Open Climate Fix is a new non-profit research and development lab, totally focused on reducing greenhouse gas emissions as rapidly as possible. Every part of the organisation is designed to maximise climate impact, such as our open and collaborative approach, our rapid prototyping, and our attention on finding scalable & practical solutions.

By using an open-source approach, we can draw upon a much larger pool of knowledge and skills than any individual company, so combining existing islands of knowledge and accelerating progress. Our approach will be to:

1. Search for ML (Machine Learning) problems where, if we solve a well-defined ML task, then there's likely to be a large climate impact. Then, for each of these challenges, we'll:

2. Collate & release data, and write software tools to make it super-easy for people to consume this data.

3. Run a collaborative 'global research project' where everyone from 16-year-olds to PhD students to corporate research labs can help solve the ML task (and, over the last 6 weeks, I've received over 300 emails from people who'd love to get involved).

4. Help to put good solutions into production, once the community has developed them, so we can be reducing emissions ASAP.

Our first area of focus: Solar Photovoltaics
Solar PV (Photovoltaics) is the largest source of uncertainty in electricity system operators’ forecasts. If a dark cloud moves across the sky, the grid can be taken by surprise and lose hundreds of megawatts of PV generation within minutes. This lost PV generation must be replaced immediately. But thermal generators take hours to spin-up from cold. The end result is that, whenever the sun is shining, the grid keeps lots of spinning-reserve online: mostly gas turbines, which are kept idling, but not generating electricity. This is expensive and carbon intensive.

The grid would require less spinning reserve if they had better PV forecasts for the next few hours. That is, better PV forecasts would reduce carbon emissions, and save money. In the UK, better PV forecasts should save £1-10 million per year (Taylor et al, 2016), and about 100,000 tonnes of CO2 per year. Scaled up globally, the carbon savings should be of the order of tens of millions of tonnes per year.

Solar PV 'nowcasting' (forecasting a few hours ahead)
We plan to build better PV nowcasts by tracking clouds from satellite images, and ‘rolling’ those images forwards in time using a combination of conventional numerical weather predictions, and machine learning.

We’ll use all available information about the winds at each cloud’s altitude, and teach our machine learning model how clouds ‘evolve’ over time.

We are excited to apply an open-science approach to the nascent area of PV forecasting to achieve a step-change improvement in forecasting accuracy, and hence to help fix climate change.

Solar PV mapping
We'd like to support the OpenStreetMap community to map the location of the world's PV panels in OpenStreetMap. (OpenSteetMap is like the Wikipedia of maps: anyone can edit the database.) We plan to use a combination of machine learning and the wisdom of the crowd to locate PV panels.

You can find more information on our website: http://openclimatefix.org