INTERANNUAL AND INTERDECADAL VARIABILITY OF …
INTERANNUAL AND INTERDECADAL VARIABILITY OF THAILAND SUMMER MONSOON: DIAGNOSTICS AND FORECASTS
by
NKRINTRA SINGHRATTNA
B.Eng., Chiang Mai University, 1995
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirement for the degree of
Master of Science
Department of Civil, Environmental and Architectural Engineering
2003
This thesis entitle:
INTERANNUAL AND INTERDECADAL VARIABILITY OF THAILAND SUMMER MONSOON: DIAGNOSTICS AND FORECAST
written by Nkrintra Singhrattna
has been approved for the Department of Civil, Environmental
and Architectural Engineering
__________________________________________
(Balaji Rajagopalan)
__________________________________________
(Kenneth Strzepek)
___________________________________________
(Thomas N. Chase)
___________________________________________
(Martyn Clark)
Date__________________
The final copy of this thesis has been examined by the
signatories, and we find that both the content and the form
meet acceptance presentation standards of scholarly work
in the above mentioned discipline.
CONTENTS
Chapter
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1. Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2. Topography of Thailand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3. Thailand Climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1. Summer Season . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2. Rainy Season . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3. Winter Season . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4. Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2 Interdecadal Variability of Thailand Summer Monsoon . . . . . . . . . . 19
1. Introduction Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2. Data Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1. Hydroclimate Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2. Large-Scale Climate Data . . . . . . . . . . . . . . . . . . . . . . . . 28
3. Diagnostic Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.1 Hydroclimatology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.2 Time Series Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.3 Relationship to ENSO . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4. The Physical Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Predictors for Thailand Summer Hydroclimatology . . . . . . . . . . . . . 56
6. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7. Correlation with Land, Ocean and Atmospheric Indices . . . . . 57
8. Correlation with Large-Scale Variables . . . . . . . . . . . . . . . . . . 59
9. Predictor Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
10. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Forecast Models for Thailand Climatology . . . . . . . . . . . . . . . . . . . . 66
2. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
1. Probabilistic categorical forecast model . . . . . . . . . . . . . 67
2. Algorithm of Categorical Forecast . . . . . . . . . . . . . . . . . . 69
3. Nonparametric Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4. Modified K-nn Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3. Model Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
1. Modified K-nn forecast . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2. Categorical Probability Forecast . . . . . . . . . . . . . . . . . . . 84
4.4 Implication of Water Management . . . . . . . . . . . . . . . . . . . . . . 87
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5 Conclusions and Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.2 Recommendations for Future Work . . . . . . . . . . . . . . . . . . . . . . 91
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . n
TABLES
Table
2-1 Correlation between six rainfall stations . . . . . . . . . . . . . . . . . . . . . . 25
2-2 Correlation between three streamflow stations . . . . . . . . . . . . . . . . . 25
2-3 Correlation between five temperature stations . . . . . . . . . . . . . . . . . 25
2-4 Correlation between rainfall and streamflow . . . . . . . . . . . . . . . . . . . 26
2-5 List of high and low ASO rainfall years . . . . . . . . . . . . . . . . . . . . . . 44
2-6 List of high and low SON streamflow years . . . . . . . . . . . . . . . . . . . 44
3-1 Correlation between monsoon rainfall and large-scale index . . . . . . 58
4-1 Conditional probabilities of ASO rainfall associated with SOI . . . . . 69
4-2 Skill scores for modified K-nn model . . . . . . . . . . . . . . . . . . . . . . . . 80
4-3 Exceedence Probabilities from K-nn results . . . . . . . . . . . . . . . . . . . . 84
4-4 Exceedence Probabilities from categorical probability results . . . . . . 85
FIGURES
Figure
1-1 Plot of monsoon rainfall averaged over six stations . . . . . . . . . . . . . 11
1-2 Thailand map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1-3 Flow chart of study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2-1 Hydroclimate station locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2-2 Correlation map between CMAP and ASO rainfall . . . . . . . . . . . . . 27
2-3 Correlation map between SATs and MAM temperature . . . . . . . . . . 28
2-4 Plot of monthly average temperature . . . . . . . . . . . . . . . . . . . . . . . . . 30
2-5 Plot of monthly mean rainfall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2-6 Composite map of vector winds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2-7 Plot of monthly mean rainfall (station-by-station) . . . . . . . . . . . . . . . 33
2-8 Plot of monthly average streamflow . . . . . . . . . . . . . . . . . . . . . . . . . 34
2-9 Scatter plot between summer SATs and ASO rainfall . . . . . . . . . . . . 36
2-10 Scatter plot of relationships among Thailand variables . . . . . . . . . . . 37
2-11 Scatter plot between ASO SATs and ASO rainfall . . . . . . . . . . . . . . 37
2-12 Plot of time series trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2-13 Plot of moving window correlations between ASO rainfall and SOI 41
2-14 Correlation maps of ASO rainfall related to SSTs and SLPs . . . . . . 42
2-15 Composite maps of ASO SSTs (high rainfall years) . . . . . . . . . . . . . . 45
2-16 Composite maps of ASO SSTs (low rainfall years) . . . . . . . . . . . . . . 46
2-17 Plot of spectral analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2-18 Composite maps of ASO SSTs and velocity potential (El Nino years) 49
2-19 Plot of moving window correlations (Thailand and Indian comparison) 50
2-20 Walker circulation mechanism (pre-1980) . . . . . . . . . . . . . . . . . . . . . 52
2-21 Walker circulation mechanism (post-1980) . . . . . . . . . . . . . . . . . . . . 52
2-22 Composite maps of SSTs and CMAP in 1997 . . . . . . . . . . . . . . . . . . 53
2-23 Composite maps of SSTs and CMAP in 2002 . . . . . . . . . . . . . . . . . . 54
3-1 Plot of moving window correlations between rainfall and indices . . 59
3-2 Correlation maps between rainfall and large-scale variables . . . . . . 61
3-3 Plot of moving window correlations between rainfall and selected indices 64
4-1 Surface plot showing nonlinear relationship . . . . . . . . . . . . . . . . . . . 76
4-2 Plot of skill scores from modified K-nn model . . . . . . . . . . . . . . . . . 81
4-3 PDFs of ensemble forecasts from modified K-nn . . . . . . . . . . . . . . . 83
4-4 PDFs of ensemble forecasts from categorical probability . . . . . . . . . 86
CHAPTER 1
INTRODUCTION
1. Background and Motivation
Thailand is a country located in Southeast Asia with a population of 61.2 million. Agriculture is a major occupation of the population that amounts to 50-60% of the national economy. In July 2003, agricultural exports are amounted to approximately 528 million dollars.[1] Most of agriculture is dependent on precipitation, either directly or via irrigation, which depends on precipitation for storage in the reservoirs. Almost all of the precipitation (80-90%) occurs during the monsoon season that spans May through October. The streamflows from the monsoon rainfall are stored in reservoirs and are subsequently, used for the following year until the next monsoon season. Thus, the main priority for water resources management is to efficiently plan and operate the reservoirs to meet irrigation schedules. Hydroelectric power contributes to 4% of total gross energy generation[2], which also adds to the constraints in water resources planning.
Erratic monsoon rainfall leading to inadequate storage can significantly impact agriculture and also create domestic water shortage, especially, in the following summer (March-June) season. Figure 1-1 shows the monsoon rainfall over Thailand during the past 50 years. It can be clearly seen that the rainfall varies over short (year to year) and longer (decadal) time scales. In addition there is a decreasing trend. Consistent variations are seen in streamflows as well (discussed in later chapters). Thailand faced drought situation during 1989 through 1993; in particular, the drought of 1989 affected all aspects of the national economy. Generally, the regions with strong impact are in the Northern, Central and Northeastern parts of Thailand. In contrast, flooding is a major problem as well. The government has to monitor streamflow levels during the monsoon season to issue flood warnings to millions of people living in the flood plains. In the last decade, the most inconsiderate flood occurred in October 1990, which inflicted extensive damage to agriculture and infrastructure costing over 150 million dollars.[3] Last year (2002), flood caused approximately 1.28 million dollars worth of damage to agricultural fields and crops over the entire country.[4] In addition, periodic tropical depressions (i.e. typhoons) cause extensive infrastructure damage and loss of human life.
[pic]
Figure 1-1: Plot of monsoon rainfall (from August to October, ASO) averaged over six stations during 1951 through 2001
Thus, drought and flood have an enormous impact on the national economy, which calls for improved water resources planning and management. This is further exacerbated by the growing population. Therefore, understanding the variability of the monsoon and consequently of the hydroclimatology is crucial to improved resources planning and management. These serve as motivation of this research. Below, the topography and climatology of Thailand is discussed, and then followed by an overview of the thesis.
2. Topography of Thailand
Thailand is located between 5o-20o N latitudes and 97o-106o E longitudes covering an area of 513,115 square kilometers (Fig. 1-2). It extends about 2,500 kilometers and 1,250 kilometers from north to south and from east to west, respectively. Thailand lies between two important water resources: the Indian Ocean to the west and the Gulf of Thailand to the east, which is connected to Pacific Ocean. The length of Thailand coastline is 1,840 kilometers approximately along the Gulf of Thailand and 865 kilometers along the Indian Ocean.
Thailand can be broadly divided into five hydroclimates:
• Northern part
• Northeastern part or Khorat Plateau
• Central part or Chao Phraya River basin
• Eastern part
• Southern part
The Central part or Chao Phraya River basin is the largest rice production region in the country, and therefore, one of the important regions for the economy. Also, the capital, Bangkok, is in this basin. Chao Phraya River originates in Northern Thailand and is 370 kilometers long before reaching the ocean with a watershed area approximately 178,000 square kilometers. There are four main tributaries of Chao Phraya that merge together to form the main Chao Phraya river: Ping, Wang, Yom, and Nan. Chao Phraya River and its tributaries supply water to sustain much of the agriculture in the country and also for domestic needs. There are several dams on Chao Phraya River and its tributaries for water and hydroelectric power generation. Consequently, Chao Phraya River is regarded as the “main bloodline” of Thailand people because of its importance and advantage.
