Nghiên cứu nâng cao chất lượng dự báo mưa, lũ phục vụ vận hành các hồ chứa trên lưu vực sông ba tt tiếng anh

MINISTRY OF EDUCATION MINISTRY OF AGRICULTURE
AND TRAINING
AND RURAL DEVELOPMENT
THUY LOI UNIVERSITY

NGUYEN VAN HIEU

THE STUDY TO IMPROVE THE QUALITY OF RAIN AND FLOOD
FORECAST IN ORDER TO SERVICE THE OPERATION OF
RESERVOIRS IN THE BA RIVER BASIN

Specialization: Hydrology
Code number: 9 44 02 24

SUMMARY OF DOCTORAL DISSERTATION

HANOI, 2020


This scientific work has been accomplished at Thuy loi University

Supervisor: Prof. Dr. Vu Minh Cat

Reviewer No. 1: Assoc. Prof. Dr. Nguyen Tien Giang
Reviewer No. 2: Dr. Nguyen Tien Thanh
Reviewer No. 3: Dr. Vo Van Hoa

This Doctoral dissertation will be defended at ……………..……….. on
date …………………………………………………………………...

This dissertation is available at:

- The National Library;
- The Library of Thuyloi University.


INTRODUCTION
1. Rationale of the research
The Ba river system is one of 9 major river systems in Vietnam. Currently,
there are 5 large reservoirs to build in the river basin namely Ayun Ha, An
Khe - Ka Nak, Krong HNang, Song Hinh and Ba Ha with a total flood control
capacity of 260.5x106 m3 – a rather small volume in comparion with the
largest total 5-day flood vulume of 2,507.3x 106 m3 occurred in Cung Son
[1]. Therefore, if accurate forecast of rainfall and flood flowing into these
reservoirs with a longer expected time will create favorable conditions for
the reservoir’s owners to operate proactively, avoiding the phenomenon of
"shock release" to the downstream (increased discharge flow suddenly from
0 m3 to the maximum flow value to the downstream, causing inundation and
damage to people and properties downstream.
Therefore, the study to improve the quality of rain and flood forecast in order
to service the operation of reservoirs in the Ba river basin" is necessary and
urgent issues for that we can find scientific arguments and methods to improve
the quality of medium-term forecast of rainfall and flood to actively operate
reservoirs according to the inter-reservoir process in the Ba river basin.
2. Research objectives and new reserch contributions
2.1. Objectives of the study
- Research to select and use appropriate numerical models for forecasting
rainfall with an expected time of 72 to 120 hours.
- Research to select and us suitable hydrological models for forecast floods
go into reservoirs in the Ba river basin for operating inter-lake systems under
Decision 878/QD-TTG of Prime Minister dated in July 18, 2018

1


2.2. New research contributions
- Through the research the meteorological models of IFS and WRFARW are
selected to forecast quantitatively rainfall with extension of the expected
time from the current 48 hours to from 72 hours to 120 hours with magnetic
resolution of (14 x14) km to (5x5) km and MIKE-NAM and HEC-HMS
hydrological models to forecast the flow goes into the reservoirs from 03 to
05 days in advance for Ba river basin.
- the successful application of the numerical rainfall models as well as the
hydrological models forecast flood goes into 04 reservoirs in the basin,
serving the reservoir operation in the Ba river basin.
3. Scope and subject of the study
3.1. Research scope
In space: forecast flood flows goes into 04 reservoirs with a capacity of over
100 million m3 in the Ba River basin including Song Hinh, Krong H’nang,
Ayun Ha and An Khe reservoirs.
In time: Increase the expected time of the forecast from the current 24 - 48
hours to 72 to 120 hours.
3.2. Research subjects
The factors that cause to create rainfall and flood flow go into 04 reservoirs
in the Ba river basin are subjected to this study.
4. Scientific and practical significances
4.1. Scientific significance
The selection, testing and verifying and application of IFS and WRFARW
meteorological numerical models to forecast rainfall with reducing a
resolution of (14 x14) km to (5x5) km in the Ba River basin, for that it can
be improving the accuracy of the forecast results and extending the expected
time from the current 48 hours to 72 hours to 120 hours.

