The Meniscus Analytics Platform (MAP) is a high performance, flexible and scalable cloud based analytics platform. It was developed around our 15+ years experience in delivering web based analytics and built for real-time processing of large data sets. MAP’s scalability allows it to deploy a solution for hundreds of thousands to millions of ‘Things’ whilst its flexibility allows properties and metadata to be added dynamically allowing an application to develop as its needs change. Finally the power of MAP delivers lightning fast calculation speeds to deliver analytics on demand – in terms of Hyperlocal rainfall MAP is processing 150,000,000 raw data points a day whilst delivering 100,000 predictions a second.
We are committed to continually improving the accuracy of these predictions over time and to this end we are adopting a number of techniques. For each Journey we also store the actual rainfall for that Journey against our predicted rainfall so that we can compare and measure the improvement in accuracy over time.
Local wind data
Initially Hyperlocal Rainfall combined various data sets to provide its predictions including local wind data obtained from various wind stations around Peterborough (around 21 stations). Whilst this data worked well when Hyperlocal started, as it was initially a project specifically for Peterborough, it was not sufficient or sustainable for the whole of the East of England or indeed the whole of the UK.
High Altitude wind data
Hyperlocal has progressed from using Peterborough wind data to using high altitude wind data for the whole of the UK. This is forecast wind data that is more accurate than ground wind data due to the fact that the clouds may move in one direction, while the below ground wind data would suggest it is moving in the opposite direction. Therefore by using high altitude wind data the tracking of rainfall data becomes more accurate which improves the Hyperlocal predictions.
In the medium to longer term, Meniscus aims to track every individual pixel of rain at 1km2 resolution to track the direction and velocity so that we can become less reliant on a forecast wind data set. This in turn will make the predictions much more dynamic to local conditions.
The principle of ground-truthing is to compare the radar rainfall data with accurate ground based weather station data and to make small adjustments to the radar data so that it more accurately matches the weather station data. This involves implementing the Marshall-Palmer Z-R relationship between radar reflectivity and actual rainfall data, Z and R respectively.
Ground-truthing is a complicated process despite its seemingly simple equation. We are currently looking to further improve our Ground-truthing efforts by carrying out more regular and automated zonal ground truthing. Additionally our current attempts at this are using a static (whole period based) relationship. Our final objective is to move to dynamic (event based) re-calibration.