Impact of pre simulation warm up period length and other factors on meteorology model output

46 Figure 4.1 Comparison trend differences with observation between 11day_warm-up and no_warm-up period of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Vietnam domain.... 51 Figure 4

Dissertation for Degree of Doctor

Supervisor: Prof Young Sunwoo

Impact of Pre-simulation Warm-up Period Length and other Factors on

Meteorology Model Output

Submitted by Nguyen Thi Huynh Tram

February, 2019

Department of Advanced Technology Fusion Graduate School of Konkuk University

Impact of Pre-simulation Warm-up Period Length and other Factors on

Meteorology Model Output

A Dissertation submitted to the Department of Advanced Technology Fusion

and the Graduate School of Konkuk University

in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Engineering

Submitted by Nguyen Thi Huynh Tram

November, 2018

This certifies that the Dissertation of Nguyen Thi Huynh Tram is approved

Approved by Examination Committee

TABLE OF CONTENTS

List of Tables iii

List of Figures iv

Abstract xi

Chapter 1 Introduction 1

1.1 Background and Objectives 1

1.2 Scope of Research 4

Chapter 2 Literature Review 6

2.1 Horizontal Grid Spacing Effects 6

2.2 Nested domain 7

2.3 Modeling interval of forecasting period 8

2.4 Different combinations of physics schemes 9

2.5 Simulated variables 10

Chapter 3 Methodology 13

3.1 Operational Method 13

3.2 Configuration for time-varying and multi-physics ensembles 15

3.2.1 Research Regions 16

3.2.2 Research Time 18

3.2.3 Physical parameterizations 20

3.3 Arakawa-C grid staggering 44

3.4 Third-order Runge-Kutta (RK3) Integration Scheme 46

3.5 Observation data 49

3.6 Verification Method 49

3.6.1 Mean Absolute Error(MAE) 49

3.6.2 Subtraction between two MAE (SMAE) 50

Chapter 4 Results and Discussion 51

4.1 Notations on Figure 51

4.2 Sensitivity of the accuracy of simulation temperature to physical scheme changes under different pre-simulation warm-up period length 52

4.2.1 Comparison trend differences with observation at Vietnam domain 53 4.2.2 Comparison trend differences with observation at Korea domain 93

4.3 The accuracy from choosing value of domain 102

4.3.1 Vietnam domain 102

4.3.2 Korea domain 103

4.4 The best-performing configuration 105

4.4.1 Physical configuration for Vietnam domain 105

4.4.2 Physical configuration for Korea domain 116

Chapter 5 Conclusions and Engineering Significance 125

5.1 Conclusions 125

5.2 Engineering Significance 127

References 128

List of Abbreviations 139

Abstract (in Korean) 141

List of Tables

Table 3.1 Configuration of time-varying and multi-physics ensembles for Ho Chi

Minh City, Vietnam Domain 15

Table 3.2 Configuration of time-varying and multi-physics ensembles for Seoul, Korea 16

Table 3.3 Research period of this study 18

Table 3.4 Cumulus schemes 22

Table 3.5 Microphysics schemes 22

Table 3.6 Longwave Radiation Schemes 23

Table 3.7 Shortwave Radiation Schemes 23

Table 3.8 Surface Layer Schemes 24

Table 3.9 Planetary Boundary Layer Schemes 24

Table 3.10 Land Surface Schemes 24

Table 3.11 120 combinations of physical options is applied for Vietnam 26

Table 3.12 80 combinations of physical options is applied for Seoul, Korea 37

List of Figures

Figure 1.1 Global ranking of risk factors by total number of deaths from all causes

for all ages and both sexes in 2016 (Source: Health Effects Institute 2018) 2

Figure 1.2 Process of WRF and CMAQ (Skamarock et al 2008) 3

Figure 1.3 Schematic diagrams for the research method for this study 5

Figure 3.1 Processing of study experiments 14

Figure 3.2 The graphical user interface of WRF Portal 14

Figure 3.3 Location of the domain 1(D1), domain 2(D2) and domain 3(D3) with the center point is Tan Son Hoa Staion (106.36, 10.79) 17

Figure 3.4 Seoul Metropolitan Government Office (37.57, 126.96), Seoul, Korea 17

Figure 3.5 Schematic of physics and their interactions 20

Figure 3.6 Illustrations of Microphysics Processes (Dudhia, 2014) 21

Figure 3.7 Arakawa-C grid staggering (Skamarock, 2008) 45

Figure 3.8 Nest grid integration sequence (Skamarock, 2008) 46

Figure 4.1 Comparison trend differences with observation between 11day_warm-up and no_warm-up period of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Vietnam domain 51

Figure 4.2 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m1_l1_s1_sf_su2_bl1 for Vietnam domain 54

Figure 4.3 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m1_l1_s1_sf_su2_bl1 for Vietnam domain 55

Figure 4.4 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m1_l1_s1_sf_su2_bl1 for Vietnam domain 56 Figure 4.5 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m2_l1_s1_sf1_su2_bl1 for Vietnam

domain 57

Figure 4.6 Comparison trend differences with observation between 11day_warm-up

and 1day_warm-up of 10 cumulus schemes for m2_l1_s1_sf1_su2_bl1 for Vietnam domain 58 Figure 4.7 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m2_l1_s1_sf1_su2_bl1 for Vietnam domain 59 Figure 4.8 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m3_l1_s1_sf1_su2_bl1 for Vietnam domain 60 Figure 4.9 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m3_l1_s1_sf1_su2_bl1 for Vietnam domain 61 Figure 4.10 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m3_l1_s1_sf1_su2_bl1 for Vietnam domain 62 Figure 4.11 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m4_l1_s1_sf1_su2_bl1 for Vietnam domain 63 Figure 4.12 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m4_l1_s1_sf1_su2_bl1 for Vietnam domain 64 Figure 4.13 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m4_l1_s1_sf1_su2_bl1 for Vietnam domain 65

