Yale Climate Connections recently described an article, Can renewable generation, energy storage and energy efficient technologies enable carbon neutral energy transition? by the Ning Zhao (Systems Engineering, Cornell University, Ithaca, NY) and Fengqi You (Systems Engineering, Cornell University and Robert Frederick Smith School of Chemical and Biomolecular Engineering, Cornell University). The study considered the New York targets and “analyzed scientific and economic data and concluded that the goals are technologically and financially feasible.” I reviewed their work and disagree. The inconsistencies between their results and other analyses done as support to the Climate Leadership and Community Protection Act (CLCPA) and omissions in their evaluation method do not make their conclusions credible.
I have summarized the schedule, implementation components, and provide links to the legislation itself at CLCPA Summary Implementation Requirements. I have written extensively in posts on implementation of the CLCPA because I believe it will adversely affect affordability and reliability as well as create more environmental harm than good which affects my future as a New Yorker. I have described the law in general, evaluated its feasibility, estimated costs, described supporting regulations, listed the scoping plan strategies, summarized some of the meetings and complained that its advocates constantly confuse weather and climate. The opinions expressed in this post do not reflect the position of any of my previous employers or any other company I have been associated with, these comments are mine alone.
The paper lists the following highlights:
- A novel bottom-up optimization framework for energy decarbonization transitions.
- Feasibility investigation on the decarbonization goals for New York State.
- Offshore wind as major electricity source by the end of the planning horizon.
- Heat pumps and geothermal technologies as main space heating methods.
- Natural gas as an important but temporary energy source at early transition stage.
An optimization framework is ultimately no more than a curve fitting exercise. Think back to a laboratory experiment where something is measured at several points, the results are plotted, and the graph is used to infer results outside the range of the observations. The theory is that if you have enough descriptive variables, can make reasonable assumptions about the range and potential effect of each variable used, and then develop a sophisticated optimization model, then it can be used to project how, in this case, the energy system could transition to zero emissions technology. The authors note that “To the best of our knowledge, there is no existing energy transition optimization study for the decarbonization of multiple energy sectors that incorporates region-level electricity generation and space heating thermal energy production, while accommodating scheduled energy system changes and climate targets”.
Of course, the problem is that the electric energy system is very complex. As a result, including all the variables and constraints is a huge undertaking. All it takes is inadvertently omitting one key constraint and the results of such a model aren’t credible. For example, consider the Integrated Planning Model (IPM) which is used by the Environmental Protection Agency to evaluate the potential impacts of proposed air quality regulations. The developers of IPM explain that it “provides true integration of wholesale power, system reliability, environmental constraints, fuel choice, transmission, capacity expansion, and all key operational elements of generators on the power grid in a linear optimization framework.” This model is so detailed that it includes “a detailed representation of every electric boiler and generator in the power market being modeled”. However, in order to be able to afford to run the model simplifications are often employed. In New York, the EPA version simplifies the transmission network so much that the fact that New York City is in a load pocket is lost and the results are not credible. There are work arounds but, in my experience, the EPA version of this model often does not work well enough to provide credible results for New York State.
Evaluation of Figure 3
In order to evaluate the conclusions, I compared the results from the Zhou and You (2020) optimization model (“Cornell Study Model”) to observations and projections made by others. The article does not make this easy. For example, a key evaluation metric are the projections of annual electric generation by source shown in their Figure 3. In order to be able to compare numbers I had to manually extract them off a blown-up version of the graph which gives a resolution of ~2,000 GWh.
As shown in Figure 3, the initial year for the study is 2019. I assume that means that they ran their optimization model using input data so it is possible to check the accuracy of it relative to observed data. I checked the model’s annual electric generation by source against the observed data that year from the New York Independent System Operator (NYISO) 2020 Load & Capacity Data report in Table III-3c Annual Net Energy Generation by Zone and Type – 2019.
Despite the low resolution possible with my interpretation of the graph it is clear that the model does not do an adequate job representing the New York electric system annual electricity generation by source relative to 2019 data. There were comparison data available for seven source categories. There was insufficient resolution or it was not clear which source category should be used for the comparison for the others. I don’t think there are any ambiguities for the nuclear, on-shore wind, and import generation categories. The Cornell Study Model over-predicted nuclear generation by 24% or nearly 11,000 GWh. Worse it exceeds maximum possible generation if the nameplate capacity operated every hour of the year by 7,106 GWh or 15%. In order to get the 2019 generation shown in the graph, the 1,985 MW of onshore wind capacity would have had to be 46% vs. the observed 25%. The Cornell Study Model under-predicted imports by 14,037 GWh or 61%. There is an 11% difference in the hydro numbers but I think that is due to the exclusion of pumped hydro in the Cornell Study Model. The remaining three categories are all natural-gas firing categories. If they are all summed up the difference is less than 10% which is close enough for this methodology.
