Observations: Solar radiation options

[DamienIrving]

The options for solar radiation are:

• AGCD (from 1990 onwards)
• SILO (from 1959 onwards)
• Multi-Source Weather (MSWX)
• ERA5
• AWRA (from 1990 onwards)

[DamienIrving]

Satellite estimates are only available from 1990 onwards, so SILO uses an blended data method (that doesn't use satellite data) to go further back in time (Zajaczkowski et al 2013)

[DamienIrving]

A non-negligible number of days in CSIRO's commercial copy of the AGCD dataset have missing values (at all grid points):

• 1990: 68 days
• 1991: 28
• 1992: 28
• 1993: 61
• 1994: 60
• 1995: 56
• 1996: 15
• 1997: 21
• 1998: 12
• 1999: 19
• 2000: 3
• 2001: 6
• 2002: 7
• 2003: 18
• 2004: 0
• 2005: 1
• 2006: 2
• 2007 onwards: 0

This isn't such a problem for bias correction methods (i.e. that look at quantile differences between model and obs), because the NaNs can just be removed and quantiles calculated from the remaining data. It is a problem for quantile delta mapping, where quantile changes from a model are applied to observations (i.e. the resulting dataset will have NaNs wherever the AGCD data has NaNs).

[DamienIrving]

The daily downwelling shortwave (solar) radiation that is produced for input into the AWRA model basically solves the AGCD missing data issue by filling missing values with the corresponding daily climatology value. The relevant technical report says:

Daily solar exposure (downward shortwave radiation) is produced from geostationary satellites (Grant et al. 2008) aggregated to the same 0.05° AWAP grid. The solar radiation record is from 1990 to yesterday, with the Himawari-8 satellite used since 23rd March 2016. Prior to that date the GMS-4, GMS-5, GOES-9 and MTSAT-1R satellites were used. Daily climatological averages (taken for each month) are used for solar radiation prior to 1990.

This is true for version numbers greater than 5 (i.e. 6 and 7). The update from version 5 to 6 involved updated input forcing data with improved site based bias correction for 1990 onwards and the adoption of the daily varying climatology used when there is missing data.

Analysis of the AWRA data has also revealed that it has less days of missing data than CSIRO's commercial copy of the AGCD data. A number of improvements to the AGCD dataset used by the AWRA modelling group (i.e. the very latest version of AGCD used in house by the BoM) has been made in recent years (including production of data for some formerly missing days) and it appears the CSIRO version of AGCD isn't the latest.

[DamienIrving]

CONCLUSION (for the moment): We could either use ARWA (v6 or v7) input data or effectively create the equivalent dataset by taking the AGCD data and filling the missing values with daily climatology values (calculated from all the non-missing days). (Applying the time shift described in #27)

[DamienIrving]

A rough comparison of the various options is shown in the following notebook:
https://github.com/AusClimateService/npcp/blob/master/obs-comparisons/solar_radiation.ipynb

[DamienIrving]

The MSWX downward shortwave radiation reference climatology was produced by bilinearly interpolating the ERA5 climatology (1979–2019) from 0.28° to 0.1° on a monthly basis, and rescaling the long-term mean to match the Global Solar Atlas (GSA) global horizontal insolation climatology (V2.0; https://globalsolaratlas.info; resampled from 0.01° to 0.1° using averaging). In the end they find basically no improvement on the ERA5 downward shortwave radiation. They say the lack of improvement is likely attributable to the small influence that the improved climatology has on the day-to-day variability (Beck et al 2022).