Figure 4.14 Comparison trend differences with observation between 1day_warm-up

and no_warm-up of 10 cumulus schemes for m5_l1_s1_sf1_su2_bl1 for Vietnam domain 66

Figure 4.15 Comparison trend differences with observation between 11day_warm-up

and 1day_warm-up of 10 cumulus schemes for m5_l1_s1_sf1_su2_bl1 for Vietnam domain 67

Figure 4.16 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m5_l1_s1_sf1_su2_bl1 for Vietnam domain 68 Figure 4.17 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m14_l1_s1_sf1_su2_bl1 for Vietnam domain 69 Figure 4.18 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m14_l1_s1_sf1_su2_bl1 for Vietnam domain 70 Figure 4.19 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m14_l1_s1_sf1_su2_bl1 for Vietnam domain 71 Figure 4.20 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf1_su2_bl1 for Vietnam domain 72 Figure 4.21 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l1_s1_sf1_su2_bl1 for Vietnam domain 73 Figure 4.22 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf1_su2_bl1 for Vietnam domain 74

Figure 4.23 Comparison trend differences with observation between 1day_warm-up

and no_warm-up of 10 cumulus schemes for m16_l1_s2_sf1_su2_bl12 for Vietnam domain 75 Figure 4.24 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l1_s2_sf1_su2_bl12 for Vietnam domain 76 Figure 4.25 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s2_sf1_su2_bl12 for Vietnam domain 77 Figure 4.26 Comparison trend differences with observation between 1day_warm-up

and no_warm-up of 10 cumulus schemes for m16_l1_s5_sf1_su2_bl12 for Vietnam domain 78 Figure 4.27 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l1_s5_sf1_su2_bl12 for Vietnam domain 79

Figure 4.28 Comparison trend differences with observation between 11day_warm-up

and no_warm-up of 10 cumulus schemes for m16_l1_s5_sf1_su2_bl12 for Vietnam domain 80 Figure 4.29 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Vietnam domain 81 Figure 4.30 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Vietnam domain 82 Figure 4.31 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Vietnam domain 83 Figure 4.32 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m16_l3_s3_sf7_su7_bl7 for Vietnam domain 84 Figure 4.33 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l3_s3_sf7_su7_bl7 for Vietnam domain 85 Figure 4.34 Comparison trend differences with observation between 11day_warm-up and no_warm-up of 10 cumulus schemes for m16_l3_s3_sf7_su7_bl7 for Vietnam domain 86 Figure 4.35 Comparison trend differences with observation between 1day_warm-up and no_warm-up of 10 cumulus schemes for m16_l5_s5_sf7_su7_bl7 for Vietnam domain 87 Figure 4.36 Comparison trend differences with observation between 11day_warm-up and 1day_warm-up of 10 cumulus schemes for m16_l5_s5_sf7_su7_bl7 for Vietnam

domain 88

Figure 4.37 Comparison trend differences with observation between

11day_warm-up and no_warm-11day_warm-up of 10 cumulus schemes for m16_l5_s5_sf7_su7_bl7 for Vietnam domain 89

Figure 4.38 The whole picture of comparison trend differences with observation

between 11day_warm-up and no_warm-up of 120 multi-physics ensembles for Vietnam domain 92 Figure 4.39 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf7_su7_bl7 for Korea domain 93 Figure 4.40 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l3_s3_sf7_su7_bl7 for Korea domain 94 Figure 4.41 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l4_s4_sf7_su7_bl7 for Korea domain 95 Figure 4.42 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l5_s5_sf7_su7_bl7 for Korea domain 96 Figure 4.43 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l1_s1_sf1_su2_bl12 for Korea domain 97 Figure 4.44 Comparison trend differences with observation between 5day_warm-up and no_warm-up of 10 cumulus schemes for m16_l3_s3_sf1_su2_bl12 for Korea domain 98 Figure 4.45 Comparison trend differences with observation between 5day_warm-up

and no_warm-up of 10 cumulus schemes for m16_l4_s4_sf1_su2_bl12 for Korea

domain 99 Figure 4.46 Comparison trend differences with observation between 5day_warm-up

and no_warm-up of 10 cumulus schemes for m16_l5_s5_sf1_su2_bl12 for Korea

domain 100

Figure 4.47 The whole picture of MAE(no_warm-up)of 2-m temperature (C) forecast at Tan Son Hoa station, Ho Chi Minh City, Viet Nam, February 2, 2011 102 Figure 4.48 The whole picture of MAE(no_warm-up) of 2-m temperature (0C) forecast at Seoul Metropolitan Government Office, Seoul, Korea, August 10, 2016 103