2


The set of the hydrological models of MIKE-NAM and HEC-HMS has been
applied for testing, combining with analysis and adjustment of affected factors
such as topography, geology, soil, and surface covers in order to quantitatively
forecast the inflows to the reservoirs for serving the operation of reservoirs and
prevent and mitigate flooding for the downstream Ba river basin.
4.2. Practical significance
Establish a flood forecasting tools with high accuracy and extended
forecasting time from the existing 48 hours to 120 hours, which can be used
practically to improve the forecasting accuracy and extend the time. expected
from the current 48 hours up to 72 hours to 120 hours and increased
resolution domain in rain forecasts as well as making the reservoir operation
more flexibly and also it hepls to have more options to achieve the highest
efficiency, ensuring reservoir safety, minimizing floods and inundation for
downstream areas.
The medium-term rainfall and flood forecasting methods in the Ba River basin
can be used as a reference’s materials for students and postgraduate students.
5. The structure of the doctoral thesis
In addition to the introduction, conclusion, list of references, appendices, the
thesis is structured into 03 chapters:
Chapter 1: Reserch Overview on medium-term rainfall and flood forecasting
Chapter 2: Building scientific basis for medium-tearm rainfall and flood
forecast in Ba river basin.
Chapter 3: results and discussions

3

CHAPTER 1 OVERVIEW OF RESEARCH ON MEDIUM-TERM
RAINFALL AND FLOOD FORECASTING
1.1. Research Overview on rainfall and flood in the world
Currently, medium-term forecasting is an integral part of all major
forecasting centers in the world. Most approaches to medium-term
forecasting are based on a combination of predictive methods in order to
capture the sources of uncertainty caused by the original field.
In 1992, the European Center for Medium-term Forecasting, EPS also used
the SV (Singular Vector) method to create initial disturbances (Palmer et al.,
1992) [4]. This EPS now has 51 component forecasts, makes daily forecasts
and provides results to countries in the European Communities that are
members of the ECMWF. This EPS system is called VAREPS (Variable
Resolution EPS) for 15 days forecast, in which the first 9 days the system
runs with TL399L62 resolution (about 50 km, 62 layers) and for the last 6
days with TL255L62 resolution (approximately 80 km, 62 layers). This is
the medium-term model with highest resolution EPS currently in the world.
At NCEP, the GFS model is run 4 times a day with every 06 hours with
selected resolution options of 0.25x0.25 degrees, 0.5x0.5 degrees and
1.0x1.0 longitude and latitude. The forecast duration of the model is up to
384 hours (15 days), that can completely meet the requirement for getting
forecast rainfall that is served as an input for the current hydrological
models.
In a study by F. Pappenberger et al., in 2008, the LISFLOOD model was also
used to study the 2007 flood event in Romania, using the input of rainfall
forecast from well known centres in the world like ECMWF, UKMO, JMA,
NCEP, CMA, CMC, BOM. The results from these centers could give alert
of the flood with 8 days in advance. The results also show that the combined
prediction based on the multi-model approach implemented by ECMWF and
UKMO had the best average characteristic of, the simulation, especially at
the tail of the distribution function, i.e the extreme value can be occured.

V. Triemig et al. (2015) used a combined flood forecasting system (AFFS)
for medium to large sized river basins in Africa with a 15-day forecast
duration. The main component of the forecasting system is the LISFLOOD
4


distributed hydrological model with GIS data and forecasted rainfall data
from the ECMWF center. The flood event in March 2003 at the Sabi River
basin (Zimbabwe) was selected to simulate, where 36 monitoring points
having observed data for cpmparison. The AFFS verification process has
achieved good results (estimated to reach over 70%). Especially, the system
has good flood forecasting results with a longer than 1 week) and basins with
large areas (over 10,000 km2). It can be seen that the application of
quantitatively forecast rainfall done by ECMWF for medium term flood
forecast will give satisfactory results.
1.2. Overview of research on of medium-term rain and flood forecasting
in Vietnam and Ba river basin
Currently, the meteorological HRM, COSMO and WRF models are being
applicable to run professionally at the National Meteo-hydrological Forecast
Center (VNNMHFC), while MM5 and WRF models have been used to
forecast rainfall at the Institute of Science and Technology of Climate
Change and RAMS model is used by the faculty for Meteo-hydrology and
Oceanography, Hanoi Natural Sciences University. In my dissertation, the
WRF model that is being run professionally at VNNMHFC is selected as the
following reasons:
+ NWP is regional and hydrostatic model that is used to run professionally
with 5km resolution.
+ WRF has a 3DVAR data synchronization system integrated coupling into
the model and it can assimilate a variety of remote sensing data such as
satellites and radars, ….

