Evapotranspiration Estimation Using NWP
By
Asnor Ishak and Dr. Dawei Han
Water Environment Management Research Centre
(WEMRC)
Department of Civil Engineering,
Faculty of Engineering,
University of Bristol,
University Walk, Clifton, Bristol
24 May 2010
7th -12th December, 2006, St Moritz, Switzerland
OUTLINE
Methodology
Result/
Discussion
Motivation/Intro
Q/A
Conclusion
Further
research
MOTIVATION
 1) what are the accuracy of the downscaled data in
comparison with the in-situ measurements?
 2) is there an improvement between the downscaled data
and the original global data?
 3) what is the accuracy of the evapotranspiration
estimated from the downscaled data?
 4) what are the impact of individual weather variables on
the evapotranspiration estimation?
 5) how could the result be further improved?
This study explores the first 4 questions and the last
question is discussed with some suggestions.
INTRODUCTION
Evapotranspiration plays a major role in the
hydrological cycle, which has great importance
in agricultural, hydrological, ecological and
climatic systems
Downscaling techniques - the performance of
the global data in ungauged catchments
Useful for water resources assessment or
forecasting by using meteorological and
hydrological model
Need to quantify/make a comparative
assessment
Comparison of the modeled and measured data
Methodology - What do we need
 Numerical Weather Model – MM5
- Regional mesoscale model used for creating weather forecasts and climate projections. –
- Maintained by Penn State University and the NCAR. (Since 1970.s)
 FAO Penman equation
The Penman-Monteith method refers to the use of an equation for computing water evaporation
from vegetated surfaces.
 ERA-40 reanalysis data
–
–
–
–
available from ECMWF website – 10x10 resolution
Variables? Rh, Wnd, Nr, Temp, Prs
6 hourly data with 4 months data in 1994 (Jan, Mac, Jul, Oct 1994)
Reanalysis data sets have become one of the most important data sets for scientific and
application communities i.e to generate “climate” data
 Observation data from HYREX study in the Brue
catchment
METHODOLOGY
Components of MM5
• long = -2.47, lat = 51.11
• grid resolution 1-km for Domain 4
• 19 × 19 horizontal grids
• Initial and boundary condition were
obtained from ECMWF
• parameterization scheme didn’t take
into account
METHODOLOGY
Components of ETo
37
0.408( Rn  G)  
u 2 (e o (Thr )(1  Rh))
Thr  273
ETo 
   (1  0.34u 2 )
where :
Rn
G
Thr

eo(Thr)
ea
u2
Rh

= net radiation at the grass surface (MJ m-2 hour-1),
= soil heat flux density (MJ m-2 hour-1),
= mean hourly air temperature (oC),
= saturation slope vapor pressure curve at Thr (kPa oC-1),
= saturation vapor pressure at air temperature Thr (kPa),
= average hourly actual vapor pressure (kPa), and
= average hourly wind speed (m s-1).
= relative humidity
= psychrometric constant (kPa oC-1),
There are 21 equations need to be calculated for estimation of ETo
METHODOLOGY
Brue Catchment
• Cover an area of 135.2
sq km (in Somerset)
• Obtained by NERC HYREX (The Hydrological
Radar Experiment)
• May 93 – 97 ext to 2000

METHODOLOGY
Hourly Evaluation : Statistical performance and Sensitivity Analysis
i1 ( yi  xi  MBE )
n
SD 
n 1
2

