We used Google Earth satellite imagery to determine the locations of 1330 dairy farms in California, by identifying metal-topped shelters alongside manure lagoons and corrals (further details given in Duren et al., 2019). We used addresses to determine the approximate location of each dairy and manually adjusted the location to the center of a dairy farm using satellite imagery in Google Earth (Duren et al., 2019; Rafiq et al., 2020). These dairy locations are publicly available as part of the Vista-CA methane mapping project on the Oak Ridge National Laboratory Distributed Active Archive Center for Biogeochemical Dynamics (https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1726, last access: 26 January 2021).

We determined which facilities are still operational by checking the data against the list of facilities that paid 2019 State Water Resources Control Board Confined Animal Facility fees (CAF fees; California State Water Quality Control Board, personal communication, 22 November 2019). We assumed that all dairies with permits and/or paying CAF fees are currently operational. This assumption is untrue, as there are an unknown number of facilities with cattle that do not pay fees (likely <20 facilities; new facilities since 2007). It is not currently possible to confidently confirm which dairies are functioning and which are not, since dairy closures are tracked by a variety of agencies, with some lag time, but this information is not accessible or consistent. Milk production statistics show that there are roughly 1400 commercial dairies in CA, including 162 dairies in northern California (CDFA, 2018).

We grouped dairies into three geographic categories by county (Table S3), North Coast (153 dairies), Central Valley (1102 dairies), and Southern California (75 dairies), to account for differences in climate, animal housing, and primary manure management styles among these three regions (Meyer, 2019).

We determined herd population sizes primarily from the 2019 CAF fees list. However, some dairies did not pay a fee in 2019, but still have animals, so for these facilities we integrated data from three sources to estimate herd numbers and demographic categories at each dairy: Regional Water Quality Control Board (RWQCB) permits, San Joaquin Valley Air Pollution Control District (SJVAPCD) permits, and individual facility documentation. The RWQCB permits are required for dairies that existed in October 2007 (California Regional Water Quality Control Board, 2013), and we used a collection of permit lists from 2014–2018 to determine the number of lactating cows. Some dairies in the Central Valley that are either new or expanded since 2007 have inaccurate or incomplete data, so we determined the number of lactating animals from the SJVAQPCD and reading individual facility documentation. The SJVAPCD permits include the maximum number of cattle in each class at a given facility in 2011, rather than the number of animals, and these dairies may have expanded since then. These data represent our best estimates, but they represent specific points in time that are not consistent between data sources. Based on milk shipments, we know that at the time of this publication, there are roughly 1.7 million lactating cows in California (Ross, 2019). Additionally, we compared our list with the 2017 United States Department of Agriculture (USDA) National Agricultural Statistics Survey (USDA NASS, 2017), which provides the number of farms and the number of cows in different dairy size classes in each county, though the NASS Census data include farms that are not commercial dairies.

For dairies with RWQCB reports, we use the number of milk cows, dry cows, heifers, and calves as the number of cattle in each class. Given that we are calculating an annual CH4 emission rate for each farm, we assume the population and demographics of each farm are constant in time, though in reality these fluctuate as cattle are sold or born. The Central Valley RWQCB assumes the population size of the lactating and dry cows varies by 15 % or less (California Regional Water Quality Control Board, 2013).

We also estimate the populations of non-lactating animals, though these data are less reliable than data for lactating animals. The RWQCB reports provide the number of dry cows, bred heifers, heifers, calves 0–3 months, and calves 4–6 months (California Regional Water Quality Control Board, 2013). From these data, we determine the median ratio of dry cows to the number of milk cows to estimate the number of dry cows for dairies without RWQCB reports. Calf and heifer populations are less reliable than mature cow populations (lactating + dry cows), as these replacement animals may or may not be at the same facility as the animals they will replace. We assume that replacement animal populations are 10 % higher than the mature cow populations (Deanne Meyer, personal communication, 7 February 2020) and are evenly distributed among the 0–23-month-old animals. For this analysis, we assume that the replacements are on the same dairies as the lactating cows in order to not double count the heifer ranches; these animals do exist but may not be present on the dairies. We also estimated the effect of this assumption on overall emissions. Enteric fermentation emissions equations also distinguish between replacement heifers <227 kg (calves) and replacement heifers >227 kg. We assume the populations are split equally between the size classes.

