2.4 Resource Assumptions and Technology Characteristics
2.4.1 Wind and Solar
US-REGEN represents wind and solar energy resources based on a detailed characterization of their spatial and temporal variation and potential scale. Variability is modeled using gridded hourly data from the NASA MERRA2 dataset. This dataset provides key meteorological variables such as wind speed, solar irradiation, and temperature, from 1980 to the present. While long-run averages are used to characterize grid cells in terms of resource quality, hourly profiles used in the model solution are based on a single representative year, typically 2015. This is the same base year as the corresponding load data, or the temperature data used to derive load in the end use model, to ensure synchronicity between wind, solar, and load and to avoid damping variance through multi-year averaging. The model does have the capability of varying the representative year to explore the implications of inter-annual variation.
Potential capacity is estimated based on a state-level dataset developed for EPRI by AWS Truepower (2012). The AWS dataset was developed using a screening process to determine the amount of land area available for wind and solar installations in each state. This screening process disallowed siting in areas such as National Parks, excessively sloping land, and developed lands. For wind, total potential capacity was calculated as the sum of potential facility siting locations. For solar, available land was ranked by resource quality, and potential capacity was approximated based on the top 1% of this available land.
2.4.1.1 Wind
Wind output profiles were derived from 2015 gridded hourly wind speed data from the MERRA-2 reanalysis dataset. We begin with the MERRA-2 variables referring to the east/west and north/south components of the wind speed vector at 50 meters, which are combined into a single vector extrapolated to various alternative hub heights using the MERRA-2 variable referring to the surface roughness coefficient, following the Monin-Obukhov specification.[1] Figure 2‑6 shows the long-run average wind speed at 100m for each grid cell, which were used to form classes based on resource quality. The number of grid cells in each class, combined with the state-level potential assessment from AWS Truepower, was used to derive total potential GW of wind capacity by resource class and state, shown in Figure 2‑7.
Wind speed at hub height is translated into power output based on assumed power curves for a range of turbine technologies. The power curve for a wind turbine reflects key design parameters such as cut-in speed, efficiency, and most importantly, the ratio of nominal generator capacity to the swept area of the turbine blades, known as specific power or specific capacity (measured in W/m2). The lower the specific power, the lower the wind speed at which output reaches nameplate capacity, and hence the higher the capacity factor, as there are more hours with output at full capacity. Conventional wisdom has held that a lower specific power is more suitable for lower quality wind sites, because there are fewer hours at high speeds to justify a larger generator. However, as wind penetration has increased, the value of generation at high wind-speed moments is reduced due to the supply glut effect, placing a premium on power capture at lower speeds. The results it that lower specific power turbines are being deployed even for higher quality sites.
In the current version of US-REGEN, the candidate power curves are based on three composite power functions estimated by AWS Truepower (2012) rather than on specific turbine models. The three curves, shown in Figure 2-8, translate roughly to: (Type 1) a 2 MW turbine with 80m blades, i.e. specific power of ~400 W/m2; (Type 2) a 3 MW turbine with 112m blades, i.e. specific power of ~300 W/m2; and (Type 3) a 2.3 MW turbines with 116m blades, i.e. specific power of ~218 W/m2. These three composite power functions correspond to a generational evolution in the types of turbines deployed in the wind industry. Older capacity installations used turbines resembling the first type, while newer wind plants look increasingly like the third. Similarly, while older wind turbines were installed at hub heights of 80m or lower, new vintage turbines typically use towers of 100m or higher. REGEN on-shore wind output profiles for a given region / wind resource class vary by vintage based on an assumed mix of turbine type and hub height as detailed in Table 2‑3.
| Vintage | Hub Height | Turbine Type |
|---|---|---|
| Existing in 2015 | 80m | Mix across Types 1,2,3, varies by state/region |
| 2015-2020 | 100m | 50/50 mix between Type 2 and Type 3 |
| 2020-2025 | 50/50 mix between 100m and 120m | 50/50 mix between Type 2 and Type 3 |
| 2025-2030 | 50/50 mix between 100m and 120m | Type 3 |
| 2030-2050 | 120m | Type 3 |
For off-shore wind, the assumed technology through 2030 is a fixed platform 120m hub height turbine of Type 2, increasing to 140m in 2035. While turbines used for off-shore are typically much larger than those used for on-shore in terms of nominal output and swept area, their specific power is generally in the range described by the Type 2 composite, rather than the lowest on-shore specific power rating described by Type 3. EPRI has data for floating turbines, but the cost is significantly higher than for fixed platform turbines.
