Where to place methane monitoring sites in China to better assist carbon management

The study addresses the urgent need for a comprehensive atmospheric methane (CH₄) observing system in China, the world’s largest anthropogenic CH₄ emitter. Despite CH₄’s major role in radiative forcing and short atmospheric lifetime, China’s in-situ monitoring network is sparse (only a few GAW sites and short-term campaigns), limiting satellite validation, emission inversion, and anomaly detection. Satellite instruments (AIRS, SCIAMACHY, GOSAT, TROPOMI) provide broader coverage but have limitations in sensitivity, precision, and lower-tropospheric sensitivity, leading to inconsistencies with ground measurements and large uncertainties in emission estimates. The research aims to: (1) simulate CH₄ spatiotemporal distributions over East Asia for 2017 using WRF-GHG constrained by satellite-derived emissions; and (2) identify optimal locations and required numbers of surface sensors using POD-based sensor placement algorithms to most accurately reconstruct surface CH₄ fields for carbon management applications.

The paper situates its work within several strands: (1) global and regional CH₄ monitoring, noting long-term records (e.g., GAW) but insufficient spatial coverage in China relative to extensive air quality networks; (2) satellite CH₄ products (AIRS, SCIAMACHY, GOSAT, TROPOMI) that enhance coverage but face retrieval precision/accuracy challenges, limited lower-tropospheric sensitivity (especially TIR instruments), and potential for small retrieval errors to propagate into large emission uncertainties; (3) the value of integrating in-situ and satellite data for emission inversion and source attribution (e.g., isotopes, ethane co-emitted species) and for evaluating satellite anomaly detection; and (4) prior use of POD-based sparse reconstruction and sensor placement in fluids, ocean temperature fields, and air pollution (PM2.5) that motivate applying MCN minimization, Extrema, DEIM, and QR pivot algorithms to CH₄ network design.

Modeling: WRF-Chem v3.9.1 with the GHG module (WRF-GHG) simulated CH₄ over East Asia for 2017. CH₄ was treated as a passive tracer (no chemistry). Emission inventory: Zhang et al. (2022), 0.5°×0.625°, including biomass burning, coal, gas, landfills, livestock, oil, rice, geological seeps, termites, wastewater, and wetlands. Domain and resolution: 115×164 grid, 36 km horizontal resolution; 29 vertical layers up to 50 hPa. Meteorology: NCEP FNL (1°×1°, 6-hourly) for initial and boundary conditions. CH₄ initial/lateral boundary conditions: GEOS-Chem (4°×5°). Observations for evaluation: (i) Surface CH₄ at Mt. Waliguan (WLG) from WDCGG; (ii) GOSAT Proxy XCH₄ v9.0 (global precision ~9 ppb) for column comparisons. Evaluation metrics included correlation, RMSE, and spatial differences by season. Sensor placement within POD framework: Data matrix from WRF-GHG comprised 365 snapshots (daily), each 115×164 pixels. Standard POD produced spatial modes; gappy POD reconstructed full fields from sparse sensors. Four algorithms selected sensor locations: (1) MCN (minimizing condition number of the reconstruction matrix M), (2) Extrema (placing sensors at maxima/minima of POD modes), (3) DEIM (recursive selection of interpolation points minimizing linear dependence error; constrained to P=n), and (4) QR pivot (QR with column pivoting to select sensor locations maximizing |M|). Experimental design: Used n=10 POD modes. Tested sensor counts P=n, 1.5n, and 2n; further sensitivity analyses with 40–300 sensors (maintaining P=2n in primary comparisons). Performed 10-fold cross-validation. Metrics for reconstruction: mean percentage error (MPE), coefficient of determination (R²), and RMSE (ppb). Potential-station scenario: Compiled locations of prior CH₄ studies in China as potential sites and compared reconstructions using those sites versus QR pivot placements.

