Daniel T. Kaffine, Brannin J. McBee, and Jozef Lieskovsky* (email@example.com)
Production of electricity from wind energy has risen rapidly in the last decade, with installed capacity roughly doubling every three years in the United States. As of the end of 2010, installed wind capacity in the United States exceeded 40 gigawatts (GW) and accounted for 2% of total electricity generation. Technological advances in wind turbine design, control and siting have led to falling costs per megawatt-hour (MWh) and increased the penetration of wind energy into the power sector. In addition, government subsidies and policies have also played an important role in encouraging wind power production. For example, a majority of states have implemented Renewable Portfolio Standards mandating that a percentage of total state electricity generation be derived from renewable sources, and the federal government provides a Production Tax Credit of $22 dollars per MWh to wind power producers.
Government support for wind power development is primarily predicated on the environmental benefits of avoided emissions, such as sulphur dioxide (SO2), nitrogen oxides (NOx), and carbon dioxide (CO2). It is these avoided emissions that form the focus of our research efforts. In particular we ask, what is the emissions savings rate for SO2, NOx and CO2 per MWh of wind power produced, and how does that savings rate vary across regions with different existing generation mixes? To answer these questions, we consider more than 50,000 hourly observations of wind generation and emissions from the territories of the Electric Reliability Council of Texas (ERCOT), California Independent System Operator (CAISO) and the Midwest Independent System Operator (MISO).
Electricity generation in the United States relies heavily on fossil fuel sources. As of 2010, coal accounts for 44% of total generation while natural gas accounts for 25% of total generation, compared to 18% for nuclear, 8% for hydropower, 2% for wind power, and < 1% each for solar, geothermal and biomass. However, research suggests that calculating the emissions savings from wind by replacing an average unit of generation and using average emission rates is an incorrect methodology. First, rather than displacing an average unit of power generation, wind is likely to displace generation from higher marginal cost sources that can easily accommodate wind power on the grid - most likely natural gas. Second, average plant emission rates may not appropriately reflect the actual emissions savings from wind generation, as rapid ramping of fossil fuel plants (known as cycling) to accommodate wind is emissions-intensive.
These concerns have even led some to claim that wind power produces little or no emissions savings. Given the widely varying assumptions and findings in previous research, there is clearly a need for a careful analysis of actual changes in emissions associated with wind generation. Such an analysis must capture both the marginal unit of generation displaced by wind as well as the marginal emissions from that displaced generation. This study helps to fill this crucial gap in the literature, and provides emission savings estimates based on large sample empirical data that will be of use to policymakers and future researchers. One advantage of our approach over much of the prior literature is that we are able to look at how emissions actually responded to changes in wind generation, as opposed to relying on dispatch and emission rate assumptions and modeling.
We estimate the emissions savings from wind generation across several Independent System Operator (ISO) territories in the United States. We exploit exogenous variation in hourly wind generation levels to identify the effect of wind generation on total hourly emissions of SO2, NOx, and CO2. Thus, our reduced-form estimation implicitly captures both the marginal unit of generation displaced by wind, as well as the marginal emissions reduction from that unit. In total, our rich data set contains over 50,000 hourly measurements of wind generation and emissions across Texas, California, and the Upper Midwest. We focus on ERCOT 2007-2009 (Texas), CAISO 2009 (California), and MISO 2008-2009 (Upper Midwest) for two reasons: first, they contain a significant portion (roughly 60%) of total wind capacity and generation in the United States, and second, these territories vary substantially in terms of their existing fossil fuel generation mix. MISO’s generation is dominated by coal, CAISO’s generation is dominated by gas, and ERCOT’s generation is roughly an even mix of both. This variation in existing generation will prove crucial in determining the emissions savings from wind generation in each territory.
We find that emissions savings across territories are less than the hypothetical savings based on average emission rate analysis. Nonetheless we do find that emissions savings from wind generation are statistically different than zero for most pollutants and vary substantially across territories. In coal dominated MISO, we find emissions savings of 4.9 lbs/MWh for SO2, 2 lbs/MWh for NOx, and 1 ton/MWh for CO2. By contrast in CAISO, where wind typically offsets gas generation, we find emissions savings of 0.0 lbs/MWh for SO2, 0.05 lbs/MWh for NOx, and 0.3 tons/MWh for CO2. Generation in ERCOT is roughly evenly balanced between coal and gas, and we find that emission savings in ERCOT fall in between MISO and CAISO, with emissions savings of 1.2 lbs/MWh for SO2, 0.7 lbs/MWh for NOx, and 0.5 tons/MWh for CO2. These results suggest that emissions savings are strongly driven by differences in existing generation mix - coal-intensive territories experience larger reductions in emissions due to wind generation.
