High nitrous oxide emissions from temporary flooded depressions within croplands

Nitrous oxide (N2O) is a potent greenhouse gas with a 100-year global warming potential 298 times that of CO2. Agricultural N2O emissions, largely driven by manure and nitrogen-based fertilisers, arise from microbial processes (nitrification, nitrifier-denitrification, and denitrification) that are strongly controlled by oxygen availability. Temporal and spatial variability, particularly hotspots and hot moments, complicate accurate emission estimates. Temporarily flooded depressions within croplands may act as hotspots for rapid denitrification, especially following fertilisation when nitrate-rich water can accumulate in poorly drained areas. Despite their potential importance, such small-scale flooded features have been largely excluded from national greenhouse gas inventories. This study addresses the knowledge gap by: (1) quantifying the magnitude of N2O emissions from flooded depressions relative to adjacent areas, (2) determining the duration of elevated fluxes after fertilisation, and (3) mapping and scaling the area contribution of such depressions across Zealand, Denmark, around early spring fertilisation.

Prior work highlights the importance of hotspots and hot moments for reactive nitrogen losses and N2O emissions. High emissions have been linked to drought-rewetting events and spring thaw periods in managed ecosystems, while recent studies emphasize hotspots in warm, well-drained organic soils. Meta-analyses show N2O emissions increase with nitrogen input, linearly up to ~150 kg N ha−1 and nonlinearly above that. Modeling studies indicate significant contributions from short-lived spring-thaw pulses to national N2O budgets. However, flooded depressions in mineral cropland soils, particularly during the non-growing to growing season transition and following fertilisation, remain underrepresented in inventories due to measurement challenges and limited spatial identification. This study situates itself within this gap by combining in situ flux measurements, laboratory incubations, and remote sensing-based hotspot mapping.

Study area and mapping: The team first identified 102 partly flooded fields (0.4–2.5 ha) within a 9000 km2 region in central Zealand (2018–2019) and then surveyed the whole island for March–April 2019. Ephemeral water bodies in croplands were mapped at 3 m resolution using PlanetScope imagery and a U-Net deep convolutional neural network for semantic segmentation, fused with a 40 cm LiDAR-based digital elevation model (DTM, DSM, slope). A training set of 1,824 manually delineated water bodies (validated with 12.5 cm aerial RGBNir orthophotos) spanning five subcategories was used, enforcing temporal ephemerality (via 7-year aerial imagery) and excluding permanent vegetation. PlanetScope (Level 3B) composites used a 3-week window (29 March–18 April 2019), selecting images with ≤20% cloud cover; a median mosaic and cloud filtering produced an average of 36 cloudless mosaics per tile. Near-infrared (NIR) thresholding estimated water fraction and temporal variability (January–July 2019, 5-day medians). Model training employed Adam (lr=0.0003), cross-entropy loss, 500 epochs, batch size 16, patch size 256, batch normalization, ReLU activations, and data augmentation (flips, rotation, Gaussian noise, pixel dropout). Post-prediction morphological filtering (6×6 structuring element) reduced artefacts and smoothed boundaries. Accuracy on an independent validation set: pixel water-body accuracy 72%, balanced accuracy 86%, precision 50%; instance-level F-score 34% (IoU≥20%), average IoU 46%; counting error 3%. Area and a 5 m rim buffer area were computed; mapping uncertainties propagated to area estimates and 95% CIs. Field flux measurements: In spring 2018/2019 (within two months after initial fertilisation and before the main growing season), N2O fluxes were measured once at each of 102 flooded depressions along transects with three positions: centre (standing water), rim (saturated, ≤0–1 cm water), and upslope/adjacent drained locations (background), each in triplicate. Static PVC chambers (24 cm diameter, 20 cm height) were inserted 2 cm into soil; lids closed for 12 min, with N2O measured using a photo-acoustic gas monitor (INNOVA 1312). Nonlinear regression of concentration rise (after 2 min) was used to estimate flux, converting ppm to µmol m−3 and applying F = (dC/dt) × V/A. Soil temperature, volumetric water content, pH, and 0–10 cm soil nitrate (ion chromatography) were measured adjacent to chambers. Laboratory incubations: Intact soil cores (12 cm length, 7 cm diameter; n=27 for treatments, 3 replicates per moisture×N level) were collected October 6, 2022 from a typical loamy Luvisol near Sorø, Denmark (55°44′N, 11°59′E) representing ambient field conditions. After ~2–4 weeks pre-incubation at 7 °C, cores were assigned to three moisture regimes: drained (air-dried to ~11% gravimetric water), ambient (~16.5%), and flooded (water-filled pores with ≥2 mm standing water). Nitrate (KNO3) was applied at three rates: Low 10 kg N ha−1, Normal 50 kg N ha−1, High 150 kg N ha−1, by 10 injections per core to homogenize the profile (total additions equivalent to 3.8, 19.2, 57.7 mg N per core). Cores were incubated at 7 °C for 75 days in 2 L sealed glass bottles. N2O flux was monitored using a Los Gatos Research Isotopic N2O Analyzer (model 914-0027), flow 143 mL min−1, 1 Hz logging, with frequent measurement schedule (daily first 17 days, then decreasing frequency). Initial 200 s of each run was discarded; fluxes computed as for field measurements. Soil properties (bulk density, total C and N, pH) were analyzed post-incubation. Duration and scaling: A near-surface wetness index mapped the persistence of flooding during March–April 2019. Scaling assumed: (i) two-month flooding window validated by remote sensing; (ii) representativeness of the 102 fields for the 20,063 mapped flooded areas; (iii) hotspots primarily within a 5 m rim of each depression; and (iv) an average rim flux 80× the adjacent background during the study period. Landscape-level contributions were estimated using mapped areas (with uncertainty) and measured flux ratios.

