---
title: "PV simulators comparison: Optical losses"
slug: "comparison-optical-losses"
description: "Explore how PV simulators model optical losses like soiling, snow, angular reflection, and spectral losses affecting effective irradiance and energy yield."
updated: 2026-08-10T14:43:10Z
published: 2026-08-10T14:43:10Z
canonical: "kb.solargis.com/comparison-optical-losses"
---
> ## Documentation Index
> Fetch the complete documentation index at: https://kb.solargis.com/llms.txt
> Use this file to discover all available pages before exploring further.
# PV simulators comparison: Optical losses
**In this document**
This article compares how the compared PV simulators handle the optical losses that occur between plane-of-array irradiance and the effective irradiance reaching the active PV cell area: soiling, snow, angular reflection, and spectral losses.
### Overview
After plane-of-array irradiance is computed and shading has been applied, four further effects reduce the radiation that actually reaches the active cell area:
- Soiling losses from dust and particulate accumulation on the module surface.
- Snow losses from snow cover on the module surface, which blocks light partially or completely.
- Angular reflection losses from non-perpendicular light hitting the glass surface.
- Spectral losses from mismatch between the incoming solar radiation spectrum and the module's spectral response.
The first two are physical attenuation effects driven by site meteorology. The third is a geometric optical effect determined by sun position relative to the module and module surface properties. The fourth is a wavelength-dependent correction determined by atmospheric composition. Each of the six simulators implements these effects with different models, default values, and application points in the simulation chain. The choice of model and the way it is applied directly influences the modeled effective irradiance at the cell and therefore the simulated energy yield.
For the test sites, system configurations, and statistical methodology shared across the series, see [Setup and test methodology](/v1/docs/comparison-setup-and-test-methodology).
### Soiling and snow losses
Accumulation of dust and other particles on the PV module surface reduces the effective irradiance that reaches the PV cell – this phenomenon is referred to as **soiling losses**. Partial or complete snow cover on the module surface similarly reduces energy output, and this is referred to as **snow losses**.
#### Soiling loss modeling
The approaches to soiling modeling fall into two categories:
- **The static approach** (Solargis Prospect, PVsyst, SAM, and SolarFarmer): Requires the user to supply a loss factor with no internal model – the result depends on the quality of the user's input data.
- **The internal-model approach**: Derives loss values from measured atmospheric and weather inputs.
Three internal soiling models exist across the simulators. The first, the [Solargis soiling loss model](/v1/docs/soiling-losses) available in Solargis Evaluate, combines particulate matter concentrations (PM2.5 and PM10), rainfall, wind, temperature, humidity, and module tilt. The second, the pvlib HSU model, uses PM2.5 and PM10 with rainfall and module tilt. The third, the pvlib Kimber model, assumes linear daily soiling accumulation that resets when rainfall exceeds a user-set cleaning threshold. Internal models provide a data-grounded estimate when no site-specific measurements are available.
##### **Bifacial rear-side soiling**
Solargis Evaluate applies rear-side soiling at 15% of the front-side value. SAM accepts independent front- and rear-side soiling factors. The other simulators apply soiling to the front side only.
#### Snow loss modeling
A separate snow model exists in Solargis Evaluate (the [Solargis snow loss model](/v1/docs/snow-losses)), pvlib (the Marion model and the Townsend model), and SAM (Marion model). PVsyst and SolarFarmer do not model snow separately; the loss can be folded into the user-supplied soiling factor. The Marion model estimates snow coverage from snow depth and module tilt, treating snow sliding as the dominant removal mechanism. The Townsend model uses monthly snow totals and snow events to estimate the percentage of DC capacity lost. Where snow is a meaningful contributor to losses, a model that responds to weather data is more informative than absorbing snow into a static soiling factor.