+ WRF has been evaluated and verified by many scientific studies and has
been assessed as good rainfall forecast model for the central region of
Vietnam (Bui Minh Tang et al., 2014).
The analytical and prediction field of ICMWF's global NWP model IFS of
14km resolution was chosen to be as the initial condition and boundary
5


conditions depending on the time of WRF model with 5km resolution. IFS
model was chosen as this is the best-rated global NWP model in the world at
present and also the highest resolution global model currently (Vo Van Hoa
et al., 2017).
Figure 1.1 below shows the diagram and briefly introduces the methods and
models used in the study for each specific objective.
The Ba River basin is a large river
in the Central part and Highlands
of Viet Nam. The density of the
monitoring network is sparse, so
the rainfall forecasting technology
used

must

be

quantitative,

objective and highly resolution,
coupled with modern flood forecat
tools. In other words, the research

direction applies highly-resolution
NWP

models,

hydrostatic-type

Fig 1.1: The diagram of research
approach

with horizontal resolution of about
5 km to be able to capture small

and medium-scale convection processes well. Therefore, the application of
the NWP models is the most suitable option for heavy rainfall forecasting in
the Ba river basin in which the horizontal resolution ranges within 5km.
Therefore, in this study, IFS and WRF-ARW models will be chosen to
forecast rainfall for Ba river basin.

6


CHAPTER 2 SCIENTIFIC
BASIS
FOR
MEDIUM-TERM
RAINFALL AND FLOOD FORECAST IN BA RIVER BASIN
2.1. Overview of the Ba river basin
The Ba River basin is one of the largest basins in the Central region Vietnam,
with a total catchment area of 13,417 km2. The length of the main river is of

374 km, originating from the Ngoc Ro mountain range (Kon Tum province)
at an elevation of 1,549 m, flowing through the territory of Kon Tum and Gia
Lai provinces in the North-South direction, turning to the Northwest Southeast from Krong Pa district (Gia Lai province) and West-East from the
territory of Phu Yen province and finally going to the East Sea at the Da
Rang estuary in Tuy Hoa city.
In this river basin with high
potential water resources in
combination with steep
slope, so the potential of
hydropower
on
the
mainstream
and
its
tributaries
is
huge.
According
to
the
hydropower
planning,
there are expected 12
hydropower
projects
Fig. 2.1: Map of the Ba River basin
namely An Khe-Ka Nak,
DakSong, Ba Thuong River, Ayun Thuong 1, 2, HChan, HMun, Ayun Ha,
Ea KRong Hnang, Ba Ba Ha, Hinh will be built with a total installed capacity

of 700 MW and an annual electricity of 2.6 billion KWh.
2.2. Establish rainfall forecast method in Ba river basin
2.2.1. Rainfall forecast numerical models for Ba River basin
The WRFARW model is used to forecast rainfall for the entire area (12ºN15ºN; 107ºE-110ºE) with a surface resolution of 5km corresponding to 90 x
7


90 grid nodes on the ground and 50 vertical levels. This model is run in a
hydrostatic form with a time step of 20 seconds as selected resolution of less
than 7 km. The input data of WRFARW model of 5km resolution is taken
from IFS model with 14km resolution.
The running procedure of WRFARW model for a single forecast is as follows:
Step 1. Download the IFS data including the analytical and forecast fields
(every 3 hours) until the 3-day term forecast.
Step 2. Decode the appropriate data domain to get enough input data.
Step 3. Interpolation of IFS data from 14km resolution to 5km resolution.
Step 4. Vertical interpolation from 26 pressure levels of IFS data to 50
vertical levels of WRFARW model.
Step 5. Set running duration and the route of input data for WRFARW model.
Step 6. Run WRFARW model.
Step 7. Decode the forecast rainfall data from WRFARW model and
interpolate the forecast rainfall data to the area with the observation stations
according to the nearest interpolation method.
2.2.2. Corrective method to determine forecast rainfall values
The research approach is determined by the following steps:
- Step 1: Assess the rainfall forecast results
- Step 2: Set the corrective coefficient and adjust the forecasted rainfall
- Step 3: Set the adjusted data as input to the hydrological modeling system.
Based on the forecast rainfall and cumulative observed rainfall at monitoring
stations, a regression equation between them for each forecast period must

be developed. With the forecast period from 0 to 240 hours, the 40
cumulative rainfall periods of 6 hours are corrected. Thus, for each prediction
period, linear regression equations are constructed separately.
8


Based on the univariate linear
regression method, the study
conducted adjusting rainfall for
each forecasting period. Rainfall
forecast is adjusted based on the
coefficients "a" and "b" of the
regression equation.