[pic]
Figure 1-2: Thailand map shows neighbors: Myanmar, Laos, Cambodia and Malaysia
3. Thailand Climate
Climatologically, Thailand has three main seasons: summer or pre-monsoon season, rainy or southwest monsoon season, and winter or northeast monsoon season.
1. Summer Season
Summer season usually takes the period from mid-February to mid-May (i.e. northern hemisphere spring). This is the transitional period from Northeast monsoon to southwest monsoon. April is the hottest month in Thailand with temperatures reaching 40 degrees Celsius or above at times. Mean seasonal temperature is around 28-30 degrees Celsius.[5] This heating of the land during the summer season is important for developing the land-ocean gradient and consequently, the strength of the Southwestern monsoon in the following season.
2. Rainy Season
Rainy season generally begins from mid-May to mid-October. Warmer land during the summer season tends to increase convection and the land-ocean gradient, resulting in cold winds from Indian Ocean and Western equatorial Pacific Oceans moving on to the land and also the Inter Tropical Convergence Zone (ITCZ). This brings moisture and abundant precipitation i.e. the Southwestern monsoon. The Southwest monsoon brings warm moist airstreams passing from the Indian Ocean toward Thailand in May. The first region getting the influence of this wind is southern Thailand especially the west coast of peninsula. Later, this wind moves to central, northeastern and northern regions. Another factor causing rainfall in this period, as mentioned above, is the movement of the ITCZ (Kripalani et al. 1995). ITCZ also arrives in May to the Southern part, and then moves northward to Southern China around June to early July. In August, the ITCZ moves southward to lie over Northern and Northeastern Thailand, and then over Central regions in September and Southern part in October. Hence, the most abundant amount of rainfall usually occurs from August to October for central Thailand although the rainy season begins in May. Statistically, most areas of the country receive approximately 1,200-1,600 mm annually. Mean temperature for this season is about 27-28 degrees Celsius.[6]
3. Winter Season
Winter season starts from mid-October to mid-February. This season is marked by dry cool airstreams from the mid-latitudes or the Siberian anticyclone in October. The first region getting the influence of this wind is Northern and then Northeastern Thailand. Normally, mean temperature during this season is around 23-24 degrees Celsius.[7] However, the temperature sometimes drops below 0 degree Celsius especially in the higher latitude and altitude areas. For central and southern regions, mild weather, and some rainfalls occur along the East coast of Southern Thailand, which is also referred as the Northeast monsoon.
Besides these two monsoons and ITCZ, Thailand also gets the effect of tropical cyclones from the Western North Pacific Ocean or the South China Sea. Much of Thailand is located away from the coast with some mountain ranges around; consequently, the wind speeds from typhoons and cyclones are reduced by the time they move into main Thailand. However, the long Southern peninsula with large exposure to both oceans has an increased risk of tropical cyclones and typhoons. Annually, 3-4 tropical cyclones hit Thailand, and they usually happen during April-May.
4. Thesis Overview
From the previous sections, it is clear that understanding the interannual and interdecadal variability of Thailand rainfall can have significant implications to agriculture and water resources management and consequently, the economy. With this motivation, this thesis seeks to address three broad questions. (i) What large-scale ocean-atmospheric features drive Thailand monsoon variability on short and longer time scales? (ii) How can this understanding be used for improved hydroclimate forecasting? And (iii) what are the implications for water resources management? The organization of the thesis is shown in Figure 1-3 and briefly discussed below.
Figure 1-3: Flow chart of study
• Chapter 2: In this chapter, climate diagnostics is performed to identify large-scale climate features that are responsible for interannual and interdecadal variability of Thailand hydroclimate. To this end, comprehensive composite and correlation analyses are performed to diagnose relationships between monsoon rainfall and large-scale climate features.
• Chapter 3: Predictors of Thailand summer rainfall are identified. Systematic correlation analysis is performed between the summer rainfall and large-scale climate variables such as sea surface temperatures (SSTs), sea level pressures (SLPs) and surface air temperatures (SATs) from proceeding months to identify the best set of predictors that can provide improved forecasts.
• Chapter 4: Here, two methods of ensemble forecasts of monsoon seasonal rainfall are described. One uses the large-scale climate predictors identified in Chapter 3 in a nonlinear regression framework, while the other uses a categorical forecast of large-scale phenomenon such as El Nino in a bootstrap framework. These methods are described and their performance assessed via three skill measures.
• Chapter 5: Summary of the results and future research work conclude the thesis.
CHAPTER 2
INTERDECADAL VARIABILITY OF THAILAND SUMMER MONSOON
3 Introduction and Background
El Nino-Southern Oscillation (ENSO) has been widely known to modulate the Asian summer monsoons on interannual and interdecadal time scales. Some examples include, relationships between ENSO and Indian monsoon (Krishna Kumar et al. 1999, Krishnamurthy and Goswami 2000), South China monsoon (Xu and Chan 2001, Lau and Wu 2001, Wang et al. 2001). ENSO also affects the onset of the monsoons – e.g., Indonesian rainfall (Hadama et al. 2002). Conventional wisdom suggests a reduction in the strength of monsoons in these regions with El Nino and vice versa (Rasmusson and Carpenter 1983, Rupakumar and Pant 1997), caused largely due to shifts in the Walker circulation. Furthermore, this dynamical relationship has been extensively used for issuing seasonal forecast of monsoons, in particular the Indian summer monsoon (Shukla and Paulino 1983, Krishna Kumar et al. 1995). Land surface processes, in particular land temperatures and vegetation structure can also play a significant role in modulating the land-ocean gradients and consequently, the ENSO teleconnections, thereby, impacting the strength of the monsoons (Chase et al. 2000, Krishna Kumar et al. 1999). Eurasian snow cover and land temperatures have also been know to influence the Asian monsoon, especially the Indian monsoon (Shukla and Bamzai 20xx, Blandford 1884).
Interestingly, in recent decades, there has been a weakening of the relationship between ENSO and Indian summer monsoon (Krishna Kumar et al. 1999) including East Asian summer monsoon (Chen et al. 2002). On the other hand, Wang et al. (2001) find a steady relationship between ENSO and Western North Pacific summer monsoon (WNPSM). Such nonstationarity in the large-scale, relationship may indicate a wider re-adjustment of atmospheric circulation and potential nonstationarities in the relationship with other monsoons in the region.
Regional processes (land surface, Eurasian snow cover etc.) and large-scale processes (tropical Pacific ocean-atmospheric conditions – e.g., ENSO, tropical Indian ocean conditions, anthropogenic warming) interact in modulating the monsoons. These are inherently complex processes, thereby, making it difficult to understand the interactions (Lau and Wu 2001).
Monsoon variability over Thailand has not been studied as widely as its counterparts in the Indian subcontinent and China. There have been some efforts to understand the relationship between rainfall over Thailand, Singapore, Indonesia with Indian rainfall based on 20 years of data before 1980 (1961-1980) by Kripalani and Kulkarni (1997, 1998, 2001). There is a weak relationship between the summer rainfall over Thailand and Indian rainfall.
Past research, as mentioned above, has focused on several regions in Asia, but there are hardly any studies investigating the variability of Thailand hydroclimate. This motivated the presented research. This research systematically investigates the temporal variability of Thailand hydroclimatology, in particular the summer monsoon. This chapter is presented as follows: data sets and Thailand climatology are first described, trends in precipitation, temperature and streamflows are next presented, understanding the temporal variability and proposal of potential physical mechanism concludes the chapter.
1. Data Summary
In this section, the hydroclimate and the large-scale climate data sets used in the study are described.
1. Hydroclimate Data
From the introduction chapter, we saw that the central region of Thailand is the most important area for Thailand’s economy – as this region produces huge quantities of agricultural products for both domestic consumption and export. Chao Phraya River passing though this central region is a key factor for the agricultural production and other economic activities. Consequently, hydroclimate data from this region have been selected for this study.
Daily precipitation and temperature data[8] have been obtained from six locations in the central Thailand region (shown in Fig. 2-1). The stations are:
• Nakhon Sawan (Nk), northernmost province of central region and the origin of Chao Phraya River
• Suphan Buri (Sp), central region
• Lop Buri (Lp), northern of central region
• Kanchana Buri (Kn), western of central region
• Bangkok (Bkk), the capital of Thailand
• Don Muang (Dm), a city in Bangkok and location of the biggest international airport in Thailand
The data spans the 1951-2001 period.
Daily streamflow data at three stations along Chao Phraya River[9] were also obtained (also shown in Fig. 2-1). The stations are:
• Nakhon Sawan (Nk), drainage area about 110,569 square kilometers, data recorded from 1956 to 2000
• Chai Nat (Cn), a province adjacent to Nakhon Sawan, drainage area about 120,693 square kilometers, data recorded from 1948 to 2000
• Ang-Thong (At), a province near Bangkok, data provided from 1976 to 1996
These data sets were obtained from the website . It has to be mentioned that the hydroclimate data for Thailand was very difficult to obtain, and its consistency was ensured by checking against other data sets.
[pic]
Figure 2-1: Hydroclimate station locations. The red, green and blue stars indicate rainfall, streamflow and temperature stations, respectively.
To check for consistency and also to come up with an “index” time series that can be used for the climate diagnostic analysis, the station data were related to each other. Annual values of precipitation, streamflows and temperatures computed from the respective daily data and the stations were correlated among themselves. Correlations among the six precipitation stations, three streamflow locations and five temperature stations are shown in Table 2-1, 2-2 and 2-3, respectively. The correlations that are significant at 95% confidence level are shown in bold. It can be seen that the streamflows and temperatures are strongly correlated as compared to precipitation. This is to be expected, as the rainfall is more variable than the streamflows. Overall, these tables suggest that the hydroclimate is fairly homogeneous. Similar correlations were found for the monsoon season – which is to be expected as almost all of the information at the annual time scale comes from the monsoon season.
