Solar data timing error skews simulation results

A critical review of the CSIRO Weather and Climate Data (Part 2 of 3)

The CSIRO weather and climate data sets discussed in our previous post are widely employed by building energy modellers, including applications to demonstrate compliance with the energy efficiency requirements of the National Construction Code (NCC). Exemplary Energy’s timely critique of the CSIRO weather and climate data sets ahead of the 2022 publication of the NCC, and highlighted three major shortcomings:

1. Reliance on weather data ending in 2015 for the characterisation of a warming climate;

2. A 30-minute error in solar data in the .epw format; and

3. A lack of coincident precipitation data despite the .epw format expressly inviting it.

Discussion of the first issue was previously covered on this blog. This article will review the 30-minute error in the solar data. In this second article of a three-part series, we investigate the 30-minute time offset in the CSIRO solar data.

The issue is likely caused by transcription error. One of the key differences between the .epw and NatHERS formats relates to the timestamp applied to solar irradiation data: the .epw format requires that the solar irradiation data represents the period prior to the timestamp, whereas the ACDB format specifies that the solar irradiation represents the hour centred on the timestamp. Failure to adjust to this time convention most notably affects the time of peak loading, and has major impacts on evaluations which incorporate on-site renewable energy generation.

Hourly plots comparing the diffuse radiation of the CSIRO .epw data with the Exemplary Energy .epw and ACDB datasets for Canberra for Days 13 to 24 in January. Note the half-hour time offset and the difference in radiation

A comparison between CSIRO’s datasets and those produced by Exemplary Energy further illuminates the differences. As an example, a detailed analysis of the Canberra climate – where the two datasets theoretically share the same source for January’s data – reveals a difference of 2.3 kWh/m² in global horizontal radiation for the month, 7.3 kWh/m² direct normal radiation, and 9.1 kWh/m² in diffuse radiation. These differences highlight the shortcomings of transcription between formats – even when applied correctly, interpolation between datapoints to account for timestamp differences will always introduce errors¹.

Applying these data to a one-month simulation of a 3-storey office building in Canberra, the Exemplary data resulted in increased cooling by 4.4% and increased peak cooling load by 3.3%, along with a 30-minute timing offset in the peak cooling load.

The opportunity for including precipitation data will be discussed in a future article.

The issues outlined herein need to be considered by policymakers and modellers alike. We have advised our colleagues at CSIRO of these findings and will continue to work with them to avoid further propagation of the errors and offer our support to improve the data going forward. We urge policymakers to be mindful of these issues as modelling inaccuracies arising now are embedded in building operations for many years to come.

In the interests of full disclosure, we note that Exemplary Energy offers high quality climate and weather data, including ersatz future climate data, that avoid the issues of the CSIRO datasets. These are available for modellers demonstrating compliance through the JV3 pathway, along with non-regulatory applications in design and optimisation and resilience testing of buildings and energy systems.

¹ Exemplary Energy’s weather and climate data products are always produced from meteorological observations at the highest-available temporal resolution. In the case of solar data, our datasets are directly based on Himawari satellite observations recorded at ten-minute intervals.