Fig 2.11 Diagram of method of
calculating rain forecast

The method used to forecast

rainfall according to the numerical model is carried out according to the
block diagram as shown in Figure 2.11.
2.2.3. Establish medium - term flood forecast method in Ba river basin
The flood forecasting using the MIKE NAM and HEC-HMS models from
rainfall forecast data presented above is conducted by the following steps:
The Ba River basin will be completely covered by a square area delimited
by 4 corners with
coordinates of 12ºN ÷
15ºN and 107ºE ÷
110ºE. The system
will

be

subdivided

into a grid of equal
sized squares of a grid
size of (14 × 14) km.
By doing this, study
area is divided into (39
× 39) squares and with
a finer grid size of an
area of (5 × 5) km, this
Fig 2.12 Diagram of process of moderate drought forecasting
process based on MIKE NAM and HEC-HMS models from
rainfall forecast data by numerical method

9

limit will have (90 ×
90) squares. This grid


system will be meshed with GIS tools and superimposed on the established
Ba river basin. To make it easier to locate the grid cells, the study carried out
the row marking (uppercase letters) and columns (ordinal numbers).
The predicted rainfall value will be determined for each square in size (14 ×
14) km or (5 × 5) km and the thesis considers this rainfall to be spread evenly
over the entire area (fig 2.14 and 2.15).

Fig 2.15 Grid division resolution
(5 × 5) km for the Ba River basin

Fig 2.14 Dividing grid resolution
(14 × 14) km for Ba River basin

The study will consider each square as part of the catchment area controlled
by its precipitation. Therefore, similar to the method of calculating the
average rainfall according to Theissen polygon, here fi is considered the area
of the square in the calculation basin. The result will be the weight of each
grid cell in size (5 × 5) km or (14 × 14) km (note that the total weight of all
grid cells is equal to 1.0). From there, it will calculate the average rainfall of
the basin before being included in the flood forecasting model.
The calibration and verification is conducted for each specific sub-basin within
the study area: An Khe sub-basin, Ayun Ha sub-basin, Krong Hnang sub-basin
and Hinh river sub-basin for specific NAM MIKE and HEC-HMS models.
Observed rainfall and evaporation data at monitoring stations in the river
basin are used for the calibration and verification.

10


Based on comparison between simulation flow and observed measurement
at the station to adjust and find the optimal parameter set on each sub-basin
to ensure sufficient reliability.
CONCLUSION OF CHAPTER 2
In order to forecast medium-term flood, the study will conduct adjustment
and testing separately for each sub-basin with two models, namely MIKE
NAM model and HEC-HMS model, in order to identify the optimal
parameters and ensure to ensure good simulation of the process of

precipitation flow in each sub-basin.
The basic input for the flood forecasting model is the projected rain value on the
grid cells. The study considers each grid cell to be a part of the catchment area
that rainfall is controlling, thereby calculating the weight for each grid cell and
finally, the average rainfall of the basin by the weighted average method.
The flood forecast results from the forecast rain data will be assessed for
errors, verification criteria, forecast quality and practical application
recommendations.
CHAPTER 3: RESULTS AND DISCUSSION
3.1. Rainfall forecast results in Ba river basin
3.1.1. Forecasting results from the model
The forecast rainfall of the two models is selevted from historical flood
events occurred at the river basin and shown in Table 3.2. With rain_wrfarw
data, the grid size (5x5km) is equivalent to (90 x 90) cells, while rain_ifs data
with the size (14x14km) km are (39x39) cells covers the entire Ba river basin.
Tab. 3.2 Estimated rainfall time using IFS and WRFARW models
Year
Time
period

2017
01-06/12
02/12/17

2016
1-5/11
15-18/12

2015
7-12/10

11

2014
27/11-1/12

2013
3-8/11

2012
3-7/10


Rainfall forecast results for the Ba River basin of the two models will cover
the entire basin with integral areas of 12ºN-15ºN and 107ºE-110ºE and total
forcast time for IFS model is 120 hours and it is only 72 hours for WRFARW
model. Format of these data is shown in figure 3.9-3.10.