The 2040 Figure 3 generation source type projections were also evaluated. Last fall the Analysis Group presented the results from their Climate Change Phase II Study for the NYISO. Importantly, the analysis looked at the generation resource requirements “that meets electricity demand in every hour all year”. Last October I evaluated their results and noted that I agree with the methodology but was worried that they had not done an adequate job defining the worst renewable resource availability case. Because they evaluated one-month periods their electric energy projections (GWh) are not comparable to the Cornell Study Model. The Analysis Group did provide capacity (MW) projections for different generation sectors. The Cornell Study report includes supplemental data with a spreadsheet (S1_Data_for_Policy_and_Geothermal) that lists capacity factors used in 2040. Assuming that the capacity of each sector equals the projected energy (GWh) in Figure 3 divided by those capacity factors and number of hours, then there is comparable capacity (MW) data. For energy storage I used information from section 6 of the paper: “For the energy transition under the scenario with carbon price policy and geothermal technologies, the electricity storage capacities in 2025, 2030 and 2050 are 2.9, 4.4 and 7.2 GW, respectively; the energy capacity for electricity storage are 3.7 GWh in 2025, 5.6 GWh in 2030, and 9.3 GWh in 2050”. For 2040 I took the average of the 2030 and 2050 projections.
Tables 9 -12 in Climate Change Impact and Resilience Study Phase II list the nameplate capacity for zero-emission resource sectors. I compared the Cornell Study Model nameplate capacities as calculated above in a summary table. The Resilience Study considered two cases: one with the CLCPA mandates and one without. The Study also compared their results with NYISO Grid in Transition study that also seeks to understand the reliability and market implications of the State’s plans to transition to clean energy sources. That study also considered two similar cases. The energy transition case study projections for both studies are markedly different than the Cornell Study. For example, in Table 9 the Resilience Study projects on-shore wind capacity 68% higher, distributed solar 41% higher, utility-scale solar 88% higher, nuclear 29% higher, and energy storage 63% higher. Of the 12 resource categories in the Resilience study only on is “close” at 11% different.
I will briefly explain why I think there are such significant differences. The biggest problem is the time-scale for the evaluation. The paper states:
The demand predictions for annual electricity and space heating thermal energy within the planning horizon for New York State are shown in Fig. 2(a) in blue and orange curves, respectively. The energy demands are expressed in an annual basis, which has been applied in previous optimization works on energy system transition considering high-penetration of variable renewable energy. The spatial resolutions for both the electric and space heating thermal energy are state level. In other words, the state-level demands as shown in Fig. 2(a) would be balanced with the energy supply in the state through optimization.
I interpret that to mean that the optimization is based on annual state-wide parameters.
If my interpretation is correct, then the entirety of the Cornell Study Model results can be ignored. On an annual basis the Texas electric system worked but when there was a short-lived extreme stress on load the result was massive blackouts. All the credible work done for CLCPA implementation determine the resource requirements based on short periods because an electric system that depends upon renewable energy has to address the period with the lowest wind and solar availability not any long-term average.
The ultimate problem is that no matter how many wind turbines and solar panels there are, when the sun isn’t shining and the wind isn’t blowing no electricity is generated. In fact, the credible studies include a special resource to address those periods. On October 8, 2020 Kevin DePugh, Senior Manager for NYISO Reliability Planning, made a presentation that lists the characteristics of this Dispatchable Emissions-free (DE) resource:
- Large quantity of DE Resource generation is needed in a small number of hours;
- DE Resource has low capacity factor (~12%) during the winter;
- DE Resource has only a 3.7% capacity factor in the summer;
- DE Resource is not needed at all during spring and fall;
- Substantial quantity of DE Resource capacity is needed, the energy need is minimal;
- DE Resource must be able to come on line quickly, and be flexible enough to meet rapid, steep ramping need;
- On an average day, storage can meet evening peaks, but the DE Resource must generate if storage is depleted and renewable generation is low; and
- In the Winter CLCPA scenario, the DE Resource output across the state must increase from 362 MW (1.1% of DE Resource nameplate capacity) to27,434 MW (85.4% of name plate capacity) in six hours of the most stressed day.
The Cornell Study Model did not address this problem because they optimized using annual parameters. Omitting this problem is a fatal flaw.
I noted other issues before I stopped looking. As noted previously the optimization model did not reproduce the 2019 resource mix. For energy storage, “the technology specification and economic data for Hornsdale Power Reserve Battery Energy Storage System that was installed by Tesla are used for battery storage systems in this study”. However, the facility is making most of its money providing Frequency Control Ancillary Services and is not being used for energy storage. I think the state-wide optimization approach smooths out all the transmission constraint issues which is a problem even in the considerably more detailed IPM system. The optimization model projected that 25,714 MW of offshore wind capacity would be needed but the National Renewable Energy Lab (NREL) has determined that New York offshore technical potential estimate is only 21,063 MW.
I have no doubt that advocates for the CLCPA will point to the Yale Climate Connections report of this study as proof that New York’s climate goals are achievable. However, even this cursory evaluation of the approach and results indicates that the claim that the goals are technologically and economically feasible are simply not credible.