#### Methodology comparison
| | Solargis Evaluate | Solargis Prospect | pvlib | PVsyst | SAM | SolarFarmer |
| --- | --- | --- | --- | --- | --- | --- |
| **Soiling internal model** | Solargis model | None | Kimber, HSU | None | None | None |
| **Soiling input** | 12 monthly LTA values | 12 monthly LTA values or 1 yearly value | Time series from chosen model | 12 monthly LTA values or 1 yearly value | 12 monthly LTA values or 1 yearly value, for bifacial independent front and rear | 12 monthly LTA, 1 yearly, or time series in input weather file (cloud) |
| **Bifacial rear-side soiling** | 0.15 × front-side | Not applicable | User-set | Front only | Independent rear-side factor | Front only |
| **Snow internal model** | Solargis model | None | Marion, Townsend | None | Marion | None |
| **Snow input** | 12 monthly LTA values | 12 monthly LTA values | Time series from chosen model | Within soiling factor | Time series from weather data | Within soiling factor |
***Table 1****: Soiling and snow modeling comparison per simulator.*
Soiling and snow losses - How does Solargis Evaluate compare
Solargis Evaluate is one of only two simulators, alongside pvlib, that derives soiling loss from actual measured atmospheric data (particulate matter, rainfall, wind, humidity) rather than a static user-supplied factor.
It is the only commercial application with this capability built in.
It is the only simulator that models bifacial rear-side soiling explicitly (at 15% of the front-side value) rather than ignoring it or leaving it to the user.
Solargis Prospect, PVsyst, SAM, and SolarFarmer all rely on the user's own loss estimate — a reasonable default for a fast tool like Solargis Prospect, but a real accuracy dependency for the others.
#### Numerical results
Results are computed for the six test sites and four system configurations defined in the [Setup and test methodology](/v1/docs/comparison-setup-and-test-methodology) article.
This comparison tests how consistently each simulator applies a soiling loss, not the quality of the underlying soiling estimate, which is addressed in the section above. To isolate this, one soiling loss profile was generated per site and used in every simulator, rather than letting each simulator's own model (or lack of one) generate its own estimate. Snow losses are not quantified here, because no simulator exposes snow as a separate loss series that could be compared.
Given the same soiling profile, all three simulators apply it almost identically. Bias and RMSE stay far below any level that would affect a yield estimate at every site and configuration (Figures 1 and 2). PVsyst is the most consistent of the three simulators. The existing small differences concentrate at Dharan and, to a lesser degree, Las Vegas, the two sites with the largest soiling losses to distribute. At both, RMSE runs above bias, which indicates the simulators differ slightly in how they spread a monthly loss factor across the year rather than in the annual total they apply.
.png)
***Figure 1****: Soiling loss bias for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
.png)
***Figure 2****: Soiling loss RMSE for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
**Per-site breakdowns of the soiling loss bias and RMSE**
.png)
.png)
### Angular losses
Angular losses arise from non-perpendicular incidence on the module glass surface. Beyond perpendicular, a larger fraction of the incoming radiation is reflected away from the cell. The losses, expressed through the incidence angle modifier (IAM), increase at high incidence angles.
#### Angular loss modeling
Two families of model handle this:
- **Empirical models** (Martin and Ruiz, ASHRAE) describe the IAM with a simple formula and one or two fitted coefficients, calibrated to measured data. They are easy to apply and transparent. With a single fixed coefficient, the model does not distinguish plain glass from anti-reflective coated glass, though the coefficient can be tuned for either case.
- **Physical models** (the De Soto Fresnel formulation, the pvlib physical model, the PVsyst Fresnel model) calculate the IAM from the optics of light passing through each layer of the module: air to glass, glass to encapsulant, and optionally an anti-reflective coating layer. They reflect the physical difference between coated and uncoated glass at the cost of requiring more material parameters.
The Sandia PV array performance model (SAPM) sits between the families, fitting a polynomial to measurements of a specific module. Across a full simulation year, energy yield differences between IAM models are typically modest – the differences are largest at low sun angles and on tilted or tracked surfaces with strong off-axis exposure.