Figure 4.49 A trend MAE of 10 cumulus schemes with configuration

m16_l1_s1_sf7_su7_bl7 of no_warm-up period for Vietnam domain 105

Figure 4.50 A trend MAE of 10 cumulus schemes with configuration

m16_l3_s3_sf7_su7_bl7 of no_warm-up period for Vietnam domain 106

Figure 4.51 A trend MAE of 10 cumulus schemes with configuration

m16_l5_s5_sf7_su7_bl7 of no_warm-up period for Vietnam domain 107

Figure 4.52 A trend MAE of 10 cumulus schemes with configuration

m1_l1_s1_sf1_su2_bll of no_warm-up period for Vietnam domain 108

Figure 4.53 A trend MAE of 10 cumulus schemes with configuration

m2_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 109

Figure 4.54 A trend MAE of 10 cumulus schemes with configuration

m3_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 110

Figure 4.55 A trend MAE of 10 cumulus schemes with configuration

m4_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 111

Figure 4.56 A trend MAE of 10 cumulus schemes with configuration

m5_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 112

Figure 4.57 A trend MAE of 10 cumulus schemes with configuration

m14_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 113

Figure 4.58 A trend MAE of 10 cumulus schemes with configuration

m16_l1_s1_sf1_su2_bl of no_warm-up period for Vietnam domain 114

Figure 4.59 A trend MAE of 10 cumulus schemes with configuration

m16_l1_s2_sf1_su2_bl of no_warm-up period for Vietnam domain 115

Figure 4.60 A trend MAE of 10 cumulus schemes with configuration

m16_l1_s5_sf1_su2_bl of no_warm-up period for Vietnam domain 116 Figure 4.61 A trend MAE of 10 cumulus schemes with configuration m16_l1_s1_sf7_su7_bl7 of no_warm-up period for Korea domain 117

Figure 4.62 A trend MAE of 10 cumulus schemes with configuration m16_l3_s1_sf7_su7_bl7 of no_warm-up period for Korea domain 118 Figure 4.63 A trend MAE of 10 cumulus schemes with configuration m16_l4_s4_sf7_su7_bl7 of no_warm-up period for Korea domain 119 Figure 4.64 A trend MAE of 10 cumulus schemes with configuration m16_l5_s5_sf7_su7_bl7 of no_warm-up period for Korea domain 120 Figure 4.65 A trend MAE of 10 cumulus schemes with configuration m16_l1_s1_sf1_su2_bl12 of no_warm-up period for Korea domain 121 Figure 4.66 A trend MAE of 10 cumulus schemes with configuration m16_l3_s3_sf1_su2_bl12 of no_warm-up period for Korea domain 122 Figure 4.67 A trend MAE of 10 cumulus schemes with configuration m16_l4_s4_sf1_su2_bl12 of no_warm-up period for Korea domain 123 Figure 4.68 A trend MAE of 10 cumulus schemes with configuration m16_l5_s5_sf1_su2_bl12 of no_warm-up period for Korea domain 124

This study makes an important contribution by improving the accuracy of meteorological model output which can be very important as input for subsequent air pollution modeling 360 time-varying and multi-physics ensembles of 120 different physics configurations are broadly classified into 3 different simulation timescales for Vietnam 160 time-varying and multi-physics ensembles of 80 different physics configurations are broadly classified into 2 different simulation timescales for Korea These ensembles were applied to investigate the detailed impact of pre-simulation warm-up period length and other factors on Weather Research and Forecasting Model output Our results show that increasing the pre-simulation warm-up period length causes bigger errors instead of improving the accuracy of the meteorology model output

A comparison of accuracy between parent domain and two nested domain outputs are investigated Based on conditions of pre-simulation warm-up period length, the present study contributes additional evidence that the value of parent domain (D1) will give more accurate results than the nested domains in case of using two-way nesting strategies

Together these two conclusions provide important insights into improving the accuracy of model output The accuracy of surface temperature forecast is improved

by 30C for the case in this study Temperature bias of the parent domain ranges from

Keywords: Weather Research and Forecasting Model (WRF), Warm-up Period length, Physics configurations.

Chapter 1 Introduction

1.1 Background and Objectives

This introductory section provides a brief overview of meteorology model Before proceeding to examine meteorology model, it will be necessary to go on to dry air Dry air in Earth’s atmosphere plays an important role in the maintenance life of ecosystem This system includes people, plants, animal, microorganism, non-living ingredients such as soil, water, etc When all animals include humans respire, they breathe in oxygen (O2), occupy 20.95% in dry air, to survive O2 is used to break carbohydrates down into carbon dioxide (CO2), water (H2O) and energy Plant uses the energy of sunlight to convert CO2 and H2O into carbohydrates and O2 by photosynthesis process Dry air also include 78.08% Nitrogen (N2), 0.93% Argon, 0.0039% CO2 and other gases such as Neon (Ne), Helium(He), Hydrogen(H2), Xenon (Xe), etc…(C Donald Ahrens(2011))

It is important to know that 90% of the atmosphere’s total volume present at the lowest layer of the Earth, troposphere This layer has a height of 7 to 20 km above sea level It is the vital place not only humans live but also all the weather go on If dry airs reach toxic levels in atmosphere, it can make threat the health of people and animals It also ruins vegetation and structures In this case, air is polluted or commonly named air pollution

It has been reported that air pollutants derive from not only natural sources but also human activities Natural sources take account of wind, volcanoes, forest fires, and so on Volcanoes and forest fires produce ash, dust, smoke, etc Wind brings dust, soot aloft or earth’s surface In addition, human activities are the main cause of air pollution, which comes from not only fixed sources but also mobile sources Fixed sources consist of industrial areas, power plants, buildings, etc Mobile sources comprise motor vehicles, ships, and jet aircraft

Air pollutants are divided into two main categories: primary and secondary air

pollutants are formed by a chemical reaction between a primary pollutant and water vapor or another pollutant

Figure 1.1 Global ranking of risk factors by total number of deaths from all causes for all ages and both sexes in 2016 (Source: Health Effects Institute 2018)

