Impacts of Soiling on Utility-Scale PV System Performance
Inside this Article
The value of energy produced from utility-scale PV systems deployed throughout the Desert Southwest depends on the systems’ ability to match the seasonal and time-of-day utility loads. During times of increased demand, some utilities charge more per kilowatt-hour—effectively increasing the value of each kilowatt-hour produced by a PV facility—whether it is realized as an avoided cost or sold. As a result, project developers, engineers, owners and utility operators focus on designing and maintaining systems that maximize production during these high-value periods.
Coincidently, the Desert Southwest experiences a long dry season that corresponds with these periods of high demand and increased energy values. In the absence of significant rain events or regular cleaning, production losses due to soiling, also known as soiling losses, increase. In 2006 at the IEEE 4th World Conference on Photovoltaic Energy Conversion, PowerLight released an article titled “The Effect of Soiling on Large Grid-Connected Photovoltaic Systems in California and the Southwest Region of the United States.” The authors confirmed “a gradual but marked decrease in system performance through the dry season for systems in arid climates” and concluded that “performance losses due to system soiling are not constant through time, rather they depend on the amount and frequency of rain that falls on the array.”
Despite PowerLight’s conclusions, several of the commonly used PV production modeling tools, and thus system owners, assess the impacts of soiling on an annual basis. Consequently, system owners have typically addressed array soiling in one of two ways—either wash the array on a regular basis to limit the losses due to soiling, or forgo array cleaning and rely on rain events to keep soiling losses to a minimum. When based on annual soiling losses—as opposed to seasonal or monthly—these decisions fail to address short-term soiling impacts that may justify the cost of cleaning an array to maximize production during the high-value kilowatt-hour periods.
In this article, I review key soiling characteristics with historical weather data to simulate site-specific soiling losses on a monthly and annual basis for two sites in the Desert Southwest. The results confirm variation in soiling losses throughout the year for both locations, with average monthly losses in June and July—the high-value period—well above the annual average. In addition, prescribed cleanings are introduced to the model to understand their impact on annual soiling losses.
Measuring Soiling Losses
Soiling can be measured as either the rate at which contaminants accumulate on the module surface or the resulting decrease in production. Ultimately, we need to determine the decrease in system performance due to soiling loss. Assuming all other factors remain constant, comparing actual production values between a control subject and a soiled array is one way to determine soiling losses for a given site.
To simulate soiling losses over time, we must determine the rate at which soiling accumulates. Although soiling rates can be calculated in a variety of ways, a soiling rate that represents the daily percent decrease in production is most valuable for the purposes of PV production modeling. Once a soiling rate for a site has been established, it can be used with rainfall data to estimate past, present and future soiling losses.
As demonstrated in the PowerLight study, the “measured soiling rate” represents the slope of a linear fit curve applied to performance data between rain events. In other words, PowerLight’s study assumes that the percent change in performance over time—in the absence of rain or cleaning—equals the percent change in soiling losses over time.
In January 2013, First Solar published a paper in the IEEE Journal of Photovoltaics titled “Direct Monitoring of Energy Lost Due to Soiling on First Solar Modules in California” that details an alternative soiling measurement technique to determine site-specific soiling rates. The technique is based on a methodology proposed in “Solar Cell Arrays: Degradation Due to Dirt,” which was published in the Proceedings of the American Section of the International Solar Energy Society in 1989, and is intended to be “practical and automated … foregoing complex equipment such as IV curve tracers.” Rather than equating soiling rates to the increase in production losses per unit of time, the method First Solar uses compares production levels among a control module, a module that is not cleaned on a regular basis and the expected performance based on typical irradiance and temperature readings.
PowerLight’s and First Solar’s techniques can be used to establish site-specific soiling rates. This is quite valuable since soiling rates can vary within a region and may depend on the incidence of nearby human activities, such as traffic, construction, or airports. Since soiling rates do not take module cleaning into account, they must be modeled with local rainfall data to determine the actual soiling losses over a defined time period. To effectively predict soiling losses based on local weather and soiling rates, we must assume how much rain is required to completely—or at least significantly—clean an array.