#### Coefficient source
The coefficient matters as much as the model. Solargis Evaluate and Solargis Prospect derive it automatically from the module surface properties, treating soiling as the dominant influence. PVsyst and SolarFarmer take the IAM definition from the loaded module PAN file, SAM selects a Fresnel variant from the module type, and pvlib leaves both the model and the coefficient to the user. Table 2 lists the available models and the rear-side treatment for each simulator.
#### Methodology comparison
| | Solargis Evaluate | Solargis Prospect | pvlib | PVsyst | SAM | SolarFarmer |
| --- | --- | --- | --- | --- | --- | --- |
| **Available models (default in bold)** | **Martin and Ruiz** with Solargis-estimated coefficient | **Martin and Ruiz** with Solargis-estimated coefficient | Martin and Ruiz (direct and diffuse), ASHRAE, SAPM, Schlick (direct and diffuse), B. Marion diffuse, physical, interpolated, user-fitted; user selects | **PV module .PAN file IAM**, Fresnel AR, Fresnel normal glass, ASHRAE, Sandia (deprecated), user-defined | **Fresnel standard glass (1-slab) or Fresnel AR (2-slab)**, both per De Soto (2004); variant selected by module type | **Custom from .PAN file**, ASHRAE, CIEMAT (Martin and Ruiz), Fresnel normal glass, Fresnel AR with PVsyst coefficients, Fresnel AR with PVEL coefficients |
| **Rear-side IAM (bifacial)** | Separate front and rear | Not applicable | User-controlled | Fresnel normal glass | Same as front | Cloud calculations only |
***Table 2****: Angular loss modeling comparison per simulator.*
Angular losses - How does Solargis Evaluate compare
SAM's physical Fresnel model and the empirical Martin and Ruiz model in Solargis Evaluate agree closely on trackers, and diverge moderately on fixed tilt.
The largest divergence, PVsyst on the bifacial configurations, traces to a difference in test input, namely the module PAN file IAM profile rather than the ASHRAE model, and not to sun-angle sensitivity.
The advantage of Solargis Evaluate here is convenience: the coefficient is estimated automatically from module properties, while PVsyst, SolarFarmer, and SAM rely on manufacturer data or a fixed model variant, and pvlib requires manual tuning.
#### Numerical results
Angular losses are the largest source of disagreement among the optical loss stages, and the disagreement is systematic. Nearly every simulator and model combination computes a smaller IAM loss than Solargis Evaluate (Figure 3). PVsyst on monofacial configurations is the exception, tracking Solargis Evaluate closely. That exception reflects the test setup rather than a model property – the IAM profile from the module PAN file was used for the bifacial configurations and the ASHRAE model was used for the monofacial ones. The PAN file profile computes 0.79 to 1.95 percentage points lower loss than ASHRAE, in the same direction in all 12 site-and-mounting pairs. pvlib and SAM apply their models consistently across all configurations, with SAM closest to Solargis Evaluate. The disagreement in general is larger on fixed tilt than on trackers, because a tracked plane holds the incidence angle near normal, where the IAM models converge. RMSE exceeds bias throughout (Figure 4), so the differences track sun position rather than acting as a flat offset.
.png)
***Figure 3****: Angular reflection loss bias for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
.png)
***Figure 4****: Angular reflection loss RMSE for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
**Per-site breakdowns of the angular reflection loss bias and RMSE**
.png)
.png)
### Spectral losses
Spectral losses arise from the mismatch between the spectral distribution of incoming radiation and the module's spectral response. The composition of the spectrum at the module depends mainly on air mass and the precipitable water content of the atmosphere. Aerosol optical depth and the clear sky index add further dependence.
#### Driver variables
Available models differ in which atmospheric inputs they use. Air-mass-only models (Sandia PV array performance model) are the simplest and require nothing beyond sun position. Two-variable models add either precipitable water (First Solar / Lee and Panchula) or the clear sky index (CREST, JRC, PVSPEC) – the first captures atmospheric humidity, the second captures cloud effects on the light spectrum. Three-variable models (Caballero) add aerosol optical depth. Richer inputs give a more responsive correction but demand more meteorological data – precipitable water and aerosol optical depth are not always available in standard weather files.