Table 2-1: Correlation between six rainfall stations: Nk, Sp, Lp, Kn, Bkk and Dm
| |Nk |Sp |Lp |Kn |Bkk |Dm |
|Nk |1.00 |0.49 |0.57 |0.54 |0.44 |0.53 |
|Sp |0.49 |1.00 |0.64 |0.61 |0.35 |0.54 |
|Lp |0.57 |0.64 |1.00 |0.58 |0.44 |0.60 |
|Kn |0.54 |0.61 |0.58 |1.00 |0.36 |0.51 |
|Bkk |0.44 |0.35 |0.44 |0.36 |1.00 |0.51 |
|Dm |0.53 |0.54 |0.60 |0.51 |0.51 |1.00 |
Table 2-2: Correlation between three streamflow stations: Nk, Cn and At
| |Nk |Cn |At |
|Nk |1.00 |0.91 |0.93 |
|Cn |0.91 |1.00 |0.98 |
|At |0.93 |0.98 |1.00 |
Table 2-3: Correlation between five temperature stations: Nk, Lp, Kn, Bkk and Dm
| |Bkk |Dm |Kn |Lp |Nk |
|Bkk |1.00 |0.27 |0.56 |0.31 |0.16 |
|Dm |0.27 |1.00 |0.44 |0.62 |0.42 |
|Kn |0.56 |0.44 |1.00 |0.74 |0.71 |
|Lp |0.31 |0.62 |0.74 |1.00 |0.79 |
|Nk |0.16 |0.42 |0.71 |0.79 |1.00 |
To check for relationship between rainfall and streamflow, seasonal streamflows (discussed about season in later chapter) from Cn and Nk were correlated with all the rainfall stations. Streamflow station At was excluded because it had data for only a short period (1976-1996). These correlations are shown in Table 2-4.
Table 2-4: Correlation (station by station) between rainfall and streamflow. Rainfall amounts are employed sum of ASO[10] values, and streamflow discharges are estimated sum of SON[11] values before correlated.
| | |Rainfall Stations |
| | |Nk |Sp |Lp |Kn |Bkk |Dm |Averg 6 |Averg 3 sta. |
| | | | | | | | |sta. |(Nk, Sp, Dm) |
|Streamflow |Nk |0.327 |0.223 |0.139 |0.205 |-0.156 |0.123 |0.156 |0.265 |
|Stations | | | | | | | | | |
| |Cn |0.473 |0.311 |0.234 |0.283 |-0.053 |0.307 |0.320 |0.443 |
The flows are generally positively correlated with the rainfall but show quite a bit of variability. This could be due to regulation and/or the quality of precipitation or streamflow data. The flow discharges at Cn are related well to Nk, Sp and Dm rainfall stations (yellow highlights). These three stations are representative of the central region of Thailand. Hence, the average of rainfall at these three stations is used as Thailand “rainfall index”. The flows at Cn are used as the “streamflow index”. Since the flows at Cn and Nk are highly correlated (see Table 2-2), it makes little difference if we selected either one of them or an average. For the temperature, we used the average at all the five stations as the “temperature index”. Henceforth, these indices are referred as Thailand precipitation, streamflow and temperature, respectively.
To further test the consistency of the above derived indices, Thailand precipitation and temperatures are correlated with gridded CMAP global rainfall data (Xie and Arkin 1997) and NCEP/NCAR Re-analysis (Kalnay 1996) surface temperature data, respectively. Figure 2-2 and Figure 2-3 show these correlation maps[12], and it can be seen that a strong positive correlation can be seen over Thailand region, in particular, over East Central part of the country. This indicates that the hydroclimate data obtained above can be considered to be representative of the country and certainly the important region of East Central Thailand.
[pic]
Figure 2-2: Map of surface CMAP precipitation correlated to ASO Thailand rainfall averaged from three stations (1979 to 2000)
[pic]
Figure 2-3: Correlation map between SATs and MAM Thailand temperatures averaged from five stations (1951-2000)
2. Large-scale Climate Data
Sea surface temperatures (SSTs), especially in the tropical Indian and Pacific Ocean region are crucial in setting up the land-ocean gradient, the strength and location of convection and also the strength and direction of winds. These are critical to the Thailand rainfall in particular and the South Asian monsoon in general. Several SST data sets have been generated by different climate centers in the US and UK with slightly different approach in interpolation from the sparse ship observations. However, in the present study, the Reynolds SST data are used. This is a gridded product on a 1° grid obtained via optimal interpolation from sparse observations of ship data (Reynolds and Smith 1994, Reynolds et al. 2002).
Atmospheric circulation variables[13] such as sea level pressure (SLP), winds, geopotential heights and velocity potential are all obtained from the NCEP/NCAR Re-analysis (Kalnay et al. 1996). These circulation variables at several vertical levels in the atmosphere are obtained on regular 2.25° grid. This data is available from 1949 onwards. These data sets are available online at .
2. Diagnostic Results
In this section, climatology and long-term trends, if any, in the hydrometeorological variables - precipitation, surface air temperature and streamflows are first presented. Followed by investigations into the links between large-scale ocean-atmospheric circulation features that drive the interannual and decadal variability in the summer monsoon rainfall and streamflows. Correlation and composite analyses are used for the climate diagnostics.
1. Hydroclimatology
Climatology of Thailand temperatures, rainfall and streamflows are presented in the form of figures. Monthly mean temperatures are shown in Figure 2-4(a), and it can be seen that the “hottest” months are during March-May (MAM), which is also the pre-monsoon season and northern hemisphere spring. Precipitation peaks during August-October (ASO) (Fig. 2-5(a)). In addition, May shows a smaller peak that occurs consistently over all the stations. This is due to the movement of the ITCZ (Fig. 2-6) as mentioned in section 1.3.2.
[pic]
(a)
[pic]
(b)
Figure 2-4: (a) Plot of monthly average temperatures over five stations showing MAM summer season and (b) histograms of three summer season months (MAM)
[pic]
(a)
[pic]
(b)
Figure 2-5: (a) Plot of monthly mean rainfall over three stations showing ASO seasonal period and (b) histograms of three seasonal months (ASO)
[pic]
(a)
[pic]
(b)
Figure 2-6: Composite maps of vector winds during 1948-2002 in (a) MAM and (b) ASO
From Figure 2-6, the Southwest monsoon that starts during May in the Indian Ocean (part of the Indian summer monsoon) brings warm moist airstreams passing from the Indian Ocean toward Thailand in May, which causes thunderstorms during this period. The ITCZ covers Southern part of Thailand in May as well; however, it then moves northward to Southern China around June to early July. In August, the ITCZ moves southward and lie over Northern and Northeastern Thailand, and then over Central regions in September and southern parts in October. This feature of a secondary peak in May can be seen at all the stations (Fig. 2-7). For streamflow discharges, it is apparent from Figure 2-8(a) that the annual peak occurs during September-November (SON), lagging a month from rainy season. Large flows are frequently found in October.
[pic]
Figure 2-7: Plot of monthly mean rainfall of six stations. The black, blue, red, green, yellow and light blue lines indicate monthly means from Nk, Sp, Lp, Kn, Bkk and Dm station, respectively.
[pic]
(a)
[pic]
(b)
Figure 2-8: (a) Plot of monthly average streamflows of Cn station showing SON seasonal flow duration and (b) histograms of three seasonal months (SON)
Probability density functions (PDFs) of the monthly and seasonal precipitation and streamflows exhibit significant skew (Fig. 2-5(b) and Fig 2-8(b)). The PDFs are estimated using nonparametric kernel density estimators. These are data-drive methods that have the capacity to reproduce any arbitrary features present in the data. For details on these methods see A.W. Bowman and A. Azzalini (1997).
Theoretically, the pre-monsoon air temperatures over the land play a key role in setting up the required land-ocean gradient, which is important to the strength of the monsoon rainfall. Enhanced land-ocean gradient leads to a stronger monsoon and vice versa. Likewise, the ASO rainfall and the SON streamflows trend to be linearly related. These links can be corroborated from Figure 2-9 and Figure 2-10. Figure 2-9 shows the scatter plot of surface temperatures over Thailand during the months of spring season with the monsoon seasonal (ASO) rainfall. A strong linear relationship can be seen indicating that higher temperatures during spring tend to favor a stronger monsoon. In all these figures, nonparametric fits based on local polynomials (Loader 1999) that smooth the scatter plot are shown in red, while linear regression fits are shown in blue. Figure 2-10 shows the scatter plot of spring temperatures and monsoon rainfall and seasonal streamflows; monsoon rainfall and seasonal streamflow and a surface plot of streamflows depending on spring temperatures and monsoon rainfall. The scatter plots show relationships as expected – i.e. enhanced spring temperature tends to favor higher rainfall and also higher streamflows; enhanced rainfall favors enhanced streamflows. In addition, an interesting nonlinear relationship is apparent between seasonal streamflows and monsoon rainfall and spring temperatures – whereby, the streamflows tend to decrease as rainfall exceeds 800 mm. This could be due to reservoir regulation or perhaps, due to enhanced evaporation and infiltration. This needs further investigation, nonetheless it is interesting to note. In Figure 2-11, scatter plots of monsoon rainfall and monsoon seasonal temperatures are shown. As expected, one would see a negative correlation - i.e. enhanced seasonal rainfall tends to cool the land and the atmosphere, thereby, decreasing the temperatures and vise versa. The figures confirm the physical consistency between the hydroclimate variables – thus, enhancing the confidence in the data set.
