 km2 (25 ha) but misses more than half of the lakes in NHDPlus below 0.15 km2 (15 ha).

Full size image

The second limitation of HydroLAKES resides in its lack of temporal resolution. In contrast to more recent remote sensing products depicting monthly surface water cover, like the European Joint Research Centre (JRC) Global Surface Water Dataset15, a large proportion of HydroLAKES polygons are a snapshot of lake size at a given time. For instance, the lake shorelines in SWBD (i.e., below 60°N outside Canada) were delineated as they appeared at the time of the data collection in February 200014; lake polygons in Europe above 60°N from the European Catchments and RIvers Network System (ECRINS) represent lake sizes in 2006 as depicted by the Corine Land Cover data (CLC2006)56. Therefore, some lakes may have shrunk, disappeared, grown, or appeared (including newly built reservoirs) since the acquisition of the source data depicting lake polygons in HydroLAKES. The temporal discrepancies in lake surface area may vary seasonally or sporadically, and may not be consistent across regions, such that users should use caution for areas that undergo large variations in surface water cover or where many reservoirs have been built.

HydroSHEDS

The quality and limitations of the underpinning hydrographic framework of drainage directions is discussed in the Technical Documentation of HydroSHEDS and related products (see https://www.hydrosheds.org). The choice of various specifications, such as the pixel resolution of 15 arc-seconds, is in alignment with previous global applications of the HydroSHEDS product45,57 to ensure compatibility of LakeATLAS with the rest of HydroATLAS and other existing studies, data, and results. The general aim of these choices is to provide data at high spatial resolution, yet without exceeding the limits of accuracy and reliability of the underpinning global datasets, while ensuring that users can process the global data without exceptional computing facilities.

Discharge estimates from global hydrological model

For many hydrological applications, the runoff and discharge estimates provided as part of the LakeATLAS dataset will be particularly important. Like all other attribute data in HydroATLAS, this information was provided by an existing source. Yet given its importance, we conducted a baseline evaluation of the discharge data. The long-term (1971–2000) mean discharge values provided in LakeATLAS were derived through geospatial downscaling45 of the 0.5-degree resolution runoff and discharge estimates of the global WaterGAP model41 (version 2.2 as of 2014). These estimates represent naturalized flow conditions without anthropogenic water use in the form of abstractions or impoundments41. After downscaling, the global total river flow into all oceans matched the original flow as modeled in WaterGAP within an error margin of 0.13%, indicating no significant distortion of large-scale totals due to the downscaling process. In addition, a validation of the downscaled discharge estimates against observations at 3,003 gauging stations globally58, representing river sizes from 0.004 to 180,000 m3 s−1, confirmed good overall correlations for long-term average discharges (R2 = 0.99 with 0.2% positive bias and a symmetric mean absolute percentage error SMAPE of 35%, improving to 13% for rivers ≥100 m3 s−1).

We also compared the modeled discharge estimates against observed discharge for a subset of 244 stations located at or near the outlet of a lake or reservoir (Fig. 5; Table 4). Gauging stations were linked to a lake if the lake’s upstream drainage area covered at least 95% of the drainage area of the gauging station. This selection resulted in a subset of stations that monitored both natural and artificial lakes, mostly clustered in North America and Europe. The sub-comparison confirmed that discharge estimates at lake outlets were about as accurate as those of other gauging stations (R2 = 0.95, SMAPE = 24%), with greater accuracy for larger lakes.

Table 4 Summary performance statistics of LakeATLAS discharge estimates.
Full size table
Fig. 5
figure 5

Comparison of modeled mean annual discharge in LakeATLAS to observed mean annual discharge at streamflow gauging stations monitoring lake outlets (n = 244). The inset map shows the location of the streamflow gauging stations that were included in the comparison.

Full size image

Preprocessing and aggregation of source data for LakeATLAS

To limit distortions and avoid the introduction of bias, the disaggregation and aggregation steps applied for the generation of LakeATLAS (see Preprocessing of attribute data in Methods section) refrain, as much as possible, from spatial interpolation methods. If original data needed to be re-projected or the resolution of original attribute datasets needed to be adjusted, the ‘nearest neighbor’ approach was applied to avoid modification of original values. This approach does not change any of the values of cells from the input raster (i.e., no averaging or median filtering is performed); it determines the location of the closest cell center on the input raster to the center of the cell in the output raster and assigns the value of that input cell to the cell on the output raster. Regional statistics and totals of the original data are thus preserved in LakeATLAS (e.g., within a lake’s drainage area).

