Quarterly Progress Update, Q3/2021
Wow, where do we start? So much has happened in the last few months! It’s definitely time to provide another progress update. Keep on reading to learn more about how our projects are doing, news about funding and how the team is growing.
Before we start with the updates though, an important announcement: We're looking for a Program Manager who will bring calm and structure to our growing research and development process and our partner engagement. If that sounds like you or someone you know, please check out our full job description. Applications close on September 17th.
We were one of the successful applicants to the Google.org Impact Challenge on Climate. Not only did this mean great financial support for us, but also that we were also invited to join Google.org’s accelerator program which allowed us to listen to experts and work through how we scale OCF. We had some great conversations and now have pages and pages of notes - reflecting how much we took away from these sessions! A big thank you to Google.org and everyone who spent time with us through the program.
We will use this experience to grow OCF with increased purpose and focus, while the funding will accelerate our work on and scaling of the Solar Electricity Nowcasting project across Europe. There is some great press coverage off the back of the announcement which you can read in for example Wired and CNBC.
Further to that, we are delighted to announce another big award: We have teamed up with National Grid Electricity System Operator (ESO), to develop and deliver our solar electricity Nowcasting work for the British electricity grid. The project will be running for 18 months and it is being funded by the Network Innovation Allowance (NIA) with around £500,000. You can read more about this in ESO’s press release, Bloomberg Green and CNBC.
Most importantly though, because of these awards and others we were able to hire our first two employees! We were blown away by the response to our first job ad - we received almost 200 applications to our job post for a Machine Learning Research Engineer. It was a very tough choice with applications showing both great skills and some really heartwarming enthusiasm for tackling climate change, and we now have two fantastic first employees: Jacob Bieker and Peter Dudfield!
Jacob previously worked at Google and the NASA Goddard Space Flight Center (where he was the first to use all of NASA's graphics processing units (GPUs) to develop a new machine learning model!) Prior to OCF, Jacob developed state-of-the-art machine learning models on petabytes of self-driving car data.
Peter has extensive hands-on experience in the energy industry through Siemens Gamesa Wind Power and Habitat Energy. He has implemented solutions in real-world energy systems and also holds a PhD in Fluid Dynamics from the University of Cambridge.
For our nowcasting project to predict solar electricity generation in the UK things have really started to progress in the last months:
- Machine Learning Research. With the expanded team we’ve been hard at work on the machine learning research, and our initial results are promising. The idea is to adapt state-of-the-art deep learning models (such as these two 2021 DeepMind papers: Perceiver IO & Skillful Precipitation Nowcasting) to predict solar electricity generation over the next few hours, using vast amounts of data from diverse sources, including satellite imagery, rainfall radar, and near-real-time solar electricity generation data from neighbouring systems.
Accessing the code. All our code is on GitHub, and we discuss our ideas and results openly in GitHub issues! If you want to get involved, jump in and help us discuss ideas in the github issue queues. It’s definitely still “research code” though, so be warned. For now, we've split our work into six main repositories:
- SatFlow: Predicting the next few hours of satellite imagery (of clouds).
- predictpvyield: Combining SatFlow's predictions with other data sources to predict solar electricity generation.
- nowcasting_dataset: Pre-preparing batches of ML training data.
- And we are implementing ML models from recently published models: Perceiver IO, Skillful Nowcasting, and MetNet. This is all very early work!
- Collaboration with Hugging Face. To reduce emissions as quickly as possible, we want to make it super-easy for people to include our machine learning models in existing products; and to enable researchers to build even better models. The wonderful folks at Hugging Face have done an amazing job at building infrastructure to share machine learning models. We're proud to announce that we're working with Hugging Face to share our machine learning models (both the code and weights) in the near future - watch this space!
- Third-Party emissions reductions assessment. The thing that is most important to us is to reduce carbon emissions and have a climate impact. To make sure we are working on the right ideas, we always estimate the potential impact of our solutions. Over the last months, Lia from Impact Forecast looked at our estimates for Nowcasting in more detail to see whether they sound realistic. They believe that our solar electricity Nowcasting work has the potential to enable a climate impact of about 1.3 million tonnes of CO2 equivalent per year across Europe.
To estimate how much photovoltaic energy will be produced, we first need to know where the solar PV systems are. At the moment this is not well known in the UK for all but the largest solar PV systems - particularly domestic systems are poorly documented.
Karan and Sol are continuing to make amazing progress on creating an app to boost the crowd-sourcing of solar panel locations. The idea is that anyone can walk around their local neighbourhood and take pictures of solar panels they see. Together with their GPS location, the pictures then get reviewed by volunteers before the extracted information is added to OpenStreetMap. They hope to release their app publicly later this year.
Over the next months we are looking forward to reviewing lots of applications for our Program Manager role, continuing to do more research on solar PV Nowcasting using machine learning, writing more funding applications, identifying user needs and creating an MVP for Nowcasting! We are also planning to release more tools and datasets for Nowcasting so that we can maybe even run a public ML competition. Stay tuned.