[pic]
Figure 2-9: Scatter plots between summer temperatures and ASO rainfall (1951-2001) (a) March temperatures, (b) April temperatures, (c) May temperatures and (d) average MAM temperatures
[pic]
Figure 2-10: Scatter plots of relationships among Thailand variables (1951-2001): (a) MAM temperatures related to ASO rainfall, (b) MAM temperatures related to SON streamflows, (c) ASO rainfall related to SON streamflows and (d) 3D plot among them
[pic]
Figure 2-11: Scatter plots between ASO temperatures and ASO rainfall (1951-2001) (a) August temperatures, (b) September temperatures, (c) October temperatures and (d) average ASO temperatures
2. Time Series Trends
In order to investigate the temporal variability and the long-term trends of the hydroclimate variables, linear trends were computed. Figure 2-12(a) and (b) show the time series of spring temperatures and monsoon rainfall. Interestingly, both these series exhibit a strong decreasing (negative) trend over the 50-year period. They are consistent in that decreasing spring temperatures lead to decreasing monsoon rainfall. The blue line is the linear regression fit to the data, and temperature and rainfall trends are –0.4 degree Celsius and –180 mm over 50 years. Besides, significant interannual and interdecadal variability can be seen in these variables. Time series of monsoon seasonal temperatures and seasonal streamflow are shown in Figure 2-12(c) and (d). An increasing trend in monsoon seasonal temperatures can be seen with a decreasing trend in the seasonal streamflows. These are consistent in that decreasing rainfall trends (Fig. 2-12(b)) tend to cool the land and atmosphere less, thereby, increasing the temperatures (Fig. 2-12(c)) and decreasing the seasonal streamflows (Fig. 2-12(d)). IPCC (2001)[14] reports a warming trend over Thailand in the spring temperatures during 1976-1999. In fact, a warming trend can be seen from 1980 until present in Figure 2-12(a); however, the overall trend from 1950 is a decreasing one. IPCC (2001) also shows a decreasing trend in precipitation over Thailand consistent with Figure 2-12(b). In summary, the trends in temperature and precipitation seen here are consistent with those reported in IPCC (2001).
[pic]
(a)
[pic]
(b)
Figure 2-12: Plot of (a) MAM temperatures (1951-2001), (b) ASO rainfall (1951-2001), (c) ASO temperatures (1951-2001) and (d) SON streamflows (1948-2000)
[pic]
(c)
[pic]
(d)
Figure 2-12 (continued)
3. Relationship to ENSO
While the long-term trends in temperature and precipitation may be linked to global trends, it is important to understand the interannual and interdecadal variations. The first step would be to investigate the relationships with ENSO. To this end, 21-year moving window correlations between Thailand monsoon rainfall and SOI (a sea level pressure based ENSO index: Tahiti – Darwin pressure) are plotted in Figure 2-13. Significant correlations are seen after 1980, and prior to 1980, there is hardly any relationship.
[pic]
Figure 2-13: Plot of 21-year moving window correlations between ASO SOI and Thailand monsoon rainfall
To further understand this relationship, the monsoon rainfall is correlated with large-scale ocean-atmospheric circulation variables for the pre- and post-1980 periods[15]. Figure 2-14(a) and (b) show the correlation between monsoon rainfall and SSTs and SLPs for the two periods. The pre-1980 period shows very weak correlations in the tropical Pacific region – in fact, a small region of positive correlation in the eastern tropical Pacific can be seen. But in the post-1980 period, a negative correlation is seen in the central to eastern Pacific and a positive in the western Pacific – reminiscent of the ENSO pattern. Similar observations can be found with SLP (Fig. 2-14(b)) – in fact, the ENSO pattern is much stronger in SLP.
[pic][pic]
(a)
Figure 2-14: Correlation maps of ASO rainfall simultaneously with (a) SSTs and (b) SLPs. The left panel indicates pre-1980 period, and the right panel indicates post-1980 period. The dark purple and red represent correlation coefficients beyond 95% of confidence level (-0.40, 0.40).
[pic][pic]
(b)
Figure 2-14 (continued)
To understand the nonlinearity of this relationship, composite analysis[16] was performed. In this, high and low rainfall years are selected and circulation features identified for these years. The high and low monsoon rainfall years during pre- and post-1980 are identified and listed in Table 2-5, while Table 2-6 indicates the high and low years of seasonal streamflows as well. The high and low years correspond to rainfall and streamflow values greater than the 95th percentile and less than 5th percentile, respectively. Figure 2-15 and Figure 2-16 show the composite SST patterns during the monsoon season in the two periods, for high and low rainfall years, respectively. It can be clearly seen that during the post-1980 period high (low) years tend to go with cooling (warming) in the central and eastern tropical Pacific. This is a clear ENSO pattern. However, in the pre-1980 period, this pattern is largely absent.
Table 2-5: List of high and low ASO rainfall years during pre- and post-1980
| |Pre 1980 |Post 1980 |
|High Years |1952, 1957, 1962 |1983, 1988, 1995 |
|Low Years |1968, 1977, 1979 |1984, 1987, 1994 |
Table 2-6: List of high and low SON streamflow years during pre- and post-1980
| |Pre 1980 |Post 1980 |
|High Years |1951, 1961, 1975 |1980, 1995, 1996 |
|Low Years |1968, 1972, 1979 |1986, 1993, 1998 |
Spectral analysis of rainfall shows a significant band around 2.5 years and a trend, consistent with observations of the time series. A moving window spectral coherence of the monsoon rainfall (Fig. 2-17 (b)) and SOI index shows significant power in the 2.5-3 years band in recent decades. Multi-taper method of spectral analysis was used. This method has been shown to perform much better than traditional spectral methods (Mann and Lees 1996, Thomson 1982). The spectral results further corroborate the recent ENSO connection to Thailand rainfall and consequently, the hydroclimatology.
[pic]
(a)
[pic]
(b)
Figure 2-15: Composite maps of ASO SSTs occur in high rainfall years during (a) pre-1980 period and (b) post-1980 period.
[pic]
(a)
[pic]
(b)
Figure 2-16: Composite maps of ASO SSTs occur in low rainfall years during (a) pre-1980 period and (b) post-1980 period.
[pic]
Figure 2-17: Plots of spectral analyses. The top and bottom indicate spectral of ASO rainfall and spectral coherence related ASO rainfall to ASO SOI index, respectively.
3. The Physical Mechanism
The interesting question is why does ENSO modulate Thailand rainfall in recent decades? Is there a change in the ENSO pattern in recent decades? If so, does this have regional impacts? What could be the physical mechanism?
To answer these, composites of SSTs and velocity potential at 200 hPa (a surrogate for the Walker circulation) during El Nino years in the pre-1980 (1957, 1965, 1972) and post-1980 (1982, 1987, 1997, 2002) period were developed in Figure 2-18. The El Nino years during pre- and post-1980 are identified based on reanalyzed SST analyses[17], which can be found at analysis_monitoring/ensostuff/ensoyears.html. Clearly a strengthening of warm SST anomalies (Fig 2-18(a)) in the eastern Pacific can be seen in recent decades and also a south eastward shift in the anomalies of 200 hPa velocity potential (Fig 2-18(b)). The shifts in the SST and velocity potential fields can be inferred as shift in ENSO related Walker circulation anomalies having implications for associated changes in regional impacts of ENSO teleconnections. Krishna Kumar et al. (1999) find a southeastward shift in the Walker circulation anomalies in the El Nino of post-1980 period – similar to that shown in Figure 2-18(b). This, coupled with midlatitude warming, they argue, could be leading cause for decrease in correlation between Indian monsoon and ENSO. Figure 2-19 shows the moving window correlation between Indian and Thailand monsoons with SOI. It can be seen that around 1980 the Indian monsoon starts to lose its correlation with ENSO while the Thailand monsoon picks up correlation.
[pic][pic]
(a)
[pic][pic]
(b)
Figure 2-18: Composite maps of (a) ASO SSTs and (b) velocity potential in El Nino years during pre-1980 (left panel) post-1980 (right panel)
[pic]
Figure 2-19: Plot of 21-year moving window correlation related either Thailand monsoon or Indian monsoon to simultaneous SOI. The red and blue lines indicate ASO Thailand rainfall and JJAS Indian rainfall, respectively. The dashed lines indicate 95% confidence level.
The hypothesis proposed by Krishna Kumar et al. (1999) in the context of the Indian monsoon seems to be responsible for the shifting correlation as seen in Figure 2-19. The hypothesis is as follows: Eastern Pacific centered ENSOs tend to contain the descending branch of the Walker circulation within the Pacific domain, thereby, reducing convection in the tropical Western Pacific regions and consequently, impacting the rainfall in these regions (including the Thailand monsoon). Under this situation, the Indian subcontinent escapes the brunt, so to speak, and the land surface processes can counter act. On the other hand, if the ENSOs are dateline centered then the descending branch of the Walker circulation has a wider reach over the Indian subcontinent, which decreases convection, and the Indian monsoon gets impacted much more than the Thailand monsoon. This is schematically shown in Figure 2-20 and Figure 2-21. Interestingly, this hypothesis is corroborated during the El Ninos of 1997 and 2002. Figure 2-22 shows the composite seasonal (June-October) SSTs and precipitation for 1997 and Figure 2-23 for 2002. It can be seen that in 1997, the year of strongest El Nino in recorded history, the SST pattern shows a strong warming in the Eastern Pacific, and the rainfall over Thailand shows a deficit while over the Indian subcontinent is close to normal. In 2002 (Fig. 2-23), on the other hand, the El Nino was mild in comparison with warming around the dateline region, and this produced a normal to above normal rainfall over Thailand region and a strong deficit over the Indian region.
[pic]
Figure 2-20: Walker circulation subsidence due to El Nino phase of ENSO during pre-1980 period
[pic]
Figure 2-21: Walker circulation subsidence due to El Nino phase of ENSO during post-1980 period
[pic]
(a)
[pic]
(b)
Figure 2-22: Composite maps of (a) SSTs and (b) CMAP precipitation in 1997
[pic]
(a)
[pic]
(b)
Figure 2-23: Composite maps of (a) SSTs and (b) CMAP precipitation in 2002
4. Summary
Thailand hydroclimatology exhibits a strong trend and interdecadal variability. The variability in recent decades (post-1980) seems to be strongly linked with ENSO. Interestingly, during the same period, Indian monsoon shows a weakening in its relationship with ENSO. These findings seem to argue for a tropical wide ENSO-related circulation shifts with more Eastern Pacific centered ENSOs in recent decades. The dominant mechanism to impact rainfall over Thailand seems to be via the Walker circulation. Eastern Pacific centered ENSOs tend to largely contain the descending branch of the Walker circulation within the Pacific domain, while the descending branch from dateline centered ENSOs tend to extend strongly over the Indian subcontinent. Two recent years, 1997 and 2002, clearly corroborate this hypothesis. The tantalizing question of what causes the shifts in ENSO related SST patterns in the equatorial Pacific remains. Perhaps, it could be a result of enhanced midlatitude warming as Krishna Kumar et al. (1999) suggest, or it could be natural variability of the tropical Pacific system (Wang et al. 2001). This needs the investigations using model simulations and data.
