Introduction to the ReEDS Model: Version 2021
So, I am going to spend a good amount of time giving kind of the whirlwind tour of ReEDS. If you've been using ReEDS—can everybody see my—is this the correct screen on here? [Crosstalk] >>Wesley Cole: That looks fine, Stuart. >>Stuart Cohen: Okay. Yeah. So, if you've been using ReEDS for a while or if you've been at one of these meetings before, a lot of this is probably going to be [a] review. But there is some new content to kind of touch on some of the newer features that other folks will go into more depth later today and discussions that we want to have after that.
So, I'll start with just a really high-level overview before digging into some of the specific features, characteristics, and unique capabilities of ReEDS. So, the general idea of ReEDS is it's a tool to do kind of "what if" scenario analysis for the U.S. electricity system, and other countries we'll hear about ReEDS—India—as well.
But the model generally is used to look out several decades into the future to understand what types of investment decisions for generation, storage, transmission, carbon mitigation are least-cost optimal for various different scenarios. So, the model is taking inputs related to the various types of technologies that contribute to providing grid services, and it also accounts for a large range of electricity system constraints and realistic limitations in the system, and so it's also taking in account things like policies and regulations, obviously demand for electricity, but also requirements for reserve capacity and ancillary services. And then, there's also transmission constraints related to the power systems themselves, resource limitations, and so on. And so, it's intended to be a really applied approach to looking at this planning problem for a realistic electricity system.
So, there are several key—I'm going to go over several kind of key inputs and outputs for the model to get a sense of what goes in, what goes out. We all have heard the—or many of us probably have heard the—modeling term, "garbage in, garbage out." We hope that we are not putting garbage in so that we get useful information out.
And so, one of the key things that we have to start with is obviously the fleet and the existing capacity, both for generation and the transmission overlay. And so, this is a pretty complex-looking figure but it's showing that we actually are getting—are able to use a unit-level database in this model accounting for all the different resources, different technologies throughout the country. And then, the black lines and the widths of those lines—sorry there's no key on there for that, but that's representing the transmission overlay that connects different balancing areas in ReEDS, and I'll go a little bit into that kind of spatial resolution later. So, obviously, the grid requirements are one of the main inputs to the model, and so that, again, includes electricity demands, for which you have various different scenarios, and that's looked at at an aggregate level in terms of how it changes over the course of many years. But then, we also have chronological profiles for electricity demand.
Here it's showing some hourly data, which is what we start with before potentially aggregating to a more aggregate time-slice resolution. And then, another key input is obviously the cost and performance of the various technologies that are available to provide these various grid services. And so, that comes in a different—several different forms depending on the technology, which shown here is the idea of we might just have one cost projection for nuclear, a couple different options for natural gas.
We can see some of the split between combustion turbine, combined cycle, and CCS. But then, for things like batteries we have several different durations that have different options for cost and performance projections. For things like wind and solar photovoltaics (PV) we have various different classes of resource. And so, there's a lot of technology resolution that goes beyond just kind of the broad category. For temporal resource availability, the model actually does start with hourly data for wind and PV that starts from a grid resource assessment data that Anthony will talk about a little bit later today, and that's in order to get obviously the heterogeneity of the resource across the United States both in time and in space.
And so, that's—this is showing kind of how different types of resources are distributed across the country, and that's one of the key things that we use this model. And a lot of these are NREL-developed data sets that we have a lot of pride about. Another key characteristic here is obviously the definitions of policies, both what exists and what could exist. And so, this slide is just kind of giving the idea that we can look at things at the national level—carbon limitations, renewable or clean energy requirements—but also state-level things like those same types of policies at the state level but also renewable portfolio standards. And our general MO is that in a given version of the model it's representing all enacted policies up to that date, and then there's several different options for if you want to test kind of "What if this was enacted? Or that?" In terms of outputs, the key high-level thing that we generally start with when we're looking at the results is what is the mix over time, both on the generation side and the capacity side? And so, we can look at how things are being built, how they're being retired, and how that balance in mix changes, and how that correlates to the relationships between technologies in terms of how they might complement or compete with each other. And it's important to look at both capacity and generation obviously because not all technologies generate near their maximum available output.
And you have certain technologies, particularly storage, that have capacity but are really net users of energy, and so if we want to understand their deployment, we have to look at the capacity basis for those things. And in addition to that aggregate build-out, we're able to look at the investment decisions at a regional level, and so we can understand where these technologies are being built, not only at what time. And then, we can look at transmission expansion as well. And so, this is a little bit more detailed look, not kind of the straight-line representation of our transmission system. And this is demonstrating that more recently we've incorporated data that actually accounts for likely transmission paths between balancing areas, and so we have a better estimation of distances and thus losses along those lines. And we can do just a better analysis of transmission capacity expansion, not only when it's built what type of transmission is built—AC, DC, and so on.
We also like to look at a lot of different economic outputs, and so that can take several different forms. One of the things that we might start with is a net present value cost across all investment and operation costs throughout the entire time period. And so, this is the discounted cost calculation, so obviously the results depend on what you're using for a discount right here. But it's a way to just get a quick breakdown of how the economics are shaking out in terms of capital expenditures, fuel for fossil fuel or nuclear fuel, other operation and maintenance expenses, transmission—this is spur line transmission—and so on.