Health Effects Institute (2018) lists ambient particulate matter (PM2.5), household air pollution, and ozone as the major risk factors of deaths In 2016, PM2.5 is generally seen as a factor strongly related to 4.1 million deaths that result from chronic lung disease, lung cancer, stroke, heart disease and respiratory infections Household air pollution can lead to 2.6 million deaths 234.000 deaths caused by chronic lung disease stem from ozone

In order to prevent air pollution problems, numerous researches have established And forecasting of air pollution is one of the effective pollution control measures method Recent developments in methods of air pollution forecasting have heightened two main methods Statistical forecast methods used statistical method to

However, a major problem with this kind of application is without knowing the mechanism of physical, chemical, or biological processes Since the 1990s, there has been an increasing interest in all of the atmospheric problems, numerical forecasting method is introduced This method use computers to simulate atmospheric physics and chemical processes Comprehensive Air Quality Model (CMAQ), a third generation air quality model, is an example of numerical forecasting method

Figure 1.2 Process of WRF and CMAQ (Skamarock et al 2008)

Moreover, air quality model are commonly provided meteorological input from meteorological model Meteorological model can also be combined with other model to simulate For example, Weather Research Forecasting Model (WRF), a meso-scale numerical weather prediction system, is designed not only to serve operational forecasting needs but also atmospheric research and dynamical downscaling of Global Climate Models (GCMs)

WRF has been developed and continuous improved by numerous of academic, research, and government organizations WRF is widely applied by both research and weather forecasting station WRF is available for free download WRF support many physics schemes, data assimilation approaches and a nesting technical for simulations from tens of meters to thousands of kilometers Therefore, WRF is the ideal model for studying different approaches WRF provide the meteorological input fields for a lot of air quality models such as CMAQ, CALPUFF, CALMET, etc… WRF can be combined with Chem model to forecast the concentration of

al, 2011) Cartellea et al (2016) established a system of WRF, CALMET and CALPUFF model to forecast environment odor Manar et al (2017) used WRF-ADMS to model atmospheric stability and turbulence with Gaussian dispersion model Abdul-Wahab et al (2011) coupled the CALPUFF and WRF to forecast a dispersion patterns and concentration of SO2

Therefore, WRF has a significant impact on the field of both meteorological model and forecasting of atmospheric The accurate of WRF has always been of interest to WRF users

The term ―time-varying and multi-physics ensembles‖ refers to a running WRF with different physic configuration over vary of period In this study, 520 time-varying and multi-physics ensembles were applied for Vietnam and Korea in order

to evaluate the impact of pre-simulation warm-up period length on the WRF output

of next evaluated day

The goal of this study is to investigate

1) Impact of pre-simulation warm-up period length on temperature output 2) Impact of pre-simulation warm-up period length on the chosen domain 3) Sensitivity of the ensembles and suggest the best-performing configuration for Vietnam and Korea

1.2 Scope of Research

The main objective of this thesis is to apply, test and evaluate the results of Weather Research Forecasting models, WRF 3.6 is investigated to find some factor that have effect to output In comparing model results with observations, this study assumes that observed values are accurate Effects of instrumental error and errors

of representativeness, the observational errors are horizontally uncorrelated

The variable compared in this study was surface temperature at 2m There are some reasons for this choice Firstly, air temperature is undoubtedly the most frequently checked element of publicly available weather forecasts Temperature is the key

the accuracy of electricity demand forecast Secondly, WRF directly output temperature at 2-m Thirdly, the main factor that cause variations in temperature are latitude, land and water distribution, ocean currents and elevation But temperature

is a main factor that causes variations in wind, humidity v.v (C Donald Ahrens, 2011) Therefore this study is firstly focus on a question how to increase the simulate temperature accurate The region studies are two megacities, Ho Chi Minh City, Vietnam and Seoul, Korea, which face air pollution challenges The accuracy

of this result is applied to CMAQ to model air quality of these regions

Figure 1.3 Schematic diagrams for the research method for this study

Chapter 2 Literature Review

The purpose of this chapter is to review the literature on Weather Research Forecast Model (WRF) WRF support a breadth of possible applications owing to a wide range options for the physics and dynamics of WRF This enables the user have to optimize WRF for specific scales, geographical locations and applications This is lead to an increasingly difficult task when the number of options increases Users have the flexibility to choose between multiple options of modeling interval (start_date to end_date), forecasting period (year, season, month, week, 10 day, etc ), number of domains (more than 1 domain), size of grid and subgrid, different physics parameterizations (e.g land surface, boundary layer, convection, cloud microphysics, radiation schemes)

Moreover, several studies investigating physic options have been carried out on no multi-physics ensembles was found to perform best for all events, variables and metrics(Jason P et al, 2011, Jankov et al 2005) That is the reason why over the past two decade most research in WRF has emphasized how to optimize WRF model in order to enhance the accuracy of WRF model output The following section reviews some studies have investigated the effects of horizontal resolution, nest domain, physical option and its effect on forecast accuracy

2.1 Horizontal Grid Spacing Effects

Horizontal grid spacing remains to be one of the most challenging problems in WRF (Mass et al., 2002; Schwartz et al., 2009; Clark et al.,2012; Abdelwares,2018; Smith et al 2018) A number of studies have examined the effects of horizontal resolution on forecast accuracy