CHAPTER 3
PREDICTORS FOR THAILAND SUMMER RAINFALL
1. Introduction
The aim of this chapter is to identify predictors of Thailand summer rainfall, which can then be used in statistical forecast models in Chapter 4 for improved predictions. Clearly, better forecasts can help significant in resources planning and management. The two main requirements for good predictors are (i) good relation with the hydroclimate variables (i.e. rainfall) and (ii) reasonable lead-time (i.e. months to season). To this end, the Thailand rainfall will be correlated with large-scale ocean-atmospheric variables (SSTs, SLPs etc.) and standard indices (e.g. SOI, Nino 3 etc.) from previous months and seasons. This will be used to identify regions with strong correlation and construct a predictor index. If one of the standard indices shows a strong correlation then that will be used as a predictor as well. This approach has been used to identify predictors for streamflows in Northern Brazil (DeSouza and Lall 2003) and in the Truckee-Carson river basins in NV, USA (Grantz 2002).
Lead-lag correlations of Thailand precipitation with standard indices are first presented, followed by correlation maps of large-scale ocean-atmospheric variables. Based on these, a suite of predictors will be identified for use in seasonal forecasting in Chapter 4.
Streamflow data, being highly regulated, and did not show consistent relationship with precipitation or large-scale climate variables. As a result, we show the results for precipitation.
2. Correlation with Land, Ocean and Atmospheric Indices
From the results in Chapter 2, we found that land temperatures during spring (that drives the land-ocean gradient) and Indo-Pacific SSTs have a strong relation to Thailand hydroclimatology, particularly, in recent decades. Therefore, as a first step, Thailand summer (ASO) rainfall were correlated with spring air temperatures and several ENSO indices such as Nino 1+2, Nino 3, SOI and Indian Ocean Dipole (IOD) index[18] (Saji et al. 1999) from previous seasons, so as to obtain as long a lead-time as possible. The correlations are computed for the post-1980 period and shown in Table 3-1. Correlation values that are statistically significant at the 95% confidence level using a t-test (Helsel and Hirsch 1995) are shown in bold numbers in the table. It can be seen that the SLP-based ENSO index, SOI, shows a strong correlation with monsoon rainfall during the concurrent season and also 1-2 seasons before - this is consistent with results in Chapter 2. Also, the IOD shows a strong correlation with the monsoon rainfall at 2-season lead-time. All these brighten the prospects for a long-lead forecast.
Table 3-1: Correlation results (post-1980 period) between ASO rainfall and indices
| |JFM |FMA |MAM |AMJ |MJJ |JJA |JAS |ASO |
|Nino 1+2 |0.41 |0.31 |0.29 |0.28 |0.25 |0.17 |0.08 |-0.06 |
|Nino 3 |0.42 |0.33 |0.15 |-0.01 |-0.13 |-0.19 |-0.24 |-0.31 |
|Tahiti-Darwin (SOI) |0.40 |0.27 |-0.07 |-0.27 |-0.44 |0.45 |0.57 |0.59 |
To check the temporal variation of these correlations, selected predictors (JFM Nino 3, ASO SOI, MAM IOD and MAM SATs) were correlated with monsoon rainfall in a moving window of 21-year, shown in Figure 3-1. It can be further confirmed that the predictors show correlations with summer rainfall only in recent decades, much as the correlations with ENSO as seen in Chapter 2.
[pic]
Figure 3-1: Plot of 21-year moving window correlations between Thailand rainfall and large-scale indices. The red, blue, green and black lines indicate MAM SAT, JFM Nino 3, MJJ SOI and MAM IOD, respectively. The dashed lines indicate 95% confidence level.
3. Correlation with Large-Scale Variables
To identify the large-scale aspects of predictors and to identify other predictors, the summer monsoon rainfall is correlated with surface air temperatures (SATs), SSTs and SLPs during prior seasons, and the correlation maps[19] are shown in Figure 3-2. The red and dark purple colors indicate correlations that are significant at 95% confidence level. Figure 3-2 shows the correlation maps of summer (ASO) rainfall and SAT, SST and SLP from preceeding seasons (MAM, AMJ and MJJ). Strong positive correlations during spring (MAM) season can be seen over Thailand with SATs (Fig. 3-2(a)). This is essentially capturing the land-ocean gradient that gets set up by the land temperatures during spring season before the monsoon, and is consistent with correlations found in Table 3-1 and also in Chapter 2. With SLPs (Fig. 3-2(b)), the correlations are strong in the subtropical region indicating that a stronger subtropical tends to enhance the easterlies and consequently, increasing the moisture transport and the rainfall. The correlations with SSTs (Fig. 3-2(c)) indicate a strong region positive correlation in the Eastern Indian Ocean and Western Pacific Ocean regions around the equator. This region is also one of the poles of the Indian Ocean Dipole index (Saji et al. 1999) and hence, the strong correlation with IOD seen in Table 3-1. Correlations with all three variables indicate persistence from the spring season. The solid box drawn around regions of high correlation show the regions from which the variables will be averaged to develop potential predictors. This will be discussed in following section.
[pic][pic][pic]
(a)
[pic][pic][pic]
(b)
Figure 3-2: Correlation maps (post-1980) between ASO Thailand rainfall and (a) SATs, (b) SLPs and (c) SSTs. The left, middle and right panels indicate MAM, AMJ and MJJ, respectively. The dark purple and red represent correlated coefficients beyond 95% of confidence level (-0.40, 0.40).
[pic][pic][pic]
(c)
Figure 3-2 (continued)
4. Predictor Selection
Based on the correlations with indices and the correlation maps with large-scale variables, predictors can be identified. Predictors with large correlations to the summer rainfall are selected. With this criterion, the selected predictors are (i) Thailand spring temperatures, i.e. MAM SATs, (ii) SSTs averaged over 10.5o-14.3o S latitudes and 108.8o-120o E longitudes and (iii) SLPs averaged over 20o-30o N latitudes and 165o-180o E longitudes.
To check the temporal variability of the strength of the predictors to monsoon rainfall, moving window correlations are shown in Figure 3-3. The predictors identified above are correlated with the summer monsoon rainfall. As expected, the predictors are largely correlated in the post-1980 period.
[pic]
(a)
[pic]
(b)
Figure 3-3: Plot of 21-year moving window correlations between ASO Thailand rainfall and large-scale variables. The red, blue, green and black lines indicate SATs, SLPs, SSTs and SOI indices, respectively in period of (a) MAM, (b) AMJ and (c) MJJ. The dashed lines indicate 95% confidence level.
[pic]
(c)
Figure 3-3 (continued)
5. Summary
Predictors from large-scale ocean, atmosphere and land variables have been identified, that have strong correlation with Thailand summer monsoon. The predictors are physically consistent. The predictors indicate a 1-2 seasons worth of lead-time. Interestingly, the predictors are related to the monsoon rainfall, only during post-1980 period when the monsoon rainfall is correlated with ENSO, as seen in Chapter 2. This suggests the tantalizing possibility that ENSO relationship could be modulating the predictability – similar to what is seen in the case of the Indian monsoon (Krishna Kumar et al. 1999). The effectiveness of these predictors, in providing useful and skillful rainfall forecast, will be tested using a nonlinear forecasting model in Chapter 4.
CHAPTER 4
FORECASTING THAILAND SUMMER MONSOON RAINFALL
1. Introduction
Seasonal forecast of Thailand summer monsoon can have significant impacts on resources planning and management. In particular, the forecast can help public sector agencies to plan for floods or droughts – e.g., reservoir operations, agricultural practices, flood emergency responses and etc. It is not clear if there is a seasonal forecast mechanism in place in Thailand. Unlike, in the case of the Indian summer monsoon, the Indian Meteorological Department is required to issue a seasonal forecast of the upcoming monsoon season by the end of May. Given the great potential utility of the forecast, this chapter develops two approaches to forecasting Thailand summer monsoon rainfall. The first is a nonparametric model based on local regressions using the predictors identified in Chapter 3. The second approach uses a coarse categorical forecast of ENSO that is typically issued several months in advance. Both these methods generate ensembles of rainfall and consequently, the probability density function (PDFs). This is very useful than mean forecast, as one can obtain threshold exceedence probabilities, key to decision making.
The two models are first described followed by criteria for skill evaluation. The results are next presented, summary and discussion of impacts of forecasts on water resources management concludes the chapter.
1. Probabilistic Categorical Forecast Model
This approach is a simple and quite useful technique, especially in the presence of coarse information. The model uses the basic concept of conditional probability and total probability theorem (Ang and Tang, 1984). The model has two steps: (i) categorical probability of the response variable (e.g., summer monsoon rainfall) conditioned upon the categorical forecast of a predictive variable (e.g., ENSO) are estimated using the total probability theorem and (ii) the historical data is resampled (or bootstrapped) using the estimated categorical probabilities – thus, generating an ensemble forecast. The description is as follows:
The probability of an event B (e.g., summer monsoon rainfall occurring in the upper percentile) occurring when some event A (e.g., occurrence of ENSO) has occurred is defined as the conditional probability of B given A (P(B(A)) and defined as
P(B(A) = P(A ( B) for P(A) > 0, or
P(A)
P(A ( B) = P(B(A) * P(A). [1]
If A has several categories – e.g., La Nina, El Nino and Neutral, then the probability of an event (B) occurring under several given events (i.e. A1, A2 and A3) can be defined using the total probability theorem as
P(B(( A1 ( A2 ( A3)) = P(B(A1) * P(A1) + P(B(A2) * P(A2) + P(B(A3) * P(A3). [2]
The conditional probabilities (P(B(A1), P(B(A2) and P(B(A3)) are typically estimated from the data, and the probabilities of occurrences A1, A2 and A3 (P(A1), P(A2) and P(A3)) are generally provided by the categorical forecast.
