Presentation by Bart van Osnabrugge (Deltares/Wageningen University) at the Symposium on catchment hydrology and WFlow, during Delft Software Days - Edition 2017. Tuesday, 24 October 2017, Delft.
2. Contents
• Introduction: Why is special attention needed to preprocessing of
forcing for gridded models?
• Method: genRE interpolation method
• Case study: Near real time precipitation grids for the wflow_hbv
Rhine model with genRE
• Hydrological model
• Research questions
• Data
• Results
• Conclusions
14 november 2017
Van Osnabrugge, B., Weerts, A.H., Uijlenhoet, R. (2017). genRE: A
method to extend gridded precipitation climatology datasets in near
real-time for hydrological forecasting purposes. Water Resources
Research. (Accepted) DOI: 10.1002/2017WR021201
3. Introduction: special attention to forcing (1)
14 november 2017
• Forcing uncertainty (quantity) remains one of the largest sources of
uncertainty in hydrological modelling
• Areal distribution of forcing has significant consequences on model
behaviour and model parameterization through calibration
• For a gridded model, every grid cell is fundamentally ungauged.
Higher resolution does not mean more information.
4. Introduction: special attention to forcing (2)
14 november 2017
• Deriving high quality near real-time gridded forcing data for
operational hydrological forecasting is a challenge
• limited availability of rain gauge data
• Task often left to the modeller / forecaster
• Real-time precipitation estimates should reflect the climatology
used for calibration of the hydrological model
• Many more historical observations (and other resources) are often
available for deriving climatological daily precipitation grids
• Aim: Find a method that makes use of the information stored in
climatological gridded datasets to interpolate near real-time gauge
data in an operational system
5. Method: genRE interpolation method
14 november 2017
Step 2: Interpolate with IDW with
additional multipliers
𝑤𝑖,𝑥 =
1
𝑑𝑖𝑥
2
1
𝑑𝑖𝑥
2
𝑛
𝑖=1
𝑚𝑖,𝑥 =
𝑀𝐵𝐺 𝑥
𝑀𝐵𝐺𝑖
𝑃𝑥 = 𝑃𝑖 𝑚𝑖,𝑥 𝑤𝑖,𝑥
𝑛
𝑖=1
𝑃𝑥 = 𝑀𝐵𝐺 𝑥
𝑃𝑖
𝑀𝐵𝐺𝑖
𝑤𝑖,𝑥
𝑛
𝑖=1
Name: generalized REGNIE (genRE)
Goal: use information from high quality
climatological grids to interpolate
operationally; and simple to implement
Step 1: Derive monthly background
grids (MBG)
6. Case study: wflow_hbv for the Rhine
14 november 2017
Old (current) model:
HBV model with subcatchments as
HRU. Calibrated for 148
subbasins (HBV-148)
New model:
wflow_hbv with 1.2x1.2 grid cell as
HRU.
Parameters HBV-148 -> wflow grid
Does this work?!
Yes, at least for modelled
discharge at main outlets little
difference
7. Case study: Research questions
14 november 2017
1. Does the use of a background grid based on a reference
climatological gridded dataset for interpolation of hourly rain
gauge observations result in comparable climatology and realistic
areal precipitation fields at different (yearly, daily, hourly) temporal
scales for operational forecasting?
2. What is the effect of station density on the quality of the obtained
gridded datasets, and how is the background grid affected by the
length of the employed climatological reference?
3. Does the use of the genRE method lead to modelled discharges
which are consistent with those obtained with the reference
climatology as forcing for a model that is calibrated on the latter?
9. Case study: climatological data
14 november 2017
genRE HYRAS EOBS RADOLAN
Time step Hour day day 15 min
Grid size 1.2x1.2km 5x5km 0.25 deg 1x1 km
Period 1996–
current
1951–2006 1950–current 2005–current
Domain Rhine
basin
KLIWAS Europe Germany
Updates Near real-
time
Sporadic Monthly Near real-time
Reference (This
presentation)
(Rauthe et al.,
2013)
(Haylock et al.,
2008)
(Bartels et al.,
2004)
15. Conclusions
14 november 2017
• generalized REGNIE (genRE) is designed to extend climatological
precipitation datasets in near real-time using (sparse)
measurement networks
• genRE successfully mimics climatological precipitation datasets
(HYRAS/EOBS) over daily, monthly and yearly time frames
• Created unique gridded 1.2x1.2 km resolution hourly precipitation
dataset for the period 1996–2016, (to be updated in real-time?),
covering the entire Rhine basin
• Differences in simulated discharges using HYRAS and genRE as
expressed in KGE scores are overall relatively small, but can be
significant for individual peak flow simulations
16. Ongoing efforts
14 november 2017
• Forcing: temperature, radiation and potential evaporation
• Data assimilation: Ensemble Kalman filter and assimilation of
water level & discharges from 700+ gauging stations
• Regionalization of model parameters: Derive grid cell
parameters from landscape characteristics (and calibrated
transferfunctions)