To better understand the effect of horizontal grid spacing of the simulation domain, Abdelwares et al (2018) examines different three tested domains Each experiment has three domains with parent grid ratios of 1:3, 1:4, and 1:5, respectively These nesting ratios are the commonly used for the WRF model

km, 10 km hozirontal grid-spacing and 30 vertical levels The results showed that higher horizontal grid resolution results in a more accurate simulated precipitation and surface temperature Accordingly, it was decided to simulate with the highest resolution as 10 km Mass et al., 2002 hold the same view

Conversely, Smith et al (2018) asserts that no improvement was found with 1,2, 4-km configurations for Great Plains NLLJ modeling NLLJ is the abbreviation of nocturnal low-level jet As Stull (1988) notes ―NLLJ is a thin stream of fast moving air, with maximum wind speeds of 10 to 20m/s usually located 100 to 300 m above the ground‖ Great Plains NLLJs plays an important role in meteorological phenomenon because of its effect on weather and climate Moreover, NLLJ have been identified as major contributing factors for the evolution of deep convection which is responsible for the warm season rainfall in the United States

Similarly, Gibbs et al (2011) holds the view that 4-km horizontal grid spacing outcome near-surface turbulent flow more accurate than 1-km spacing Kain et al (2008), Schwartz et al (2009), and Clark et al (2012) also claim that 2-km horizontal grid spacing did not produced enough convection forecasts value compared with 4-km configuration Additionally, although 4-km grid is too coarse

to occupied convective scale circulations, without using convective parameterization does not produce unrealistic precipitation forecasts (Bryan et al 2003) This is consistent with the findings of Weisman et al (1997) Together, these studies outline that 4-km horizontal grid spacing was recommend in order to decreasing computational burden

2.2 Nested domain

The Weather Research and Forecasting (WRF) support two-way nesting technical

to increasing resolution for region via concurrent grid nesting With grid nesting, information from the coarse domain or outer domain (d1) is interpolated and provided as lateral boundary conditions to first nested domain (d2) and from d2 to innermost domain (d3) Information from the d3 gives feedback to d2 and d2 gives

of 27, 9 and 3km and concluded that in the case of the smallest domain was the strongest variability E.Naabil et al (2017) designed two main with the outer domain with a 25km horizontal resolution Inner domain is at 5 km horizontal resolution This study show that precipitation estimates at the inner domain is different with the outer domain This study claims that the reason may be come from different domain, grid size, the boundary conditions, physical parameterization schemes In this study, the inner domain is run with no cumulus parameterization scheme Therefore, precipitation of outer domain is compared with ―observed‖ precipitation

California Air Resources Board San Joaquin Valley Air Pollution Control District (2016) used domain 03 to drive the CMAQ simulations In an analysis of ensemble data assimilation, Tom, R.D and G.J Hakim (2009) found that outer domain have results equilibrium with the observations By contrast, an error of minimum pressure

in inner domain is up to 70% In short, this research attempts to answer the question: which is domain that user should extract result?‖ This study will answer this question

2.3 Modeling interval of forecasting period

Skamarock et al (2008) suggested a method that user should run WRF with a forecast period because this gives WRF a chance to spin-up The pre-forecast could

pre-be range from 0 to 6 hours In this time, WRF will adjust topography, produce cloud files by hour Unfortunately there is a lack of occurrence on the spin-up time that provides the best results for the simulation

Franz-Georg et al.(2016) analyzed the data from four independent experiments and concluded that 12-h spin-up time of WRF is assumed that provides more realistic atmospheric phase screen (APS) prediction, which is about twice as good for the 12-h spin-up APS prediction as for the 6-h spin-up APS prediction

Kleczek et al (2014) studied the effects of 12, 18 and 24h spin-up on the model result and concludes that longer spin-up time increased the temperature bias but improved the vertical wind-speed profiles and water vapor mixing ratio

Much of the research up to now has been simulated with this large spin-up period Abdelwares et al (2018) simulated for 2 years with 6-month spin-up Although biases in the simulated temperature biases were much less than the precipitation biases but this simulated temperature biases is typically nearly 30C Smith et al (2018) set up 24-h period The first 12 h of each simulation were used to accommodate model spin-up effects Zittis G et al (2014) used the length of each simulation is 2 years with one month of spin-up time Dyer et al (2016) use 24h spin-up The simulations were done from 1999 to 2003, using 1999 as a spin-up period (E.Naabil et al 2017)

In this study, spin –up was therefore defined in term of pre-simulation warm-up period length Evaluated period can be 2 year, one year, month, 10 days One forecasting period can be a part of evaluated period This study suggests that accumulation periods are the same meteorological periods, if NCLL is 6-h accumulation periods The impact of a spin-up time on temperature mitigation has not yet been studied and is the main objective of this study

2.4 Different combinations of physics schemes

In the WRF system, users have to select from many different physics parameterizations (e.g land surface, boundary layer, convection, cloud microphysics, radiation schemes) This type of uncertainties has received much attention over the last years In particular, numerous studies discuss the WRF sensitivity to various parameterized physical processes for different simulation timescales Jason P et al (2011) conclude that no multi-physics ensembles were found to perform best for all events, variables and metrics Mellor-Yamada-Janjic planetary boundary layer scheme and the Betts-Miller-Janjic cumulus scheme were suggested

Smith et al (2018) conducted using three common boundary layer parameterization schemes: the Mellor- Yamada Nakanishi Niino (MYNN), the Yonsei University (YSU), and the Quasi-Normal Scale Elimination (QNSE) schemes