If B has several categories as well (e.g., below normal, normal and above normal: B1, B2 and B3), then using the total probability theorem (Equation [2]), the conditional probability of each of these categories of B given the categories of A is given as [3]
P(B1(( A1 ( A2 ( A3)) = P(B1(A1) * P(A1) + P(B1(A2) * P(A2) + P(B1(A3) * P(A3) [3-1]
P(B2(( A1 ( A2 ( A3)) = P(B2(A1) * P(A1) + P(B2(A2) * P(A2) + P(B2(A3) * P(A3) [3-2]
P(B3(( A1 ( A2 ( A3)) = P(B3(A1) * P(A1) + P(B3(A2) * P(A2) + P(B3(A3) * P(A3). [3-3]
2. Algorithm of Categorical Forecast
1. From chapter 3, the ASO SOI was found to have a strong correlation in recent decades with the summer monsoon rainfall. Also, SOI is a good indicator of ENSO and several agencies around the world provide categorical forecasts of ENSO and some particularly the SOI index. The SOI index is divided into three categories – “low”, if the SOI value is below the 33rd percentile; “high”, if the SOI value is above the 66th percentile and “neutral”, otherwise. Likewise, the summer rainfall is also divided into three categories at the same percentiles.
2. The conditional probabilities of the rainfall categories given the SOI categories (Table 4-1) are estimated from the historical data.
Table 4-1: Conditional probabilities of ASO rainfall categorized and associated with ASO SOI
| |
| | |ASO Rainfall |
| | |Low (RL) |Neutral (RN) |High (RH) |
|ASO |Low (SL) |0.57 |0.29 |0.14 |
|SOI | | | | |
| |Neutral (SN) |0.38 |0.25 |0.37 |
| |High (SH) |0.00 |0.57 |0.43 |
| | | | | |
For example, when SOI index indicates a low value, the probability of the rainfall occurring in the low (or below normal) P(RL(SL) is 0.57. A strong relation between SOI and rainfall, particularly in the extreme categories is apparent.
3. When a categorical forecast of ENSO (i.e. SOI) for a given year is issued, the probabilities P(SL), P(SN) and P(SH) are provided. Using the conditional probabilities from data (Table 4-1) and Equation [3] the probabilities P(RL), P(RN) and P(RH) of rainfall being in the low, neutral or high categories are estimated. Of course, by definition, P(RL) + P(RN) + P(RH) = 1.
4. Now, the historical rainfall data is bootstrapped with the probabilities P(RL), P(RN) and P(RH). For example, if these probabilities are 0.3, 0.5 and 0.2, respectively. Then, with 20% probability rainfall values in the low category will be resampled, 50% probability the rainfall in the neutral category and 20% probability in the high category. A traditional bootstrap resamples historical data giving equal weight to all the data points, whereas, here the data points are weighted based on their categorical probabilities. Hence, this can also be thought of as a “weighted bootstrap”. The bootstrap samples, thus generated will provide the ensemble forecast.
5. Repeat steps 3 and 4 for each year.
Categorical forecasts of ENSO (SOI) are issued since the mid 1990s. There are not enough forecasts to test. Therefore, a “typical” categorical forecast was selected for demonstration of this approach. It is known that a forecast of El Nino or La Nina phase has a higher skill than the neutral phase. Hence, the categorical forecast for El Nino is assumed to be 0.15, 0.15 and 0.70; La Nina as 0.70, 0.15, 0.15 and neutral as 0.30, 0.40 and 0.30, respectively for P(SL), P(SN) and P(SH).
This method is simple and a powerful tool for generating ensemble forecasts quickly. Because of its simplicity, it has been widely used in ensemble forecast of precipitation, temperature (Wilks 1995, 1998, Wilks and Wilby 1999). However, the main drawback is that since the historical values are resampled – values not seen in the history cannot be generated. Therefore, the variety of the ensembles is limited.
3. Nonparametric Model
Traditional models fit a regression, often linear, between the response variable (i.e. summer rainfall) and the independent variables (i.e. predictors). They are of the form,
Yt = a1 * x1t + a2 * x2t + a3 * x3t + … + ap * xpt + e [4]
Where the coefficients a1, a2,…, ap are estimated from the data
e is the error which is assumed to be normally distributed with mean 0 and variance sigma2, also estimated from the data.
Implicitly, the variables are also assumed to be normally (or Gaussian) distributed. If not, they are generally transformed to a normal distribution (e.g., log or power transform) before the model is fit. Once the model is fit (i.e. the coefficients estimated) then for any new value of the predictors the model is used to estimate the response variable and the error distribution (in this case Gaussian) provides the uncertainty. There is a rich literature for fitting and testing such models and software are extensively available (Helsel and Hirsch 1992). Such models have been widely used for hydroclimate forecasting in the US (e.g., Liu et al. 1998, Piechota et al. 2001, Cordery and McCall 2000), for Indian monsoon forecasting (Krishna Kumar et al. 1995).
In the above model (Equation [4]), if the independent variables happen to be past values of the response variable itself, then it forms a time series model of Auto Regressive Moving Average framework. These are linear regression based frameworks as well and hydrologists have developed and used such models for streamflow simulation and forecast (Salas 1985, Yevjevich 1972, Bras and Iturbe 1985).
The main drawbacks of traditional linear regression based models are (i) assumption of Gaussian distribution of data and errors, (ii) assumption of linear relationship between the variables, (iii) higher order fits (e.g., quadratic or cubic) require large amounts of data for fitting and (iv) not portable across data sets (i.e. sites).
Nonparametric methods provide an attractive alternative with their flexible and powerful framework. In this approach the model is:
Y = ((x1, x2, x3, … xp) + e [5]
which is same as the traditional model (Equation [4]). But the function ( is estimated locally. In that, the value of the function at any point ‘x’ is obtained by fitting a polynomial to a small set of neighbors to ‘x’. This is the gist of nonparametric approach – “local” estimation. Once the neighbors to ‘x’ are identified then there are couples of options:
i) The neighbors can be resampled with a weight function that gives more weights to the nearest neighbor and less to the farthest; thus, generating an ensemble forecast or simulation (Lall and Sharma 1996, Rajagopalan and Lall 1999, Yates et al. 2003, DeSouza et al. 2003)
ii) A polynomial can be fit to the neighbors that can be used to issue a mean value (or mean forecast) – also used for spatial interpolation (Rajagopalan and Lall 1998)
Thus, the parameters are estimated the size of the neighborhood and the order of the polynomial. Being a local estimation scheme, it has the ability to capture any arbitrary local features. Furthermore, unlike the parametric alternative, non-prior assumption is made as to the functional form of the relationship (e.g., in the parametric model, Equation [4], a linear relationship is assumed to begin with). These are two major differences that offer flexibility to nonparametric methods. There are several nonparametric approaches, such as kernel-based (Bowman and Azzalini 1997); Splines, K-nearest neighbor (K-nn) local polynomials (Rajagopalan and Lall 1999, Owosina 1992); Locally weighted polynomials (Loader 1999). The K-nn local polynomials and the Local weighted polynomial (LOCFIT) approaches are very similar. Owosina (1992) performed an extensive comparison of a number of regression methods both parametric and nonparametric on a variety of synthetic data sets. He found that the nonparametric methods handily outperform parametric alternatives.
In this research a modified version of LOCFIT is adopted also known as the modified K-nn. The LOCFIT method is simple and easy to implement, hence, this was chosen. The modification was developed by Prairie (2002), Prairie et al. (2003) and applied for streamflow and salinity modeling. Later, Grantz (2003) demonstrated the use of this approach for streamflow forecasting on the Truckee-Carson basin in Nevada, USA. The methodology is described in the following section.
4. Modified K-nn Model
The method is described in the following steps:
1. For a given data set, the best choice of neighborhood size (K) and the order of polynomial (p) is obtained using objective criteria such as Generalized Cross Validation (GCV), likelihood or such measures.
2. At each observed data point, xj, K nearest neighbors are identified and a local polynomial of order p is fitted. This is then used to estimate the value of the dependent variable (i.e. the conditional mean) at the observed point. The residual (ej) is then computed. This is repeated at each data point, thus obtaining the residuals at all of them. This can be called as the “fitting” process.
3. For a new data point xnew at which a forecast is required, the conditional mean value Ynew is obtained using the step 2.
4. Now select one of the neighbors of xnew, say xi and select the corresponding residual ei, this is now added to the mean forecast Ynew + ei, thus, obtaining one of the ensemble members. The selection of one of the neighbors is done using a weight function
1/j
W (j) =((((((
k
(1/j [6]
j=1
As can be seen, this weight function gives more weight to the nearest neighbor and less to the farthest neighbors. The number of neighbors to be used to resample the residuals need not be same as the number of neighbors used to perform the local polynomial in step 1. In practice, square root of (n-1) is used to resample the residuals.
5. Repeat step 4, as many times as required. This will result in the ensemble forecast.
6. Repeat steps 3 to 5 for each forecast point.
The main advantage of this modified K-nn approach is the ability to capture local error structure that can be non-Gaussian. Also, it improves upon straight bootstrap (Lall and Sharma 1996, DeSouza and Lall 2003), in that; values not seen in the historical record can be produced. Unlike in traditional bootstrap only historical values are resampled.
The relationship between independent variables (i.e. SLPs and SSTs predictors) and summer monsoon rainfall are shown in Fig. 4-1. Significant nonlinearity can be seen, thus, highlighting the utility of the nonparametric approach.
[pic]
Figure 4-1: 3D plot showing nonlinear relationships between ASO rainfall and independent variables (MAM SST and MJJ SLP)
2. Model Evaluation
Both the models are verified in a cross-validated mode. In that, the data (rainfall and the predictors) for a given year is dropped out of the model and the model is applied to generate ensemble forecast for the dropped year. This is repeated for all the years. Apart from visual inspection, the ensembles are evaluated on three criteria:
i) Correlation between the observed value and the median of the ensemble forecast. This is much like evaluating the mean forecast that would come from a standard linear regression model.
ii) Likelihood function (LLH) (Rajagopalan et al. 2002). This evaluates the skill of the model in capturing the PDF.
iii) Rank Probability Skill Score (RPSS) (Wilks 1995). This evaluates the skill of the model in capturing the categorical probabilities.
