search

ai4er-cdt / flood_risk_shipston

The purpose of this project is to investigate whether we can establish the effectiveness of natural flood management (NFM) interventions undertaken in the British town of Shipston-on-Stour during 2017 to 2020 from publicly available meteorological data and private data from the river gauge in Shipston.
1 stars 0 forks source link

Shipston forcing data required for LSTM #17

Closed shmh40 closed 3 years ago

shmh40 commented 4 years ago

We would like forcing data for Shipston for the following features. Ideally we would have an average daily value for each feature over the whole Shipston basin, from ~1970 to 2020.

Currently in the model:

Potential to be added to the model

shmh40 commented 4 years ago

List of possible data sources (thanks Simon M/Luke/Marc)

UK water resources portal: https://eip.ceh.ac.uk/hydrology/water-resources/ [No data for the area - Ira] Centre for Environmental Data Analysis (CEDA): https://www.ceda.ac.uk/ Environment Agency data search portal: https://data.gov.uk/search?filters%5Bpublisher%5D=Environment+Agency National River Flow Archive: https://nrfa.ceh.ac.uk/ [Similar data to Wisky but more coarse in time - Ira] Land Use: UK gov land use registry/Aerial Photography

Other resources: Google Earth Engine: https://earthengine.google.com/ [looking into it - Ira, Arduin] QGIS software: https://qgis.org/en/site/ LiDAR (?)

herbiebradley commented 4 years ago

Update on the most useful features.

I ran a 5 layer LSTM model on several different feature combinations, both with a random single basin from CAMELS-GB and with multiple basins. All feature combinations were repeated 3 times with different random seeds and the results were averaged.

Features Single-Basin NSE % Improvment on Temp. + Precip. Multi-Basin NSE % Improvment on Temp. + Precip.
Temp. + Precip. 0.95467 0 0.839267 0
Temp. + Precip + Humidity 0.9668 1.27 0.8474 0.969
Temp. + Precip + Windspeed 0.9624 0.81 0.849 1.16
Temp. + Precip + Shortwave Rad 0.964 0.98 0.8536 1.708
Temp. + Precip + Peti 0.9669 1.28 0.8465 0.862
Temp. + Precip + Peti + Humidity 0.9677 1.365 0.8532 1.66
Temp. + Precip + Humidity + Shortwave 0.8556 1.946
Temp. + Precip + Humid. + Short. + Peti + Windspeed 0.9619 0.757 0.8684 3.47

Although the results have slight differences with single basin vs multiple basin, I believe we can conclude:

  1. The variance in basin discharge is almost completely explained by temperature and precipitation - adding other forcings only gives marginal improvements.
  2. Humidity and Peti give basically the same performance improvement, so we only really need humidity for Shipston.
  3. Shortwave radiation and Windspeed may give slight performance improvements over Temp. + Precip + Humidity, so may be useful for Shipston. If a nearby weather station has these values that would be good.

In addition, since we have found surprisingly good results (NSE of 0.96) using only a single-basin model, this may be the most practical way of predicting Shipston's flow with decent accuracy in the next few weeks.