Characteristics of Soiling
The PowerLight and First Solar studies, along with a study by Arizona State University (ASU), identify some key characteristics of soiling on PV arrays. In addition, McCarthy Building Companies has observed these trends and characteristics for several of the sites it monitors in the Desert Southwest. These trends are useful for estimating how soiling impacts PV production. Several of these characteristics serve as key assumptions in models used to predict soiling loss, also known as soiling estimating models.
Rain events. A paper based on a study conducted by ASU titled “The Effects of Soiling on PV Module and Radiometer Performance” concluded that 0.2 inch of rain is nearly equivalent to physically cleaning the modules and typically restores production levels to 99.5% of a cleaned module. In other words, after 0.2 inch of rain, soiling losses are reduced to 0.5%. First Solar was not able to validate this assumption in its recent study, and PowerLight found that “significantly more rainfall” is required on some systems “to completely clean the modules.” Factors like humidity and small, dusty rain events tend to negatively impact this metric— meaning more rain may be needed where such conditions exist.
Soiling rates in the Desert Southwest. The PowerLight study found that systems in California and the Desert Southwest experienced an average of 0.1% to 0.3% decrease in efficiency per day due to soiling. McCarthy’s analysis of several larger systems in the Phoenix metropolitan area revealed soiling rates between 0.04% and 0.07% per day. Lastly, First Solar observed “fairly constant” soiling rates in low desert regions of southeastern California without agricultural activities, averaging 1% per month or 0.03% per day. The study also shows that sites with more agricultural activity have significantly higher soiling rates.
Soiling rates increase with human activity. Although soiling rates remain somewhat constant per site in the Desert Southwest, rates vary from one site to the next. In the Central Valley, PowerLight observed daily soiling rates between 0.1% and 0.2%, with the higher rate occurring in areas with more human activity, such as urban environments, highways and airports.
Angle of incidence. Soiling losses increase as the angle of incidence between the sun’s rays and the module surface increases. For example, a 25° incidence angle has twice the losses of a normal angle, and a 60° angle has four times the losses. Soiling losses tend to be highest in the morning and evening, when the angle of incidence is greatest.
Tilt angle. The ASU study shows that soiling rates are not significantly impacted by the module tilt angle nor by modules mounted on tracking arrays. That said, the tilt angle does impact the ability of rainfall to clean the module surface. To maximize the cleaning effect when it rains, tracking systems should be stowed at a minimum of 5°, and preferably more, to allow the water to sheet off the surface of the modules. Fixed systems with a tilt angle of less than 5% experience relatively higher soiling rates than those with a greater tilt angle.
Estimating Soiling Losses
Combining what we know of the characteristics of soiling on PV arrays with local weather data enables us to estimate soiling losses on a monthly or even daily basis. System operators can use soiling estimating models to better understand the impacts on production over a specific time period, calculate the predicted loss in revenues due to underperformance and decide when the cost of cleaning the array is worthwhile. McCarthy utilizes a Microsoft Excel–based estimating model that assumes an incremental increase in soiling losses for each day between rain events. The soiling estimator relies on user-assigned daily soiling rates, typically set between 0.05% and 0.1%, and resets soiling losses to 0.0% after a quarter-inch of rainfall.
Site-specific analysis. A high concentration of utility-scale PV systems are installed near Gila Bend, Arizona, including First Solar’s 290 MW Aqua Caliente Solar Project, which is expected to come on line in 2014. A review of 20 years of weather data can be modeled with an assumed soiling rate of 0.1% per day to plot soiling losses on a monthly basis and show the variation in soiling losses per season. Understanding this variation provides system owners and operators the ability to identify short-term losses due to soiling. These losses can then be assessed against the corresponding electric rates to determine the impacts on ROI and whether cleaning the array is justified.
Graph 1 (above) plots monthly soiling losses over a 20-year period. The highest soiling losses occur in the summer months, peaking in June and reaching the annual low in September—presumably once the dry season ends. The average annual soiling loss for the time frame evaluated is 5.2%, but losses during the summer months are much higher. In this example, assuming a 5.2% decrease in production due to soiling during June and July could result in underproduction relative to monthly goals. Given that most utilities pay more for kilowatt-hours produced during periods of high demand, the financial losses to soiling during these months would be greater.