Fig. 3.9 Format of forecasted rain

Fig. 3.10 Format of forecasted rain data

data from IFS (39x39) grid pattern

from WRFARW (90x90) grid pattern

3.1.2 Assess the forecasting skills of the model
To evaluate the rainfall forecasting skill for WRFARW model in the central
region, 8 heavy rains from 2013 to 2017 are selected. A total of more than
50 samples were collected for evaluated purposes. On average, each heavy

rain lasts for 3 days with the average total rainfall of about 250-300mm. To
assess whether the dynamic downscaling approach based on the WRFARW
model is really effective, the following assessments are also made for rainfall
forecasts from the IFS model. Rainfall forecast values from IFS and
WRFARW models are interpolated to the location of the monitoring station
in the study area according to the nearest interpolation method (the rain
forecast value at the station will be the projected valuem, showing rainfall at
the nearest grid point to the station). In the study basin, a total of 12 rainfall
observed stations and rainfall results are compared with these stations.
The accumulated 24h rainfall and threshold values of 50mm in 24h and
100mm in 24h rain were used to evaluate forecasting skills between IFS and
WRF models. The 24, 48, and 72 h forecast durations are put into evaluation,
in which the accumulated rainfall from 0 - 24 h is called the first day of rain
12


forecast, from 24 - 48 h is called the second day and 48 - 72 h is called the
3rd day, ... Specifically, to evaluate rainfall prediction skills, the average
errors (ME), absolute errors (MAE) and the square errors (RMSE) were used
to assess.
To assess the skills to forecast heavy and very large rainfall, indicators of
BIAS, POD, FAR and ETS are also used. The BIAS indicates whether the
bias is high (BIAS> 1) or low (BIAS <1). The POD and FAR indicators,
respectively, indicate the correct forecasting rate of the occurrence (the
closer the POD is, the better) and the short forecast (the closer the FAR is,
the better). The ETS index is a composite index that indicates the entire phase
prediction skills and covers aspects of the BIAS, POD and FAR indicators,
the closer the ETS is to 1, the better.
Tab. 3.4 Results of evaluation and comparison of rainfall forecast quality
for IFS model (14km) and WRF model (5km)

IFS Model (14x14) km
WRF Model (5x5)km
Rainfall
ME
MAE
RMSE
ME
MAE
RMSE
forecast (mm/24h) (mm/24h) (mm/24h) (mm/24h) (mm/24h) (mm/24h)
Day 1st
Day 2nd
Day 3th

- 12
-19
-22

22
32
41

29
46
52

-8
-12
-16

16
21
30

20
25
32

Similar to table 3.4, tables 3.5 and 3.6 provide the calculated results of BIAS,
POD, FAR and ETS respectively for heavy rain thresholds. From this result,
it can be seen that the trend of low bias forecasting occurs in both IFS and
WRF but WRF still has better prediction skills than IFS, especially at the
threshold of heavy rain. Regarding the skill of correctly forecasting the
occurrence of rain (through POD), it is clear that the WRF model has better
skills than IFS, especially at the 48 and 72h forecast limits at both heavy rain
thresholds. Similarly, the rate of short prediction (FAR) is also significantly
reduced after dimming the dynamic scale by WRF model. In general, from
the results of the evaluation and comparison on tables 3.3 and 3.4, it is easy
13


to see that the WRF model has a better quality of rainfall forecast and very
large than the IFS model (see ETS index).
Tab. 3.5 Results of evaluating and comparing the rain forecast quality for IFS
model (14km) and WRF model (5km) with heavy rain threshold (> 50mm /24h)
IFS Model (14x14) km
WRF Model (5x5)km
Rainfall
forecast BIAS POD FAR ETS BIAS POD FAR ETS
Day 1st

Day 2nd

0.33
0.15

0.42
0.28

0.32
0.49

0.21
0.10

0.65
0.38

0.75
0.56

0.25
0.33

0.34
0.28

Day 3th

-0.08

0.13

0.62

0.09

0.24

0.34

0.42

0.19

Tab. 3.6 Results of evaluation and comparison of rain forecast quality for IFS model
(14km) and WRF model (5km) with very heavy rain threshold (> 100mm /24h)
Rainfall
IFS Model (14x14) km
WRF Model (5x5)km
forecast BIAS POD FAR ETS BIAS POD FAR ETS
Day 1st
Day 2nd