To quantify uncertainties caused by the polygon to raster conversion of lakes, we tested the extent to which the representation of lake polygons as 15 arc-second grid cells introduced spatial inaccuracies due to misrepresentation of small (sub-pixel) lakes, overlap of multiple lakes in one pixel, or discrepancies of inside/outside associations along shorelines. For this purpose, we compared the lake surface area as calculated from the polygons against the lake surface area as calculated from the grid that defines ‘inside’ lake allocations (using latitude-corrected pixel sizes). Results showed that for lakes ≥1000 km2 in polygon area (n = 178), the mean overestimate was 10.8% — i.e., the sum of grid cell areas was larger than the polygon area. For lakes in the range 100–1000 km2 (n = 1,530), grid cell area exceeded polygon area on average by 24.1%; for lakes in the range 10–100 km2 (n = 14,981) by 44.0%; and for lakes in the range 1–10 km2 (n = 168,492) by 85.8%, which for a lake of 1 km2 in size corresponds to ~4 additional grid cells. The percent mismatch keeps rising for even smaller lakes, with lakes 10 ha in size showing a default overestimation of ~150% even if represented by a single 15 arc-second grid cell (~25 ha). Despite these notable mismatches, all overestimations of lake surface extent occur along the shoreline of lakes; i.e., the overestimations are spatially constrained to the distance of a single 15 arc-second grid cell and thus fall within the presumed spatial uncertainties inherent in most attribute source grids. We therefore consider these discrepancies (i.e., overestimations) to be reasonable, in particular given the intended goal of reducing the potential noise of singular pixel outliers for small lakes, which form the majority of LakeATLAS. Finally, a small number of HydroLAKES polygons fall entirely outside the land mask of HydroSHEDS (n = 793 lakes, i.e., <0.1% of all records, mostly representing lagoons or parts of estuary systems) and were thus assigned noData results for all attributes.

The quality of spatial statistics, like average air temperature within a lake or the dominant land cover class within a lake buffer, can be affected by arbitrary artefacts or outliers in the underpinning attribute data, causing uncertainties. Overall, uncertainties are largest for very small lakes that are represented by only one or few attribute pixels. Also, some attribute grids, such as soils or land cover, may inherently be affected by the presence of lakes. If lakes that are represented in a source attribute grid are not spatially congruent with the polygon for that lake in HydroLAKES (e.g., showing the same lake yet with a slight spatial offset), the determination of the true land cover or soil characteristics found in the buffer zone around the lake polygon may be affected by the ‘false’ presence of a lake water surface. Finally, the use of ‘majority’ assignments can introduce statistical bias when the results get aggregated at different scales due to an issue known as ‘modifiable areal unit problem’ (MAUP)59. A particularly relevant case of the problematic and scale-dependent interpretation of ‘majority’ attributes is presented in the association of each lake to a country. For countries with boundaries that are not crossed by lakes, such as Australia or any island nation, the country association of each lake is straightforward. In contrast, lakes at land borders can extend over multiple countries. For example, the Caspian Sea straddles the borders of five countries, yet the country it was assigned to is Kazakhstan as it is the country with which it overlaps the most. These types of problems need careful consideration by the user before aggregating or interpreting lake statistics for larger spatial units or different scales.