And so, you can kind of compare these aggregate costs across a scenario to see sort of the long-run economics in that way. We also like to look at a lot of marginal cost metrics, or—which can be interpreted in some ways as sort of price metrics, although there's a lot of caveats there. We're not representing all of the market rules, for example, of different ISOs, and so the price formation in ReEDS, it is based on kind of the shadow prices and the marginals off of the different constraints for things like demand and capacity. But there's no expectation that these prices are going to match what's in real markets.
The idea is that we can compare these types of marginal price and cost metrics across scenarios and see what's higher, what's lower, what are the different components of each? And so, we can break these prices down into things like the energy price per load or the capacity price for planning reserves. We can also take these costs and look at them in a more raw level and break them down over time. And you could look at these annualized, discounted, undiscounted.
And then, these costs and prices can—more recently we've created a module that does postprocessing and actually is intended to estimate kind of the conversion between some of these more wholesale marginal cost metrics and think about what would be the additional costs for transmission, distribution, administration, and so on in order to come up with an estimate of actual retail rates. And so, this is fairly new, but the point here is that obviously wholesale market prices are different than retail prices and a lot of people will care about a metric that is "What will show up in my monthly bill?" And so, we developed this to just provide another option for an economic metric to look at. We also like to look at a lot of the impacts of the scenarios on the environmental side. And so, obviously we're optimizing generation and dispatch in this model, and so we can look at the emissions implications of that. And more recently, and I think you'll hear more about this later, but we also have a new module to actually convert some of these air emissions into monetary health impacts using some literature data on kind of mortality, morbidity, and things like that.
So, that's a real quick high-level look at just some of the basic features and ideas involved in the ReEDS model. And so, now I'm going to dive just kind of the next level deeper into what are some of the specific characteristics of the model and what I like to say: "What makes ReEDS special?" because there are a lot of these tools out here, and we thank you for being here and choosing ReEDS. So, how does it really work? So, ReEDS is an optimization problem. If you're already very familiar with optimization problems, this is probably pretty straightforward. But the idea is you start with an objective—so, what is the one quantity that you want to maximize or minimize? In our case, it's minimize the total capital and operating costs of the electricity system.
And it does that subject to several different constraints on that system. I mentioned a lot of these already: energy balance, capacity requirements for planning and operating reserves, policy requirements, availability of renewable resources, limitations on how energy reserve or even things like renewable energy credits can get traded around, how the physical constraints on the power systems themselves and fuel supply, and also things like what—prescribed builds and retirements and so on. So, there's a lot of different constraints. Whoever has looked at the code knows this very well.
But the idea is you've got this overall objective: minimize the cost. You've got a lot of different constraints. We already talked about a lot of these inputs, and those involve not only the data but also the structures. And so, I'll talk about some of those structures a little bit later in more detail but we have, for example, 134 zones for which we balance supply and demand.
So, that's kind of the spatial resolution for representing a lot of the characteristics. And so, all of these different things—cost and performance, resource availability, the initial capacity mix—go as an input. And then, we talked about these outputs already, where we get the capacity mix, transmission. I haven't talked a whole lot about operations, but we'll get to that a little bit later, where there is a component of the optimization that involves a dispatch as well, and so we are actually optimizing the energy generation versus something a little bit simpler like assuming a capacity factor that people would do, like in an LCOE. So, ReEDS does do that. We actually optimize the dispatch as well.
And then, we talked about emissions and cost metrics. So, to break down what goes into the objective, we have all the different capital, fixed and variable O&M costs, fuel, kind of opportunity costs for providing ancillary services, policy payments. So, all of these things are being summed up. The objective is formulated as a 20-year net present value. And so, what it's doing is it's minimizing the costs in a specific model year but doing it with a 20-year economic life present value calculation.
And so, what that kind of means is if you're running the model kind of one step at a time, 1 year at a time, it's looking at that model year as if it's sort of a static, myopic representation of what the next 20 years would be like. So, there are some limitations to that but that's one of the ways in which we run the model, and really the more typical way. And so, the reason for that is that you can actually account for kind of a reasonable economic life, and that has—also incorporates things like the financing of new capital investment. And essentially, in the model it boils down to how you convert—how you would convert overnight capital costs to installed capital costs by accounting for things like interest during construction, any kind of tax rates and policies, debt/equity ratios.
And we can have these—some of these parameters be technology specific, and they can change over time. Like, if there's a tax policy that's enacted that has some kind of sunset, we can account for that change and how that would influence financing over time. I mentioned this briefly on the last slide, but one of the things we really want to make sure everyone understands about the ReEDS model is that it doesn't use LCOE at all. It's actually optimizing the operation. An LCOE number requires you to assume a capacity factor, so you're assuming up front how—what percentage of the year you're running, but ReEDS is actually doing that in the optimization. And so, we can calculate LCOE as an output, but it's not simply just looking and comparing LCOEs across technologies to make decisions.
And then, it also—another thing to be clear of is electricity prices or any kind of marginal price or cost metric, these are also outputs. There are not inputs. And so, this is not a price-taker model in any form. It's simply a least-cost model subject to constraints, and those constraints effectively define what the prices of the different kind of services on the grid would be. So, that's what the objective is doing. The way we typically run the model—and this is the only thing I'm really going to go over here—is this sequential solve mode.