Our dataset consists of over 50,000 hourly observations of total wind generation in MWh and total emissions in pounds of SO2 and NOx and tons of CO2 in ERCOT (2007-2009), MISO (2008-2009), and CAISO (2009).
Hourly emissions data is sourced from the Environmental Protection Agency’s (EPA) Continuous Emission Monitoring Systems (CEMS) program, which requires coal and gas power units with over 25 MW of capacity to submit hourly data on SO2, NOx and CO2 emissions. These emission reports are required by the EPA to monitor compliance with emission regulations, and strict quality assurance standards are in place to guarantee the accuracy of emission measurements. However, emissions per territory are not explicitly reported under CEMS. To determine which units operated in a given area, each unit is spatially referenced using latitude/longitude against the spatial footprint of each operating territory, obtained through the operating territory’s website. Units that fall under the spatial footprint of the territory are assumed to provide generation to the corresponding territory and the emissions from that plant are included in the territory’s total emissions. Thus, an observation consists of the total hourly emissions of each pollutant by territory, representing the sum of emissions from all units.
The hourly wind generation data is acquired from each operating territory (ERCOT, MISO, CAISO) and represents total electricity generation from wind turbines operating in the territory. This publicly available data, directly reported by the operating territory, is posted on the operators’ websites. It should be noted that the availability of hourly wind generation data is the primary limiting factor of our analysis, both in terms of the time period and territories over which data is available. Wind generation data is available for ERCOT from 2007, for MISO from 2008, and for CAISO from 2009. We collected this data for each of these three territories through December 31st, 2009. The 50,000 hourly observations of wind generation in our dataset thus provide a detailed look at actual wind generation levels across the three territories. Furthermore, the three territories we study account for over 60% of total wind capacity and generation in the United States.
Temperature is a key determinant of electricity demand and thus emissions. Temperature data for all territories is taken from the National Oceanic and Atmospheric Administration’s (NOAA) hourly temperature database, which is available through subscription to NOAA’s hourly surface data. A population-weighted average is created for each operating territory utilizing the major population centers within the territory’s footprint. These average hourly temperatures are used throughout the analysis.
During the 2007-2009 period, average yearly total generation in ERCOT was 306.3 million MWh, with wind power representing 4.7% of total generation. Coal accounted for 37% of total generation and gas accounted for 43% of generation. ERCOT average emission rates across all forms of generation was 2.63 lbs/MWh for SO2, 0.72 lbs/MWh for NOx, and 0.64 tons/MWh for CO2. Figure 1 displays the ERCOT generation mix from November 5th through November 12th in 2008. This figure reveals substantial variation in wind power produced at any given point in time. During high load periods (middle of the day), substantial gas generation is online, and variation in wind power is accommodated by gas cycling. By contrast, during low load periods (overnight), limited gas generation is available, and variation in wind power is accommodated by coal cycling (as evidenced by the drop in coal generation relative to the base level output during periods of large overnight wind generation).
Figure 1. Generation Mix in ERCOT (November 5-12, 2008)
During the 2008-2009 period, average annual total generation in MISO was 566.2 million MWh, with wind power representing 2% of total generation. MISO relies primarily on coal generation with 80% of total generation coming from coal and only 2.7% from gas. In coal dominated MISO, average emission rates are substantially higher than in ERCOT, at 5.74 lbs/MWh for SO2, 2.15 lbs/MWh for NOx, and 0.86 tons/MWh for CO2.
In 2009, total generation in CAISO was 178.6 million MWh, with wind power accounting for 3.2% of total generation. CAISO has no coal plants in their territory, while 35% of total generation came from gas. Due to the lack of coal plants, average emission rates in CAISO were much cleaner than ERCOT or MISO, at 0.00 lbs/MWh for SO2, 0.37 lbs/MWh for NOx, and 0.16 tons/MWh for CO2. The heterogeneity in emission rates and generation sources across these three territories will prove important in understanding the emission savings from wind emissions.