Comparison between LakeATLAS, LakeCAT and GLCP

Given the importance of a lake’s upstream drainage area on its water quality and functioning, it is crucial to correctly delineate lake watersheds. For LakeATLAS, we used network routing algorithms to determine the hydro-environmental characteristics of the entire area upstream of each lake’s pour point based on the drainage direction grids of HydroSHEDS (see Calculation of ‘upstream’ statistics in Methods section). A similar approach was applied in LakeCAT to compute the characteristics of lake watersheds for the contiguous United States, albeit using drainage direction grids at a much higher resolution (1 arc-second for LakeCAT21 vs. 15 arc-seconds for LakeATLAS). To compare LakeATLAS with LakeCAT, we spatially joined lake polygons from both datasets, matching lakes that overlapped for at least 90% of their respective area. Based on this subset of lakes, we found that watershed areas delineated for LakeATLAS using lake pour points and HydroSHEDS drainage direction grids corresponded relatively well to watershed areas in LakeCAT (Fig. 6a; log-log least-square regression R2 = 0.61, SMAPE = 51%, n = 3,975). The largest discrepancies were typically caused by different interpretations of lake pour points that were located near confluences between streams of substantially different size categories. In these situations, even a minor spatial mismatch can assign a lake to the small drainage basin of the tributary stream instead of the large basin of the mainstem river, or vice versa. Small lakes were more frequently affected by this issue.

Fig. 6
figure 6

Comparison of estimates of lake watershed area and human population count in lake watersheds (or ‘lake basins’) between LakeATLAS, LakeCAT and GLCP. Only lake polygons that overlap for at least 90% of their respective surface areas across datasets were included in the comparisons with LakeCAT (a,b, d,e; n = 3,975). Because LakeATLAS and GLCP both rely on lake polygons from HydroLAKES, we compared watershed area and population attributes for all lakes between these two datasets (c, f; n = 1,422,499). All axes are logarithm-transformed and black lines represent the identity line (1:1 line of equality) for each plot.

Full size image

Like LakeATLAS, the GLCP database24 relies on the lake polygons from HydroLAKES. In contrast to LakeATLAS and LakeCAT, however, GLCP assigns climate and population statistics to ‘lake basins’ that were not strictly computed from the lakes’ actual hydrologic watersheds (i.e., the upstream areas that drain into each lake). Instead, GLCP uses a spatial proxy for each lake that is defined by the smallest sub-basin that encloses the lake in its entirety24, based on the sub-basin geometry provided in the HydroBASINS dataset45. As HydroBASINS subdivides watersheds based on stream confluences rather than lake outlets, the spatial representativeness of the resulting ‘lake basins’ remains ambiguous. On the one hand, substantial parts of the associated basin may be located downstream of the lake; this issue is particularly important for headwater lakes with small drainage areas (note that even for the smallest lake size class, the median ‘lake basin’ size in GLCP is 156 km2). On the other hand, for lakes with large watersheds, the basins used in GLCP may not comprise the entire upstream drainage area but only part of it. Therefore, the physical meaning of the ‘lake basin’ associated with each lake in GLCP is inconsistent across lake sizes and landscape configurations, and neither represents the full hydrologic drainage area nor a clearly defined area-of-influence within a given buffer distance around the lake. This potential mismatch is illustrated by comparing the area of ‘lake basins’ identified in GLCP to the hydrologic drainage areas of lakes determined from LakeCAT and LakeATLAS. Differences between GLCP and LakeCAT (Fig. 6b; log-log least-square regression R2 = 0.04, SMAPE = 146%, n = 3,970) result from both differences in lake polygons and basin association methods. Differences between GLCP and LakeATLAS (Fig. 6c; log-log least-square regression R2 = 0.02, SMAPE = 171%, n = 1,422,499; see ref. 24 for details on the 5,189 HydroLAKES polygons excluded from GLCP) only stem from differences in basin association methods as these two datasets both rely on HydroLAKES surface polygons. For both comparisons (Fig. 6b,c), GLCP ‘lake basins’ tend to overstate actual lake drainage areas for lakes with watersheds smaller than ~100 km2, which reflects about the smallest average size of sub-basin units available in HydroBASINS. In contrast, GLCP tends to understate actual lake drainage areas for larger watersheds, as GLCP only picks the smallest sub-basin unit that entirely contains the lake rather than the entire drainage area.