We estimated enteric fermentation in three ways, which have previously been used to estimate emissions at the state or national levels: (E1) according to the method used in California's greenhouse gas emission inventory (Charrier, 2016), (E2) a method used for estimating emissions for the continental US (Hristov et al., 2017) and (E3) a method suggested by recent research done in California (Appuhamy, 2018). These three methods increase in their complexity: method E1 is based solely on the population and a statewide emissions factor, E2 is based on a statewide emission factor and diets, and E3 is based on diet as well as the quality of milk provided. We performed each of these calculations with lactating cows only (subscript l) and total cattle, including calves, replacement heifers, and dry cows (subscript t).

The first method, E1, is based on the calculations used by CARB for the official statewide greenhouse gas emission inventory (Charrier, 2016), which is in turn based on the IPCC Tier 1 Guidelines (IPCC, 2006). For this method, we estimate total enteric emissions (CH4,e1) based on the number of cattle (n) and a standard emission factor for each cattle type (Eq. 1). Method E1 assumes enteric fermentation emissions (ef1) are 114.61 kg CH4 per lactating dairy cow per year (ef1,l; Table 3). 1CH4,e1l=ef1l⋅nl For all cattle, the total enteric emissions are the sum of the product of the number of cattle (n) and the emission factor (Eq. 2). Method E1 assumes that the emissions factors are 11.63 kg CH4 per dairy calf per year (ef1c), 43.53 kg CH4 per replacement heifer aged 7–12 months per year, and 65.71 kg CH4 per replacement heifer aged 12–24 months per year (Charrier, 2016). We use a weighted mean of 58.32 kg CH4 per replacement heifer per year (ef1h). Here i represents the classes of cattle, including milk cows, calves, and replacement heifers. The CARB inventory does not provide an emission factor for dry cows, so we exclude those from this analysis (Charrier, 2016). 2CH4,e1t=∑ief1i⋅ni The second method, E2, is based off of calculations in Hristov et al. (2017). For this method, we estimate the total enteric emissions (CH4,e2) as the product of the number of cattle (n), dry matter intake (DMI), and an emission factor (ef2; Eq. 3) (Hristov et al., 2017). Method E2 assumes DMI is 22.9 kg per head per day for lactating cows, 12.7 kg per head per day for dry cows, 8.5 kg per head per day for dairy replacement heifers, and 3.7 kg per head per day for calves, and emission factors are 436, 280, 161, and 70 g per head per day for lactating cows, dry cows, dairy replacement heifers, and calves, respectively (Table 3). 3CH4,e2t=∑ini⋅DMIi⋅ef2e,i The third method, E3, is based on calculations by Appuhamy (2018). For this method, we estimate the total enteric emissions including the number of cattle (n), dry matter intake (DMI), neutral detergent fiber (NDF) in the diet, and milk fat (mf) (Appuhamy and Kebreab, 2018; Table 3). We also include factors for DMI (fDMI), NDF (fNDF), and milk fat (fmf). Here, emissions are the sum of emissions due to DMI, neutral detergent, and milk fat content (Eq. 4). 4CH4,e3l=nl⋅(fDMI,l⋅DMIl+fNDF,l⋅NDFl+fmfl⋅mfl) Note that Appuhamy (2018) consider mature cows to be dry cows for 60 d of the year (16.4 %) and lactating cows the remainder of the year, while we count the dry and lactating cows separately. For the other cattle classes (i, including dry cattle, replacement heifers, and calves), the E3 emissions are the product of DMI and a DMI factor (fDMI), times 365 d per year as in E2 (Eq. 5). 5CH4,e3t=nl⋅fDMI,l⋅DMIl+fdNDF,l⋅NDFl+fmf⋅mf⋅365+∑i(ni⋅fDMI,i⋅DMIi)⋅365

