Climatic influence on PLR uncertainty

Introduction

PV systems experience gradual performance reduction over their lifetime due to component aging and other effects. Many technology- and climate-specific mechanisms contribute to this performance loss, making it difficult to simulate and predict from first principles. Instead, it is common to extract empirical performance loss rates (PLR) from performance measurements from fielded systems using statistical techniques.

Many statistical techniques for PLR estimation have been proposed. What has emerged as a de facto standard is the “year-on-year” (YOY) method as implemented in RdTools. From a multi-year production timeseries dataset, the YOY method extracts an estimated PLR and associated confidence interval using a sequence of data filtering, normalization, aggregation, and bootstrapping steps. The accuracy and uncertainty of the PLR estimate depend on the ability of these data processing steps to suppress the “noise” introduced by effects like resource variability and identify the underlying performance trend. The longer the dataset being analyzed, and the more stable the climate, the more certain the PLR estimate becomes.

Since these estimated PLRs affect the financing of new PV systems, understanding the uncertainty of these estimates brings significant value. This chapter examines the YOY method’s performance across climates, with the central question: how many years of data are needed to achieve a given level of certainty, and how does it vary across climates?

Methodology

Synthetic PV performance datasets based on NSRDB PSM3 data and an assumed PLR of -0.75 %/year were simulated for a grid of locations across the United States. The datasets vary in length from 2-10+ years, and are repeated in each location to span the full history of the PSM3 dataset so that long-term resource variability is captured. The year-on-year method is then applied to each dataset and the 95% confidence interval for the estimated PLR is recorded. This gives a set of confidence intervals for each location and dataset length. These confidence intervals are then examined to determine, for each location, how many years of data are required to achieve a given level of certainty in the PLR estimate. This process can be repeated for different simulation parameters (e.g. PV technology).

Using these idealized simulated datasets, we can produce a “best case” estimate for PLR estimation uncertainty. In real datasets, the true uncertainty will be larger due to other sources of noise like system outages, sensor error, and array soiling.

For full details on the simulation and analysis methodology, see [2].

Scenario 1: c-Si

First, we examine the results for a simulated system using crystalline silicon (c-Si) PV modules.

This map visualizes the minimum dataset length (in years) for the PLR confidence interval to have a 90% chance of being smaller than ±0.05 %/year (results for other thresholds can be viewed using the layer control in the corner of the map).

Calculations

Calculate and show the difference between two layers:

Layer 1:

Layer 2:

This map shows that fewer years of data are needed in the desert southwest than elsewhere in the country to achieve a given uncertainty level. However, relaxing the required certainty level reduces the required dataset length, with almost everywhere in the U.S. requiring only 2-3 years of data to achieve PLR estimates of ±0.1 %/year (under the idealized conditions described above).

Scenario 2: CdTe

Do these results depend significantly on the technology being examined? The following maps show the same analysis as above, but assuming First Solar Cadmium Telluride (CdTe) PV modules instead of crystalline silicon.

Calculations

Calculate and show the difference between two layers:

Layer 1:

Layer 2:

The results in this case show broadly the same pattern as in the c-Si map, although any given location might require one more or one fewer year of data.

References

1. Michael G. Deceglie, Kevin Anderson, Daniel Fregosi, William B. Hobbs, Mark A. Mikofski, Marios Theristis, and Bennet E. Meyers. Perspective: performance loss rate in photovoltaic systems. Solar RRL, June 2023. doi:10.1002/solr.202300196.

2. Dirk C. Jordan, Kevin Anderson, Kirsten Perry, Matthew Muller, Michael Deceglie, Robert White, and Chris Deline. Photovoltaic fleet degradation insights. Progress in Photovoltaics: Research and Applications, 30(10):1166–1175, April 2022. doi:10.1002/pip.3566.

3. Marios Theristis, Kevin Anderson, Julián Ascencio-Vásquez, and Joshua S. Stein. How climate and data quality impact photovoltaic performance loss rate estimations. Solar RRL, December 2023. doi:10.1002/solr.202300815.

Data files

The geographic datasets shown on this page are available in the GeoTIFF files listed below:

GeoTIFF file	Description
	Description: Years of data required to achieve +/- 0.0375%/year uncertainty with 50% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.075_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.0375%/year uncertainty with 90% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.075_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.075%/year uncertainty with 50% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.15_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.075%/year uncertainty with 90% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.15_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.05%/year uncertainty with 50% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.1_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.05%/year uncertainty with 90% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.1_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.1%/year uncertainty with 50% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.2_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.1%/year uncertainty with 90% probability for year-on-year PLR estimates for c-Si PV modules.  can275_0.2_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.0375%/year uncertainty with 50% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.075_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.0375%/year uncertainty with 90% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.075_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.075%/year uncertainty with 50% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.15_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.075%/year uncertainty with 90% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.15_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.05%/year uncertainty with 50% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.1_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.05%/year uncertainty with 90% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.1_90.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.1%/year uncertainty with 50% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.2_50.tiff - [22.3 KiB]
	Description: Years of data required to achieve +/- 0.1%/year uncertainty with 90% probability for year-on-year PLR estimates for CdTe PV modules.  fslr_0.2_90.tiff - [22.3 KiB]