Stochastic 1-minute data generation

In this document

Some applications require 1-minute solar radiation data that cannot be derived directly from satellite imagery. A solution to produce such data is to use a stochastic generator that increases the temporal resolution from the native satellite resolution (10, 15 or 30 minutes) to 1-minute time step.

Overview

A stochastic generator is used to enhance the temporal resolution of solar irradiance data from 10, 15, or 30 minutes to 1-minute intervals. This approach generates data that accurately captures local solar variability while preserving statistical consistency with the original dataset. Additionally, the method produces 1-minute DNI data, enabling detailed, high-resolution analysis for PV system performance.

The availability of 1-minute data is essential for optimizing PV systems, including refining the DC/AC ratio, accurately estimating clipping losses, and meeting grid compliance standards. While the generated data is statistically representative rather than an exact minute-by-minute reflection, it provides critical insights for applications such as battery sizing and variability analysis, supporting advanced system design.

Representation of sample 1-day series of PVOUT with different time granularity

Stochastic generation of 1-minute irradiance

A stochastic generator enhances the temporal resolution of solar irradiance data from native satellite intervals (10, 15, or 30 minutes) to 1-minute time steps, offering a detailed representation of solar irradiance variability.

The synthetic generator utilizes satellite-derived global horizontal (GHI) and direct normal irradiance (DNI), along with their corresponding clear-sky components (GHIc and DNIc). The process is carried out in two key steps:

  1. Interpolation to 1-minute resolution: The input satellite data, available at 10-, 15-, or 30-minute intervals, is interpolated to a smooth 1-minute time series using mean-preserving spline interpolation. This ensures that the resulting "large signal" maintains full consistency with the original satellite data while preserving the mean values. Unlike standard interpolation methods, this approach guarantees that the average of the interpolated 1-minute values remains equal to the input satellite data.

  2. Enhancing small-scale variability: To accurately capture the short-term fluctuations observed in real-world solar irradiance data, a high-variability "small signal" with a zero mean is added to the large signal. This adjustment ensures that the synthesized 1-minute time series mimics the fine-scale variability seen in measured irradiance data, particularly during cloud-enhancement events, which are often undetectable at coarser time resolutions (beyond 10 minutes). The zero-mean property of the small signal preserves the consistency of the synthetic data with the original satellite input.

The small signal time series is derived from an extensive repository of 1-minute solar irradiance observations collected at over a hundred radiometric sites. It is primarily constructed from observed GHI and DNI values, from which the 10-, 15-, or 30-minute mean has been detrended to create a zero-mean series. Additionally, this repository is enriched with collocated satellite-based GHI and DNI estimates.

For each satellite-derived input value, the generator selects the most similar small signal patch from the repository based on prevailing sky conditions. This selection considers factors such as solar position, clearness index, and diffuse fraction. For example, in the case of 15-minute satellite resolution, a corresponding small signal patch consists of 15 consecutive values. By integrating these high-variability patterns, the Solargis synthetic generator produces realistic 1-minute irradiance time series that closely resemble ground-based measurements.

Use cases and limitations of stochastic generation

Applications such as PV system optimization and grid compliance require high-resolution 1-minute solar irradiance data that cannot be directly obtained from satellite imagery.

One significant use case is optimizing the DC/AC ratio, where 1-minute data provides a precise understanding of PV electricity production, allowing designers to better match inverters to system needs. It also aids in accurately estimating clipping losses, which occur when inverters cannot handle excess power from the PV array, reducing inefficiencies.

Furthermore, 1-minute data improves site selection by offering granular insights into areas prone to intermittent cloud cover. It also supports compliance with grid operators' technical standards, helping reduce penalties and enhance grid reliability. Such data is essential for applications requiring detailed analysis of solar resource variability.

A key limitation of the stochastic method is that the generated data is statistically representative rather than reflecting actual minute-by-minute variations at a specific location. Consequently, it is unsuitable for applications demanding precise modeled vs. real-data comparisons. However, for tasks such as battery sizing or analyses requiring statistical characteristics like variability and ramps, Solargis' 1-minute stochastically generated data is a reliable and valuable resource.

Further reading