MBE 
n
i 1
( ymi  yo i )
n
• The SD and MBE are also expressed as percentage of mean value of
corresponding meteorological parameters.
• Not meant as a pass/fail test but to put modeling results in the
proper perspective
RESULT
The statistics of different
meteorological variables
derived from the mesoscale
model (MM5)
BIAS (%)
Surface pressure
Surface temperature
Relative humidity
Wind speed
SD (%)
MM5
Reanalysis
MM5
Reanalysis
Jan-94
0.04
0.08
0.15
0.11
Mar-94
-0.073
0.08
0.38
0.07
Jul-94
0.15
0.18
0.08
0.07
Oct-94
0.19
0.22
0.1
0.1
Jan-94
-3.54
13.48
24.1
19.9
Mar-94
6.01
7.8
22.5
14.1
Jul-94
9.35
3
11.7
10.5
Oct-94
8.81
8.99
18.0
14.2
Jan-94
-5.47
-5.55
7.8
7.6
Mar-94
-9.23
-9.26
7.8
9.1
Jul-94
-21.09
-13.61
11.1
10.5
Oct-94
-6.69
-8.2
9.7
10.9
Jan-94
260
216
148
121
Mar-94
217
202
87
98
Jul-94
301
273
145
151
Oct-94
419
332
194
160
RESULT
The comparison
between the ERA-40
reanalysis data and
the MM5 data
RESULT
Ten day time series of surface pressure between the reanalysis, MM5 and the observed
Ten day time series of surface temperature between the reanalysis, MM5 and the observed (left: winter; right: summer)
RESULT
Ten day time series of relative humidy between the reanalysis, MM5 and the observed
Ten day time series of wind speed between the reanalysis, MM5 and the observed
RESULT
The results of the ETo derived from MM5
BIAS
SD
(mm/hr and %)
(mm/hr and %)
Jan
0.018 (46%)
0.024 (64%)
Mar
0.023 (27%)
0.032 (37%)
July
0.088 (44%)
0.085 (43%)
Oct
0.029 (29%)
0.044 (43%)
MM5 downscaled data impact on ETo
CONCLUSIONS
o Atmospheric pressure can be estimated very accurately from the
downscaled data (with less than 0.2% error)
o Wind speed is the worst weather variable to derive (with a huge
discrepancy of around 300%)
o Air temperature is quite reasonable (< 10%). The net radiation and
relative humidity have about 10~20% error
o In comparison with the original reanalysis data, the downscaled data
are generally better except wind speed.
o The ETo estimation from the downscaled data has about 30%~40%
error compared with the estimation from the observed weather
variables.
CONCLUSIONS
o The sensitivity analysis has shown that the most important weather
variables are net radiation and relative humidity
o The dominant weather variables are net radiation (during the warm
period) and relative humidity (during the cold period)
o It is interesting to note that albeit the huge discrepancy in wind
speed (around 300%), its impact to ETo is insignificant
o This study provides hydrologists with valuable information on
downscaled weather variables and further exploration of this
potentially valuable data source by the hydrological community
should be encourage so that useful experience and knowledge
could be accumulated for different geographical and climatical
conditions
FUTURE STUDIES
o Several approaches to improve evapotranspiration
estimation from the downscaled data:
o 1) For example, wind speed from the MM5 downscaled data is
generally overestimated with large bias;
o If a temporary weather station is set up at the investigation site to collect
short term in-situ measurements;
o A mathematical model could be developed from the concurrent in-situ
measurements and the downscaled data to correct the overestimated
wind speed;
o If surrounding catchments have in-situ measurements, a correction
model could be developed from those nearby sites and then applied to
the investigation site using regionalisation principles;
o This model could be applied for long term water resources assessment.
FUTURE STUDIES
o 2) data assimilation using multiple sources could improve the downscaled
data.
o Remote sensing using satellites is able to measure some weather variables
such as solar radiation.
o Interpolation from the nearby weather stations could provide useful
information for the investigation site.
o It may be useful to build a data assimilation model to assimilate various
information sources so that the integrated data have the maximised utilisation of
all the information available;
o 3) modern artificial intelligence (AI) technology provides us with many
useful tools to model complex physical processes.
o If a short in-situ measurements are available at the investigation site, an AI
model such as Artificial Neural Networks or Support Vector Machines could
be developed to map the downscaled weather variables to ETo estimation.
o Further on, the AI model may be used to model the catchment runoff directly if
rainfall data are available;
o 4) numerical weather models such as the MM5 have many parameterisation
schemes and it may be useful to explore these schemes to find an overall
optimal scheme for evapotranspiration estimations.
Thank you
DEFINITION
Evapotranspiration :
- includes all processes by which water at the
earth’s surface is converted to water vapor.
- It includes evaporation from the plant
canopy, transpiration, sublimation and
evaporation from the soil.
- roughly 62% of the precipitation that falls on
the continents is evapotranspired.
- Evapotranspiration exceeds runoff in most
river basins and on all continents except
Antarctica (Dingman, 1994)