We estimated manure emissions for each dairy three ways: (M1) according to the method used in California's greenhouse gas emission inventory (Charrier, 2016), (M2) a method used for estimating emissions for the continental US (Hristov et al., 2017), and (M3) a method suggested by recent manure management research done in California (Meyer, 2019; Fig. 1). Methods M1 and M2 are based on average statewide manure management, while method M3 is based on facility-level or regional manure management. We perform each of these calculations first with milk cows only and then including calves, dry cows, and heifers. All three methods follow the same general equation, though they have differences in the specific variables used in Eq. (6). 6CH4,m,l=nl⋅ρCH4⋅VSprod⋅Bo⋅∑iMCFsystem⋅fsystem In this equation, n is the number of cows, ρCH4 is the density of CH4, a conversion factor of m3 CH4 to kg CH4, which is a constant 0.662 kg CH4/m3 as reported by CARB and the IPCC (IPCC, 2006). VSprod is the total amount of volatile solids (VSs) produced per animal (kg per head per day), Bo is the maximum methane production capacity per unit of VS in dairy manure (0.24 m3 CH4/kg VS), MCF is the methane conversion factor for each system, and fsystem is the fraction of manure going into each manure management system (Table 4). The different systems include pasture, daily spread, solids, liquid/slurry, lagoon, and drylot (Fig. 1; IPCC, 2006).

The first method, M1, is based on the method used in the CARB greenhouse gas inventory. For M1, methane emissions from manure management are calculated for each dairy facility based on the fraction of manure in each management system, the total VS production, the CH4 density, Bo, and the methane conversion factor for each system (CARB, 2014; IPCC, 2006; US EPA, 2017). For method M1, we assume that a constant proportion of manure is in each management type on each dairy according to statewide proportions, which are described in Fig. 1 (CARB, 2014). The methane conversion factor for each system is shown in Table 4 (Charrier, 2016). VS production is an animal-specific constant among management types.

The second method, M2, is based on the methodology used by Hristov and colleagues (Hristov et al., 2017). These are the product of the VSs excreted, the methane generation potential, the waste management system distribution in the state, the methane conversion factor (MCF) for the state, and the methane density (Eq. 7). The percentages of waste entering daily spread, solid storage, liquid slurry, and anaerobic lagoon are shown in Table 1, with corresponding MCFs shown in Table 4. VS excreted and Bo are defined in Appendix A.

For the third method, M3, we estimate manure management based on data from the SJVAPCD air quality permits and regional differences in manure management as follows below (Eq. 7) and shown in Fig. 1. CH4 emissions for each manure management system were determined according to CARB emission factors described above and summed for each farm. As described previously, only dairies in the San Joaquin Valley with >500 cows in 2011 have SJVAPCD permits. For these dairies, we estimate manure emissions based on the reported dairy management practices documented in permits, though this information represents facilities inconsistently. These permits report the presence of corrals or freestalls as housing types; flush, scrape, or vacuum systems for manure collection; and mechanical separator, settling basin, or weeping wall as solid–liquid separator systems (Table S1 in the Supplement). Housing type typically determines the fraction of manure that is processed by the manure handling system, which can be quantified as the percentage of time cows spend on concrete. For dairies with corrals or freestalls present, we assume time on concrete to be 70 % (Meyer, 2019). For dairies without freestalls, we assume time on concrete to be 30 % (Meyer, 2019). We assume that time in the milking parlor is 12.5 % of total time, which is almost always flushed or hosed out into a liquid manure handling system (i.e., liquid/slurry or lagoon). For the remainder of the time on concrete, we assume that for facilities with scrape or vacuum systems reported, the manure is stored as solids; for facilities with only flush systems reported, we assume that this manure is flushed into lagoons. We assume that the remaining manure (time not spent in housing) is not collected and remains as solids in the open lot or pasture. For dairies with solid–liquid separator systems reported, manure that is flushed to lagoon is diverted to solid storage based on the mechanical separator efficiency (0.05 for mechanical separator; 0.225 for settling basin; 0.25 for weeping wall). We also estimate the effect of using manure solids as bedding. The majority of manure solids are used as bedding, as it is a cost-effective and easily available option to keep the animals comfortable, though some solids are land applied or removed off farm (Chang et al., 2004). Previous research suggests that solid manure loses roughly 33 % of its C as CO2 in the first month (Ahn et al., 2011); we assume that on dairies with lagoons, solid manure remains in the manure pile for at least 1 month to dry out and that half of the remaining 67 % of the manure C returns to the housing facility and ultimately ends up in the lagoon. The fraction of manure entering the lagoon, fbed, is therefore 33 %. We assume that all heifer manure is scraped, though in reality some heifer lanes may be flushed (Table S2). 7CH4,m3,l=nl⋅ρCH4⋅VSprod⋅Bo⋅[(flagoon+fsolid⋅fbed)⋅MCFlagoon+fsolid⋅(1-fbed)⋅MCFsolid+fliquid⋅MCFliquid+fpasture⋅MCFpasture] Given that air district data only exist for the San Joaquin Valley, we made assumptions about housing and manure management in the other regions in California for method M3. For the remaining Central Valley dairies without air quality permits, we used the mean partitioning of solid vs. liquids from permitted dairies in each county. In the Southern California dairies, open-lot-style farms are predominant (Deanne Meyer, personal communication, 7 February 2020), and most do not even flush the feed lane. On these dairies, we assume that only the milking parlor is flushed, at 12.5 % of the time, and the rest of the manure is either dry scraped or remains in the open lot. In the North Coast, pasture dairies are prevalent, though many dairies have some housing for cows. Here, we assume time on concrete is 39 %: on average 2 months inside in the winter and 30 % of the rest of the year. During the winter months, the manure is scraped into pits. In the summer, the manure is dried and stacked. In the North Coast, we assumed that only the milking parlor was flushed (12.5 %). Nevertheless, even with accurate accounting, the different climatic, animal housing, manure management, and biogeochemical factors in each dairy affect the actual CH4 emissions at any given time (Hamilton et al., 2006).

