THE IMPACT OF SOLAR UNCERTAINTY ON PROJECT FINANCEABILITY: MITIGATING ENERGY RISK THROUGH ON-SITE MONITORING

Transcription

1 THE IMPACT OF SOLAR UNCERTAINTY ON PROJECT FINANCEABILITY: MITIGATING ENERGY RISK THROUGH ON-SITE MONITORING Marie Schnitzer Christopher Thuman Peter Johnson AWS Truepower, LLC 463 New Karner Road Albany, NY ABSTRACT As projects increase in size and installed cost, a greater emphasis is being placed on reducing the uncertainty assigned when evaluating a project s long-term performance. For 11 locations, at least one year of highquality ground measurements were used to estimate the long-term solar resource and associated uncertainty. In parallel, the long-term solar resource and uncertainty were estimated using available modeled data. These results were quantified and evaluated to determine how the availability of on-site data affects the uncertainty in the long-term solar resource. Results from this study indicate a difference in estimated solar resource of up to 6% when using on-site measurements compared to satellite-modeled data, and corresponding to a general reduction in uncertainty of approximately 2% to 6%. 1. INTRODUCTION There are risks associated with the development of any energy project, including legal risks from project contracts, permitting, site control, equipment risks, and system performance. These areas influence the ability for a project to obtain financing. With solar photovoltaic (PV) projects, a major area of risk is quantifying the expected annual energy production. An accurate preconstruction energy estimate offers insight into the financial risk of a solar PV project, which is quantified through the amount of uncertainty associated with the energy estimate. While there are several areas of uncertainty, one of the most significant drivers is the uncertainty in solar input data. While the industry has historically relied on modeled data to estimate the on-site solar resource, uncertainty in energy estimates can be reduced by on-site monitoring and an indepth analysis of all available data sources. This report outlines the sources of solar resource data and through the use of on-site data collected at multiple sites across the United States, demonstrates the benefit that onsite monitoring has to reduce uncertainty in energy estimates. This is shown by comparing energy production uncertainty for two scenarios: with and without on-site solar data. The case study confirms that on-site monitoring supports high confidence energy estimates (i.e., P90, P99) for bankable energy assessment. 2. BACKGROUND 2.1 Motivation for Case Study As the solar industry matures, lessons learned from the wind industry can be applied to increase the financial success of solar projects. Like in the solar industry, early wind projects were less reliant on on-site meteorological data, and many projects suffered through overly optimistic performance expectations and unrealized revenue streams. Early wind practices were relinquished in favor of more sophisticated methods, specifically the reliance on site-specific measurements to assist in energy estimation with higher accuracy. As with the wind industry, the need for site-specific data to accurately characterize solar energy production is becoming apparent 2, increasing the magnitude of financial risk for

2 investors. In most cases, publicly-available measured and modeled reference data sources lack the desired accuracy or representativeness of the project site to sufficiently mitigate the energy production risk for larger projects. 2.2 Sources of Solar Data Since project cash flows are tied to energy estimates derived from solar resource estimates, accurately quantifying the onsite radiation is important. This is accomplished using measured and/or modeled data sources. An optimal energy assessment includes a thorough evaluation of all reference data available. Evaluation of multiple data sets allows for selection of the most appropriate data source(s) to be used to estimate the long-term irradiation at the site. Additionally, the advantages and disadvantages of each can be leveraged Modeled Data Multiple sources of public and private modeled solar data exist. Although model-based irradiance data may have lower accuracy compared to other sources and are limited by grid-cell resolution, modeled data sets are advantageous in that they are available for any location. Modeled data sets also are more likely to have a longer period of record when compared to on-site data. A common source of publicly available modeled solar data is the U.S. National Solar Radiation Database (NSRDB). This database consists of 15+ years of modeled hourly data across the U.S. as well as TMY3 data sets, which provide an annual time series of hourly solar data at specific locations in the U.S. Although this data source has historically been relied upon by the solar industry for site prospecting, it has limited value as a data source for directly estimating the long-term solar resource. This is because the NSRDB TMY3 data setss are comprised of a combination of satellite-modeled, numerically-modeled, and back-filled data, which increases their uncertainty. As a result, the National Renewable Energy Laboratory (NREL) states that NSRDB meteorological data may not be suitable for climatological work. The meteorological fields in the NSRDB should be used only as ancillary data for solar deployment and sizing applications Surface Reference Stations Several international, nationwide, regional and state-wide measurement networks collect solar radiation data in the form of surface-based measurements. These networks exhibit various periods of record and varying degrees of data quality. While much of this data is publicly available and can have greater accuracy than modeled data, it is rare to find high-quality solar data near a project site due to spatial diversity of reference networks and poor station maintenance On-Site Measurements The collection of high-quality on-site measurements at the project site can be useful for understanding site-specific characteristics. These data can be used to in conjunction with longer period-of-record data sources to estimate the long-term solar irradiation while minimizing the uncertainty associated with this estimate. 3. SOLAR DATA AND UNCERTAINTY 3.1 Impact of Solar Data on Uncertainty Uncertainty is the risk associated with how an observed or calculated value may differ from the actual value. Although there are multiple areas of uncertainty during a project s development process, uncertainty in the solar resource estimate is one of the most significant sources of the overall energy production uncertainty, as shown in Table 1. Because of this, selection of an appropriate reference data source is critical for reducing uncertainty in the solar resource estimate, which in turn increases the confidence in the project s energy estimate. TABLE 1. COMPONENTS OF PROJECT ENERGY UNCERTAINTY Uncertainty Source Typical Uncertainty Range Annual Degradation 0.5-1% Transposition to Plane of Array 0.5-2% Energy Simulation, Plant Losses 3-5% Solar Resource Estimate 5-17% Solar resource uncertainty can be broken down into four categories, shown in Table 2. The measurement component of solar resource uncertainty is often the greatest contributor, and can range significantly depending on the data source used. Modeled data sources have measurement uncertainties in the 8 to 15% range. On-site measured data usually has a measurement uncertainty between 2 and 7%, depending on the quality of instrumentation and the frequency of on-site maintenance: a regular schedule for leveling and cleaning solar irradiation sensors is necessary to achieve the lowest measurement uncertainty.