Model performance: WRF-GHG reproduced seasonal/spatial patterns of GOSAT column CH₄, with differences <10 ppb in autumn/winter and some overestimation in spring/summer. Daily surface CH₄ at WLG showed correlation r=0.67; summertime/autumn high biases likely reflect missing chemical loss (OH sink) and WLG’s background setting. Spatial CH₄ features: High columns over Sichuan Basin (~1930 ppb in summer) and eastern China; surface CH₄ hotspots linked to coal mining (Shanxi, Guizhou) and rice paddies/energy, with surface CH₄ up to ~2700 ppb. Algorithm comparison (n=10 modes): • P=n: MCN (MPE 23.43%, R² 0.27, RMSE 752.89 ppb), Extrema (31.47%, 0.18, 1010.64), DEIM (5.01%, 0.71, 124.92), QR pivot (4.94%, 0.74, 122.35). • P=1.5n: MCN (4.69%, 0.56, 156.55), Extrema (7.65%, 0.34, 266.25), QR pivot (3.88%, 0.79, 99.47). • P=2n: MCN (3.61%, 0.67, 119.16), Extrema (5.44%, 0.44, 188.41), QR pivot (3.46%, 0.81, 90.14). Under oversampling (P=2n), all algorithms captured main spatial features; QR pivot concentrated sensors in high-CH₄ regions and achieved the lowest errors. Scaling with sensor number: Reconstruction improved with more sensors (except DEIM constrained by P=n). With 100 sensors, QR pivot achieved RMSE 69.82 ppb and R² ~0.86, similar to MCN with 300 sensors. Potential-site scenario: Using locations from prior studies yielded MPE −3.14%, R² 0.69, RMSE 104.16 ppb—slightly better than MCN but worse than QR pivot (by 0.32% MPE and 14.01 ppb RMSE), with notable errors over central/eastern China (coal regions). Optimal count: QR pivot performance improved markedly from 20 to 160 sensors. At 160 sensors, RMSE was 58.56 ppb with minimal regional bias; further increases to 200 and 300 sensors gave marginal gains (55.96 and 48.46 ppb, respectively).

The study demonstrates that a data-driven, POD-based network design can substantially enhance the reconstruction of surface CH₄ fields in China, directly addressing gaps left by sparse in-situ coverage and retrieval limitations. By validating WRF-GHG against GOSAT and WLG, the authors establish a credible dynamic baseline for testing sensor placement. Among algorithms, QR pivot consistently selects sites that prioritize high-CH₄ regions and strong gradients, yielding the best trade-off between accuracy and computational efficiency. The findings imply that well-placed, moderately sized networks can rival or exceed the performance of more evenly distributed or ad hoc networks. The analysis further shows that relying solely on existing/potential sites underrepresents coal-mining regions, degrading reconstructions over central/eastern China. An optimized network designed via QR pivot closes these gaps, supporting more reliable satellite evaluation, emission inversion, and anomaly detection—all critical for carbon management strategies.

This work provides a model-informed, algorithmic framework for planning CH₄ monitoring networks in China. WRF-GHG simulations (2017) and POD-based sensor placement show that QR pivot site selection best reconstructs surface CH₄, particularly in high-emission regions. A network of about 160 optimally placed sensors offers a strong balance of performance (RMSE ~58.6 ppb) and cost, with marginal improvements beyond this size. The approach supplies concrete guidance for future network expansion to support satellite validation, emission inversion, and policy-relevant CH₄ management. Future research should improve model chemistry (e.g., explicit OH-driven CH₄ sinks) and integrate more observations to refine dynamical evolution of CH₄, further enhancing data-driven placement and reconstruction accuracy.

Key limitations include: (1) CH₄ treated as a passive tracer in WRF-GHG without chemical loss, contributing to overestimation in regions with strong OH (e.g., western China); (2) sparse surface observations in China limit comprehensive model evaluation; (3) sensor placement algorithms are data-driven and sensitive to input data quality and representativeness, so model biases can propagate into site selection; and (4) the study period is limited to 2017, which may not capture interannual variability in emissions and transport.