0.21
-0.09

0.36
0.22

0.46

0.59

0.13
0.08

0.46
0.30

0.45
0.38

0.22
0.41

0.28
0.20

Day 3th

-0.24

0.09

0.72

0.02

0.21

0.20

0.48

0.15

3.1.3 Correct and compute the forecasted rainfall value
The steps for making corrections as well as methods for re-calculating the
forecasted rainfall values are presented in chapter 2. Based on the forecasted
results and cumulative rainfall at monitoring stations, the thesis has
developed a regression equation for each forecast period. With the forecasted
time from 0 to 240 hours, equivalent to 40 periods of cumulative rainfall of
6 hours is adjusted. Thus, for each prediction period, linear regression
equations are constructed separately.
3.1.3.1. Assessing rainfall forecast results that has not yet adjusted the
statistics
In Figure 3.3 presents the calculated results of ME (mm) between forecast
rainfall and actual observation data at some stations in the study area. It is
found that the forecast rainfall value from models has a tendency to lower
than that in observed cases with the popular ME index from -4.85mm to
about 0mm. In particular, the forecast error is most obvious at An Khe and
14


Buon Ho stations, with the popular ME index less than -2mm, especially at
An Khe station. In contrast, the predicted error is rather small at Kon Tum
and Pleiku stations, with the common ME index ranging from -1mm to less
than 1mm. In addition, the forecast results also show that the forecast error
is larger in the short prediction period and for the longer the forecast period,
it seems the better quality.

Fig. 3.17 ME index of accumulated rainfall
forecast for 6 hours with observed data (mm)

Fig. 3.18 Correlation coefficient (FA)
between forecast rainfall and observed data

In Figure 3.18, correlation coefficients between rainfall forecast and
observed data are presented. The results showed that the correlation
coefficients were positive in most cases. This shows that the forecast results
partly indicate the actual trend of rainfall. However, the correlation
coefficient between forecast and observed rainfall is quite low. In particular,
the highest correlation coefficient for the forecast period of less than and
equal to 78 hours is about value of 0.2. In contrast, when forecast time of
greater than 78h, the correlation coefficient is very low, around value of 0.1.
From the above analysis results, it can be found that the forecast models get
fit in term of time series with observed ones, but having large difference in
peak value of rainfall when forecast duration is less than 78 hours.
Conversely, the model does not forecast well when forecast time is increased
with greter than 78 hrs (from 78 to 240 hours) in term of time series, but
lower errors in term of peak value.
3.1.3.2. Assessment of forecast quality after statistical adjustment
Based on the univariate linear regression method, the thesis has adjusted
rainfall for each forecasting period. Rainfall forecast value is adjusted based
on the coefficients "a" and "b" of the regression equation.
15


Corrected rainfall forecast results get rather low errors, with the common ME
index ranging from -1mm to about 0 mm. In general, the adjusted rainfall
forecast results tend to be lower than the actual observed data. In which the

forecast error is largest at stations An Khe and Buon Ho. However, in
comparison with uncorrected forecast rainfall, the revised forecast results
have significantly lower errors.
The value of the correlation coefficient between forecast rainfall after being
corrected with observed data shows that after adjustment, rainfall forecast is
positively correlated with observed data in all forecasting cases. It means that
the forecast rainfall after adjustment reflect the good trend in comparison
with actual observed data at stations. The most significant high correlation
coefficient of above 0.4 is found when forecast time is less than 78 hours.
When forecast time is greater than 78 hours, the correlation coefficient is
gradually reduced, getting value of about 0.2. So it can be said that statistical
adjustment helps to improve forecast quality with rainfall series is more
consistent with reality.
3.1.3.3. Forecast rainfall field for serving calculation
The forecast rainfall results after corrected by linear regression are
significantly improved in term of peak errors and good fit in term of time
series. It means that the forecast error of corrected rainfall is quite low, with
the common ME index from -1mm to 0mm. With the forecast time less than
78h, the correlation coefficient of the forecast has been corrected with
observation is quite good with value of bigger than 0.4, some cases up to 0.8
(at Pleiku station at the time of forecast 18-24h and 42-48h). However, with
a longer forecast period (over 78 hours), although the correlation coefficient
has improved, but the value is still quite low, common in 0.2.
Rainfall data after adjustment are used for further flood simulation. In the study,
it shows that in case of forecast time of less than 78 hrs the forecast time series
value is a better fit with the actual data. For hydrological simulation, coefficients
a and b have been spatially interpolated for the study sub-basins.