#### Technology dependence
The magnitude of the spectral effect depends strongly on the module technology. Crystalline silicon (c-Si) and copper indium selenide (CIS) modules have a broad spectral response and a small annual spectral correction. Higher-bandgap thin-film technologies, particularly cadmium telluride (CdTe) and amorphous silicon (a-Si), are more sensitive. Spectral correction is therefore most consequential for thin-film deployments and for sites with extreme atmospheric conditions, very humid or very dry, or with high aerosol load.
#### Methodology comparison
| | Solargis Evaluate | Solargis Prospect | pvlib | PVsyst | SAM | SolarFarmer |
| --- | --- | --- | --- | --- | --- | --- |
| **Available models (default in bold)** | **First Solar (Lee and Panchula)** | **First Solar (Lee and Panchula)** | Caballero, First Solar, JRC, Polo, PVSPEC, SAPM No default - user selects | **CREST** (a-Si) and **SAPM** (Sandia database modules), auto-applied, First Solar available but disabled | **SAPM** (when Sandia module model is selected) | **First Solar** (when precipitable water is available) |
| **Driver variables** | Air mass, precipitable water | Air mass, precipitable water | Model-dependent | Model-dependent | Air mass | Air mass, precipitable water |
| **Rear-side (bifacial)** | Same as front | Not applicable | User-controlled | Not specified | Not specified | Not modeled |
***Table 3****: Spectral loss modeling comparison per simulator.*
Spectral losses - How does Solargis Evaluate compare
Solargis Evaluate applies a physically grounded spectral correction (First Solar model) automatically for every module technology.
PVsyst neglects spectral correction by default for crystalline silicon, the technology used in most installations, so a PVsyst user has to know to turn it on.
pvlib offers more model choices but applies none of them by default, leaving the decision entirely to the user.
SAM and SolarFarmer both apply reasonable defaults conditionally. The practical gap is between the out-of-the-box coverage of Solargis Evaluate, SAM, and SolarFarmer, and the opt-in approach of PVsyst and pvlib.
#### Numerical results
Spectral loss can be compared only for Solargis Evaluate and pvlib. PVsyst applies its spectral correction on the electrical side rather than as an optical loss, and SAM does not export the term. Both simulators were run with the First Solar model, so this comparison isolates implementation differences within a single model rather than differences between models. The pattern follows atmospheric humidity, as the model design predicts: pvlib computes a slightly larger spectral loss than Solargis Evaluate at the humid tropical sites and a slightly smaller one at Las Vegas, the driest site (Figure 5). Mounting configuration makes no measurable difference, which is consistent with a correction driven by the atmosphere rather than by array geometry. RMSE exceeds bias at every site (Figure 6), so the two implementations diverge more at time-step level than the annual totals suggest. All differences stay small.
.png)
***Figure 5****: Spectral loss bias for pvlib against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
.png)
***Figure 6****: Spectral loss RMSE for pvlib against Solargis Evaluate, as a percentage of the front-side GTI entering the step, by mounting configuration across the six test sites.*
### Further reading
#### Solargis knowledge base
- "[Argus optical simulation overview](/v1/docs/argus-optical-simulation-overview)": Solargis
- "[Argus PV simulation chain](/v1/docs/argus-pv-simulation-chain)": Solargis
- "[Comparison setup and test methodology](/v1/docs/comparison-setup-and-test-methodology)": Solargis
#### Soiling models
- ["Soiling model for PV applications: improved parameterizations"](https://ieeexplore.ieee.org/document/10359694): by Lara-Fanego, V., Gueymard, C., Micheli, L.