The large data set for this area also allows us to compare annual soiling loss values over the same 20-year period. Using the same assumptions as the previous model, the estimated annual soiling losses range from 2.5% to nearly 15%.
As illustrated in Graph 2, the annual soiling losses vary quite a bit from one year to the next. The majority of the years have soiling losses between 3% and 6%; yet in 2002, when it rained only twice within a 9-month dry stretch, estimated soiling losses were greater than 14%.
Variation by location. Designers frequently assume that soiling characteristics and the associated impacts on PV systems are consistent throughout the Desert Southwest. We can effectively compare two distinct areas using McCarthy’s soiling estimator and identify local or site-specific trends. In Graph 3, McCarthy compares the average monthly soiling losses between Gila Bend, Arizona, and Bullhead City, Arizona. Bullhead City is located along the border between Arizona and Nevada, just south of Boulder City, Nevada, and, similar to Gila Bend, is experiencing a significant amount of PV development.
Compared to Bullhead City, Gila Bend averages 1.5 additional rain events per year. The rain events are also more consistent and evenly distributed throughout the year. As a result, Bullhead City has an average annual soiling loss of 9.4%—4% greater than Gila Bend—with losses exceeding 10% in August, when kilowatt-hour values are at a premium.
The Value of a Good Wash
Soiling losses not only vary from one location to the next within the Desert Southwest, but also vary from one month to the next. Cleaning utility-scale arrays can be costly and may not be worth it. Soiling estimation models are an effective way to simulate a prescribed cleaning and determine if the reduction in soiling losses justifies the expense.
For example, for the Gila Bend area, we can simulate prescribing a single wash in early July so that the soiling loss returns to 0.0%. As we can see from Graph 4, with a single wash per year, the estimated annual soiling losses decrease. There is a significant reduction in loss in 2002, from nearly 15% to approximately 7.5%. Although a single wash reduces soiling losses in 2002 and reduces the average annual losses due to soiling by 1% over the 20-year period, certain years experience minimal, if any, improvements. Depending on the cost of cleaning relative to the increase in revenue, the model suggests that it might have been worth cleaning the array only in 2002.
If it rains immediately before or after a cleaning, that can nullify or minimize the benefits of a cleaning. When this occurs, as it did in 2000, cleaning the array does not improve soiling losses for the year. Although system operators could opt to forego cleaning an array after a rain, unfortunately, until we have foolproof weather forecasting, it is impossible to guarantee it will not rain shortly after cleaning an array—which makes choosing if and when to wash even more challenging. That said, Graph 4 shows that a single wash in July 2002 would have had a significant impact on production and therefore warrants a cost-benefit analysis.
When applying the same simulation to the Bullhead City region, the annual soiling losses are decreased by 3% and an additional 1.3% if a second wash is prescribed. Thus, cleaning an array in the Bullhead City region twice per year is estimated to reduce the annual losses due to soiling by almost 5% and may be worth the investment. Understanding the increase in production due to cleaning allows system owners to calculate the associated increase in production revenue, and compare that with the cost of cleaning the array.
Using the Soiling Estimation Model for Operations
Soiling estimation models are also valuable tools during the design and operational phases of a project. Modeling soiling losses based on local weather data and assumed soiling rates provides valuable information for production modeling, especially when systems are intended to meet seasonal or monthly energy demands. Likewise, the model is quite valuable once the system is in operation. You can simulate a prescribed wash to estimate how much it will increase production. With this information, you can conduct a cost-benefit analysis to determine if washing the array is worthwhile. Finally, you can design a soiling estimating model to notify system operators during periods of abnormally high soiling loss due to extended dry periods. Such notifications could prove valuable, such as at Gila Bend in 2002 when soiling losses neared 15%.
—Scott Canada / McCarthy Building Companies / Tempe, AZ / mccarthy.com