And so, ReEDS really consists of kind of two major model optimizations. One is the supply module, and that's the one that is making all of the investment decisions, and it has a simpler form of representing electricity dispatch, at least the way we normally run the model because of computational limitations. But because of those challenges and the limitations involved in simplifying time resolution to make investment decisions, we have this separate module called Augur that functions to calculate capacity value of variable renewables and storage, curtailment of variable renewables, and energy arbitrage for storage, and that's actually using—and I'll go into this on the next slide—7 years of hourly data.
And so, the balance there is that that module has a lot higher—that part of the optimization has a lot higher time resolution and can do a lot better job characterizing these specific parameters that then inform the supply module and how it can make decisions around those variable renewable storage and all the other technologies in combination. And so, it steps through in time and each of these supply modules solves—it's minimizing that same objective before getting—feeding the existing—the resulting system into this Augur module that then calculates these parameters to inform the supply module. And the reason we do this is because these parameters depend on the electricity mix—the demand profiles, net load—and that changes over time, changes what you build. And so, it's important to account for that transition. And we'll look a little bit more later into sort of what that actually looks like.
So, for the time resolution—so, this is showing graphically what I've already talked a little bit about. So, in the main investment of the supply module we typically use this 17-time-slice resolution. And so, there's four time-slices per season plus this super peak to capture reserve requirements. And that at least allows us to capture seasonality of resources and chronological, diurnal profiles of resources to some extent. But obviously, that's fairly coarse. We have a lot of ongoing work where we're—that Patrick is going to talk about later—where we're actually building capability to define any time resolution you want.
But there's a lot of computational challenges with adding time resolution, and so we're still kind of working on what that will look like in kind of default solve mode going forward. But that's why we have this Augur module that's accounting for these hourly profiles, and it does a simplified dispatch across all of these hours, 7 years' worth, to assess these parameters that I mentioned before. Spatial resolution also has a lot of different forms. So, I'm just going to get all of this on there.
For renewables, we actually do have some capabilities where if you want you can actually represent some site-level renewable data. Again, always a trade-off between computational complexity and sort of accuracy. But we also generally have 356 resource regions for wind and CSP. These 134 zones or balancing areas is generally the resolution that we represent most technologies.
But again, there's subcategories of technologies and things like that that I'll go into on the next slide to get more resolution there. But this is where we're enforcing these constraints on things like electricity demand, planning reserves, operating reserves. And then, kind of going up, we can aggregate from there.
And so, there's—these boundaries all follow county boundaries, and so they're easily aggregated to states, a little bit less easily aggregated to things like ISO regions, NERC census divisions. But we have representations of those that are as close as we can get so that we can enforce constraints or define parameters at different levels of resolution in the model. On the technology side, we have quite a bit of technology differentiation in the ReEDS, in the model. I don't really think I'll go over—I'm not going to list everything here. These slides will be made available.
But we have various different types of the fossil technologies on that side. We're working on with the new version having even more options for things like different CCS capture rates—carbon capture storage, for those that are less familiar with that. We've got some also work looking at flexible configurations of CCS. For renewables, it's not like just in each region there's a cost for wind and a cost for solar. No, we have various different subcategories and resource classes for all the different renewable resource types. And then, I guess that'll go—this is classified incorrectly on the bullets, but we also are—now have hydrogen fuel combustion turbine technologies represented for storage, different battery durations, pumped hydro, and compressed air energy.
One of the newer additions is PV-battery hybrids, and so that allows you to share some components and potentially have some cost savings there. And then, some other more recent stuff that Brian is going to talk about a little bit later is some of these more demand-side technologies. So, we now have capabilities to actually invest in electrolyzers and steam methane reformers to produce hydrogen that's then used by these combustion turbines. We also have air capture, direct air capture technology in the model. And so, these are things that are getting added to start to have better—more options for representing decarbonization scenarios and true net-zero-carbon scenarios.
We also have upgrades available so you can upgrade between technologies. And then, one of the things that's near and dear to my heart, maybe not as many other people on the ReEDS team, but if you're interested in any energy-water interactions, there are some switches there where you can actually further distinguish all of these fossil technologies and CSP by its cooling technology and water source. And so, you can track where the power system is getting its water, how much it's withdrawing, how much it's consuming, and there are some options there as well. Even further than that, for the thermal fleet, the fossil and nuclear technologies, we also differentiate those by vintages. And this is actually user input.
By default—I honestly can't remember the number of vintages we have for each technology, but you can specific how many you want to represent, particularly for the existing fleet all the way up to a unit-specific representation of the existing fleet. And so, this doesn't necessarily preserve all of the unit-specific characteristics, but if you want, you can create enough vintages in a region such that every coal plant has its own vintage in that region and you can track that. And then, over time you can define vintages for new builds, like I want 2020 through '25 to be vintage "new two;" I want '25 through '30 to be "new three" and so on like that. And so, you have a lot of ability to track and even influence what happens to individual vintages if that's something that you want to—that you care about. In terms of the capital stock more generally—so, I talked about before we have a unit-level database primarily based on EIA data from the NEMS model that defines our kind of initialized fleet, that still the initialized year is 2010.