Abdelwares et al (2018) analyzed data from 16 WRF different physics

boundary layer schemes, CAM for the longwave and shortwave radiation, NOAH for the land surface scheme, BMJ for the cumulus scheme, and WSM6 for the cloud microphysics are the optimal configuration This configuration improved the simulated surface air temperature and precipitation by an average of 4% and 47.5% Kleczek et al (2014) evaluated the performance of the planetary boundary layer such

as (PBL) Yonsei University (YSU), VH96, Asymmetric Convective Model version

2 (ACM2), Mellor–Yamada–Janjic (MYJ), Mellor–Yamada–Nakanishi–Niino (MYNN25), Quasi-Normal Scale Elimination (QNSE) and Bougeault–Lacarrere (BOUL) schemes and found that all model simulation have the same structure for the potential temperature which bias is 2K in the upper PLB

Zittis et al (2014) suggested a configuration with the YSU planetary boundary layer scheme, the WSM6 microphysics scheme, KF as cumulus scheme, the NOAH land surface model and CAM for long- and shortwave radiation The differences between his configurations include the planetary boundary layer and the cumulus scheme These schemes have effects on the simulated precipitation compared to other parameterizations The impact of the PBL scheme on temperature is less negligible than microphysics scheme at desert areas

Together, these studies outline that WRF configuration remains to be one of the most challenging problems in numerical modeling of the atmosphere and climate

2.5 Simulated variables

Teisberg et al (2005) showed that the NWS forecast produces a total benefit of

$166 million per year, corresponding to the persistence forecast For 10C improvement in accuracy, the benefit is about $59 million per year Frei (2010) demonstrated that the saving cost of tourism, outdoor events is $362 million per year,

of road transport is $ 78–96 million, of the hydroelectric sector is $ 98 million per year, of the nuclear when the accuracy of weather forecasts are significant

Simulating of precipitation was much more sensitive than surface air temperature

to the change in physics parameterization The most sensitive to changing the

planetary boundary layer schemes and the cumulus parameterization schemes was precipitation And it is also least sensitive when the microphysics scheme changed Previous studies have shown that WRF often under-predicts rainfall Isidora Jankov(2006) 1500x1500 km region centered over the south-central United States

a matrix of 18 WRF model configurations, created using different physical scheme combinations, was run with 12 km grid spacing for 8 International H20 Project (IHOP) MCS cases Skill score measures averaged over all 8 cases for all 18 configurations indicated that no one configuration was obviously best at all times and thresholds The greatest variability in forecasts was found to come from changes

in the choice of convective scheme, although notable impacts also occurred from changes in the microphysics and planetary boundary layer (PBL) schemes Specifically, the forecast of system average rain rate is affected by changes in convective treatment Choices of microphysics and convective treatment influence forecasts of total domain rain volume The impact of interactions (synergy) of different physical schemes, although occasionally of comparable magnitude to the impacts from changing one scheme alone (compared to a control run), varied greatly among cases and over time, and was typically not statistically significant

Abdelwares(2018) investigated the simulated precipitation and surface air temperature These results are compared to observational data to emphasize that changing the physics parameterization can effects on the simulated precipitation fields Precipitation was least sensitive to modifying the microphysics scheme and most sensitive to changing planetary boundary and cumulus scheme Changing the longware radiation is more significant changes than shortwave radiation scheme However, the simulated precipitation is less realistic than the simulated surface air temperature The biases in the simulated precipitation were much more than the surface air temperature biases Zittis G et al (2014) investigated 12 different physics configurations to assess the performance in simulating minimum and maximum surface air temperature, total precipitation The results show that the precipitation is more sensitive to cumulus scheme and difficult to model However, the simulated surface air temperature is most sensitive to microphysics scheme Precipitation is

Generally, it is difficult to model realistically precipitation The performance of the WRF in air temperature fields was more realistic than the simulated precipitation fields

Chapter 3 Methodology

In this study, WRF version 3.6 is used for simulation WRF used non-hydrostatic equation for a limited area This equation has multiple options for multi-physics parameterizations schemes The term ―time-varying and multi-physics ensemble‖ refers to a running WRF with different physic configuration over vary of period This chapter describes and discusses the methods used to investigate the accuracy of time-varying and multi-physics ensembles in this study The first part is presented the process of experiment The second part moves on to describe in greater detail the configuration of time-varying and multi-physics ensembles

The third part is represented Arakawa-C grid staggering, which is employed for the horizontal grid In this study, the operation of Arakawa-C grid staggering was found to cause the accuracy of simulation The fourth part is described the third-order Runge-Kutta scheme This equation is a main reason of explosion the error of WRF output for longer simulation Two next parts are observation data and verification method

3.1 Operational Method

In order to generate temperature time series forecast, this study was conducted in the form of a series of seven primary software components

Figure 3.1 Processing of study experiments

The first software is WRF Portal, which is used for the purpose of creating parameters for namelist.wps file WRF Portal has a number of attractive features such as graphical user interface (GUI) for configuring WRF domain, easy to select

or locate domain on map, create namlist.wps file follow the complicated rules of WPS Therefore, WRF Portal is a useful tool for new user WRF Portal has been developed and will be maintained by the the Advanced Computing Section (ACS) of Earth System Research Laboratory (ESRL), part of NOAA, the National Oceanic and Atmospheric Administration

Figure 3.2 The graphical user interface of WRF Portal

After creating, namelist.wps is commonly used for geogrid.exe, ungrib.exe and metgrid.exe(4) There are three programs of WRF Preprocessing System (WPS), which is used to prepare domains for WRF Geogrid.exe is used to interpolate static geographical data to the modeling domain that draw in WRF Portal Ungrib.exe is used to extracts meteorological fields that are supplied by GRIB formatted files