The likelihood function (LLH) is applied to measure the skills of forecast models. Its process is to categorize forecasted values to three divisions: below, normal and above normal. The ensemble forecasts falling into these three categories are compared to historical data and then develop a skill score. The likelihood skill score for any given year of forecast is defined as:
N
[ ( Pf ]1/N
t=1
LLH = (((( [7]
N
[ ( Pc ] 1/N
t=1
where Pf is probability of ensemble forecasts in observed category for any given year
Pc is the climatological probability of in the observed category in the given year – since we divided the rainfall into three categories at the 33rd and 66th percentile, the probabilities of each of the categories is 1/3.
n is length of time series
The LLH values vary from 0 to numbers of categories (i.e. 3 in this study). The score of zero indicates lack of skill, a score of greater than 1 indicates that the forecasts are better than climatological forecast.
The ranked probability skill score (RPSS) is also applied to quantify the skills of forecast models against the climatology. This method evaluates the probability of ensemble forecasts falling into many categories (i.e. in this study: below average, average and above average) and compared to historical data. The RPSS score for any given year is defined as:
k i j
RPS = 1 [ ( ( ( Pn - ( dn)2] [8]
((( i=1 n=1 n=1
k -1
RPSS = 1 – RPS (forecast) [9]
RPS (standard)
for k mutually exclusive and collectively exhaustive categories (in this case we have three categories, so k = 3). The vector d (d1, d2, …, dk) represents the observation vector such that dn equals 1 if the observation fell in category n or 0 otherwise.
The RPSS scores vary from +1 to -( (i.e. perfect skill to bad skill). Scores above 0 indicates improvement to climatological forecast. The key difference between RPSS and LLH is that in RPSS the
3. Results
Ensemble forecasts are made at the beginning of each month starting from April. The skill of the forecasts is evaluated using the three skill scores described in the previous section and are presented in this section. Skill scores of the forecasts from the two models described above are presented. Skills are also compared during high (wet) and low (dry) years. Threshold exceedence probabilities during the extreme years are estimated and the PDFs of ensembles of a few representative years are also presented.
1. Modified K-nn Forecast
From the set of predictors (SST, SAT and SLP based) identified in Chapter 3, the optimal subset was found by the combination that gave the best forecast skill. Several formal methods are available for subset selection – such as, subset regression or cross-validation metrics etc. Since there are only three predictors in this case almost all combinations were tried out to find the optimal predictor set. For summer monsoon rainfall, the best set of predictors were found to be SLP and SST based. The land temperatures (SAT) did not seem to improve the skill much. MJJ SLPs and the MAM SSTs, these predictors were used in the modified K-nn model, and a forecast was issued at the beginning of each month starting April 1st for each year. The predictors are the average values from the preceeding season (i.e. preceeding three months). The skill scores are shown in Table 4-2. It can be seen that the skill increases significantly as the forecast lead-time decreases. This is intuitive and consistent with expectations. There is significant skill from May 1st onwards providing a 4-month lead-time that can be very useful for resources planning and management. This is quite impressive.
Table 4-2: Skill scores of modified K-nn model for ASO rainfall in all years of post 1980
| | |Skill Scores |
| | |Correlations |LLH |RPSS |
|Forecas|Apr 1st |0.47 |0.68 |-0.61 |
|t Date | | | | |
| |May 1st |0.39 |0.95 |0.03 |
| |Jun 1st |0.59 |1.57 |0.51 |
| |Jul 1st |0.57 |1.36 |0.28 |
| |Aug 1st |0.65 |2.09 |0.79 |
Figure 4-2 shows the three skill measures for forecast issued at the start of each month from April through August, for all the years (Figure 4-2(a)), wet and dry years (Figure 4-2(b) and (c)). It can be seen that the model exhibits very good skill as the lead-time decreases and also the skills are better in the wet years relative to dry.
[pic]
(a)
[pic][pic]
(b) (c)
Figure 4-2: Plot of skill scores in (a) all years, (b) wet years and (c) dry years of post-1980 for ASO rainfall from modified K-nn model
PDFs of the ensemble forecasts during selected wet and dry years made on August 1st are presented along with the climatological PDFs in Figure 4-3. The observed value is shown as the dotted line. It can be seen that the ensemble forecasts show a shift in the PDF in the right direction – more so in the case of wet years.
One of the key variables for decision makers is the probability of threshold exceedences. We chose 700 mm (the 90th percentile of the data) as a surrogate for wet (or flood) conditions and 400 mm (the 10th percentile of the data) for dry conditions. From the PDFs of the ensembles the exceedence probabilities are computed for the selected wet and dry years and shown in Table 4-3. It can be seen that for the wet years the climatological exceedence probability is 0.16, while the ensembles indicate a very high probability of exceedence of this threshold, thus indicating a wet condition. This information can be very helpful in flood emergency response planning and management. For the dry exceedence, only in year 1994, the ensembles could indicate a small exceedence, consistent with the earlier observation that the skills in wet years is better than dry. Similar estimates were obtained from forecasts issued in June and July.
[pic]
(a)
[pic]
(b)
Figure 4-3: PDF of ensemble rainfall forecasts of (a) high years and (b) low years during post-1980 from modified K-nn model. The dashed lines indicate the observed values in those years.
Table 4-3: Exceedence probabilities for (a) wet and (b) dry years compared the forecasts from modified K-nn against the climatology
| |(a) | |
|Year |Climatology |K-nn |
|1983 |16.6% |89.0% |
|1988 |16.3% |82.9% |
|1995 |16.2% |25.1% |
| |(b) | |
|Year |Climatology |K-nn |
|1984 |84.7% |84.1% |
|1987 |87.5% |100.0% |
|1994 |88.8% |39.5% |
2. Categorical Probability Forecast
For this approach the conditional probabilities of the categories of summer rainfall were computed (Table 4-1). These were computed from summer monsoon rainfall and the ASO SOI index. For each year a categorical forecast of ENSO (i.e. ASO SOI) was assumed, resulting in conditional probabilities of categories of rainfall (using Equation [3]) and consequently, ensemble forecasts. In this approach we make only one forecast, assuming that the categorical forecast of ENSO is obtained sufficiently ahead (1-3 months). Of course, in reality the actual probabilistic forecast will be used to generate the ensembles. The skills scores are 0.07, 1.09 and 0.06 respectively, for correlation, LLH and RPSS measures. A modest skill is seen. However, these skills will improve with improvements in the categorical forecast of ENSO. The skills for the wet and dry years are 1.14, 0.18 and 1.18, 0.16 for LLH, RPSS, respectively. The skills in the extreme years are much better.
The PDFs of the ensembles for the wet and dry years (same as in Figure 4-3) are shown in Figure 4-4. The ensembles get the shift but they are not as good as the shifts seen from the modified K-nn model (Figure 4-3). Table 4-4 shows the exceedence probabilities for the wet and dry years. Here too, the exceedence probabilities show considerable skill relative to climatology but below what is seen from the modified K-nn model in Table 4-3.
The probabilistic categorical approach was presented as a quick and easy method to obtain ensemble forecasts in the presence of coarse information (such as categorical probabilistic forecast of ENSO).
Table 4-4: Exceedence probabilities for (a) wet and (b) dry years compared the forecasts from categorical probability against the climatology
| |(a) | |
|Year |Climatology |Model |
|1983 |15.6% |10.4% |
|1988 |16.9% |7.9% |
|1995 |16.8% |13.0% |
| |(b) | |
|Year |Climatology |Model |
|1984 |87.7% |90.9% |
|1987 |87.9% |86.7% |
|1994 |90.3% |86.3% |
[pic]
(a)
[pic]
(b)
Figure 4-4: PDF of ensemble rainfall forecasts of (a) high years and (b) low years during post-1980 from categorical probability model. The dashed lines indicate the observed values in those years.
4. Implication to Water Management
Increased stress on the Chao Phraya River basin due to increased population growth is resulting in water quantity and quality problems. To mitigate this, effective planning and management of water resources is called for. In the short term this requires a good idea of the upcoming monsoon season – i.e. good seasonal forecast and in the long term it needs realistic projections of scenarios of future variability and change. There is no known long-lead forecast of Thailand summer monsoon or streamflows as a result, much of the water resources planning in Chao Phraya basin is near term – i.e. responding to near term weather forecast. Given the need, the ensemble forecasts of monsoon rainfall from the models presented above can provide critical information to water managers for effective decision making. Some of the potential applications include:
i) The threshold exceedence probabilities can provide the water managers with the probabilistic knowledge of potential drought or flood conditions in the upcoming season. This can be used to effectively plan annual and seasonal reservoir, emergency response preparedness, flood plain management, cropping strategies, conservation measures etc.
ii) The threshold exceedence probabilities can also be used as a surrogate for wetness or dryness and provide probabilistic information on flooding potential, land slides, etc. and develop optimal response strategies.
iii) The ensembles of rainfall can be used to drive a water balance model and generate ensembles of streamflows. These can then be used in the water resources decision making to identify optimal responses.
The long-lead forecasts can provide significant information to government and public sector agencies to mobilize resources to meet contingencies. The forecasts will provide a useful and powerful tool to water managers in long-term planning that is currently lacking.
5. Summary
Two models for ensemble forecasts of Thailand summer monsoon are offered. One is a quick and dirty approach to obtain ensembles when only categorical information of large-scale climate (e.g., ENSO) is available. This is based on estimating the conditional probabilities of categories of rainfall and subsequently, bootstrapping the historical data with these probabilities. The other is a nonparametric method based on local polynomials that uses large-scale climate predictors from prior seasons that are identified apriority. The nonparametric method has the ability to capture nonlinear features and non-Gaussian error features that might be present in the data. The nonparametric model in particular, provides significant skill at lead times of 3-4 months. This has tremendous implications to water management, early warning and preparedness and also for resources planning in general.
CHAPTER 5
CONCLUSIONS AND RECOMMENDATIONS
This chapter summarizes the findings from the research and recommendations for future research.
1. Summary
As motivated (in Chapter 1), the Thailand economy approximately 50-60% depends on the agricultural activities within Chao Phraya river basin. Most of the agriculture consume water from precipitation and reservoirs – i.e. also depends on precipitation for storage. Almost all of the annual precipitation comes during the monsoon season occurring from May through October. The runoff from the monsoonal rains is stored in reservoirs for use in the dry periods. Year to year variability of rainfall coupled with inadequate storage leads to frequent floods and droughts. Thus, a need to understand this multiyear variability and also to develop tools for rainfall forecast at long lead-times. These pressing needs motivated this research, which lead to the following:
• Decreasing trend in Thailand monsoon rainfall during the 1950-2000 period with a slight increasing trend in the post-1980 period is seen. This is consistent with a decreasing trend in the spring season temperatures, which is important in setting up the land-ocean gradient. These trends are also consistent with IPCC (2001) findings.