Our identification strategy hinges on exploiting the exogenous and stochastic variation in hourly wind power generation. The reduced-form model presented below captures the systematic response of conventional generation (and thus emissions) to hourly fluctuations in wind generation. Total emissions Eirt of pollutant i in territory r at hour t are separately regressed by territory against total hourly wind generation in each territory Wrt (in MWh), average hourly temperature Trt and its square Trt2 in each territory, and a vector of other control variables Xt:
The coefficient of interest is βir, which represents the marginal change in emissions in each territory due to a change in wind generation. Thus, for every MWh of wind generation produced in hour t in territory r, this coefficient represents the reduction in lbs/lbs/tons of SO2/NOx/CO2. The remaining covariates control for trends in wind generation and emissions that may be correlated, leading to erroneous interpretations of βir. Due to heating and cooling needs, temperature is a strong driver of electricity demand and emissions and thus is explicitly included as a covariate along with the square of temperature, to account for non-linearities. The remaining covariates in the vector Xt are fixed effects to account for other sources of variation in emissions. Hourly fixed effects are included to account for diurnal wind variation over the course of the day, which can be correlated with changes in the electricity demand profile. On average, winds are strongest in the early morning hours when electricity demand and emissions are at their nadir, and therefore failing to control for this hourly variation would lead to an overestimate of the emissions reductions from wind.
Over the sample period, wind capacity steadily increased, which may be correlated with changes in demand and emissions driven by macroeconomic effects unrelated to wind generation. To account for these longer-run trends, month-year fixed effects are included, leading to identification of the effect of wind generation on emissions through within-month variation. Finally, though wind generation is not correlated with the day of the week, day-of-week fixed effects are included to capture within-week variation (primarily between weekdays and weekends) in electricity demand and emissions.
Emissions savings estimates
The estimates of the emission savings from wind generation in ERCOT, MISO, and CAISO are presented in Table 1. The reported coefficients in the first row can be interpreted as the lbs/lbs/tons of SO2/NOx/CO2 emissions reduced per MWh of wind generation. The first panel represents the emissions savings by pollutant due to wind power in ERCOT from 2007-2009. Each MWh of wind generation in ERCOT on average reduced SO2 by 1.235 lbs, NOx by 0.739 lbs, and CO2 by 0.484 tons. All coefficients are very statistically significant. It should be noted that our estimates for ERCOT are similar to those in Novan (2010) who employs an estimation strategy comparable to that presented here.
Table 1. Estimation results for emissions reductions from wind generation by territory
The next panel represents emission savings in coal dominated MISO from 2008-2009, where each MWh of wind generation in MISO reduced SO2 by 4.890 lbs, NOx by 1.995 lbs, and CO2 by 1.025 tons. Again, all coefficients are statistically significant and are larger than the estimated emissions savings in ERCOT. By contrast, in the final panel for gas dominated CAISO, we find emissions savings in 2009 of 0.008 lbs/MWh for SO2, 0.054 lbs/MWh for NOx, and 0.299 tons/MWh for CO2, with significant coefficient estimates for NOx and CO2.
The estimated emission savings in ERCOT using average plant emission rates found in Cullen (2010) provide a useful reference point. Cullen calculates that 3.15 lbs of SO2, 1.05 lbs of NOx, and 0.79 tons of CO2 were avoided per MWh of wind power in ERCOT from 2005-2007. By contrast, our estimates for ERCOT (2007-2009) above find substantially smaller emission savings rates of 1.235 lbs/MWh for SO2, 0.739 lbs/MWh for NOx, and 0.484 tons/MWh for CO2. This difference is likely driven by emissions associated with cycling – relying on average emission rates for emissions savings calculations will likely overestimate emissions offset per MWh of wind power.
The importance of the generation mix can be seen by comparing estimates of emission savings across territories. Figure 2 displays emission savings per MWh against the percentage share of coal generation in each territory (fit with a quadratic polynomial). Each pollutant exhibits an upward trend with respect to coal share, with emissions savings from SO2 displaying the steepest increase. The stronger dependence of SO2 emission savings on coal share is driven by the fact that coal is the only source of SO2, while NOx and CO2 are also produced by gas. Each pollutant also exhibits a convex response to coal share. Territories with low to moderate coal share typically have a substantial number of gas plants, and it is these gas plants that are used to accommodate wind on the grid, and thereby relatively smaller emission savings are generated. As coal share increases and gas share decreases, the ability of gas to accommodate wind is also diminished, which in turn implies that base load coal is cycled more frequently to accommodate wind, increasing emission savings.