16

Tab. 3.7 The results of calculating the dependent coefficient (a) and
freedom (b) in the regression equation for predictive products by IFS model
Value coefficients a and b of linear regression equation
Forecast
An Khe
Buon Ho
Kon tum
Pleiku
period
00-06h
06-12h
12-18h
18-24h
24-30h
30-36h
36-42h
42-48h
48-54h
54-60h
:
234240h

a
1.02
1.89
1.26
1.13
2.05
1.09

0.53
1.24
2.64
1

:

b
4.01
0.82
4.03
3.04
1.52
2.14
4.73
2.98
0.6
2.24

:
0.96

3.06

a
0.05
0.28
0.95
-0.11
0.27

0.25
0.5
0.31
0.16
-0.04
:
0.07

b
1.86
1.38
4.57
2.75
1.46
1.51
5.31
2.1
1.67
1.84

:

a
1.65
0.49
0.01
0.23
0.14
1.09
0.43

0.15
0.1
0.1

:
0.06

b
1.18
1.9
2.01
0.12
1.94
1.08
1.29
0.19
1.91
2.12

:
0

a
0.82
0.96
0.55
1.33
0.43
1.07
0.19

0.81
0.21
-0.13

:
0

b
0.61
0.68
0.92
-0.29
1.11
0.39
0.79
-0.27
1.28
1.75

:
0

0.01

Tab. 3.8 The results of calculating dependency coefficients (a) and freedom
(b) in the regression equation for forecast products by WRF model
Value coefficients a and b of linear regression equation
Forecast
An Khe
An Khe

period
00-06h
06-12h
12-18h
18-24h
24-30h
30-36h
36-42h
42-48h
54-60h

a
2.08
1.55
0.01
0.52
0.7
0.28
1.87
1.11
0.98

b
0.4
-0.21
4.18
5.05
2.72
3.25
1.27

2.98
0.35

a
0.08
0
0.51
-0.12
-0.24
0.01
0.42
0.31
-0.21

1.91
2.58
6.14
3.89
2.73
3.02
5.99
3.15
2.84

17

a
0.47
0.85
0.31

-0.02
1.56
0.2
0.44
0.16
0.53

1.75
0.77
2.07
0.54
1.28
2.43
2.36
0.73
1.88

a
0.49
0.31
0.53
0.11
-0.11
0.07
0.27
0.11
0.28

0.61
0.94

0.92
0.87
2.44
1.84
1.29
1.16
1.19


Fig. 3.19 The ME index of the accumulated
rainfall forecast for 6 hours has been
corrected with observed data (mm).

Fig. 3.20 The correlation coefficient
between rainfall forecast has been adjusted
statistically with observed data

Thus, based on the method of calculating, evaluating and adjusting the
rainfall forecast value by two models of IFS and WRF, it can say that the
forecast rainfall value after adjustment is fitted with observed value. This
predicted rainfall value will be used as an input in the hydrological
forecasting models in the Ba River basin.
3.2. Flow forecast results in Ba river basin
3.2.1. Calibrating and verifying flood forecasting models
As presented in Chapter 2, the measured rainfall data in the past with
different flood peak levels is selected to calibrate and verify and find the
model's parameter set for both NAM MIKE and HEC-HMS models for each
sub-basin (An Khe, Ayun Ha, Krong Hnang and Hinh river).
Tab. 3.9 Table of Results of Mike NAM and HEC-HMS model parameters
for An Khe and Ayun Ha sub-basins

18


Tab. 3.10 Results of MIKE NAM and HEC-HMS model parameters for
Krong Hnang sub-basin and Hinh river

3.2.2. Flow forecast results to the reservoirs in the Ba River basin
3.3.2.1. Setting forecast rain data as input for the hydrological models
Forecasted rainfall values according to IFS and WRF models were
determined in content 3.2. The predicted rainfall value has also been adjusted
to match the actual figures and ensure reliability. Each grid cell of the studied
sub-basins has been identified for both (5x5) km and (14 x 14) km types as
shown in Figures 3.45 - 3.46 and Tables 3.19 - 3.20.

Fig. 3.46 Location of grid cells
with the size of 14x14km

Fig. 3.45 Locations of grid cells of
size (5x5) km

The amount of precipitation at the specific grid cells for each 04 basins of
reservoirs will be detailed and converted to calculate the forecasted rainfall
values as tables 23 and 3.24.

19


Tab. 3.19 Grid diagram of resolution
(5x5) km of An Khe sub-basin

Tab. 3.20 Grid diagram resolution
(5x5) km Ayun Ha basin.

