A physical-empirical model for site-specific soiling accuracy

The PVRADAR soiling model enables accurate prediction of lifetime soiling losses for utiltiy-scale PV power plants:

Combination of satellite data and local measurements: The model integrates airborne particle concentration data such as PM2.5 and PM10, precipitation, and other meteorological variables. By combining satellite data with measurements from the nearby weather stations, it captures historical patterns and seasonal impacts.
Location-specific model calibration: Model parameters, including the rain cleaning threshold, are estimated using soiling measurements from multiple geographies and climate zones. This ensures the model is calibrated to the specific environmental conditions of each site.
Consideration of plant technical parameters: For realistic loss assessment, the model factors in key plant technical parameters, such as mounting structure type, maximum tilt angle, and night stow position. Electrical constraints, including clipping and curtailment, are also incorporated to reflect the impact on actual energy losses.

Types of soiling

Soiling refers to the accumulation of substances or materials blocking part of the incoming irradiance and thereby reducing the electrical output of PV modules. Soiling losses vary by location, influenced by climatic conditions, vegetational cycles, human activity, and other factors, and usually result from the combined effect of various pollutants.

Climatic effects

Dust

Dust is the primary source of soiling in dry and arid environments, such as deserts. Originating from mineral sources, dust can be effectively cleaned by rain. However, after extended dry periods, losses can be significant.

Snow

Snow soiling occurs when snowfall covers PV modules. Although this type of soiling can be severe, it is typically short-lived because the snow tends to slide off, melt, or be blown away by the wind. The impact of snow soiling is highly dependent on the orientation and mounting structure of the modules. While cleaning machines can remove snow, such an approach is usually not cost-effective.

Pollen

Pollen soiling, caused by nearby vegetation, is highly seasonal. Although rain can wash away some of the pollen, a significant portion may adhere to the modules in the presence of high humidity, forming persistent layers that require wet cleaning. Initially, the impact of pollen soiling may be minimal, but it can accumulate over the years, leading to a substantial reduction in production efficiency.

Agriculture

Industrial and agricultural activities can create super-localized soiling effects, which change over time as land use evolves. These effects are challenging to predict and must be factored in by engineers. Other examples of localized soiling sources include livestock, mines, and unpaved roads.

Birds

Bird droppings is a form of highly localized soiling that is very difficult to predict. Bird droppings can create hot spots on PV modules, leading to potential damage and efficiency loss. In areas with large bird populations, regular maintenance and cleaning schedules are essential to mitigate these effects.

Dust

Snow

Pollen

Agriculture

Birds

PVRADAR soiling model has been validated by industry leaders

To make fast, confident decisions, we need models we can trust. This collaboration gave us a clear picture of what’s reliable and what isn’t.

Brett Pendleton

Director - Performance & Design

Discover independent studies and real-world success stories that demonstrate the accuracy and reliability of the PVRADAR Soiling Model across diverse environments.

TotalEnergies validates PVRADAR as the most accurate soiling model across PV sites in the US

TotalEnergies conducted a blind validation of the PVRADAR soiling model across 9 U.S. PV sites. Using only location and mounting configuration inputs, PVRADAR achieved the highest accuracy (median RMSE = 1.9%), outperforming pvlib’s HSU (2.3%) and Kimber (3.1%), as presented at the 2025 PVPMC Workshop in Cyprus.

Leeward benchmarks soiling models at California PV site to guide cleaning

Leeward Renewable Energy and PVRADAR conducted a blind comparison of soiling models using I-V measurements at two PV plants in California. Results confirmed the accuracy of the PVRADAR model and its value for portfolio-wide decision-making.

35% lower error compared to pvlib HSU

Since 2017, our engineers have been actively involved in measuring, predicting, and preventing soiling losses. The PVRADAR team is constantly working to improve the performance of its soiling models by building an extensive database of soiling measurements from locations worldwide.

Here are 10 examples from locations in the USA, showing a comparison of the PVRADAR model with the HSU and KIMBER models available in pvlib. The data for this example has been gathered by NREL [1] and is available in the Duramat Datahub. Results for pvlib HSU and pvlib Kimber were calculated using their default inputs.

Model example is not loaded yet.

[1] Micheli, L., Ruth, D., Deceglie, M. C., & Muller, M. (2017). Time Series Analysis of Photovoltaic Soiling Station Data (NREL/TP-5J00-69131). NREL.