- Mapping: 20,063 individual flooded depressions identified across Zealand (March–April 2019); average size 802 ± 150 m²; total flooded area 23.1 ± 1.3 km², equaling 0.50% of 4,973 km² farmed area. The 5 m rim zones around depressions total 13.2 ± 0.8 km² (0.26% of farmed area). - Flooding duration: Near-surface wetness analysis indicated stable area-integrated flooded extent over the two-month period, despite changes in individual depression size. - Field fluxes: Rim zones exhibited mean N2O fluxes of 652 ± 118 µg N m−2 h−1, over 80× higher than adjacent drained field areas (8 ± 2 µg N m−2 h−1). Centre areas had generally lower but variable fluxes relative to rims. Soil temperatures during measurements were 2–7 °C; rim and centre were water-saturated (45–52% vol. water, near porosity), while adjacent areas averaged 32 ± 8.3% vol. water. Nitrate concentrations positively correlated with high rim fluxes. - Laboratory incubations: Under Normal N (50 kg ha−1), flooded cores averaged 410.1 ± 295.3 µg N2O–N m−2 h−1, comparable in magnitude to field flooded conditions; Ambient averaged 1.1 ± 2.9; Drained 0.7 ± 2.8 µg N2O–N m−2 h−1. Flooded vs Ambient ratios: 370× (Normal N), 29× (High N). Flooded vs Drained ratio under High N reached 536×. Across treatments, mean fluxes ranged from ~0.7 to 1,072 µg N2O–N m−2 h−1. Soil properties: bulk density 1.4 ± 0.1 g cm−3; SOC 1.59 ± 0.09%; total N 0.18 ± 0.01%; C:N ≈ 8.85; pH 6.25 ± 0.03. - Landscape scaling: Considering only the 5 m rim (0.26% of farmed area) during the two-month flooding window and an 80× emission factor relative to adjacent areas, rim hotspots accounted for 30 ± 1% of total cropland N2O emissions in that period (uncertainty from mapped area only). Fluxes remained high for at least 8 weeks post-fertilisation. - Emission factors: Incubations suggest >5–8% of added nitrate can be released as N2O under flooded conditions over weeks, exceeding the commonly used 1% default in inventories (contextualized by the limited spatial extent of flooding).

The study demonstrates that small, temporarily flooded depressions within croplands act as potent N2O hotspots, particularly at the rim zones where nitrate-rich runoff accumulates under saturated, anoxic conditions conducive to denitrification. The observed 80× higher rim emissions versus adjacent field areas, sustained over weeks post-fertilisation, explain a disproportionate contribution of these features to field- and landscape-scale budgets, despite their <1% area share. Spatial gradients around depressions likely reflect nitrate depletion as water traverses the rim to the centre, with the rim receiving nitrate inputs from upslope soils. Laboratory incubations corroborate field magnitudes and sensitivities, showing strong moisture and nitrate controls on N2O production and indicating that a substantial fraction of added nitrate can be emitted as N2O under flooded conditions. Scaling using deep-learning-derived maps of ephemeral water bodies reveals that the 5 m rim areas alone can contribute ~30% of the cropland N2O budget during the two-month spring window. These results underscore the need to include hotspot processes in emission inventories and to target hotspot management for mitigation. Hydrologic routing, timing and intensity of precipitation, and fertilisation practices collectively drive hot moments; thus, high-frequency monitoring and hydrologic modeling are essential to capture episodic peaks and inform mitigation strategies.

Temporarily flooded depressions in croplands, especially their rim zones, are extreme N2O hotspots following spring fertilisation. Field measurements across 102 sites and laboratory incubations show fluxes 80–140 times higher than adjacent drained soils, persisting for at least eight weeks. Mapping and scaling across Zealand reveal that these rim hotspots, covering only 0.26% of cropland area, can account for ~30% of N2O emissions during the two-month flooding period. The study introduces an integrative framework combining high-resolution remote sensing and deep learning with in situ and laboratory flux measurements to identify, quantify, and scale hotspot contributions. These findings indicate a strong mitigation potential by managing or retiring periodically flooded sub-field areas, offering climate and agronomic co-benefits. Future research should: (i) expand measurements across soil types, climates, and management systems; (ii) employ high-temporal-resolution flux monitoring to resolve bursts; (iii) couple hydrological and biogeochemical models to quantify nitrate mobilization, routing, and residence times; (iv) evaluate interannual variability in flooding extent and duration; and (v) assess targeted land-use changes and agronomic practices to mitigate hotspot emissions.

- Spatial and temporal representativeness: Field fluxes were snapshot measurements within a two-month window in 2018/2019, and year-to-year variability in flooding extent, duration, and environmental conditions may alter emissions. - Incubation generality: Laboratory experiments used one soil type (Luvisol) from a single site; results may not generalize across different soils and climates. - Mapping uncertainty: Pixel-level precision (50%) and instance-level F-score (34%) indicate uncertainty in delineating small and dynamic ephemeral water bodies, especially <150 m². Temporal compositing may miss short-lived events. - Scaling assumptions: Assumes 102 measured depressions are representative of 20,063 across Zealand and that the two-month flooding period and 5 m rim dominate emissions. Using only rim areas likely yields a conservative estimate; including entire depressions may increase contributions but with higher variance. - Methodological constraints: Closed-chamber measurements can introduce biases, though addressed with non-linear fitting. Competition with plants during the growing season may reduce N2O production relative to non-growing season conditions. - Emission factor context: High laboratory-derived emission factors (>5–8%) represent hotspot conditions and limited area; they do not directly imply changes to field-scale default inventory factors without broader validation.