Evaporative water loss of 1.42 million global lakes

Lakes, covering ~5 million km² of land, are crucial for ecosystems, biodiversity, and human water resources. Owing to strong vapor pressure gradients at the water–atmosphere interface, lakes can lose substantial water via evaporation, which depends jointly on evaporation rate and open-water area. Both components vary geographically and respond to climate drivers (temperature, radiation, humidity, wind) and hydrologic conditions (droughts, runoff), as well as lake ice phenology. Despite their importance, global evaporative water loss from lakes has not been comprehensively quantified due to limitations in consistent global datasets for surface area dynamics, ice cover, and evaporation rates. Previous global efforts largely emphasized changes in evaporation rate and not the total evaporative volume, neglecting the critical roles of changing lake area and freeze–thaw cycles. This study aims to quantify the spatiotemporal patterns, long-term trends, and drivers of global lake evaporation volume and to provide a global lake evaporative volume (GLEV) dataset for improved water resources assessment and climate impact evaluation.

Past global studies have predominantly analyzed lake evaporation rates, constrained by challenges such as cloud contamination in optical satellite imagery, difficulties in estimating lake heat storage and circulation, and the scarcity of consistent, high-resolution, long-term lake area and ice phenology datasets. As a result, evaporation volume—which integrates evaporation rate with dynamic open-water area and ice cover—has been underexplored. Prior model-based global estimates suggested ~1300 km³ year⁻¹ of lake evaporation, but lacked explicit coupling with observed area dynamics and detailed ice-cover representation. Recent advances in Landsat-derived surface water datasets, reanalysis meteorology (e.g., TerraClimate, ERA5, GLDAS), and improved algorithms for lake evaporation and ice phenology provide a foundation to overcome these limitations and enable volumetric assessments at global scales.

Study domain and lake masks: The analysis covers 1.42 million global lakes and reservoirs (≥10⁵ m²), excluding the Caspian Sea, mapped using HydroLAKES. Among these, 6715 reservoirs were identified by HydroLAKES via mapping and GRanD cross-referencing. A buffering procedure was applied to reduce geometric inconsistencies.

Evaporation volume computation: Monthly evaporation volume for each lake (1985–2018) was computed as Ve = Ek × Ak × (1 − fu/100), where Ek is monthly lake evaporation rate (mm d⁻¹), Ak is monthly lake surface area (m²), and fu is the percentage of time within a month when the lake is fully ice covered. Monthly results were aggregated to annual values for trend analysis.

Evaporation rate (Ek): Ek was estimated with a physically based algorithm derived from the Penman combination framework, explicitly accounting for net radiation, lake heat storage, and aerodynamic terms (including wind speed). Modern meteorological inputs from TerraClimate, ERA5, and GLDAS were used. The algorithm includes enhancements to better represent deep lakes with substantial heat storage and was validated against diverse environmental conditions.

Lake surface area dynamics (Ak): Monthly lake surface area time series were reconstructed by integrating Landsat-based Global Surface Water Dataset (GSWD) annual water classification maps with HydroLAKES geometries and a temporal interpolation/reconstruction scheme (e.g., Ay,m = Aw + Wm × Ay−1). An image-enhancement and contamination mitigation workflow reduced cloud/shadow impacts. Reconstructed areas were validated against the Global Reservoir Surface Area Dataset (GRSad) and G-REALM water level data for reservoirs, showing generally satisfactory relative bias and rRMSE distributions.

Ice cover fraction (fu): Lake ice duration was modeled using empirically derived freeze-lag and thaw-lag relationships that account for thermal memory and meteorological conditions (e.g., relationships with ice depth and air temperature). Validation using in situ phenology records showed strong agreement: for 76 lakes with both ice-on/off dates, R² = 0.93 and SD of absolute bias = 9.2 days year⁻¹; monthly simulations for 14 North American lakes had an average bias of ~7 days year⁻¹. Daily binary ice/no-ice estimates were aggregated to monthly fractional ice cover to reduce errors for large lakes.

Trend and variability attribution: Long-term trends and interannual variability of Ve were attributed to three drivers: evaporation rate (Elake), lake surface area (A), and ice duration (lid). Detrending experiments for each driver produced alternate Ve time series isolating contributions. Relative contributions were computed as the percentage of each driver’s contribution to the total trend/variability.

Uncertainty: Total uncertainty in global Ve was estimated from component uncertainties (Ek, Ak, fu), yielding an overall uncertainty of ~9.95% (≈10%).

Data and code: HydroLAKES shapefiles; meteorology from TerraClimate, ERA5, GLDAS; GSWD for surface water; validation datasets GRSad and G-REALM; processed GLEV dataset (monthly lake open areas, evaporation rates, and Ve for 1.42 million lakes) available at Zenodo (10.5281/zenodo.4646621); code at https://github.com/gchaowater/lakeFup.

