---
title: "PV simulators comparison: Conclusion"
slug: "pv-simulators-comparison-conclusion"
description: "Compare PV simulators like Solargis Evaluate, PVsyst, SAM, and pvlib to understand energy delivery differences and their impact on tool choice."
updated: 2026-08-10T14:42:48Z
published: 2026-08-10T14:42:48Z
canonical: "kb.solargis.com/pv-simulators-comparison-conclusion"
---

> ## 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: Conclusion

**In this document**

This concluding article of the Comparison of PV simulators series compares the energy delivered to the grid across the full simulation chain. It identifies which stages drive the differences in the final result, and what those differences mean for the choice of simulation software.

### **Overview**

The previous articles in this series isolate one stage of the PV simulation chain at a time, and analyze the differences between the simulators. This article compares the final energy delivered to the grid and traces how the per-stage differences combine to produce that result.

Quantitative results cover pvlib, PVsyst, and System Advisor Model (SAM) against Solargis Evaluate. Solargis Prospect and SolarFarmer are excluded from the quantitative comparison and appear in qualitative comparisons based on documented methodology only. For the comparison setup (sites, system configurations, version numbers, and bias/RMSE definitions), see [Setup and test methodology](/v1/docs/comparison-setup-and-test-methodology).

#### Key findings

- **Median bias at the grid connection point stays within 1.1% for all three simulators**, and 44 of the 72 cases fall within 2%.
- **Solargis Evaluate reports the highest energy delivered to the grid in most cases.** All three compared simulators read lower at the median, and lower in 50 of the 72 individual cases.
- The **conversion from plane-of-array irradiance to DC power** at the array output contributes more to the final bias than the whole rest of the chain.
- **Bias partly cancels along the chain**, but RMSE roughly doubles between front-side irradiance and the grid connection point.
- Differences **downstream of the DC array output are negligible**. DC-side losses, inverter conversion, and the entire AC side each move the final bias by 0.25 percentage points or less.

> [!TIP]
> **Note**: Solargis Evaluate is used as the mathematical reference series for bias and RMSE calculations across this comparison series. These statistics require a chosen reference, and Solargis Evaluate is one suitable choice. This convention does not imply Solargis Evaluate is more accurate than the other simulators. Differences indicate where simulators diverge in methodology, not where any one is correct.

### Energy delivered to the grid

Median bias is −0.93% for pvlib, −1.09% for PVsyst, and −0.45% for SAM. All three read lower than Solargis Evaluate more often than higher, though none is consistently low: pvlib and PVsyst are negative in 18 of 24 cases, SAM in 14. Individual cases land within 1% of Solargis Evaluate in 8 cases for pvlib, 9 for PVsyst, and 10 for SAM.

Fixed-tilt bifacial systems agree most closely for all three simulators, and monofacial trackers agree least for pvlib and PVsyst. Sulov is the extreme case, carrying the highest RMSE for every simulator (pvlib 18.0%, PVsyst 14.1%, SAM 24.1%). Dharan carries the largest single bias excursions on the bifacial tracker (pvlib −6.2%, SAM −7.7%).

Figures 1 and 2 present these results by mounting configuration, and the accordion below them holds the breakdown by individual site.

![](https://cdn.document360.io/ae2d502f-6c0d-4865-a68e-43ad8da61149/Images/Documentation/ac_grid__summary__bias(3).png)

***Figure 1****: Energy delivered to the grid, bias for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the Solargis Evaluate value, by mounting configuration across the six test sites.*

![](https://cdn.document360.io/ae2d502f-6c0d-4865-a68e-43ad8da61149/Images/Documentation/ac_grid__summary__rmse(1).png)

***Figure 2****: Energy delivered to the grid, RMSE for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the Solargis Evaluate value, by mounting configuration across the six test sites.*

**Per-site breakdowns of the energy delivered to the grid, bias and RMSE**

![](https://cdn.document360.io/ae2d502f-6c0d-4865-a68e-43ad8da61149/Images/Documentation/ac_grid__breakdown__bias(1).png)

![](https://cdn.document360.io/ae2d502f-6c0d-4865-a68e-43ad8da61149/Images/Documentation/ac_grid__breakdown__rmse(1).png)

### Main sources of differences

Table 1 shows the contribution of each stage in the simulation chain to the cumulative bias.

The bias in the optical chain (above the DC conversion step in bold) drifts toward positive. The transposition stage starts all three simulators between +0.29% and +0.61%, and the angular loss step adds a further +0.52 to +1.20 percentage points, because the compared simulators mostly compute smaller angular losses than Solargis Evaluate (PVsyst with monofacial configurations, using the ASHRAE model, computes a marginally larger loss). The DC conversion step then reverses that drift and overshoots it, subtracting 1.54 to 3.01 percentage points. The final bias for each simulator is relatively small, compared to the movements in some of the stages.

The near-agreement in the final row is therefore partly coincidental. Two independent sets of model choices happen to push in opposite directions across these test cases. The transposition and angular models push the bias upward; the DC conversion models push it downward. Neither responds to the other, and nothing in the chain enforces a balance between them.