3 TABLE 2. COMPONENTS OF SOLAR RESOURCE UNCERTAINTY Uncertainty Source Typical Uncertainty Range Spatial Variability 0-1% Representativeness of Monitoring Period 0.5-2% Inter-annual Variability 2-5% Measurement Accuracy 2-15% 3.2 Uncertainty and Probability of Exceedance In order to quantify the energy risk for a project, developers rely on the concept of probability of exceedance, or the level of confidence that a project s actual energy production will achieve at least a certain value based on the energy production uncertainty. For example, the energy value that can be expected with 90% confidence is called the P90. Figure 1 helps to demonstrate this concept. According to Fitch Ratings, Moody s Investors Service, and Standard and Poor s, P90, P95, and P99 confidence intervals are commonly used to evaluate the relative risk of solar PV projects. into energy assessment models, and satellite data has greater accuracy than most numerical models. The on-site data was collected with LI-200 pyranometers with each site having a period of record of a year or more. At 9 of 11 sites, regular on-site maintenance was conducted, increasing the confidence in the measurements from these sites. Annual GHI was evaluated from each data source to show how results from modeled data may differ from on-site data. For the modeled data, the long-term mean was calculated directly from a 13-year period of record. For the on-site data scenario, the shorter period of record was adjusted using a long-term reference data source and the measure-correlatepredict 4 (MCP) methodology. Since a percent difference in solar resource approximately translates to the same percent difference in energy production, the difference from these data is useful to understanding the impact of the selected solar input data source on energy production estimates. In addition to comparing the magnitude of the solar resource from different data sources, two uncertainty estimates were developed in parallel, one using satellite-modeled data and the other using on-site data. A comparison of the resulting uncertainty was conducted to show how confidence can be increased by utilizing on-site measurements. P90 energy estimates were compared for a typical modeled data and onsite data scenario to show how site-specific measurements influence the P90 energy estimate. 4.2 Results Magnitude of the Solar Resource Fig. 1. Characterization of a Project s Uncertainty and Probability of Exceedance The value of the P90, P95, and P99 are directly related to the amount of uncertainty in the energy production estimate. The higher the uncertainty, the lower the P90 when compared to the P50. Therefore, reducing uncertainty will maximize the P90 energy and minimize investment risk. For the 11 sites considered, the satellite-modeled long-term GHI was compared to the estimate derived through the use of on-site measurements. The results, displayed in Figure 2 and Figure 3, show a mean difference of 2.8% between the satellite model and on-site estimates, ranging from 0.1% to 6.2%. When translated to energy, this difference can have a similar percentage impact on the projected energy output. No consistent bias was observed between the mean GHI from the two data sources. 4. CASE STUDY 4.1 Approach To demonstrate how solar input data affects uncertainty in energy production estimates, a case study was conducted for 11 sites in the United States using modeled and measured data. The modeled data used in the analysis was satellitemodeled with a 13-year period of record. This data source was used with the assumption that much of the solar industry s current prospectors use modeled data sets as input Fig 2. Annual GHI Estimate using Modeled and On-Site Data