We estimated facility-level uncertainty in the number of cows as 20 %, as suggested by the IPCC (IPCC, 2006, Sect. S1 in the Supplement). We estimated facility-scale uncertainty for enteric fermentation emissions for each of the three methods (Table 2, Sect. S1). The methods for calculating the standard errors of each variable are shown in the Supplement. For E1, we calculated the standard error in ef1 and n. For method E2, we calculated the standard error in DMI, n, and ef2. For E3, we calculated the standard error in DMI, NDF, milk fat, fDMI, fNDF, and fmf for lactating cows and DMI only for nonlactating animals. We propagated the standard error of each variable through the emissions calculation equations, assuming the errors were uncorrelated (Sect. S1.1).

We estimate the facility-scale uncertainty in manure management emissions by propagating uncertainty in the terms ncows, fraction of time on concrete, VSprod, methane conversion factor (MCF), Bo, and fbed. We did not address uncertainty in ρCH4, as it is considered to be a constant (US EPA, 2017). Uncertainty for time on concrete was determined from variance observed in a recent study (Meyer, 2019) that describes four Central Valley dairies: two with freestalls and two without freestalls. We assume for our analysis that the time on concrete is equal to the fraction of manure produced that passes through the lagoon (flagoon). We also assume that the remainder of the manure (1-flagoon) is stored as a solid in the Central Valley, in pasture in the North Coast, and drylot in the Southern dairies. We assumed that the North Coast dairies had freestalls or loafing barns for the winter, and the Southern dairies had no barn housing; however, there are exceptions to these generalizations we did not consider as we have little systematic data on dairies outside of the Central Valley apart from expert knowledge. We estimated the uncertainty in the VS production rate based on the variability reported for lactating cattle and heifers over 13 years (2000–2012) in the CARB inventory (CARB, 2014). We calculated the mean and standard error for VS production for each of these two populations. We estimated the uncertainty of the MCFs using data reported by Owen and Silver (Owen and Silver, 2014). We estimated the uncertainty of Bo, the theoretical maximum methane production, using data from a meta-analysis (Miranda et al., 2016). We estimated the error uncertainty of fbed to be 100 %, as this value may range from including no manure as bedding to including all solid manure as bedding. To propagate the errors in total for the manure management system, we rearranged Eq. (8) with two factors to be as follows, where MCFx is the MCF for either solids, pasture, or drylot, and given that flagoon+fx=1. 8CH4,m=nl⋅VSprod⋅Bo⋅ρCH4⋅(flagoon⋅MCFlagoon-flagoon⋅MCFx+MCFx) We used the sum of the squared partial derivatives of each variable times the variance of that variable to propagate the uncertainty in facility-scale manure emissions (Sect. S1.1). To determine the relative effect of manure and enteric emissions from E3 and M3 on facility-level emissions, we propagated the uncertainty associated with the two emissions in quadrature.

