University of California at Berkeley The capability of different satellite observing configurations to resolve fine-scale methane emissions Alexander J. Turner1,2, Daniel J. Jacob2, Joshua Benmergui2, Jeremy Brandman3, Laurent White3, & Cynthia A. Randles3 1UC Berkeley, 2Harvard University, 3ExxonMobil Research and Engineering Company 217 AGU Fall Meeting Funded by ExxonMobil, DOE ARPA-E, and the Miller Institute at UC Berkeley December 13, 217
The importance of fine-scale methane sources Contribution to US emissions (%) Emissions (tons h -1 ) Gridded EPA inventory Quantile Top 1% of grid cells make up ~3% of emissions in the EPA inventory Jacob, Turner, et al. (216)
The importance of fine-scale methane sources Emissions (tons h -1 ) Contribution to US emissions (%) How can different satellite observing systems resolve fine-scale sources? 35 N 34 N 33 N 32 N 31 N Gridded EPA inventory EDF Barnett Shale Methane Inventory Quantile.5 Top 1% of grid 3 N cells make up ~3% of emissions 1 W 99 W 98 W 97 W 96 W 95 W in the EPA inventory Jacob, Turner, et al. (216) 5. 4.5 4. 3.5 3. 2.5 2. 1.5 1. Methane flux (μmol m -2 s -1 )
Details of the WRF-STILT modeling 4 nested WRF domains with nudging to NARR (in outermost domain) Hourly STILT trajectories from every 1.3 km 12 vertical levels (including a surface level) for STILT trajectories
Resulting footprints for the satellite observations Use these footprints to construct the H matrix that maps from emissions to concentrations
Footprints for the whole observing system
Simulating methane column enhancements 35 N EDF Barnett Shale Methane Inventory 5. 4.5 CH 4 = enhancement 34 N 33 N 32 N 31 N 4. 3.5 3. 2.5 2. 1.5 1..5 Methane flux (μmol m -2 s -1 ) 3 N 1 W 99 W 98 W 97 W 96 W 95 W
Simulating methane column enhancements footprint CH 4 = Hx enhancement emissions
Quantifying the information content of the observing system cost function (Bayesian with Gaussian errors): J (x) = 1 2 (y Hx)T R 1 (y Hx)+ 1 2 (x x a) T B 1 (x x a ) posterior solution: ˆx = x a + H T R 1 H + B 1 1 {z } posterior covariance matrix posterior error covariance matrix: Q =(H T R 1 H {z } observations Fisher information matrix: F = H T R 1 H + B 1 {z} ) 1 prior H T R 1 (y Hx)
Quantifying the information content of the observing system Fisher information matrix: F = H T R 1 H Example cost functions Bayesian: Least-squares: LASSO: Tikhonov: (y Hx) T R 1 (y Hx)+(x x a ) T B 1 (x x a ) (y Hx) T R 1 (y Hx) (y Hx) T R 1 (y Hx)+ P i x i (y Hx) T R 1 (y Hx)+ x T x Eigenvalues of F can tell us about the information content of the observing system
Comparing different satellite observing configurations Flux threshold (µmol m -2 s -1 ) 1 1 Information content for constant sources 1 1-1 1-2 Eigenvalues of EDF inventory Info 5 98 286 961 2221 EPA inventory Configuration TROPOMI GeoCARB (daily) GeoCARB GeoCARB (hourly) hi-res 5 1 15 2 25 Ranked flux patterns Flux threshold (µmol m -2 s -1 ) Information content for variable sources 1 1 1/21/213 1 1-1 EDF inventory EPA inventory Info 2 8 54 458 Configuration GeoCARB (daily) GeoCARB GeoCARB (hourly) hi-res 1 2 3 4 5 Ranked flux patterns Large scales (basin-scale) Small scales (~1.3 1.3 km 2 ) Can directly compare different observing systems
Can interrogate the importance of various design parameters Information content (weekly) Information content (daily) 25 2 15 1 5 1 4 1 3 1 2 1 1 1 Constant sources 1 returns per day median 1-σ 2-σ 2 4 6 8 1 12 14 Instrument precision (ppb) Temporally variable sources 1 returns per day 2 4 6 8 1 12 14 Instrument precision (ppb) Constant sources 4 ppb precision 2 4 6 8 1 Return times per day Temporally variable sources 4 ppb precision 2 4 6 8 1 Return times per day Quantifies the importance of precision and sampling frequency 25 2 15 1 5 1 4 1 3 1 2 1 1 1 Information content (weekly) Information content (daily) *GeoCARB-like resolution
The capability of satellite observing systems to resolve fine-scale emissions Emissions (tons h -1 ) Contribution to US emissions (%) Gridded EPA inventory Quantile Flux threshold (µmol m -2 s -1 ) 1 1 Information content for constant sources 1 1-1 1-2 Eigenvalues of EDF inventory EPA inventory Info Configuration 5 TROPOMI 98 GeoCARB (daily) 286 GeoCARB 961 GeoCARB (hourly) 2221 hi-res 5 1 15 2 25 Ranked flux patterns Information content (daily) 1 4 1 3 1 2 1 1 1 Temporally variable sources 1 returns per day median 1-σ 2-σ 2 4 6 8 1 12 14 Instrument precision (ppb) Important points from this work: 1) Fine-scale sources make up a large fraction of the anthropogenic emissions 2) A week of TROPOMI obs can constrain the mean emissions in the Barnett Shale 3) GeoCARB constrains constant sub-basin scale sources 4) Quantifying fine-scale, transient sources will require better than 6 ppb precision