2.4.1.2 Solar
US-REGEN considers three types of central station solar PV technologies (fixed tilt crystalline silicon, single-axis tracking, and double-axis tracking), concentrated solar 'power tower' technology with endogenous determination of the amount of thermal storage, and fixed tilt rooftop solar PV. Note that the adoption of rooftop PV is modeled within the end-use model, and the resulting output profile is an exogenous input into the electric sector model.
Solar output profiles were derived from gridded hourly radiation flux data from the MERRA-2 reanalysis dataset. We begin with the MERRA-2 variable referring to surface incident shortwave radiative flux, which we interpret as the global horizontal irradiance (GHI) in each hour / grid cell. From this flux we derive two related quantities, the diffuse component of GHI and the corresponding direct normal irradiance (essentially the direct component of GHI adjusted by the angle of incidence). The diffuse share is estimated by comparing GHI to a hypothetical horizontal flux at the top of the atmosphere, which is calculated based on the angle of the sun in a given hour and the coordinates of the grid cell. This ratio, always less than (or equal to) 1, is interpreted as a clearness index and is used to estimate the diffuse component. Essentially, the lower the clearness index, the higher the share of surface incident irradiance that is assumed to be diffuse. The estimation function is based on Ridley et al (2010). Once the diffuse component is estimated, the direct horizontal component is calculated as the difference between GHI and the diffuse component, and DNI is subsequently calculated by adjusting for the angle of incidence. DNI is always greater than (or equal to) the direct horizontal component because it reflects irradiance incident to a plane perpendicular to the incoming flux. Figure 2‑9 shows the long-run average GHI for each grid cell, which were used to form classes based on resource quality. The number of grid cells in each class, combined with the state-level potential assessment from AWS Truepower, was used to derive total potential GW of solar capacity by resource class and state, shown in Figure 2‑10.
Diffuse and direct irradiance are translated into output for a variety of solar photovoltaic technologies that are specified in terms of the orientation and tilt of the panels, which may be fixed or tracking the sun's position across the hours of the year. For fixed panels, we consider five distinct orientations (E, SE, S, SW, W) and five distinct tilt levels. The tilt levels are defined in terms of a proxy for optimal tilt, which is a function of latitude (roughly, the higher the latitude, the higher the optimal tilt angle). The five tilt levels are defined as 100%, 75%, 50%, 25%, and 0% of the optimal angle, where 0% refers to a flat horizontal panel (for which orientation is irrelevant). For tracking panels, we consider a single-axis configuration which has fixed tilt at the optimal angle and orientation matching the azimuth angle of the sun, and a double-axis configuration in which both tilt and orientation are adjusted so that the panel is always perpendicular to the incoming solar flux. For each grid cell and each hour, we calculate captured irradiance for a panel in all fixed and tracking combinations.
Captured irradiance has three components: direct, diffuse, and reflected. Each depends on the angle of the panel and the previously calculated direct and diffuse components of GHI. Direct captured irradiance is calculated based on the direct component of the incoming flux and the angle of the panel relative to the angle of incidence. Diffuse captured irradiance is calculated as a share of incoming diffuse component. Because the diffuse flux is coming down from all directions, the captured share is equal to 1 for a flat panel and less than 1 for tilted panels. The higher the tilt angle, the larger the "blind side" to the diffuse component, up to a hypothetical perfectly vertical panel, which would capture only half of diffuse irradiance. Finally, we assume that 20% of GHI is reflected (this fraction may vary depending on the albedo of the surface) and the direction is diffuse but with opposite sign. Hence a flat panel captures none of the reflected irradiance, and tilted panel would capture some, up to a hypothetical perfectly vertical panel, which would capture half of reflected irradiance.