But then, it also has prescribed retirements and prescribed builds up until now and the near future of what we—what's already been announced. In addition to these prescribed retirements, there are also some options where you can do some scenarios around shortening, extending coal lifetimes, different substances around nuclear, relicensing and lifetimes. And you can always go into the unit database and play around with the numbers and have that show up in the model if you want to look at different scenarios for when things might retire. But in addition to that, the model does also endogenously retire if you don't recover a certain fraction of FOM costs. And so, this fraction can also be user-defined and it's just a factor that's there to add a little bit more flexibility in how you might want to think about retirement decisions are actually made in the real world.
I see there's a couple of chat questions, or chat messages. If somebody wants to cut me off to answer those in real time, let me know. But I'm not monitoring those as I go. >>Wesley Cole: We're good there, Stuart. >>Stuart Cohen: Okay.
>>Wesley Cole: I think—yeah. >>Stuart Cohen: So, on the fuel price side, I'll kind of start from the bottom because that's simpler. So, right now we're just using exogenous assumptions for coal and uranium prices. These come from the EIA Annual Energy Outlook. But for natural gas, since its usage and price formation is more tied to the electric sector, we have a lot more options there.
And so, we have different scenarios around resources availability that give you effectively high-, mid-, and low- priced natural gas. But we also have different approaches to representing formation of natural gas prices. So, ReEDS actually generally uses a supply curve representation, and the way that works is if you choose an AEO, say the AEO reference natural gas case and ReEDS over time uses exactly the same amount of natural gas as in the EIA AEO reference case, then the prices will be the same as the EIA AEO reference case. But if ReEDS has a different model, it's probably going to do different things.
And so, we actually use the price quantity pairs over time for this case to find the linear supply curve, and what that means is if ReEDS uses less gas in that particular case, prices will be lower. If it uses more gas, prices will be higher, and so on. And then, these linear supply curves—we've tested several different ways of parameterizing them, really in levels of aggregation.
And so, these are some of the different options that are available. I'm not going to go into everything in a lot of detail, but essentially, this last one we had supply curves with kind of two different components to the linear supply curve. One is a slope on the national side; one is on the census division side. And that's what we use by default, but you can look at these other approaches if you want. Or you can actually have just fixed exogenous prices if that makes more sense for your analysis. And that could be the case if you're running a scenario where you're forcing natural gas prices to zero.
These supply curves could do weird things when you get to near zero fuel usage. So, it might be important to think about some of these other options in that case. On the electricity demand side, again, some of our base options are based on the Annual Energy Outlook, but we also have several different electricity demand scenarios from NREL's Electrification Futures Study, and so that includes both the degree of electrification—those scenarios involved assumptions around how quickly electrified transportation and building energy systems are adopted and how fast the technology improves. And so, that essentially defines different magnitudes of electrified demand growth. But there's also options for defining how much of that electrified demand is assumed to be flexible. And so, this flexible demand actually gets optimized within the ReEDS dispatch, and so it's kind of a carved-out portion of demand you can shift around.
And so, there's different scenarios for that. They're largely derived on essentially how flexible you think that we can manage electric vehicle charging. Just other things to mention on the load side. We're accounting for losses on the transmission systems and in storage systems, round trip efficiency losses. And we talked about the 7 hours of hourly load data.
And the demand-side technologies have come up as well. And so, if you build these things, their effect on electricity demand also get accounted for in the model. On the reserve side—so, first I'll talk about operating reserves. And so, this is kind of the ancillary service products that we represent in the model. In the ReEDS, there are three we represent separately—flexibility, spinning, and regulating—that are defined by kind of how fast—or how long—how fast you need to be able to supply a certain amount of capacity, as well as the quantities of renewable energy and load. And so, for example, flexibility reserves are a function of the generation from wind and the capacity from PV.
And all of these numbers come from a literature publication that came out several years ago. And so, for example, flexibility is only based on the renewables. Spinning reserve requirements, based on a percentage of load. Regulations, a little bit of both.
And then, in terms of what technologies can do, we use technology-specific ramp rates to define how much they could provide toward these different ancillary services. And then, we also—the objective function has opportunity costs associated with providing those reserves. And again, the way the constraints are set up are that you can only provide these if you have free capacity that you're not using for energy. And so, that's for meeting electricity demand, and so that's kind of as well.
We do also allow you to essentially use available transmission capacity to contribute to reserves. So, that's like saying if a neighboring balancing area has excess capacity, they could send through a line to you if you absolutely needed it. That's accounted for as well. And the same thing with storage: It can provide reserves but you need to make sure that there's sufficient energy available. Next, we have the planning reserves.
And so, this is actually a relatively important constraint. If you look closely at sort of the example price, marginal price data that I showed previously, you would have noticed that this planning reserve requirement is the second-biggest contribution to kind of the total marginal cost metric. And so, this is the requirement for capacity to ensure resource adequacy.
Essentially, can you meet demand in the worst-case peak net load situation? And so, that's based on the electricity demand plus an assumed reserve margin that comes from NERC data for a given region. And we're applying this to each specific balancing area and season. And in order to characterize this constraint, you need to assume what each generation, storage, and transmission technology can be credited for this kind of firm capacity or its capacity credit. So, for things that are fully flexible and dispatchable, like natural gas, that's going to be 100%.
