Home

TOWARDS the goal of improved weather and climate forecasting using Big Data and Artificial Neural Networks (ANNs), I recently incorporated the Climate Lab Pty Ltd. Most of the work at the Climate Lab is done offline – it is not referenced or mentioned on blogs. It is sometimes published in the peer-reviewed scientific literature.

In contrast, this weblog, jennifermarohasy.com, is a place for me to occasionally make public comment. This is almost exclusively about issues that should be of public interest, and for which I have some experience and expertise as a scientist who has worked continuously for 35 years – variously in the private and public sectors including at universities, always on issues concerning the natural environment. For example, I’ve just recently responded to criticism of my paper with John Abbot, as published in the journal GeoResJ last year (2017).

I have also recently blogged on Great Barrier Reef issues in solidarity with my colleague Peter Ridd. Dr Ridd was to fronting the Federal Court in Brisbane on 12th November, this is now likely to be on 25, 26, 27 March 2019. My first blog post outlining some of the related issues can be read/downloaded here.

Jennifer Marohasy BSc PhD
Noosa, Queensland, Australia
14 November, 2018

****
Since 2007 the contents of this website and blog have been archived each year by the National Library of Australia.

P.S. Despite a significant inter-governmental investment in climate-related research over several decades, there has been no improvement in the skill of medium to long-range weather and climate forecasts. The current approach attempts to simulate actual physical processes, while assuming a dominant role for carbon dioxide as a driver of climate change.

I’m of the opinion that the future is in a radically different approach based on a new paradigm, already made possible by the advent of Big Data and Artificial Intelligence (AI) – specifically Artificial Neural Networks (ANNs). ANNs can be used to mine historical climate data for patterns, construct statistical models, and then using these to forecast.

This is a radically different approach to forecasting because it does not involve the coding of equations that describe weather processes. Rather it involves mining historical data, and building statistical models.

Measurements of any variable associated with weather and climate, when arranged chronologically at the appropriate scale, show patterns of recurring oscillations. The oscillations may not be symmetrical, but they will tend to channel between an upper and lower boundary – over and over again.

Most of these oscillations can be decomposed into sine waves of varying phase, amplitude and periodicity. These oscillations may, or may not, represent real world phenomena that can be explained in terms of atmospheric physics and chemistry, and/or the gravitational-pull of the Moon, and/or variations in the electromagnetic field of the Sun, and of course its changing declination relative to the tilt of the Earth, et cetera – but they exist.

As long as the relationships embedded in the complex oscillation continue into the future – and it is only these relationships that we model – a skillful forecast is theoretically mathematically possible.

Of course, the quality of data used to construct the arrays inputted into the ANNs is important. Thus, my longstanding concerns about the appropriate calibration of equipment used to measure surface temperatures by the Australian Bureau of Meteorology – and also the inappropriate remodelling of data through a technique known as homogenisation. I’ve detailed these concerns in a letter to the Chief Scientists, which can be downloaded and/or read here.