Metgrid.exe will be merged the meteorological fields extracted by ungrib.exe and the model grids defined by geogrid.exe

Once metgrid.exe is successfully run, obsgrid.exe uses the output files from metgrid.exe as input and incorporate information from observations for the purpose

of improving meteorological analyses

Following the correction for meteorological input, output of obsgrid.exe is feed up

which is used to atmospheric research and operational forecasting needs For this study, WRF was used to simulation of temperature and other meteorological factors These temperatures were then verified with observation data Finally, the conclusion is taken out

3.2 Configuration for time-varying and multi-physics

ensembles

In this study, 480 time-varying and multi-physics ensembles of WRF 3.6 were applied for Vietnam and Korea In this part, the configurations of these ensembles are described below

Table 3.1 Configuration of time-varying and multi-physics ensembles for Ho Chi

Minh City, Vietnam Domain

VietNam

Outer Domain 1 (D1)

First nested Domain 2 (D2)

Innermost Domain 3 (D3)

Location Tan Son Hoa Staion (106.36,10.79) Ho Chi Minh city, Vietnam

Start time

End time

11day_warm-up: 2017/ 1/ 22 ~ 2017/2/ 3 1day_warm-up: 2017/ 1/ 31 ~ 2017/2/ 3 no_warm-up: 2017/ 2/ 1 ~ 2017/2/ 3

Evaluated day: 2/2/2017

Physical option 120 member multi-physics ensemble (Table 3.9)

Initial data NCEP FNL(Final) Operational Global Analysis data

Table 3.2 Configuration of time-varying and multi-physics ensembles for Seoul,

Korea

Korea

Outer Domain 1 (D1)

First nested Domain 2 (D2)

Innermost Domain 3 (D3)

Location Seoul Metropolitan Government Office (37.57, 126.96) , Korea

Physical option 80 member multi-physics ensemble(Table 3.10)

Initial data NCEP FNL(Final) Operational Global Analysis data

3.2.1 Research Regions

In this study, WRF3.6 over two domains is applied to generate temperature time series forecast for this study Atmospheric observations from meteorological towers

at the center of study locations were used to calculate temperature at every hour and

to verify the WRF output

The first selected study location was Ho Chi Minh City, Vietnam and Seoul, Korea The extend of the Ho Chi Minh City, Vietnam is presented in Table 3.1 and Figure 3.3 The meteorological tower is located at Tan Son Hoa Station (106.36, 10.79), Ho Chi Minh City, Vietnam

Figure 3.3 Location of the domain 1(D1), domain 2(D2) and domain 3(D3) with the

center point is Tan Son Hoa Staion (106.36, 10.79)

The second study domain selected is Seoul, Korea is presented in Figure 3.4 and Table 3.2 The meteorological tower is located at Seoul Metropolitan Government Office (37.57, 126.96), Seoul, Korea

Figure 3.4 Seoul Metropolitan Government Office (37.57, 126.96), Seoul, Korea

In this study, WRF was designed for a triple domains using two-way feedback

increasing resolution for region via concurrent grid nesting With grid nesting, information from the coarse domain or outer domain (D1) is interpolated and provided as lateral boundary conditions to first nested domain (D2) and from D2 to innermost domain (D3) Information from the D3 sends feedback to D2 and D2 sends feedback to D1 D1, D2 and D3 interact with each other through two-way strategies

The D01 and D02 grids were used to resolve the larger scale synoptic weather systems, while the D03 grid resolved the finer details of the atmospheric conditions and was used to drive the air quality model simulations

Predicted values are compared to measured data from meteorology stations through the domain 1 Then, model ensemble with MAE in predicting the selected parameters and with simulation time is identified

3.2.2 Research Time

In order to evaluate the impact of pre-simulation warm-up period length to WRF output of next evaluated day, this study is designed by 3 different simulation timescales for Ho Chi Minh city, Vietnam and 2 different simulation timescales for Seoul, Korea

Table 3.3 Research period of this study

(day)

Pre-simulation Warm-up period (day)

Evaluated day (day)

Vietnam

11day_warm-up 12

day

11 day (1/21/2017-2/1/ 2017

The term ―11day_warm_up‖ refers to 11 day of pre-simulation warm-up period length With the evaluated day is February 2, 2017, this is a result which is simulated during a continuous 12-day from January 21, 2017 to February 2, 2017

―1day_warm_up‖ can be defined as pre-simulation warm-up period length is 1 day With the evaluated day is February 2, 2017, this is a result which is from January 31,

2017 to February 2, 2017

The term ―no_warm_up‖ has come to be used to refer to pre-simulation warm-up period length is 0 With the evaluated day is February 2, 2017, this is a result that is simulated from February 1, 2017 to February 2, 2017

In case of Korean domain, 5day_warm_up‖ can be defined as pre-simulation warm-up period length is 5 day With the evaluated day is October 8, 2016, this is a result which is consecutively simulated from October 2, 2016 to October 8, 2016 The term ―no_warm_up‖ has come to be used to refer to pre-simulation warm-up period length is 0 With the evaluated day is October 8, 2016, this is a result which is simulated from October 7, 2016 to October 8, 2016

The model is run with a base time of 00 coordinated universal time (UTC) with output generated each hour from each hour from 00 to 2300 UTC