But for variable renewables and storage, where you don't necessarily have enough resources available at a given time to provide the full energy or full power output, rate of power output of the systems, we have to do a calculation to assume what that capacity credit would be. And then, again, we can use this sort of available transmission capacity as well. And so, what that kind of looks like is you've got your peak demand, some reserve margin, you've got what the firm capacity is of these more flexible dispatchable technologies, and then you're adding onto that what is the—you're looking at the variable capacity and the net load of these other systems, and then doing a calculation for the capacity credit of wind, solar, and energy storage.
And this is one of the things that Augur module is doing with that 7 years of hourly data. Just to really briefly get into a little bit more what that looks like: So, it's essentially using—it's taking—it's starting with the load duration curve. It's using the variable renewables at the time to come up with a net load duration curve, and then it's using that to look at not only what the capacity value of what's already been built is but also a marginal capacity value for the next quantity of variable renewables that would be built potentially in the next solve year.
And so, since this is a nonlinear thing, that's one of the reasons why we do this process, so that we can kind of dynamically update this marginal capacity value, because it really changes over time. The solar you build today probably has nearly 100% capacity credit at today's peak net load, but 30 years from now when the peak net load is in the evening because of how much solar exists, then the next amount of solar that gets built can't really provide any more capacity at peak net load. And so, that's one of the things that you exactly see if you look at a high share of solar scenario.
This is just an example where since this is done for each region there's a distribution, but very clearly as you're building out more solar that capacity credit, or the marginal capacity credit drops off. And then, eventually it's very close to zero. And so, that's what these types of calculations are intended to capture.
And then, on the curtailment side it's similarly using this—it's calculated the net load. Here's our classic duck curve—and using that to make an assumption for what effectively is your curtailment rate. And the reason that—another reason we have this module, it's doing an hourly dispatch, it's because it's also dispatching storage. So, you can use storage potentially to reduce curtailments, and so that module can account for that and then calculate curtailments also based on some of that diurnal storage optimization. And then, ultimately what that looks like is that you can estimate the distribution of curtailment rates across regions and how that changes with the shares of those technologies. A little bit more on just a few other things, and then I will be done with my slides and we can open it up for some questions.
But a couple other things that I wanted to highlight just based on kind of what are the hot topics and things of interest today. So, obviously battery storage is—there's a lot of talk about battery storage and what that can do, what that's going to do, how that's going to change the grid and how it can fit in with variable renewables. We are looking at other storage technologies as well—I'll present on pumped hydro in a little bit. But I just wanted to point out that all storage technologies can provide various different value streams.
I mentioned all of these already, but just to kind of pull it all together, it can perform energy arbitrage, it can produce curtailment and provide reserves. And we represent all of these different storage durations, and the reason for that is because there's a—it's kind of a balancing act in terms of—for storage deployment in terms of what you would want to build based on how the capacity value, energy value changes over time. And it also depends obviously on the technology parameters as well.
So, the duration of that storage, cost and performance of that storage, and that includes the round-trip efficiency. And so, here this is just an example scenario that's pretty favorable towards construction, particularly battery storage. So, it starts off you've got the existing pumped hydro fleet. That kind of just marches along.
But if you look closely, you can see how it starts off building 2-hour storage, then it starts building four, and then six, and then eight, and then ten. So, this is correlating to an increase in variable renewable shares of generation, and so we can really clearly see how the preferred battery duration or a mix of—not battery, just storage generation in general actually changes over time depending on the grid mix. And that's because of the relationship between all of these factors.
And so, that's something that we've been improving ReEDS' ability to account for. Some other things that I just want to point out that, like I said, Brian is going to talk about a little bit later are that we have a lot of new features to have a better look at decarbonization analysis. So, that includes for—on the hydrogen side—so, we have these hydrogen fuel turbines. They can be built new or they can be upgraded from existing combined-cycle or natural gas turbines. And we can represent this in a couple ways. The simplest way is to have an exogenous hydrogen price.
In that case, you're not accounting for investments in hydrogen-producing technologies and how that might affect the hydrogen price. But we do now have the ability to represent those production technologies endogenously. And so, you can be building steam-methane reforming or electrolysis alongside the generation technologies that use that hydrogen, and then effectively the fuel price for hydrogen is based on which of these producing technologies you're building and certainly the cost assumptions for those technologies.
So, that's another thing you could do, shift these around. On the carbon dioxide removal side, I mentioned before that we're expanding our representations of carbon capture technologies. There's also now going to be a biomass CCS, so you can have a carbon-negative CCS technology.
That's tied to—I didn't mention this before, but we do have biomass fuel supply curves, and so bio-CCS would be tied to that. So, you can't just build indefinitely; you do have a finite fuel resource. At least, that's assumed in the model. And then, you can have not only new builds but upgrades to CCS from coal and gas. We have a couple different options on direct air capture for the cost and performance projections, and so you can see how that could play a role. And then, another feature that we're incorporating into the model is going to be an endogenous CO2 transport and storage network.
And so, that allows you to kind of further constrain the build-out of CCS systems and account for not only the cost and performance of capturing the carbon at fossil fuel facilities but also the limitations around where you would have to then inject that carbon in an underground storage site and the transportation costs to do that, and the region-specific costs of injection and maintaining that CO2 storage or sequestration. Last couple of slides. I also just wanted to mention again there's a wide range of policy options that are available in the model.