A site or system that changes the size of one movement without changing the other will report a larger difference. Las Vegas with fixed-tilt bifacial configuration shows this for pvlib: near shading and the angular model carry the bias to +4.27% by the end of the optical chain, DC conversion then adds a further 3.40 percentage points instead of subtracting, and the final bias after electrical losses reaches +5.73%, against a median of -0.93%. Table 1 suggests the simulators agree to within about 1%, but across the 72 individual cases the bias differences run from −7.66% to +6.71%.

Differences of opposite sign cancel in bias but add in RMSE, which is why RMSE rises at nearly every step: from a median near 2% at front-side irradiance to 4.29% for pvlib, 3.85% for PVsyst, and 3.37% for SAM at the grid connection point. For pvlib the rise is monotonic in all 24 cases.

| Simulation chain stage | pvlib: change | pvlib: cumulative | PVsyst: change | PVsyst: cumulative | SAM: change | SAM: cumulative |
| --- | --- | --- | --- | --- | --- | --- |
| Front-side GTI transposition | +0.31 | *+0.31* | +0.61 | *+0.61* | +0.29 | *+0.29* |
| Horizon shading | +0.15 | *+0.60* | −0.12 | *+0.36* | not separated | *not separated* |
| Near shading | −0.01 | *+0.48* | +0.34 | *+0.39* | −0.05 | *+0.34* |
| Soiling | 0.00 | *+0.48* | 0.00 | *+0.39* | 0.00 | *+0.32* |
| Angular (IAM) | +1.20 | *+1.98* | +0.52 | *+1.00* | +0.52 | *+1.21* |
| **DC conversion** | **−3.01** | ***−1.09*** | **−2.52** | ***−1.66*** | **−1.54** | ***−0.69*** |
| DC-side losses | −0.05 | *−0.90* | +0.11 | *−1.29* | −0.02 | *−0.36* |
| Inverter conversion | −0.04 | *−0.95* | −0.02 | *−1.31* | −0.05 | *−0.44* |
| AC side to the grid | −0.02 | *−0.93* | +0.23 | *−1.09* | +0.08 | *−0.45* |
| **Energy delivered to the grid** | – | ***−0.93*** | – | ***−1.09*** | – | ***−0.45*** |

***Table 1****: Contribution of each stage of the simulation chain to the bias against Solargis Evaluate, in percentage points, as a median across the 24 site and configuration cases. The change column gives the median change contributed by that stage; the cumulative column gives the median bias carried at the end of it. The two columns are computed independently, so the change values do not sum to the cumulative value. SAM does not separate horizon from near shading, and its near shading figure therefore covers both.*

#### DC conversion stage components

The DC conversion stage includes the rear-side irradiance contribution for bifacial systems, the cell temperature model, and the module conversion model. The individual components are not exposed by every simulator, so the step can be decomposed only partially. The modeling differences are covered in the [Electrical modeling of the DC side](/v1/docs/comparison-electrical-modeling-dc-side) article and [Detailed results and analysis](/v1/docs/pv-simulators-comparison-detailed-results-analysis#dc-conversion-stage) does the partial decomposition.

### Implications for software choice

The stage that has the largest impact on the final result is not fixed. At the Sulov site, terrain horizon drives the divergence, and pvlib and PVsyst differ by up to 4.5 percentage points at the horizon shading step alone, in opposite directions. At the Dharan site with trackers configurations for SAM and pvlib, front-side GTI transposition contributes about 2 percentage points of bias before any loss is applied. Sites with significant terrain horizon, a high diffuse fraction, or temperature extremes are correspondingly harder to anticipate.

> [!NOTE]
> Agreement between simulators at the end of the chain does not imply agreement along it. A simulator can report yield within 1% of Solargis Evaluate at the grid connection point while differing by 3% at the PV array input. A comparison made only on final yield hides this fact.

**Software choice therefore matters more as project complexity rises.** On simple fixed-tilt systems at unobstructed sites the simulators converge to within about 1% at the median. The spread widens with tracking, complex terrain and horizons, which is where the modeling depth of the chosen tool has the largest effect on the reported yield.

### Further reading

#### Solargis knowledge base

- "[PV simulators comparison: Overview](/v1/docs/pv-simulators-comparison)": Solargis
- "[Setup and test methodology](/v1/docs/comparison-setup-and-test-methodology)": Solargis
- "[Irradiance modeling](/v1/docs/comparison-irradiance-modeling)": Solargis
- "[Optical losses](/v1/docs/comparison-optical-losses)": Solargis
- "[Electrical modeling of the DC side](/v1/docs/comparison-electrical-modeling-dc-side)": Solargis
- "[Electrical modeling of the AC side](/v1/docs/comparison-electrical-modeling-ac-side)": Solargis
- "[Detailed results and analysis](/v1/docs/pv-simulators-comparison-detailed-results-analysis)": Solargis

#### General simulator comparison

- ["Cross-validation of PV system simulation software"](https://www.researchgate.net/publication/335842590_Cross-validation_of_PV_System_Simulation_Software): Driesse, A., Patel, N.