While captured irradiance is calculated for all combinations of these variables, solar PV output profiles for the model are based on four distinct configurations: rooftop PV, fixed-tilt utility-scale PV, and single- and double-axis tracking PV. Rooftop solar PV is assumed to have fixed orientation and tilt, with a distribution over angles (see Table 2‑4). We assume that while rooftop installations attempt to configure panels in a south-facing orientation with optimal tilt, this is not always possible, resulting in a majority of installations being sub-optimal in one or both dimensions. We also represent fixed-tilt utility-scale PV, which is assumed to have south-facing orientation with optimal tilt. Single-axis and double-axis tracking are also represented as distinct configurations as described above.
| Flat | 25% | 50% | 75% | Optimal | |
|---|---|---|---|---|---|
| E | 0.01 | 0.01 | 0.01 | 0.01 | 0.06 |
| SE | 0.01 | 0.01 | 0.01 | 0.01 | 0.06 |
| S | 0.06 | 0.06 | 0.06 | 0.06 | 0.36 |
| SW | 0.01 | 0.01 | 0.01 | 0.01 | 0.06 |
| W | 0.01 | 0.01 | 0.01 | 0.01 | 0.06 |
Finally, captured energy at the panel is adjusted for temperature impacts on module efficiency, non-linear inverter losses, and a gross de-rating factor reflecting a range of factors not otherwise captured. Module temperature is estimated as a function of ambient temperature and incoming radiation. In full sun, module temperature can be 15-20 degrees C hotter than ambient temperature. Efficiency loss is calculated as a linear impact with coefficient 0.0045 per degree C delta between module temperature and an assumed rated temperature of 25 degrees C. This can result in impacts of roughly 10-15% in either direction for extreme hot or extreme cold conditions. Inverter are losses are assumed to be non-linear, such that losses are lowest in percentage terms when module output is highest but increase exponentially as output drops. This happens because a portion of inverter power loss remains constant regardless of input/output. The relationship between inverter loss and module output is calibrated such that losses at rated output are roughly 6%, increasing to roughly 26% at 10% of rated output. The implied average inverter efficiency is approximately 90%. In addition to inverter losses, we assume a gross de-rating factor of 0.9, i.e. additional 10% loss applied uniformly to the module profile, to reflect factors such as panel soiling, degradation, mismatch effects, and system availability.
Note that we currently assume inverter nominal AC capacity equal to solar module nominal DC capacity. In some installations, inverter capacity is sized lower than module capacity, which results in a higher nominal capacity factor (and potentially higher cost per nominal kW) because the ratio of nominal output to input is greater than 1. A smaller inverter may be used because losses on the DC side result in lower than nominal peak output, but it may also be used even when it implies curtailing or "clipping" some module output during peak conditions. Because these conditions occur only a few hours of the year, it may not be cost effective to size the inverter to capture fully all output. However, for utility-scale tracking configurations in high-quality resource areas, peak DC output as calculated by REGEN can exceed nominal output by as much as 15% (assuming nominal rating based on 1000 W/m2). In this situation, setting inverter capacity equal to nominal module capacity may be more suitable.
We also represent a concentrating solar power (CSP) technology with integrated thermal storage. The input profile for this technology is based on the DNI component only, as mirrors are assumed to track the sun's position and diffuse and reflected irradiance are not captured. Because of the high capital cost of CSP, it is assumed that this technology is only eligible for deployment in regions with sufficiently high resource quality, which we define as long-run average annual DNI of 2,600 kWh/m2. Figure 2‑11 shows the geographic area where this threshold is met, encompassing only seven states in the Southwest (CA, AZ, NV, NM, UT, CO, and TX). Because high-quality resource areas for CSP likely coincide with high-quality solar PV resource areas, the potential capacity constraint for solar within a given region / resource class is applied jointly across PV and CSP technologies.