Like I said, we are always making sure that we're accounting for all of the existing policies, at least the best that we can. So, that includes all of the tax credit policies, any state-level clean energy policies or regional-level carbon trading, or state level in the case of SB100. We can account for air emissions policies, capacity mandates of different technologies, but there's also lots of options around playing with emission caps, rate limits, tax, clean energy or renewable energy requirements. And this is all defined very generically, so you have a lot of options for what types of policies you might want to consider. And so, just to kind of close off here, there's a lot of things that we've got in the works that we're really pushing the boundaries of complexity and what we can do computationally. And I'm well aware that on the user side that probably creates a lot of challenges, but the goal here is to have a lot of flexibility in how we represent things like temporal resolution, spatial resolution, transmission system options, generation options, and then that just gives a lot of options for the user to consider different types of systems and tailor the analysis for what they want to look at and what they can do.
And so, just some examples. We really want to be able to run ReEDS without Augur at an—well, maybe not everybody, but without Augur at—but at an hourly resolution, because then you could have everything fully encompassed in the same model. But that's hard to do. We're working on spatial flexibility, upgrading transmission in a couple different ways.
I mentioned some of the CO2 network stuff that's going on, but also potential for adding hydrogen transportation network representation as well. Some of you all might be familiar with the NARIS study that included Mexico and Canada. One of the challenges with that study is it used some proprietary data and now somewhat outdated data, so we've got some ongoing work to at least build the capability to where you could choose the options to run it for North America, or the U.S. and Canada. And it would run—there might be some data limitations, but we want to have that be possible for the public model going forward. And then, also things like looking at different weather years to account for different load and renewable energy profiles.
So, really high-level summary. ReEDS has a lot going on, but the reason for that is because there's a lot of really unique things. You saw how big the team is of people that are working on this model. We're really trying to be at the cutting edge of representing emerging technologies in the electricity sector, variable renewable storage, and everything else.
And so, our goal is to keep expanding the capabilities so that we're as relevant as we can be to the current grid-panning questions. So, that is all I have. >>Wesley Cole: Thanks, Stuart.
Question for you from the chat. Can you talk a little bit about units and being able to provide electricity operating reserve, planning reserve—can you do all of those, just one of those? How does that work with assigning different services to a single generator? >>Stuart Cohen: So, you have a certain amount of capacity for a given generation block or capacity block, so we're not necessarily representing individual units. But say there's 100 megawatts of gas in a region. It can only provide reserves at all if it's got available capacity that's not being used for energy. So, there's constraints in there that are essentially saying energy plus flex reserves plus spin reserves plus regulation is less than the max capacity.
So, that's how that's constrained. So, you can't kind of double dip in your capacity. Is that the—does that answer the question, Wesley? >>Wesley Cole: Yeah, but then you also have planning reserves too, though, that does…? >>Stuart Cohen: Yeah. Yeah. So, planning reserves is treated separately.
And so, planning reserves is not an operational constraint. It's in the name: it's a planning constraint. So, it's not based on what you're doing at a given amount of time; it's based on what is your essentially available capacity at a given time.
And so, it's independent of the operating reserves and energy constraints. >>Wesley Cole: Thanks, Stuart. So, any other questions, drop them in the chat. Or you can unmute yourself and ask them. We've got a couple of minutes before we're going to head to break.
>>Stuart Cohen: Yeah, see, there's a bunch in here. I don't know if there's any you want me to look at specifically. >>Wesley Cole: No, I responded to all of them except the one you just answered, so I think we're in good shape there.
If anybody wants to follow up on a question we answered in the chat, you're free to unmute or welcome to follow up in the chat. Okay. So, we don't have—we won't make you sit here and wait. We'll go—you're welcome to put stuff in the chat through the break. We'll respond to things there. Stuart, if you want to read this one and maybe respond to it in a second? So, we will break after Stuart answers this next question that showed up and then we'll come back on at 10:15 Mountain Time and start our next round of presentations there.
But Stuart, do you want to go ahead and answer this question? >>Stuart Cohen: Yeah, Wayne, I think I understand your question, and the answer is yes. So, yeah, I mean, you've got one constraint that says energy plus operating reserves across all three has to be less than the rate of capacity, but then you have a whole ‘nother constraint set, is requiring—that's requiring planning reserves that the provision of—the contribution of planning reserves is completely independent of the contribution to energy and operating reserves. So, I think that—is that a clear answer to your question? Okay. >>Wesley Cole: And that's consistent with the way things operate in real life.
So, people are operating in a capacity market; they're looking at their planning. That's separate from their operations, so those things are treated independently and we've done the same thing with the model. >>Stuart Cohen: Yeah, because the planning reserve idea is for long-term planning. It's not to be based on the—it's to keep a buffer of capacity for long-term purposes. And in ReEDS it's not like we're representing extreme events.
But the whole purpose is to ensure that ReEDS is investing enough capacity so that in the real world it would be available for heat waves and frozen electricity grids and stuff like that. So, that's what that planning reserve constraint is intended to do. But yeah, we're not actually representing those events in ReEDS. We do other work where we take a ReEDS solution, we downscale it, run it into our PLEXOS production cost models.
That runs at an hourly resolution. You can play with lots of different extreme event scenarios there, and that's the kind of stuff that we do to sort of stress test our ReEDS build-outs. ReEDS… Clean Air Act 111 B and D. >>Wesley Cole: So, Sandra, I think your question about the Clean Air Act pieces—that would probably be an offline piece. We're getting pretty far into the weeds here. So, feel free to follow up, Sandra.