Due to the large number of dairies, propagating the facility-level uncertainty to the state level using standard methods produces unrealistically low statewide uncertainty estimates (<1 %). This suggests that the uncertainties at the facility level are not independent. Therefore, we used previously published estimates for state-scale uncertainties for each of the six methods, from the EPA (E1, M1; US EPA, 2017), Hristov et al. (2017) (E2, M2), and the IPCC (E3, M3, IPCC, 2006).

We performed a sensitivity analysis on each of the methods. We calculate sensitivity δx|y of emissions (x) to each parameter (y) as 9δx|y=∂x∂y⋅σy, where ∂x∂y is the partial derivative of emissions (x) with respect to each variable (y) in the emissions equation and σy is the uncertainty in each parameter y (i.e., fractional uncertainty times the value). We calculate fractional uncertainty as each uncertainty divided by the sum of all uncertainties, as in Eq. (12). 10δ=δx|y∑iδx|y We also determined the relative sensitivity of total emissions to manure and enteric emissions.

We converted the Vista-CA dairy database into a raster image using R (R Core Team, 2013). We then convert the image to a 0.1∘ × 0.1∘ grid in WGS84 to match CALGEM (Jeong et al., 2012) and the Spatial EPA (Maasakkers et al., 2016) inventories. We subtract the values from the CALGEM, Hristov, and Maasakkers emission inventories from the Vista-CA map to observe spatial variations between inventories.

The 2017 USDA Dairy Census reports the number of milk cows in California to be 1 750 329. We report a total of 1 712 229 mature cows in Vista-CA, including 1 455 395 lactating cows and 217 400 dry cows, distributed across 1330 dairy farms. We also report a total of 1 380 040 heifers, and 1 639 966 calves. We assume a 20 % error in our uncertainty in the number of cattle, as recommended by the IPCC (2006).

The data regarding the number of cows are proprietary information that are not consistently reported by any one agency, and the agencies do not communicate with each other. Regulatory agencies should strive to identify facilities and the number of animals in a timely manner, and they should communicate with each other. Further, the number of milk cows varies interannually (as they only lactate for part of the year and are considered dry cows the remainder of the year), and the animals are sold and traded. These factors make this information surprisingly difficult to estimate.

Total enteric emissions for all cattle are 310.0±22.9 Gg CH4/yr for method E1, 376.7±31.2 Gg CH4/yr for method E2, and 354.8±71.0 Gg CH4/yr for method E3. We did not find statistically significant differences between the three methods of calculations of enteric CH4 emissions for either milk cows or all cattle in the state (Table 2, Fig. 2a). Statewide enteric emissions for milk cows only are 210.5±15.6 Gg CH4/yr for method E1, 231.1±19.2 Gg CH4/yr for method E2, and 215.3±43.1 Gg CH4/yr for method E3. We found relatively consistent proportions of enteric fermentation CH4 emissions of milk cows to total cattle. Milk cows account for 71 %, 67 %, and 61 % of total enteric emissions based on methods E1, E2, and E3, respectively. The difference in enteric emissions between the milk cows and total cows is due to the fact that non-milk cows produce significant amounts of enteric methane emissions.

Total manure management emissions for all cattle are 315.1±30.6 Gg CH4/yr based on M1, 340.0±111.2 Gg CH4/yr based on M2, and 368.0±110.4 Gg CH4/yr based on M3, the farm-specific method. We did not find statistically significant differences in manure management emissions between the methods of calculations for either milk cows or all cattle (Table 2, Fig. 2b). Total manure management emissions for milk cows only are 310.9±30.2 Gg CH4/yr based on M1, 334.9±109.5 Gg CH4/yr based on M2, and 359.8±107.9 Gg CH4/yr based on M3. The fraction of manure emissions that comes from the milk cows is greater than 98 % for all three methods. This is because the manure of non-milk cows is primarily managed in ways with very low methane emissions, including daily spread, on drylots, or on pasture. The difference between the emissions from milk cows alone and emissions from the total dairy herd is smaller than the uncertainties in manure emissions.