(Class 1 is defined as > 2,800 kWh/m2, Class 2 as > 2,600 kWh/m2)
2.4.2 Biomass
Regional agricultural and forestry cellulosic biomass supply feedstocks are derived from the Forest and Agriculture Sector Optimization Model with Greenhouse Gases (FASOM-GHG)[2]. With FASOM-GHG, we estimate regional supply curves over time for the delivered cost of biomass for energy production, which could be electricity generation or a variety of industrial and transportation applications. The model endogenously and simultaneously accounts for food, feed, co-product, and other bioenergy market feedbacks and implications. The resulting biomass supply curves are inputs to the electric model for fueling potential dedicated new and coal retrofit biomass generation, as well as co-fired generation. Figure 2‑12 illustrates the regional differences in delivered biomass costs for electricity production. At the lowest prices, the South and Plains dominate supply with forestry residues. As price rises, dedicated energy crops in particular, in the Midwest, South, and Plains, become economic to produce, store, and deliver, as are modest supplies of agricultural and forestry residue feedstocks. At higher prices, more substantial supplies of agricultural residues and forestry residues, including whole trees, become available. Because bioenergy products can be substitutes and complements, bioenergy markets are not independent. Therefore, sets of consistent biomass supplies are being developed for the alternative uses of biomass across the end-use model (i.e., electricity, industry, transportation). For more details on the forestry and agricultural biomass supply modeling, see Appendix B.
2.4.3 Hydro and Geothermal
New additions of hydroelectric power are limited to planned or underway projects through 2015 (see Section 2.6.1). New additions of geothermal power capacity are constrained based on estimates of discovered and undiscovered conventional sites in the western regions by NREL (2007). Total new potential additions amount to roughly 30 GW by 2050. We also assume an improving capacity factor for geothermal power over time as a result of technical progress, from roughly 50% today to roughly 80% by 2050.
2.4.4 Fossil Fuels
Fossil fuel prices (for coal, fuel oil, and natural gas) and availability are specified exogenously in the electric sector model. Fuel prices are based on a selected EIA Annual Energy Outlook case projection; defaulting to the High Oil and Gas Recovery case corresponding to lower fuel prices.
2.4.5 Energy Storage
US-REGEN represents the key structural features of energy storage. The model endogenously determines investment levels and dispatch of storage on an hourly basis. Storage investment, charge, and discharge decisions are co-optimized with other technology investment and dispatch decisions. By default, US-REGEN represents four possible storage technology classes, and other technologies can easily be added. These storage classes include pumped hydro, compressed air, concentrated solar power, and a variety of configurations of battery storage. Each technology carries a charge penalty, which represents the ratio of input to output energy. Only compressed air has greater output than input because it also takes a natural gas input (1.44GJ/MWh) that converts at a loss to electricity. Each technology has an assumed lifetime for new and existing builds. No new pumped hydro is considered, due to lack of resource data. Concentrated solar includes the costs for the power block as well as the solar field and receiver and the storage unit must charge from its own field.
EPC stands for 'Engineering, Procurement, and Construction'. CAGR stands for 'Compound Annual Growth Rate'.
US-REGEN's characterization of battery costs is based on the EPRI Energy Storage Technology and Cost Assessment of the state of energy storage costs (EPRI, 2018) which includes a comparison of multiple analyses and specifies cost ranges for lithium ion and flow batteries. Battery cost projections include costs of the system, grid integration equipment, fixed and variable maintenance, and system decommissioning. As illustrated in Figure 2‑13 above, the report finds that total costs for a 4-hour battery system could fall to around $840/kW in 2030, with further declines over time. In US-REGEN, battery storage costs are represented as a linear function of both nominal output or power capacity (the "door"), expressed in $/kW, and nominal energy capacity (the "room"), expressed in $/kWh, so the total cost projections from the report are disaggregated into these "door" and "room" costs for use in the model. Unless specifically fixed for a given technology, the model can endogenously choose the ratio of energy capacity to output capacity, i.e. the duration of the system. Carried charge is assumed to have a loss rate of 10 percent per month.