4 Fig. 4. Difference in Energy Production Uncertainty using Modeled and On-Site Data Fig. 3. Percent Difference in Annual GHI Estimate using Modeled and On-Site Data Impact on Uncertainty A comparison of the resulting energy uncertainty estimates shows how confidence can be increased by investing in onsite instrumentation. For both scenarios, combined energy uncertainty was calculated by combining the following uncertainties in quadrature: Measurement uncertainty Inter-annual variability Representativeness of the monitoring period Spatial variability For both the on-site data and satellite-modeled data scenarios, uncertainties were applied uniformly with the exception of the measurement component. Based on published uncertainties for satellite-modeled GHI, a measurement uncertainty of 8.5% was assigned for the satellite data. This led to a combined uncertainty range of 8.7% to 9.5% for the 11 sites, with a mean uncertainty of 9.2%. Alternatively, measurement uncertainties between 4% and 6% were assigned for the assessment using on-site data. Each site s measurement uncertainty was based on the instrumentation installed on-site and the frequency of site maintenance. For the on-site scenario, this led to a combined uncertainty range of 4.5% to 7.3%, with a mean uncertainty of 5.7%. Figure 4 shows the difference in uncertainty from each scenario. On average, on-site monitoring reduced the project s expected uncertainty from 9.2% to 5.7%, or approximately 3.5%. Two of the 11 sites (10 and 11) exhibited poor or undocumented maintenance practices, which increased measurement uncertainty by 2% for these sites. Excluding these two sites as outliers, the mean reduction in uncertainty was 3.9% Impact on Probability of Exceedance A greater solar resource uncertainty results in an even a greater impact when applied at higher confidence intervals. The difference in measurement uncertainty has an increasingly significant impact on high-confidence energy estimates such as the P90, P95, and P99. To demonstrate this, probability density plots were generated for a typical uncertainty assessment using satellite-modeled data and a typical uncertainty assessment using on-site data. Using the combined uncertainties, the P90 energy estimate was calculated for the satellite-modeled and on-site data scenarios, depicted in Figure 5. In the scenario with satellite-modeled data, the P90 is much lower when compared to the P50 due to the greater amount of uncertainty in the GHI data. Fig. 5. Probability Distribution Showing the P50/P90 Difference Modeled Data (top) and On-Site Data (bottom) Scenarios The difference between high-confidence energy estimates is even greater for the P95 and P99 estimates, as shown in Figure 6. The average P50 difference of all 11 sites is approximately 2.8%. However, as the level of confidence in

5 each estimate increases, the difference between the estimated GHI increases, resulting in more than a 5% difference at the P90 confidence level and more than a 10% difference at the P99 confidence level. The results of the case study demonstrate the importance of on-site data collection for all project stakeholders. With project financial returns dependent on reliable energy production estimates, minimizing energy risk is an important factor to consider early in the development process. A significant portion of energy risk can be mitigated by on-site solar resource monitoring. Executing diligence in the design and implementation of a monitoring campaign will likely result in financial benefits in the longrun through the reduction in overall project uncertainty and the procurement of highly accurate and defendable data sets. For this reason, on-site data collection is expected to become an increasingly important aspect of pre-construction planning, especially as the solar industry matures to the development of multi-megawatt projects. 6. REFERENCES Fig. 6. Comparison of Confidence Energy Estimates as Percent of P50 for Modeled and On-Site Data 5. CONCLUSION The case study of 11 sites in the United States demonstrated the importance of on-site data collection in the following ways: Magnitude of the Solar Resource. On average, the long-term GHI calculated from modeled data differed from the long-term GHI estimated using on-site data by 2.8% on average, with one site differing by as much as 6.2%. When translated to energy, this difference can have a similar impact on the projected energy output. Impact of Uncertainty. On average, on-site monitoring reduced the overall uncertainty from 9.2% to 5.7%, or approximately 3.5%. This is a significant reduction when considering the application to probability-of-exceedance energy estimates. Value of On-Site Maintenance. Lack of regular onsite maintenance to clean and level instrumentation can increase measurement uncertainty by as much 2%. For sites with regular on-site maintenance, the average uncertainty reduction over modeled-data was 3.9%. Impact on Probability-of-Exceedance Energy Estimates. A greater solar resource uncertainty results in an even a greater impact when applied at higher confidence intervals. Using typical uncertainty estimates, on-site monitoring may increase the P90 by over 5% and the P99 by over 10%. (1) Schnitzer, M. et al (2011). Reducing Uncertainty in Bankable Solar Resource and Energy Assessments through On-Site Monitoring Proceedings of the 2011 ASES National Solar Conference, American Solar Energy Society, (2) Rating Criteria from Solar Power Projects, Fitch Ratings, : Fitch looks for a minimum of one year, hourly, well-maintained, onsite data for a complete solar resource supply assessment. Shorter data periods than one year will not capture the full seasonal and diurnal characteristics of solar irradiance at a particular site, and would be considered either midrange or weaker. (3) National Solar Radiation Database User s Manual. National Renewable Energy Laboratory (4) Thuman, C. et al (2012). Quantifying the Accuracy of the Use of Measure-Correlate-Predict Methodology for Long-Term Solar Resource Estimates. World Renewable Energy Forum, Denver, CO, 2012.