Managing Australia’s Ene...


Managing Australia’s Energy Mix Using Japanese Satellite Images

Managing Australia’s Energy
The Silicon Review
27 September, 2022

Recent advances in space-borne image sensors and database technology can play a big role in improving green energy efficiency on a nationwide scale. In this particular example, researchers at Axibase, a database software company based in Cupertino, California, demonstrate how weather observations from the Bureau of Meteorology Australia (BMA) and the Japanese Himawari satellite can be utilized to predict and optimize nationwide photovoltaic (PV) grid output in Australia.


National weather agencies, such as BMA, rely on automated ceilometers and sky imagers to gather cloud density as a daily input into weather forecasting methods.

While there are close to 8,000 functional weather stations in Australia, only 400 stations collect cloud cover, of which 50 collect samples at least four times per day. The stations measure cloud cover in oktas, or 1/8 increments, from 0 (sunny) to 1 (all clouds).

While the small number of stations is sufficient for weather forecasting, it is not enough for estimating power generation at thousands of solar power farms, such as the UQ Warwick Solar Farm displayed below, where power produced is a function of nominal capacity and irradiation (sunlight).


At the same time, the advances in imaging equipment make it possible to measure cloud cover using imagery collected by powerful image sensors installed on Himawari 8 - a geostationary weather satellite operated by the Japan Meteorological Agency and located 35000 kilometers away from Earth.


The Himawari 8 satellite provides high quality images of the planet in 16 frequency bands every ten minutes. To determine cloud cover from Himawari images we analyzed infrared band 13 with wavelength equal to 10,400 nm.

Using the geographical coordinates of each station, each weather station was mapped to a pixel on the the satellite images.

Note: Stations too close to the coast line and coordinate grid overlays were excluded from the analysis.

A simple method to detect clouds was used. Since clouds are cooler than the surface of the earth, clouds are rendered white on the satellite images, and the surface of the earth is black. Therefore, the brightness of the pixels in the images reflects cloudiness. Hence, we calculated cloud cover for a given location as the average pixel brightness over a 3 x 3 square of pixels, centered over the station.

Since one pixel on the image, depending on the location, covers an area from 5.5 x 3.9 to 5.5 x 5.6 square kilometers; the determination of cloudiness of an area sized about 230 square kilometers is analyzed.

To improve the correlation between stations and satellite measurements, an adjustment was made. Rather than using black as the ground color, we determined a shade of grey for each station that the earth, without cloud cover, reached during the day and used this color as the baseline. This logic is used because the temperature of the surface of the earth changes throughout the day, meaning that the brightness on the satellite images changes as well. The brightest point occurred during the night, when the earth is coolest. All darker shades are considered cloudless, and lighter shades are considered cloudy.

The chart below illustrates that the cloud cover measured by Himawari and actual PV output have a high degree of correlation.

An increase in calculated cloud cover leads to decrease in power output and vice versa. Note that no power is generated during night hours when sun altitude is at zero.



Data Flow

The data from the BMA was loaded into Axibase Time Series Database in JSON format. ATSD is a non-relational database optimized for storing and analyzing time series. Satellite images were downloaded from JMA web service, pre-processed in R using EBImage, oce, and geosphere packages, and then also stored in ATSD as cloud oktas measurements.



When Himawari went live in 2015, we knew it was a game changer. Our contribution at the time was the idea of merging satellite images and PV data. It's amazing how many forward-looking projects are building on this foundation a few years later. - Sergei Rodionov, CEO, Axibase

Given the time lag for satellite data is only 10 minutes, the data is recent enough to enable near-term forecasting of PV generation on a nationwide or regional level. To accomplish this task, a wide range of SQL functions in ATSD can be used to aggregate PV capacity of active solar stations adjusted for the observed cloud cover level at each station. With the continuing increase of solar power in the energy mix, the aggregate expected PV output can be proactively estimated by conventional power stations to adjust energy generation from non- renewable sources.