We report the statewide uncertainty in enteric emissions to be 7.4 %, 8.3 %, and 20 % for E1, E2, and E3, respectively (Table 2). The facility-level standard errors for enteric fermentation we calculated are 21.3 % for E1, 33.5 % for E2, and 35.6 % for E3. We find that sensitivities in enteric fermentation differ between the three methods (Table 3). E1 is most sensitive to the number of cows (n) at a facility. E2 is equally sensitive to n and ef2, followed by the DMI of lactating cows. E3 is most sensitive to DMI, followed by n.

We report the statewide uncertainty in manure emissions to be 9.7 %, 32.7 %, and 30 % for M1, M2, and M3, respectively (Table 4). The facility-level standard errors for manure emissions we calculated are 51.0 % for M1, 47.8 % for M2, and 73.5 % for M3. While the uncertainty for M1 is smaller than M2 and M3, this is due to the relative simplicity of the equation, with fewer propagated errors, rather than being the inherently best model. A recent report determined that the CARB methodology underestimates manure methane emissions (NASEM, 2018). Here, all three of our methods are most sensitive to the lagoon MCF (45.0 %–80.2 %), followed by n cows (7.2 %–15.3 %), except for M3, which is sensitive to the fraction of manure allocated to bedding (41.5 %) (Table 4). Our data on MCF for lagoons are only based on nine observational studies from outside California (Owen and Silver, 2014), so more measurements are needed to reduce this uncertainty. Further, there is little information on the amount of manure used for bedding. Overall, our uncertainty analysis is based on limited data from very few dairies.

Total uncertainty in CH4 emissions at the facility scale (E3 + M3) is 35.6 %; 81.9 % of the uncertainty is due to uncertainty in manure emissions, while 18.1 % of the uncertainty is due to enteric emissions. The higher uncertainty in the manure emissions than enteric emissions is due primarily to our uncertainty in facility-level manure management practices and the limited information on lagoon MCF.

Using the farm-specific method (E3 and M3), the two largest sources of CH4 from California dairy farms are manure emissions from lagoons (34.8 %) and enteric fermentation (51.5 %) statewide. Of manure management CH4 emissions, 96.3 % came from lagoons statewide, 1.8 % from solid storage, 0.4 % from liquid/slurry, 0.2 % from pasture, and 0.0 % from drylot and solid spread. Of the three geographic regions, the majority of manure management CH4 emissions came from the Central Valley (94.9 %), with only 1.7 % of manure emissions from the North Coast and 3.4 % from Southern California. Per-cow manure management emissions were also highest in the Central Valley (0.24 Tg CH4 per milk cow per year) due to the predominance of lagoons as manure management practice, compared to the North Coast (0.12 Tg CH4 per milk cow per year) and Southern regions (0.15 Tg CH4 per milk cow per year). In the 153 North Coast dairies, the 46 931 cows encompassed 1.7 % of calculated manure emissions and 3.2 % of calculated enteric emissions. The 75 dairies with a total of 77 122 cows in the Southern dairies made up 3.4 % of calculated manure emissions and 5.3 % of calculated enteric emissions.

With these emissions data, we also calculated enteric–manure ratios, which can be useful for methane mitigation planning. Mitigation strategies for dairy methane generally target either enteric or manure emissions, affecting this ratio. Manure management emissions per cow are much more variable than enteric emissions regionally, as manure practices vary more than feeding regimes. Therefore, differences in enteric–manure emissions are likely due to differences in manure management. The enteric–manure ratio of CH4 emissions in the North Coast is the highest, at 2.0, the enteric–manure ratio in the Southern dairies is 1.7, and in the Central Valley the ratio is 1.0 (Fig. 4). These differences are primarily due to the differences in manure management and cow housing type across regions: the Central Valley primarily uses flush systems, storing a large percentage of manure in lagoons, while North Coast and Southern California dairies tend to have scrape systems and drylots, respectively. Because lagoons have the highest MCF, the Central Valley has the highest per-cow emissions and lowest enteric–manure CH4 ratios. The CARB inventory also shows a statewide enteric–manure ratio of 1.08, which is primarily influenced by the large number of dairies in the Central Valley (CARB, 2014). The enteric–manure ratio also has implications for verifying mitigation effectiveness, as strategies that reduce either enteric or manure emissions should alter this ratio. If emission signatures of enteric fermentation differ from those of manure management, such as the δ13C–CH4 isotopic signature, it may be possible to use downwind or regional measurements of these signatures and their changes with mitigation to quantify enteric–manure ratios.