| Pumped Hydro | Compressed Air | Concentrated Solar Power | Battery | |
|---|---|---|---|---|
| Indicative 2030 “Door” Cost [$/kW] | N/A | $1,500 | $900 (power block) | $100 |
| Indicative 2030 “Room” Cost [$/kWh] | N/A | $100 | $15 (storage); also costs for solar field and receiver | $100 |
| Duration [hours] | 20 | Endogenous | Endogenous | Endogenous |
| Charge Penalty | 1.2 | 0.8* | 1% loss per day | 1.1 |
| Lifetime [years] | 100 | 30 | 60 | 20 |
| Notes | No new investment considered | Natural gas input (1.44 GJ/MWh) | Must charge from its own field | Loss rate for carried charge (10%/month) |
*Additional natural gas input means <1MWh electricity needed for 1MWh of output
US-REGEN's dynamic formulation uses an aggregation of representative hours into load segments to significantly speed up computation.[3] This creates challenges for representing storage effectively, as chronology is lost by construction. Storage technologies are therefore typically only turned on for static model runs that use all 8760 hours. EPRI is currently testing possible approaches to represent storage in a smaller number of segments. However, the static model with full 8760 hourly resolution continues to offer the best assessment of the constraints and economic potential of integrating storage.
2.4.6 Hydrogen
The electric sector model includes several technologies for producing hydrogen fuel which can be either consumed for power generation or supplied directly as an end-use fuel. Hydrogen is an energy carrier produced by primary processes using other energy inputs. US-REGEN includes three pathways for hydrogen production, the first based on steam methane reforming using natural gas as the hydrogen source (the most common currently deployed technology), one based on coal gasification in which coal is the hydrogen source, and one based on electrolysis in which water is the hydrogen source and electricity is the fuel input. These technologies are summarized in Table 2‑5.
| Technology | Capital Cost ($/kg per day) | Fixed O&M ($/kg per day per year) | Variable O&M ($/kg) | Fuel input per output H2 (btu basis) |
|---|---|---|---|---|
| Steam-methane Reforming | 875 | 40 | 0.04 | Gas 1.1 Elec 0.03 |
| Steam-methane Reforming with CCS | 1300 | 50 | 0.06 | Gas 1.21 Elec 0.04 |
| Coal Gasification | 1584 | 63 | 0.04 | Coal 1.18 |
| Coal Gasification with CCS | 2376 | 86 | 0.06 | Coal 1.36 |
| Electrolysis | 1000 | 75 | 0.01 | Elec 1.42 |
Similar to electricity, the model chooses both the level of production capacity and the dispatch of hydrogen production, as well as its consumption for either generation or end-use, in each load segment, in order to explicitly represent the capacity factor of the production technology. Supply and demand of hydrogen are required to balance on an annual basis, based on an assumption of negligible storage costs to avoid explicitly representing chronology of storage balance. Costs of electrolytic hydrogen depend endogenously on the capacity factor of the electrolyzer and the price of the electricity during the production segments.
Enabling multiple pathways for hydrogen production and use allows US-REGEN to simulate its role as an energy carrier serving multiple purposes in the energy system. When hydrogen from coal gasification is used for power production, it approximates an IGCC generation process, though substituting hydrogen for syngas. Hydrogen from electrolysis that is converted back into electricity functions as a form of seasonal storage. Though electrolysis requires a significant source of water, water availability is currently assumed and not tracked directly as a limiting factor. Figure 2‑14 illustrates the pathways represented in US-REGEN for hydrogen supply and disposition.
The implication of Monin-Obukhov Similarity Theory is that wind speeds at different layers in the atmosphere are related according to:
speed(h) = speed(50) * ln(h / z0) / ln(50 / z0), wherehrefers to hub height in meters andz0refers to the surface roughness coefficient for the grid cell. The smoother the surface, the less the relative change in wind speed at higher layers. ↩︎FASOM-GHG is maintained by Bruce McCarl at Texas A&M University. Documentation is maintained at https://agecon2.tamu.edu/people/faculty/mccarl-bruce/FASOM.htmlopen in new window. ↩︎
See Section 2.5 Design of Aggregated Segments for more details. ↩︎