3.2.3 Physical parameterizations

Figure 3.5 Schematic of physics and their interactions

Cumulus scheme is used to re-distribute air in gird columns to explain vertical convective fluxes for these columns that totally contain convective clouds Most cumulus schemes detrains cloud and ice at cloud top (except Betts-Miller-Janjic scheme) These are used by microphysics schemes

Figure 3.6 Illustrations of Microphysics Processes (Dudhia, 2014)

Microphysics is a process that controls the formation of cloud droplets and ice crystals, the process of grow up and fall out as precipitation Microphysics schemes will be provided atmospheric heat and moisture tendencies, microphysical rates, surface resolved-scale rainfall Microphysics schemes base on aggregation, accretion, growth or fall-out process of particle types such as cloud (Qc) water vapor(Qv) , rain drops(Qr) , snow(Qs) , graupel(Qg) , ice crystal (Qi) to calculate latent heat release from condensation, evaporation, deposition, sublimation, freezing, melting

Table 3.4 Cumulus schemes

Abbreviation cu_physics Name of Scheme

D1 D2 D3

c110 1 1 0 Kain-Fritsch (new Eta) scheme (Kain, John S (2004)) c220 2 2 0 Betts-Miller-Janjic scheme(Janjic, Zavisa I (1994)) c330 3 3 0 Grell-Freitas ensemble scheme (Grell, G A and

Freitas, S R.(2014)) c440 4 4 0 Old GFS Simplified Arakawa-Schubert (SAS)(Pan, H

L., and W S Wu., 1995) c550 5 5 0 New Grell scheme (G3) (Grell, Georg A., 1993)(Grell,

G A, D Devenyi, 2002) c660 6 6 0 Tiedtke scheme (ARW only) (Tiedtke, M., 1989,

Zhang et al (2011) ) c14140 14 14 0 New GFS SAS from YSU (Han et al (2011)

c84840 84 84 0 New SAS (HWRF) (Han et al (2011) )

c93930 93 93 0 Grell-Devenyi ensemble scheme (Grell et al (2002)) c99990 99 99 0 Previous Kain-Fritsch scheme(Kain et al (1990)) Cumulus parameterization in the innermost region (D3) was switched off because the fluxes can be explicitly resolved at resolutions under 10 km (Skamarock et al 2008)

Table 3.5 Microphysics schemes

Abbreviation mp_physics Name of Scheme Mass Variables

(Kessler, E., 1969)

Qc Qr Warm rain (i.e.no ice) m2 2 Lin et al scheme

Table 3.6 Longwave Radiation Schemes

Table 3.7 Shortwave Radiation Schemes

al 2004)

(Hong et al 2004)

Qc Qr Qi Qs Mesoscale (5arrays, no graupel)

microphysics, operational High-Resolution Window (NOAA, cited 2001)

Qc Qr Qs (Qt*) Cloud-scale single-moment

6-class scheme (Lim et

al 2010)

Qc Qr Qi Qs Qg Double-moment (8-13arrays)

Abbreviation ra_lw_physics Name of Scheme

l1 1 rrtm scheme (Mlawer et al 1997)

l3 3 CAM scheme (Collins, William D., et al.,

2004) l4 4 rrtmg scheme (Iacono et al., 2008)

l5 5 (New)Goddard scheme (Chou et al., 1999)

Abbreviation ra_sw_physics Name of Scheme

s1 1 Dudhia Shortwave Scheme(Dudhia, J., 1989) s2 2 (old) Goddard shortwave scheme Chou et al.,

1999,Chou et al., 2001 s3 3 CAM scheme (Collins, William D., et al., 2004)

Table 3.8 Surface Layer Schemes

Table 3.9 Planetary Boundary Layer Schemes

Table 3.10 Land Surface Schemes

s4 4 rrtmg scheme (Iacono et al., 2008)

s5 5 (New)Goddard scheme (Chou et al., 1999)

Abbreviation sf_sfclay_physics Name of Scheme

(Jimenez, renamed in v3.6) (Jimenez, et al., 2012)

2006)

Abbreviation bl_pbl_physics Name of Scheme

sf_sfclay_physics=1 or 7

sf_sfclay_physics=1

Abbreviation sf_surface_physics Name of Scheme

su2 2 unified Noah land-surface model (Tewari et

al., 2004)

with Pleim-Xiu surface and ACM2 PBL (Noilan et al., 1989)(Pleim, J E., and A Xiu, 1995)(Xiu, Aijun, and J E Pleim,

2001)(Pleim, J E., and A Xiu, 2003)(Pleim,

J E., and R Gilliam, 2009)(Gilliam, R C., and J E Pleim, 2010)

Initial and boundary conditions (IC/BCs) for the WRF modeling were based on the 32-km horizontal resolution North American Regional Reanalysis (NARR) data that are downloaded at the National Center for Atmospheric Research (NCAR) Boundary conditions to WRF were updated at 6-hour intervals

In this study, a time-varying and multi-physics ensemble is distinguished when only a single parameterization scheme is changed For exemple one type of cumulus scheme while another parameterization (Microphysics scheme, Planetary Boundary Layer scheme, Longwave Radiation scheme, Shortwave Radiation scheme, Surface Layer scheme, Land Surface scheme) is identify Table 3.9 shows a list of physical options for each ensemble 120 combinations of physical options is applied for 3 kind of pre-simulation warm-up period (see Table 3.3) A total of 360 time-varying and multi-physics ensembles are applied for Ho Chi Minh City, Vietnam Similar, Table 3.10 shows 80 combinations for 2 kind of pre-simulation warm-up period (see Table 3.3) A total of 160 time-varying and multi-physics ensembles are applied for Seoul, Korea