Flood Forecasting part 2

A new project has been undertaken by several UK Councils using MAP to deliver specific localised flood monitoring and is currently being tested by Cambridge City Council.

Flood forecasting

Flood forecasting

This service combines actual radar rainfall and forecast rainfall data and uses it to calculate the maximum rainfall return period for a specific location for the next 36 hours as well as the Antecedent Precipitation Index (API) as an indicator of moisture levels in the soil (this calculation uses soil type data).

The solution also displays river level gauging data from the Environment Agency and the historic and forecast information is displayed in a mobile-friendly dashboard. The next stage in this project is to implement a learning based model to correlate increases in these rainfall metrics with the increases in the river gauging height.

All the above working examples demonstrate some of the capabilities of MAP in terms of integrating a broad variety of datasets into models built inside MAP as well as the capability to use MAP to deliver aggregated datasets into third party models run externally to MAP.

MAP makes extensive use of RESTful Web Services making it possible for approved third parties to easily load new datasets, create new Entities and extract aggregated/calculated datasets from MAP. An Entity is any unique identifiable “Thing” which in this context might be a community, a property, a catchment etc. Each Entity contains internal ‘hooks’ to the calculations/models required to turn any raw dataset into the required calculated metrics. MAP provides the framework to allow these Entities to be readily extended making it easy to change the underlying models/calculations and adding in new database properties and metadata.

It is also envisaged that MAP will use additional datasets to enhance the capabilities of existing models and potentially use these to improve the inputs into current models. These additional datasets might include:

  • Satellite based imagery. MAP already has the capability to process images into gridded datasets and this is already used to capture soil type information. Such imagery might improve estimates of permeable/impermeable contributed areas to name one such use.
  • Topological data or potentially LIDAR data to improve the topographical data fed into models.
  • Using data from third party weather stations to include companies who operate a network of weather stations providing soil moisture sensors at various depths, humidity and evaporation sensors. Meniscus is already looking to integrate such data.

It is envisaged that a final solution may comprise an enhanced mobile app and associated mobile friendly dashboard delivering location specific flood information derived from this broad range of datasets. Importantly this flood information will use machine learning to improve flood predictions based on the relationship in the actual data and will not rely upon complex deterministic models – unless they are already available.