Figure 2. Emissions savings per MWh of wind power against the fraction of coal generation.
Plotted points indicate estimated emissions savings rates by pollutant in CAISO, ERCOT and MISO (left-to-right). Plotted lines represent fitted quadratic polynomials for each pollutant.
In the preceding sections, we provided estimates of emissions savings from wind power in Texas, California and the Upper Midwest. Our reduced form approach leverages the exogenous variation in hourly wind production to identify the impact of wind power on system-wide emissions. Looking to the future, accommodation of wind onto the grid will become an increasingly important issue, as wind was the second largest new source of installed capacity in the U.S. in 2008 and 2009. We show that the emissions savings corresponding to this growth in wind power will vary substantially depending on the fuel source displaced by wind. In particular, the share of coal in the existing generation mix strongly influences emissions savings from wind. This suggests that there may be benefits to adjusting the existing Production Tax Credits to reflect the regional emission savings (or a proxy thereof) from a MWh of wind power.
Based on current trends, several competing forces will influence emissions savings from wind power in the future. First, gas is the leading source of new generation capacity in the U.S. This would tend to increase the gas offset by wind power and reduce the emission savings associated with wind (although of course electricity generation from gas itself is less emissions-intensive than coal). Second, as wind capacity grows, the ability of existing gas generation to accommodate wind power will diminish, leading to increased cycling of coal plants, potentially increasing emissions savings. Finally, increasing wind capacity will likely require an increase in ramping of thermal generation, as the magnitude of shifts in wind speed is amplified into larger swings in aggregate wind generation. This increased cycling of thermal generation (in magnitude and potentially frequency) may erode the emissions savings per MWh of wind power as thermal generation is utilized less efficiently to accommodate wind. While it is unclear which of these effects will win out, it is clear that the resulting emission savings of wind power will depend critically on the factors highlighted in this study. As such, this research provides a transparent empirical framework for updating and refining emission savings estimates as data on wind generation in more territories and across longer time periods becomes available.
Callaway, D. and M. Fowlie (2009). Greenhouse gas emissions reductions from wind energy: Location, location, location? Working paper.
Cullen, J. (2010). Measuring the environmental benefits of wind-generated electricity. Working Paper.
Kaffine, D.T., B. McBee, and J. Lieskovsky (2011). Emissions savings from wind power generation: Evidence from Texas, California and the Upper Midwest. Working Paper.
Novan, K. M. (2010). Shifting wind: The economics of moving subsidies from power produced to emissions avoided. Working paper.
* Kaffine - Division of Economics and Business, Colorado School of Mines. McBee- Research and Development, Bentek Energy LLC, Lieskovsky - Research and Development, Bentek Energy LLC. For further details on the research discussed in this article, see Kaffine, McBee, and Lieskovsky (2011) “Emissions savings from wind power generation: Evidence from Texas, California and the Upper Midwest.”
 For further discussion on the limitations of existing methodologies, see Callaway and Fowlie (2009) “Greenhouse gas emissions reductions from wind energy: Location, location, location?”; Cullen (2010) “Measuring the environmental benefits of wind-generated electricity”; and Novan (2010) “Shifting wind: The economics of moving subsidies from power produced to emissions avoided.”
 It should be noted that this measure of total generation, as reported by CAISO, also includes net imports, which constitute over a quarter of the reported total generation. The total generation reported above for ERCOT and MISO also include net imports, though they are much smaller as a percentage than CAISO (1% and 5% respectively).
 Standard errors for all estimations reported below correct for heteroscedasticity and autocorrelation. Newey-West standard errors are reported with a 5-day lag for SO2, 1-day lag for NOx, and 3-day lag for CO2.
 Alternative specifications with month and year fixed effects or flexible polynomial time trends yielded estimates nearly identical to those presented below, as did estimations with month-hour fixed effects. In addition, estimations were run with heating-degree day and cooling-degree day specifications instead of temperature, generating coefficients and standard errors that differed only trivially from those reported below.
 See Kaffine et. al (2011) for further discussion of the potential impact of imports/exports on emissions savings, discussion of the impacts of wind volatility on emission savings, estimations based on daily aggregations of wind power and emissions, estimations including load measure and other robustness checks, back-of-the-envelope calculations of national emissions savings based on generation mix and wind generation by state, and calculations of marginal benefits of avoided emissions by territory.