The differences in watershed delineation (or association) between the three datasets also have implications for the computation of watershed-level statistics. This is for example the case for estimates of human population counts within the lake watershed (or ‘lake basin’) in 2010, an attribute that all three datasets share, although LakeCAT contains statistics from a different source of population data than LakeATLAS and GLCP. LakeCAT relies on the United States decennial census60 while LakeATLAS and GLCP rely on the Gridded Population of the World (GPW) dataset61, version 4. Despite relying on a different data source, lake polygons, and resolution, LakeATLAS provides a relatively accurate estimate (as compared to LakeCAT) of the population count within lake watersheds (Fig. 6d; log-log least square regression R2 = 0.66, SMAPE = 92%, n = 3,975). GLCP, within its ‘lake basin’, overstates the population count within the actual hydrologic drainage area (i.e., upstream) compared to LakeCAT (Fig. 6e) for 85% of lakes, particularly for lakes with smaller watersheds (log-log least square regression R2 = 0.35, SMAPE = 148%, n = 3,970). Similarly, despite relying on the same source of data as LakeATLAS, GLCP estimates of population counts in ‘lake basins’ are not comparable to LakeATLAS estimates in the lakes’ actual drainage areas (Fig. 6f).

Usage Notes

LakeATLAS offers a large variety of hydro-environmental attributes intended for a broad range of user applications. It remains the user’s responsibility to decide whether certain attributes, statistics, or scales are meaningful and appropriate. For example, the association of a large lake to a single country based on spatial majority may be adequate for a lake that is entirely or mostly within the country but can be highly misleading for a transboundary lake spanning multiple countries. Similarly, the association of coarser scale attributes, such as national GDP values, to small lakes may be meaningful for statistical assessment purposes, yet it will not realistically represent small-scale spatial patterns. More generally, users are expected to inform themselves on the meaning, quality, and relevant uses of the source data by consulting the primary literature associated with each attribute.

Beyond the existing attribute columns contained in LakeATLAS, users can extract a variety of inherent information by applying their own post-processing algorithms and cross-calculations. For example, attributes can be analyzed by comparing results across different scales, such as comparing lakes with relatively unpopulated shorelines but high population densities in their watershed to lakes with densely populated shorelines but a sparsely populated watershed — identifying the importance of local versus distal factors. Similarly, attributes can be summarized by other attributes, such as the percentage of lakes within protected areas per country or per freshwater ecoregion. Finally, attributes can also be normalized using the existing information of multiple columns. For example, discharge can be divided by upstream watershed area in order to calculate ‘specific discharge (per km2)’; or by upstream population numbers in order to calculate ‘water availability per person’ in the lake’s drainage area.

The BasinATLAS and RiverATLAS vector datasets are both derived from the same underpinning hydrography of HydroSHEDS that was used for the identification of lake pour points. Therefore, all three components of HydroATLAS (BasinATLAS, RiverATLAS, and LakeATLAS) are mutually linkable via uniquely defined spatial relationships. For example, every river reach is associated with one or more lakes (one-to-many relationship), every lake’s pour point is associated with a single river reach (one-to-one relationship), and every river reach or lake falls within a sub-basin (many-to-one relationship). Through these relationships, a lake can be associated with the river network characteristics of the river reach at its pour point, such as the distance from the upstream headwater source or from the final downstream pour point (i.e., at the ocean or at the most downstream pixel of an endorheic basin). Similarly, LakeATLAS is generally compatible with the growing list of other raster and vector datasets that are built from, or linked to, the hydrographic framework of HydroSHEDS. Examples of such datasets encompass a global assessment of the free-flowing status of rivers, including an estimate of their sediment transport62, and a range of aquatic species compilations, including continental maps produced by IUCN63,64. Other datasets that rely on HydroLAKES as a geospatial foundation, including GLCP, can also be linked to LakeATLAS via lake polygon IDs.

Intensive efforts have been made to verify the licenses of the underpinning source datasets, and specific permissions were obtained from individual authors if needed, in order to release all derived attribute columns of LakeATLAS (version 1.0) under either a Creative Commons Attribution 4.0 International License (CC-BY 4.0) or an Open Data Commons Open Database License (ODbL 1.0), both permitting reuse of the data for any purpose including commercial. LakeATLAS users are requested to honor the individual reference suggestions of the source data providers; hence citations and acknowledgements should be made to both the original data source(s) and the LakeATLAS compendium. For example, the following template illustrates a reference to precipitation data sourced from LakeATLAS: “Precipitation data from the WorldClim v1.4 database (Hijmans et al. 2005) has been used in the spatial format of LakeATLAS v1.0 (Lehner et al. 2022).” Detailed information regarding the license and reference(s) for each attribute column is provided in the Technical Documentation of LakeATLAS and in Table 2.