Loading [MathJax]/jax/output/CommonHTML/fonts/TeX/fontdata.js

Precipitation Event Week

Are satellite rainfall products useful for hydrological simulations?

Design of the exercise

The exercise is divided in four different steps:

1) We will calibrate the MILc rainfall-runoff model (lumped version of the MISDc) over a basin in South Africa using the GPCC rainfall product. Being based on rain gauge data, we will assume this product the most accurate one. For the calibration we will use the Particle Swarm Optimization algorithm, optimizing the Kling-Gupta Efficiency index (KGE).

2) We will run the MILc over the basin using the calibrated parameters from GPCC and different rainfall products as inputs: ERA5, SM2RAIN, H23;

3) We will show the potential of improving flood simuation by using simple bias correction and recalibration of H23

4) We will show the potential of improving flood simuation by using an integrated product: H64

In particular, rainfall products used will be:

All the data are stored in different text files and are named considering the different rainfall products.

Import the necessary python libraries

Plot Africa basin

Let's take a look to the rainfall and discharge time series

STEP1: calibration of the MILc rainfall-runoff model (lumped version of the MISDc) over a basin in Africa using the GPCC rainfall product

Model calibration with Pyswarm optimization tool (PSO)

see the documentation and installation at https://pypi.org/project/pyswarms/0.1.9/

PSO https://link.springer.com/article/10.1007/s00500-016-2474-6

We are going to calibrate 8 model parameters: https://onlinelibrary.wiley.com/doi/abs/10.1002/hyp.8042

The figure shows, in the upper plot, relative saturation and rainfall. In the bottom plot the temporal comparison between observed and simulated river discharge is shown. The scores are displayed in the title. We obtain a satisfactory KGE of 0.73 with the calibrated parameters considering the entire calibration period

STEP2: running the rainfall-runoff model with GPCC-calibrated parameters considering different precipitation inputs

Model run using GPCC-calibrated parameters and input precipitation from ERA-5 rainfall dataset

Model run using GPCC-calibrated parameters and input precipitation from SM2RAIN rainfall dataset

Model run using GPCC-calibrated parameters and input precipitation from H23 rainfall dataset

Let's control what leads to poor performance of H23 rainfall product

So bias is the main driver of the poor model simulation...

Step 3: we might try to correct its bias and re-run the model with the bias corrected product

Correcting the bias of the rainfall is effective!! But let's see if recalibration is better...

Let's try to re-calibrate...

Results are now even better...

This is true also for SM2RAIN? Does recalibration improve the performance?

Yes, recalibration boost the performance of the model as parameters are able to take into account of the bias of the the rainfall observations

Step 4: What about integrating SM2RAIN and H23? Let's calibrate with H64 product...

 We obtain really good results. Let's see if we can do better than ERA5

Calibration using ERA-5 rainfall dataset

Well, in this basin H64 is even better than ERA5. So are H-SAF rainfall products useful? YES THEY ARE!

Lab done!!!