We compare this spatially explicit facility-level database with three other existing bottom-up spatial inventories, the spatially explicit EPA model (Maasakkers et al., 2016; comparable to E1 + M1), the Hristov model (Hristov et al., 2017, comparable to E2 + M2), and the CALGEM model (Jeong et al., 2012, 2016), by aggregating these estimates to 0.1∘ × 0.1∘ resolution to match the spatial scale of these other products (Fig. 4). The EPA model and the Hristov model were both developed for the contiguous United States, while CALGEM was developed for California only. First, we note that there are no significant differences in the statewide total methane emissions or methane emissions on a per-cow basis amongst the three products. However, there are differences in how manure is treated. CARB estimates that 76 % of manure is stored as a liquid, in either lagoon or liquid/slurry, while Hristov assumes that all manure is in lagoon or liquid/slurry, which are the manure treatments with the two highest emissions factors (Hristov et al., 2017). Thus the Hristov estimates are consistently higher than those of CARB and this farm-scale estimate.

We determined Pearson's correlation coefficients using R to test differences in spatial patterns between inventories. CALGEM is the closest to Vista-CA emissions (E3 + M3), with a Pearson's correlation coefficient of 0.77. The estimate of Hristov et al. (2017) is the second closest, with a Pearson's correlation coefficient of 0.58, but it tends to overestimate emissions in the Central Valley, including hotspots of methane emissions. The estimate of Maasakkers et al. (2016) matches the least, with a Pearson's correlation coefficient of 0.25, and tends to underestimate the hotspots of methane emissions in the Central Valley. The other models also have emissions in areas where Vista-CA does not have dairies (shown in gray in Fig. 4). The estimate of Hristov et al. (2017) includes the largest emissions area where Vista-CA does not show dairies, mostly in the lower Central Valley and Southern regions, though also in the North Coast. The estimate of Maasakkers et al. (2016) follows, with additional emitting areas primarily in the lower Central Valley. CALGEM has the fewest areas that are not in Vista-CA, mostly in the North Coast and Southern regions of California.

We found that existing solid separators reduce statewide manure CH4 emissions by 26.2 Gg/yr (8.0 %). This estimate assumes that half of all separated solids are used as bedding, and one-third of the C of separated solids is emitted as CO2, rather than CH4, as with other solids.

We estimated the effects of anaerobic digesters on CH4 emissions at 100 dairies in the Central Valley that have or are scheduled to have anaerobic digesters in 2017–2019 (CDFA, 2020b, Fig. 5). Following the USEPA, we assume a 75 % efficiency in anaerobic digesters (Lory et al., 2010; Charrier, 2016). We predict a total reduction of CH4 emissions by 36.5 Gg CH4/yr. This represents a 51.9 % decrease in manure emissions and a 25.9 % reduction in total (manure + enteric) emissions from dairies with these digesters, resulting in a 10.9 % decrease in statewide manure emissions and a 5.3 % decrease in total (enteric + manure) statewide dairy emissions. However, limited data exist on farm-scale emissions before and after digesters, or on the efficiency of digesters.

Our estimate provides a baseline against which the effectiveness of digester systems to reduce CH4 emissions can be assessed. Current top-down measurements of CH4 emissions in California are associated with large uncertainty and are not likely to capture signals of this magnitude. Jeong et al. (2016) inversion modeling posteriors suggest a 25 % error in CH4 emissions in the California Central Valley, but grid-scale error is much higher. The 95 % confidence intervals for the Central Valley are 1020–1740 Gg CH4/yr (Jeong et al., 2016), where the uncertainty in emissions is an order of magnitude larger than the reduction we expect to see from the digesters.