Methane emissions from agricultural ponds are underestimated in national greenhouse gas inventories

Aquatic systems contribute roughly half of global natural and anthropogenic methane emissions, and very small freshwater bodies (<0.1 ha) emit disproportionately high methane per unit area, contributing about 37% of lentic methane emissions despite comprising less than 10% of global lake and pond surface area. Many of these small systems are human-constructed agricultural ponds (farm dams, impoundments, dugouts), typically 0.01–1 ha, which experience elevated nutrient inputs (fertilizer and manure runoff), shallow depths, and rapid warming, all of which enhance methanogenesis. Prior work indicates farm ponds can produce over three times the CO2-equivalent emissions per area compared with reservoirs. Because emissions from agricultural ponds are anthropogenic, they should be included in national greenhouse gas inventories. IPCC guidelines have recently been refined to encourage inclusion of these systems as Other Constructed Waterbodies, yet data gaps on pond abundance and distribution hinder robust accounting. This study aims to provide a first-order, continental-scale assessment of methane emissions from agricultural ponds in the United States and Australia by integrating pond maps, a meta-analysis of methane fluxes with temperature standardization, and partitioning of diffusive and ebullitive fluxes, and to compare results against national UNFCCC inventory reports.

Background studies show that small lentic waterbodies are major contributors to inland methane emissions relative to their area. Agricultural ponds often receive high nutrient loads from livestock manure and fertilizers, leading to high methane production; they can be warmer and shallower than natural lakes, further enhancing emissions. Ollivier et al. reported farm ponds emit 3.43 times more CO2-equivalent per area than reservoirs. The IPCC 2019 Refinement recommends reporting methane from constructed waterbodies (including agricultural ponds <8 ha) in national inventories, but emission factors (e.g., 183 kg CH4 ha⁻¹ yr⁻¹ for freshwater and brackish ponds) are temperature-independent and may underpredict in warmer climates. There is limited global information on the abundance and distribution of agricultural ponds, complicating inclusion in inventories. Prior datasets and studies on lakes and reservoirs provide insights into temperature dependence of methane fluxes and the relative roles of diffusion and ebullition, which can inform estimates for ponds.

Spatial datasets and pond mapping: The study combined large-scale pond mapping for Australia and the United States. For Australia, AusDams.org (N ≈ 1.7 million) identified farm ponds using AI applied to high-resolution satellite imagery (scale 1:25,000–1:250,000), estimated to capture ~90% of Australian farm ponds. For the U.S., the National Hydrography Dataset High Resolution (USGS; N ≈ 7.8 million) was used (scale 1:20,000–1:100,000). Ponds between 0.01 and 1 ha were retained, and only those within agricultural-relevant land covers (crops, open forests, shrubs, herbaceous, or bare land) were included based on the Copernicus Global Land Service 100 m land cover map. Manual inspection suggested >95% of mapped waterbodies were artificial agricultural ponds. IPCC guidance of ±10% uncertainty was assumed for remote sensing products. Temperature covariates were derived from 10 years (Jan 2010–Jan 2020) of MODIS Terra Land Surface Temperature (MOD11A1.006) at 1 km resolution to compute median annual daily temperatures. Meta-analysis of methane fluxes: A literature search (Web of Science, April 29, 2022; query for methane and agricultural pond terms) yielded seven datasets for agricultural ponds, with 12 subtropical Australian, 154 temperate Australian, 101 semi-arid Canadian, and 8 tropical Indian records. Two Swedish cropland pond observations were excluded due to low representation. New measurements from 11 temperate agricultural ponds in Victoria, Australia (April 2021) were added, following Malerba et al. protocols. In total, methane flux data were compiled for 286 pond observations across climates. Most studies measured diffusive fluxes via short-duration chamber deployments or headspace extraction; Grinham et al. captured both diffusive and ebullitive fluxes with longer chamber measurements. To account for ebullition, a dataset compiled by Rosentreter et al. for lakes and reservoirs was used to estimate the temperature-dependent partitioning between diffusive and ebullitive fluxes, given the paucity of pond-specific ebullition data. The diffusion fraction decreases with temperature: approximately 72% of total methane at 5 °C and 12.5% at 30 °C. Temperature dependence of ebullition was similar between lakes and reservoirs, supporting transferability to ponds. Temperature standardization and scaling: To compare fluxes across climates, daily methane emission rates were standardized to 15 °C using a Boltzmann–Arrhenius relationship with temperature sensitivity (E_M) drawn from the Rosentreter et al. dataset for lakes and reservoirs (reported N up to 313). Standardized total fluxes (diffusion + ebullition) at 15 °C were derived for all 286 observations. For spatial predictions, the relationship was inverted to adjust the 15 °C standardized flux by local median annual temperature for each grid cell containing ponds, yielding temperature-adjusted annual methane emissions per unit pond area. Pond area density (pond ha per landscape ha) at 5 arcmin resolution was multiplied by the temperature-adjusted flux to obtain cumulative emissions (kg CH4 yr⁻¹ ha⁻¹ of land), then integrated to national totals. Uncertainty analysis: Non-parametric bootstrapping (1000 iterations) propagated uncertainties from: (i) the meta-analytic distribution of standardized fluxes, (ii) the temperature sensitivity (E_M), (iii) the temperature-dependent partitioning of diffusion vs. ebullition, and (iv) pond density/area (±10% for remote sensing). At each iteration, all steps were recalculated using resampled inputs to produce distributions and 95% confidence intervals. Coefficients of variation identified dominant uncertainty sources (flux means, temperature sensitivity, ebullition contribution; CV ~20–28%), with pond distribution estimates more certain (CV ~10%). Assumptions included pond size bounds (0.01–1 ha), applicability of lake/reservoir temperature dependencies to ponds, representativeness of median annual temperatures, and independence of seasonal variability beyond the 10-year averages. All analyses were conducted in R 4.2.2 with standard spatial and statistical packages; data and code are publicly available via Mendeley Data.

