1 of 10

Week 5 – Rainfall Gens

2 of 10

For more information on what are timeseries models and how a weather generator is built – refer to Time Series Models and Time Series Models for Continuous Simulation strongly recommend to understand the theory behind a weather generator

In the course we are going to focus on the practical aspect of running a weather generator

3 of 10

What is a rainfall generator?

With these statistics a rainfall generator will replicate this rainfall timeseries

Synthetic rainfall timeseries

You have a rainfall series – this series will have a lot of stats:

  • Mean
  • Standard deviation
  • Variance
  • Skewness
  • Probability dry/wet
  • Seasonal variations
  • AMAX
  • Autocorrelation between timesteps
  • If data is spatial – correlation between different cells

Rainfall Generator

4 of 10

Rainfall generators vs hydrological models

  • A rainfall generator is just another type of model, hence there are lots of similarities with hydrological models
  • As in a hydrological model, you can also calibrate your rainfall generator

Trying to tweak the model and re-calibrating it would be a very long exercise on its own

      • You can change the parameters that it uses
        • You can change the fitting method

5 of 10

Rainfall generator validation

As in a hydrological model, you can also validate your rainfall generator!

6 of 10

Why do we need a rainfall generator?

Scotland - SEPA

Ile de France - France

England - EA

Ministerio para la Transición Ecológica y el Reto Demográfico - Spain

It doesn’t matter where you are, if you are a hydrologist you will be working with very long return periods!

7 of 10

Why do we need a rainfall generator?

30 year rainfall dataset

Unsuitable for calculation of very long return periods

Rainfall Generator

Generate a 10000 year rainfall dataset

Calculate with more confidence a very long return period rainfall/flood

8 of 10

Return periods - Issues with fitting short timeseries

Bad fitting – Not enough extreme values in a timeseries

These examples are from fitting a 25 year timeseries

Large gaps in the fit

Even when the fitting is appropriate, how to guarantee that extrapolation is good?

9 of 10

Alternative solution -Pooling Groups�

1. Find catchments that have similar characteristics to the one you are studying

  • Area
  • Mean Rainfall
  • Storage
  • Floodplain extent
  • Urban extent

2. With all the additional data derive a curve/formula to extrapolate the flow from the studied catchment

3. Use this formula on a 2yr return period event to derive a 100 or 200 year event.

10 of 10

Tasks for the week/next week

    • Install anaconda/miniconda on your computer – The weather generator can only run on a local machine – it is not possible to use Collab

    • By Monday 15/01 I will upload a video tutorial on how to use the rainfall generator on google drive – you can have a go on your own this week if you want ☺

    • Once you’ve ran it analyse the graphs given on the validation process :

  • Is your generator accurate? Does it accurately represent the rainfall in all months?
  • Is there any bias in your model? Does you model systematically over/underestimate rainfall processes?