• Global magnitude: Annual global lake evaporation volume (excluding the Caspian Sea) is 1500 ± 150 km³ year⁻¹ (1985–2018), about 15.4% higher than a prior model-based estimate of ~1300 km³ year⁻¹.
• Long-term trend: Ve increased at 3.12 km³ year⁻¹ (≈31.2 ± 24 km³ decade⁻¹), or 2.1 ± 1.6% decade⁻¹.
• Trend attribution: Increases in Ve are attributed to increasing evaporation rate (58%), decreasing ice coverage (23%), and increasing lake surface area (19%).
• Reservoirs: 6715 reservoirs contribute 16% (≈235 km³ year⁻¹) of global Ve while comprising only ~5% of storage capacity and ~10% of total lake surface area; reservoir Ve increased at 5.4% decade⁻¹.
• Spatial patterns: Above 46° latitude, lakes account for ~69% of global Ve, with ~83% of that occurring June–November due to minimal ice and higher Ek. The Great Lakes and African Great Lakes contribute ~8.8% and ~15.7% of global Ve, respectively, compared to ~9.1% and ~6.2% of global lake area, highlighting the role of Ek and ice.
• Evaporation contribution to land ET: Globally, Ve represents ~2.37% of terrestrial evapotranspiration volume (2001–2018), with strong regional variability—e.g., Great Lakes Basin ~12.2%, Tigris–Euphrates ~13%, Amazon ~0.5%.
• Regional trends: All nine thermal regions show increasing Ek (p < 0.05), with the Northern Frigid (NF) region rising ~3.7% decade⁻¹; NF also shows strong increases in open-water area due to ice decline and exhibits one of the largest Ve increases (~4.5% decade⁻¹; ~0.6 km³ year⁻¹ in figure context). Area trends align with regional hydroclimate (e.g., sharp area decline in Southern Warm region during the Australian Millennium Drought; steady area increase in NF due to warming-driven ice reductions).
• Category analysis: For ~77% of lakes, Ve variability is dominated by non-precipitation climate forcings (temperature, humidity, wind) via Ek and ice; for ~23%, Ve variability is controlled by hydrologic conditions and reservoir operations via area changes (all reservoirs fall in this category).

Findings demonstrate that global lake evaporative water loss is substantial, increasing, and controlled by combined changes in evaporation rate, dynamic open-water area, and ice phenology. Using evaporation volume rather than rate alone provides a more accurate climate-impact metric for water resource assessment because it integrates spatial mismatches between Ek and area, and accounts for freeze–thaw effects. High-latitude and high-altitude regions, where warming is amplified, experience greater open-water durations and higher heat uptake, accelerating Ve increases and potentially altering regional humidity and hydroclimate. Reservoirs contribute a disproportionate share of evaporative losses and are rapidly increasing, implying significant implications for water availability under growing demand. These results call for improved representation of lake–atmosphere exchanges, ice processes, and dynamic lake area in Earth System Models and for incorporating volumetric evaporation into basin-scale water management and planning.

This study provides the first global, long-term (1985–2018), monthly lake evaporative volume dataset (GLEV) for 1.42 million natural lakes and reservoirs, integrating satellite-derived area dynamics, modeled evaporation rates, and ice phenology. We quantify a mean global lake evaporation of 1500 ± 150 km³ year⁻¹ and a significant increasing trend of ~31 km³ decade⁻¹ (2.1% decade⁻¹), driven primarily by rising evaporation rates, declining ice duration, and expanding open-water areas. Reservoirs account for 16% of global Ve and are increasing faster than natural lakes. The results advocate for using evaporation volume as the primary indicator for climate impacts on lake systems and water management. Future work should refine ice fraction estimates for large lakes, better constrain lake heat storage and aerodynamic parameters, assimilate higher-resolution and multi-sensor observations for area dynamics, represent reservoir operations explicitly, extend the record beyond 2018, and couple volumetric lake evaporation with Earth System Models to assess feedbacks on regional hydroclimate.

Uncertainties arise from (1) meteorological forcing and parameterizations in the evaporation algorithm (e.g., net radiation, heat storage, wind function); (2) reconstruction of monthly lake area from annual Landsat classifications, including residual cloud/shadow contamination and reliance on interpolation; and (3) simplified ice modeling using binary daily ice states and empirical freeze/thaw lag relationships, which may underrepresent partial ice cover on large lakes. Component uncertainties propagate to an estimated ~10% uncertainty for global Ve. Additional limitations include exclusion of the Caspian Sea, potential biases where in situ validation is sparse, and incomplete representation of reservoir operations and rapid hydrologic alterations at monthly scales.