- Meta-analysis indicates large variability in total methane fluxes from agricultural ponds (<1 to >10³ kg CH4 ha⁻¹ yr⁻¹). At 15 °C, mean emissions are 204 kg CH4 ha⁻¹ yr⁻¹ (95% CI: 83–521; median 157.7), close to but slightly above the IPCC emission factor (183 kg CH4 ha⁻¹ yr⁻¹), which lacks temperature dependence. At 30 °C, predicted mean emissions are 405 kg CH4 ha⁻¹ yr⁻¹ (95% CI: 164–1037; median 314.2), about twice the IPCC factor. - United States: 2.56 million agricultural ponds covering 420.9 kha emit an estimated 95.8 kt CH4 yr⁻¹ (95% CI: 61–157). - Australia: 1.76 million agricultural ponds covering 291.2 kha emit an estimated 75.1 kt CH4 yr⁻¹ (95% CI: 47–123). - Using 100-year GWP of 28, combined methane emissions equate to about 4.79 Mt CO2-eq yr⁻¹ (95% CI: 3.01–7.86) across the two countries. - Comparison to national inventories (UNFCCC 2020, Other Constructed Waterbodies <8 ha): U.S. reported 173.1 kha and 43.75 kt CH4; Australia reported 316.4 kha and 40.73 kt CH4. These reported values are lower than this study’s agricultural-pond estimates by approximately 46% (U.S.) and 54% (Australia), suggesting underestimation in inventories. - Dominant uncertainty arises from average flux estimates, temperature sensitivity, and ebullition contributions (CV ~20–28%), whereas pond distribution estimates are relatively robust (CV ~10%).

The findings indicate that agricultural ponds are a larger methane source than currently recognized in national greenhouse gas inventories, with emissions likely underestimated by about half in both the United States and Australia. The commonly used IPCC emission factor for constructed waterbodies does not account for temperature, leading to underprediction in warmer regions where ebullition increases and diffusive contributions decline. Incorporating temperature-dependent fluxes and the ebullition component yields substantially higher estimates. Given agriculture already accounts for a large share of national methane emissions, excluding agricultural pond emissions understates the sector’s footprint. The study underscores the need for explicit inclusion and improved accounting methodologies for agricultural ponds in UNFCCC reporting, clarifying overlaps with manure management categories. Beyond inventory implications, the work highlights that management interventions targeting nutrient inputs (e.g., reducing livestock access, vegetated buffers) can meaningfully curb methane emissions while offering co-benefits for water quality, biodiversity, and farm productivity. However, additional measurements across climates and seasons and better partitioning of flux pathways are needed to refine estimates and guide effective mitigation.

Agricultural ponds are pervasive, anthropogenic sources of methane with emissions that are likely undercounted in national GHG inventories. By integrating high-resolution pond maps, a meta-analysis of methane fluxes with temperature standardization, and temperature-dependent diffusion–ebullition partitioning, this study estimates that ponds in the U.S. and Australia emit about 95.8 and 75.1 kt CH4 yr⁻¹, respectively—approximately double the amounts implicitly considered in current inventory reporting. Management offers practical mitigation: excluding livestock from ponds has been shown to halve methane emissions and improve water quality, and vegetated buffers may further reduce nutrient inputs, though trade-offs require evaluation. Future research should prioritize: (1) expanded, year-round pond measurements across diverse climates to refine mean fluxes and temperature sensitivities; (2) pond-specific ebullition quantification and drivers; (3) explicit accounting of other GHGs (CO2, N2O) and seasonal dynamics; (4) improved global mapping of agricultural ponds; and (5) integrating pond emissions into agricultural sector assessments and exploring incentives (e.g., carbon credits) for management interventions.

Key limitations include the scarcity and geographic bias of pond-specific methane measurements, leading to uncertainty in average fluxes and temperature sensitivity. Ebullition contributions were inferred from lakes and reservoirs due to limited pond data, assuming transferability. Temperature adjustments relied on 10-year median annual land surface temperatures, which omit seasonal and short-term variability in pond area and temperature. The analysis focused on methane and did not include CO2 or N2O, potentially underestimating total GHG impacts or missing sinks. Remote sensing-based pond counts and areas carry ±10% uncertainty, and manual validation suggests but does not guarantee that all mapped waterbodies are agricultural ponds. National inventory comparisons may be affected by category definitions (e.g., separation of pond emissions from manure management) that were not fully resolvable from available documentation.