We're happy to chat there, especially getting the right people involved in there. But let's go ahead and take a break here and we'll come back in 15 minutes and pick things back up for more of a deep dive into what's changed in the 2022 version. So, thanks, everybody. [Break in audio] >>Wesley Cole: Okay.
Well, welcome back, everybody. I'll just answer a question in chat real quick before we get started here. The new capacity factor constraint in the model is there to ensure we don't have—especially in zero-carbon scenarios—we don't have sitting around providing firm capacity that can actually provide that firm capacity because they'd have to generate to provide emissions. And we applied it at the RTO level rather than the plant level just as a way to get more resolved, but we didn't want to go down to the plant level because some units do sit around and don't really operate much while others do. So, some of the oil units in some years, they won't run very much; other years they'll run a lot. So, we—anyway, we've done a number of tests there to see what happens.
It doesn't change the answer much but there's other ways you can represent it. Happy to talk more there if you have other questions. So, let's go ahead and start with our 2022 model update section.
So, we are going through changes we've made over the last year for the 2022 model version that will be part of this year's standard scenarios. So, I'm going to kick things off with just some general model updates. These are just our relatively routine things that we do every year but are pretty important for making sure that the model stays up to date. And then, as you can see in the agenda, we'll step through a number of things.
It's a little bit of a lightning round kind of thing. They're relatively short presentations, quite a bit of changes we pack in here. So, we're certainly interested in the feedback if you have thoughts on "Hey, it would be more helpful to get this information in this way or in other ways." We're open to that.
But this is the same framework we used last year, so we'll go ahead and continue with it here. So, let me go ahead and jump in. So, I'm going to cover updates in these five things: so, the policies and tax credits, projections we pull out of the Annual Energy Outlook, cost and performance, plant database, and a variety of miscellaneous things.
So, these are all input-focused things. we're not really changing the model really for these; it's just changing inputs. So, the first piece is we've updated our RPS requirements based on Galen Barbose's 2022 RPS report. It's not out yet; it's forthcoming.
So, that included Illinois, North Carolina, Oregon, and their updates. Rhode Island had an RPS update after we got this from him, so we also included that Rhode Island RPS update. We've added state policies for states that don't allow new nuclear. So, there's a number of states that have legislation that says new nuclear is not allowed in those states, so we've explicitly added that. These have been changing. There's a couple of states that had disallowed new nuclear that just repealed that legislation or changed it so that new nuclear is allowed.
So, we'll keep tracking that. We've also added a basic representation of corporate RECs or voluntary procurement, depending on your—how you want to classify this. It is basic but we have something in there right now that captures that based on Jenny Heeter's various market assessments for how much those entities procure. We've also changed the safe harbor for offshore wind to be 10 years per the recent IRS guidance. And we've updated offshore wind required builds based on state mandates.
The state mandates continue to evolve in the coastal states, and so we've updated those. And then, finally, we've changed the long-term electric sector cap, CO2 cap for California in what was the AB32 and then the SB100 so that—they have their zero-carbon targets and the CPUC puts out projections for how much that would apply to the electric sector, and so we've updated ours there. This is a tighter cap, so it reduces emissions in California more rapidly than their prior cap. And then, the second piece in here to talk about is AEO update.
So, AEO produces their things every year. We use a lot of their inputs. We use fuel prices. You can see the coal prices and how those have changed—not much there. Natural gas prices over the long term are pretty similar but over the near term they're quite a bit higher. I don't think they're as high as they actually are this year, but they're still higher than last year in that near-term period.
And then, demand growth is also—the rebound from COVID is quite a bit faster in this year's projections relative to last year. They projected a slower return to growth and this year it's more rapid. And then, in terms of cost and performance, so we use the Annual Technology Baseline as our cost and performance inputs.
As you can see here, most of the technologies are pretty similar in terms of their—what the changes are. So, the R&D row is just the basic cost with fixed financials. The market row includes things like the tax credits. So, you can see the impact on offshore wind from extending the safe harbor window.
You've got a lower cost because of that. But the costs are pretty similar. So, in most cases the costs are pretty much the same. You'll see a bigger delta in the utility-scale PV plus battery costs, but that's because of a change in the definition of configuration more than a change in the underlying costs of the technologies. And then, in terms of plant database, we've updated to the latest plant database from NEMS based on the AEO 2022.
And the biggest change that you'll see there is that the Illinois nuclear plants had stated their intention to retire at the end of 2021. Illinois passed some legislation to keep those plants online, so those plants are no longer retiring. So, there's quite a bit more—well, I mean, a decent chunk of additional nuclear staying online in this year's database. EIA now has a PV-plus-battery technology captured in the database, and there's quite a bit of planned potential in the database, so that's now all showing up in the near term, showing up in the '21, '22, '23 builds. And then, there's a little bit less near-term natural gas capacity. And then, the other piece that is at least useful for us is the database is now linked directly with ReEDS.
Before, there were some heritage pieces with an older version of ReEDS where it took—there were a couple steps between the database and the data coming into ReEDS, so if you changed the database that's in the model, it wouldn't necessarily reflect those changes in the model. Now, if you go and you change the database, it will reflect those changes directly in ReEDS. So, if you want to change a retirement date in the database or change a plant from gas to coal or add a new row with a new plant that you want to put in there, those will automatically make their way into ReEDS now, which is pretty convenient. And then, last thing is the—just miscellaneous updates. So, a variety of these first. We've added winter capacity to the model.