- ["The effect of soiling on large grid-connected photovoltaic systems in California and the southwest region of the United States"](https://www.semanticscholar.org/paper/The-Effect-of-Soiling-on-Large-Grid-Connected-in-of-Kimber-Mitchell/1de41e5ee4e3dcdc7da22c9f1f7def353ffd5c83): by Kimber, A., Mitchell, L., Nogradi, S., Wenger, H.
- ["Simple model for predicting time series soiling of photovoltaic panels"](https://ieeexplore.ieee.org/document/8735892/): by Coello, M., Boyle, L.
#### Snow models
- ["Measured and modeled photovoltaic system energy losses from snow for Colorado and Wisconsin locations"](https://www.researchgate.net/publication/273438187_Measured_and_modeled_photovoltaic_system_energy_losses_from_snow_for_Colorado_and_Wisconsin_locations): by Marion, B., Schaefer, R., Caine, H., Sanchez, G.
- ["Photovoltaics and snow: an update from two winters of measurements in the SIERRA"](https://ieeexplore.ieee.org/document/6186627/): by Townsend, T., Powers, L.
#### Angular reflection models
- ["A new model for PV modules angular losses under field conditions"](https://www.researchgate.net/publication/233335961_A_new_model_for_PV_modules_angular_losses_under_field_conditions): by Martin, N., Ruiz, J. M.
- ["Improvement and validation of a model for photovoltaic array performance"](https://www.researchgate.net/profile/Churchill-Agutu/post/Any-idea-how-to-determine-PV-cell-reference-charavertistics/attachment/59d625e26cda7b8083a22482/AS:468653582884864@1488747129041/download/Improvement+and+validation+of+model.pdf): by De Soto, W., Klein, S. A., Beckman, W. A.
- ["Determination of the optimum orientations for the double exposure flat-plate collector and its reflections"](https://www.sciencedirect.com/science/article/abs/pii/0038092X66900041): by Souka, A. F., Safwat, H. H.
- ["An inexpensive BRDF model for physically-based rendering"](https://onlinelibrary.wiley.com/doi/10.1111/1467-8659.1330233): by Schlick, C.
- ["Numerical method for angle-of-incidence correction factors for diffuse radiation incident photovoltaic modules"](https://www.osti.gov/servlets/purl/1350025): by Marion, B.
- ["Photovoltaic array performance model" (SAND2004-3535)](https://www.osti.gov/servlets/purl/919131/): by King, D. L., Boyson, W. E., Kratochvil, J. A.
#### Spectral correction models
- ["Spectral correction for photovoltaic module performance based on air mass and precipitable water"](https://www.semanticscholar.org/paper/Spectral-correction-for-photovoltaic-module-based-Lee-Panchula/9f7ce02c2d2cad525b4033febfa3dec206d3d08f): by Lee, M., Panchula, A.
- ["Spectral corrections based on air mass, aerosol optical depth and precipitable water for PV performance modeling"](https://ieeexplore.ieee.org/document/8254355/): by Caballero, J. A., Fernández, E., Theristis, M., Almonacid, F., Nofuentes, G.
- ["A simple model for estimating the influence of spectrum variations on PV performance"](https://www.researchgate.net/publication/256080247_A_simple_model_for_estimate_the_influence_of_spectrum_variations_on_PV_performance): by Huld, T., Sample, T., Dunlop, E.
- ["Development and testing of the PVSPEC model of photovoltaic spectral mismatch factor"](https://www.researchgate.net/publication/348262055_Development_and_Testing_of_the_PVSPEC_Model_of_Photovoltaic_Spectral_Mismatch_Factor): by Pelland, S., Beswick, C., Thevenard, D., Cote, A., Pai, A., Poissant, Y.
- ["Spectral irradiance correction for PV system yield calculations"](https://www.researchgate.net/publication/237353389_Spectral_Irradiance_Correction_for_PV_System_Yield_Calculations): by Betts, T. R., Gottschalg, R., Infield, D. G.
- ["Development of spectral mismatch models for BIPV applications in building façades"](https://www.sciencedirect.com/science/article/pii/S0960148125004823): by Polo, J., Sanz-Saiz, C.