+ Weight of grid cells:
Tab. 3.23 Weighted values for grid cells (5x5) km in An Khe sub-basin
Cell

M10

N2

N3

N4

N5

N6

N7

N8

N9

N10

N11

W

0.004

0.007

0.012

0.012

0.017

0.020

0.012

0.004

0.013

0.027

0.008

Tab. 3.24 Weighted values for grid cells (14x14) km in An Khe sub-basin
Cell

E1

E3

G4

G3

E4

E2

F1

F3

F5

F4

F2

W

0.007

0.035

0.161

0.063

0.052

0.039

0.007

0.244

0.001

0.252

0.139

3.3.2.2. Flood forecast results for the Ba River basin
The expected rainfall with an estimated time of 6 hours, used to forecast
floods in the sub-basins are: 13/11/2013 06:00 to 22/11/2013 12:00; 27 / XI
/ 2014 06:00 to 05/11/2014 06:00; 07 / X / 2015 06:00 to 19 / X / 2015 00:00;
30 / X / 2016 06:00 to 11/11/2016 06:00.
Calculated results by Mike-NAM model for 2013, 2015, 2015 and 2016 based
on rainfall forecast data with grid cell value of (5x5) km and (14x14) km

Fig 3.48 forecast flood process to
An Khe Lake XI / 2014

Fig 3.47 forecast flood process to
An Khe Lake XI / 2013
20


Fig 3.49 forecast flood process to

An Khe Lake X / 2015

Fig 3.50 forecast flood process to
An Khe Lake X / 2016

Results calculated by HEC-HMS model for 2013, 2015, 2015 and 2016
according to rainfall forecast data (5x5) km and (14x14) km

Fig 3.51 forecast flood process to
An Khe Lake XI / 2013

Fig 3.52 forecast flood process to
An Khe Lake XI / 2014

Fig 3.53 forecast flood process to
An Khe Lake X / 2015

Fig 3.54 forecast flood process to
An Khe Lake X / 2016

+ Evaluation of flood forecast errors for the two models MIKE NAM and
HEC-HMS is shown in Table 3.31 and Table 3.32.
21


Tab. 3.31 Criteria for evaluating flood forecasting quality according to
MIKE NAM model for 4 sub-basins on Ba river

Tab. 3.32 Criteria for evaluating flood forecast quality according to HECHMS model for 4 sub-basins on Ba river

The medium term flood forecast by MIKE NAM model based on the rainfall
forecast in grid size of (5x5) km and (14x14) km has good results with
satisfactory level and small difference.
Similarly, the results of medium-term flood forecasting by HEC-HMS model
based on rainfall forecasted in grid size of (5x5) km and (14x14) km have
good results. Although the guarantee level is not met, the index is also
approximately satisfactory level and small error.

22


CONCLUSION OF CHAPTER 3
The rainfall forecast results, corrected by linear regression, have significantly
improved errors and fitted in time series values compared to uncalibrated
results. In particular, the error of corrected forecast rainfall is quite low, with
the common ME index from -1mm to 0 mm. With the forecast time less than
78h, the correlation coefficient between simulated and observed data is quite
good, commonly on 0.4, some cases up to 0.8. However, when forecast
period is greater than 78 hours, although the correlation coefficient has
improved, but the value is still quite low, common around 0.2.
Corrected rainfall data will be used for further flood simulation. In particular,
the forecast results with the time less than 78h have a better fit on the trend
of the actual data. For hydrological calculations, the coefficients a and b have
been spatially interpolated for the 4 stations namely An Khe, Yaun Ha,
Krong H’Nang and Hinh river sub-basins.
On the basis of the optimal parameter set, combined with the forecasted rain
correction results under the two grid size options of 5km and 14km. The
results show that the ability to apply both MIKE NAM and HEC-HMS
models to forecast moderate rainfall in 2 grid cell sizes is quite good
according to Nash criteria and flood peak error.

Proposal to use flood forecasting model as follows: Using rain forecast
model with grid size (5x5) km as input data for flood forecasting model will
give better results than rain forecast data grid size (14x14) km. Applying
NAM MIKE model to forecast flood will be best for Yaun Ha and An Khe
sub-basins.
CONCLUSIONS
The main contents conducted in the study:
Overview on the study situation of rainfall and flood forecasting in the world
as well as in Vietnam.
Correction and construction of quantitative forecast rain field after model.
Establishing a high-resolution quantitatively rainfall forecast problem in the
Ba river basin by using the IFS and WRFARW meteorological models with
23