So, the CAP variable and all the outputs, they're still reporting net summer capacity just like they always have, but the model now includes a winter capacity so that especially thermal plants will often—they often have more capacity in the wintertime because you get higher mass flow rates through the turbines, so you can put out more energy. So, that's now reflected in the model, reserve margin, an energy that those plants can generate in the wintertime. Our planning reserve margin can now be applied to a region's coincident peak rather than just the individual peaks of each subregion.
I think Patrick's going to talk about that a little bit. And then, we have—if you've ever—if you haven't gotten in the code, this doesn't matter. But if you've been in the code, we've had separate indices, r regions for the model, BA regions, the 134 regions that Stuart talked about, RS indices for the 356 wind and solar regions. We had some parameters that had both—were indexed by the both of those, and it was really annoying to work with those. We've gotten rid of all of those now and just have a single—each parameter only has a single region set, so you don't have two region indices on any parameter, so that makes dealing with them a lot nicer.
And then, the last piece is the retail rate module that we released in last year's version, we've updated it to now account for negative emission technologies so that those costs get accounted by states more appropriately. So, that's everything I have. So, I'm happy—if folks have questions, I'm happy to take them. And then, so I'll—if anything turns up in the chat, otherwise I'll turn it over to Stuart, who's going to go next on pumped storage.
>>Stuart Cohen: Yeah, Wesley, there is one question about the REC policy definitions. >>Wesley Cole: Yeah, so the RECs cred—there is a REC variable that applies to every state and actually includes a source and a sink. So, we're tacking RECs generated in one state transferred to another state to account for technology eligibility rules, trading rules, out-of-state fractions, and those kind of things.
So, it's—it is a fairly sophisticated representation of RECs to try and capture all the nuances in the state RPS requirements. They're certainly not perfect and it's really hard to keep up with 50 states who are changing rules all the time. But yeah, it—so, it's not just a generic REC; it's—there are specific sources of RECs for each technology and state. >>Stuart Cohen: Should I go ahead? >>Wesley Cole: Yep, take it away, Stuart. >>Stuart Cohen: All right. So, yeah, I mean, this is—so, one of the newer data sets that NREL has put together on the resource side is some new supply curve data for pumped storage hydropower, specifically closed-loop systems.
And so, I just wanted to spend a little bit of time introducing that to the folks on here so you understand what it is, what's available in ReEDS, and how you might use it to look at different storage technology alternatives. So, for—starting really from the top, this is kind of a cartoon of what a pumped storage system looks like. So, you essentially have an upper reservoir, a lower reservoir, a powerhouse at the bottom where gravity feeds water down into, it spins a turbine, produces electricity, and then there's also a pump component in here with which you could send the water back up to the upper reservoir. So, it's a big energy storage system that is using water as the storage media for potential energy. A couple key things to think about is systems can use reservoirs that are on river systems or off river systems.
And so, most of the existing facilities today are open-loop and use reservoirs that are on rivers—at least one of them. But as you'll see, what we're looking at with this new data set is specifically closed-loop systems, and the reason for that is because even though you have to build two new reservoirs, you're not interacting with the river system. And so, there are some reduced environmental impacts associated with that, at least on the—from the fission hydrology standpoint. And so, pumped storage is, as some folks might know, is really the largest current player in energy storage in the U.S. and the world. And one of the differences between pumped storage systems and things like the battery, utility-scale batteries is really just the size and the scale. And so, yeah, you might only have the similar storage duration, but the quantity of energy storage in some of these reservoirs is—really is quite large.
And so, they—one of the things that we're looking at is how you could use these larger systems over longer time scales to take advantage of that. And so, as I mentioned, we recently released a data set that's a resource assessment for where pumped storage can be located throughout the United States. So, I'm just going to talk a little bit about what goes into that. But this was done for the contiguous United States, which is all that gets used in ReEDS because that's at least what the U.S. version of the model is scoped for.
But we do have data for Alaska and Hawaii and Puerto Rico as well. You can look at this data set online. It's incorporated into the ATB, so that's one way you could look at some aggregated data. But you can actually go onto this website, play around, look at different sites, zoom in, see what the reservoirs look like, see what they—see what some of the characteristics are, head height, power capacity, costs, and so on. And so, this is just another set of data along the same vein as some of the wind and solar data, like Anthony is going to present on next.
Just as a quick overview of how we come up with this, this is a geospatial analysis that's looking at the topography of the entire United States and essentially looking at where you have what essentially are dry gullies that you could put a dam on one side to create a reservoir for pumped hydropower. And so, it uses some kind of technical specifications and constraints to identify all possible reservoir locations, and then it will eliminate any reservoirs that intersect with things like critical habitats, national parks, urban areas. And then, as I said before, this is specifically looking at closed-loop, and so it ensures that there's no intersection with existing waterways. And then, once it's filtered the reservoirs down in that way it then looks at kind of the maximum distance and minimum head height between the reservoirs to find all possible pairs. And then, it eliminates overlapping systems by essentially plucking out the pairs that have the lowest costs across overlapping systems. And