• Thailand monsoon rainfall shows a strong relationship with ENSO in the post-1980 period. In exactly the same period the Indian summer monsoon shows a dramatic weakening of relationship.
• A hypothesis was proposed that involves shifts in Walker circulation depending on the SST pattern in the tropical Pacific. In that, anomalous SST patterns in the Eastern-Pacific tend to contain the descending branch of the Walker circulation over the South East Asian region thereby impacting Thailand monsoon. On the other hand, SST patters in the central-Pacific tend to extend the descending branch well over the Indian subcontinent, thereby having less impact on the Thailand summer monsoon. In recent decades the ENSOs have tended to be Eastern-Pacific centered and hence, the association with Thailand summer monsoon and weakening with the Indian.
• Large-scale ocean-atmospheric variables were identified as predictors of Thailand summer monsoon in the post-1980 period. In particular SSTs over a region East of Indonesia and SLPs over the subtropical Pacific around dateline provided the strongest correlations with summer rainfall at 1-2 season lead-time. These are part of the ENSO system.
• Two forecasting methods were used for generating ensemble forecasts of summer rainfall using the identified predictors. The methods provide skilful forecasts that have the potential for impacting resources planning and management.
2. Recommendation for Future Works
There are several extensions that will greatly enhance the understanding of the Thailand monsoon and also enable its application to resources management and planning. To this end here are some recommendations:
i) Forecast Improvement
Although the forecast methods described in this research did a good job overall in providing a skilful forecast, it is based only on a short period of historical data (post-1980). To provide more confidence in the methods, these have to be applied to data from other regions of Thailand and also for a longer period.
ii) Streamflow Forecasting Model
The streamflow data that was obtained in this study appears to be regulated, hence the relationship with large-scale ocean-atmospheric variables was inconsistent to what was obtained from the precipitation data. One of the key steps is to obtain unregulated streamflows and repeat the analyses presented in this research and test the forecast models. This will have immediate impact in reservoir operations and flood plain management.
iii) Causes for ENSO Shifts
As seen from the analyses, an Eastern Pacific shift of SSTs due to ENSO tends to impact Southeast Asian regions, while a Central Pacific shift of SSTs impacts the Indian subcontinent. The key question that needs further investigation is what drives these shifts in recent decades? Is it anthropogenic or natural variability? Investigating, this is key to a better understanding of interdecadal variability and also to obtaining better predictors for improved hydrologic predictions. Global Climate Models will have to be used to analyze competing hypothesis to obtain better insights.
iv) Statistical-Physical Forecasting Model
As part of GEWEX/GAME-2 program physical watershed models are being developed for the Chao Phraya river basin. It would be good to incorporate climate information into these models. For example, climate information could drive stochastic weather generators (Rajagopalan and Lall 1999, Yates et al. 2003) to produce scenarios of daily weather, consequently driving the physical watershed models to produce ensembles of streamflow forecasts and also other variables such as stage, flood plain delineation, etc. Such statistical-physical approaches can be a powerful tool for flood plain management.
v) Decision Support System
The true test of the forecasts will be their performance in the context of decision making. This requires identifying existing decision support systems or developing new systems for water and other resources management and evaluating the performance of the ensemble forecasts in making the right decisions.
BIBIOGRAPHY
Abraham, Bovas and Johannes Ledolter. Statistical Methods for Forecasting. New York: John Wiley and Sons, 1983.
Ang, A. H-s and W. H. Tang. Probability Concepts in Engineering Planning and Design. New York: Wiley publication, 1984.
Ayyub, Bilal M. and Richard H. McCuen. Probability, Statistics and Reliability for Engineers and Scientists. 2nd ed. Florida: CRC Press LLC, 2002.
Bamzai, A. S. and J. Shukla 1999: Relation between Eurasian Snow Cover,
Snow Depth, and the Indian Summer Monsoon: An Observational Study.
Journal of Climate, 12(10), pp. 3117–3132.
Blanford, H. H., 1884: On the Connection of Himalayan Snowfall and Seasons of Drought in India, Proc. R. Soc. London, 37, 476-487.
Bowerman, Bruce L. and Richard T. O’Connell. Forecasting and Time Series. 3rd ed. California: Duxbury Press, 1993.
Bowman, A. W. and A. Azzalini. Applied Smoothing Techniques for Data Analysis. Oxford: Clarendon press, 1997.
Bras, R. L. and I. Rodriguez-Iturbe. Random Functions and Hydrology. Massachusetts: Addison-Wesley Publishing, 1985.
Carroll, R. J., D. Ruppert and L. A. Strefanski. Measurement Error in Nonlinear Models. London: Chapman and Hall, 1995.
Chase, Tom N., Pielke Sr., Roger A., Greg Marland, Richard Betts, Joseph L. Eastman, John O. Niles, Dev Niyogi and Steve W. Running, 2002: The Influence of Land-Use Change and Landscape Dynamics on the Climate System – Relevance to Climate Change Policy beyond the Radiative Effect of Greenhouse Gasses, Phil. Trans. of the Roy. Socie. of London (series A), 360, 1-15.
Chen, Junye, Barbara E. Carlson and Anthony D. Del Genio, 2002: Evidence for Strengthening of the Tropical General Circulation in the 1990s, Science, 295, 838-841.
Chou, C., 2003: Land-Sea Heating Contrast in an Idealized Asian Summer Monsoon, Climate Dynamics, 21, 11-25.
Climate Diagnostic Center. CDC Interactive Plotting and Analysis Pages. [Online] Available , January 2002.
Climate Diagnostic Center. Create a monthly/seasonal mean time series from the Dataset. [Online] Available at series1.pl, March 2002.
Climate Prediction Center. Monthly Atmospheric and SST Indices. [Online] Available , March 2002.
Climate Prediction Center. Monthly Atmospheric and SST Indices. [Online] Available , March 2002.
Climate Predictor Center. Cold and Warm Episodes by Season. [Online] Available .
Cordery, I. and M. A. McCall, 2000: A Model for Forecasting Drought from Teleconnections, Water Res. Res., 36(3), 763-768.
Dabang, Jiang, Wang Huijun, Drange Helge and Lang Xianmei: Instability of the East Asian Summer Monsoon-ENSO Relationship in a Coupled Global Atmosphere-Ocean GCM, Norwegian Research Council.
Department of Agricultural Extension. Flood 2002. [Online] Available . doae.go.th/dac/, September 2003.
DeSouza, Filho, F. A. and U. Lall, 2003: Seasonal to Interannual Ensemble Streamflow Forecasts for Ceara, Brazil: Applications of a Multivariate, Semi-Parametric Algorithm, Water Res. Res., in press.
Durbin, J. and S. J. Koopman. Time Series Analysis by State Space Methods. New York: Oxford University Press, 2001.
Electricity Generating of Thailand. Welcome to Electricity Generating Authority of Thailand. [Online] Available , September 2003.
Fasullo, J. and P. J. Webster, 2002: Hydrological Signatures Relating the Asian Summer Monsoon and ENSO, Journal of Climate, 15, 3082-3094.
GAME-T. Long Term Discharge Observation by RID. [Online] Available , March 2002.
Goddard Institute for Space Studies. Surface Temperature Analysis. [Online] Available , September 2003.
Goswami, B. N. and R. S. Ajaya Mohan, 2001: Intraseasonal Oscillations and Interannual Variability of the Indian Summer Monsoon, Journal of Climate, 14, 1180-1198.
Grantz, K., B. Ragagopalan and E. Zagona, 2002: Forecasting Spring Flows on the Truckee and Carson Rivers, EOS Trans. AGU, 83(47), Fall Meet. Suppl. Abstract H12G-04.
Hamada, Jun-Ichi, Manabu D. Yamanaka, Jun Matsumoto, Shoichiro Fukao, Paulus Agus Winarso and Tien Sribimawati, 2002: Spatial and Temporal Variations of the Rainy Season over Indonesia and their Link to ENSO, J. of the Meteo. Soc. of Japan, 80, 285-310.
Hansen, J., R. Ruedy, J. Glascoe and M. Sato. GISS Analysis of Surface Temperature Change. [Online] Available RuedyG.html, September 2003.
Helsel, D. R. and R. M. Hirsch. Statistical Methods in Water Resources. Amsterdam: Elsevier Science B. V, 1995.
Huet, Sylvie, Annie Bouvier, Marie-Anne Gruet and Emmanuel Jolivet. Statistical Tools for Nonlinear Regression. New York: Springer-Verlag, 1996.
Intergovernmental Panel on Climate Change. Trends and Variability in Key Climate Variables. [Online] Available , November 2003.
Japan Marine Science and Technology Center. Indian Ocean Dipole (IDO) Home Page. [Online] Available , July 2002.
Kalnay, E. and Coauthors, 1996: The NCEP/NCAR 40-year Reanalysis Project, Bull. Amer. Meteor. Soc., 77, 437-471.
Kottegoda, Nathabandu T. and Renzo Rosso. Statistics, Probability, and Reliability for Civil and Environmental Engineers. New York: McGraw-Hill, 1997.
Kripalani, R. H. and Ashwini Kulkarni, 1997: Rainfall Variability over South-East Asia – Connections with Indain Monsoon and ENSO Extremes: New Perspectives, Journal of Climatology, 17, 1155-1168.
Kripalani, R. H. and Ashwini Kulkarni, 2001: Monsoon Rainfall Variations and Teleconnections over South and East Asia, Inter. J. of Cli., 21, 603-616.
Kripalani, R. H., S. V. Singh, Nalini Panchawagh and M. Brikshavana, 1995: Variability of the Summer Monsoon Rainfall over Thailand-Comparison with Features over India, International Journal of Climatology, 15, 657-672.
Kripalani, R.H. and A. Kulkarni, 1998: The Relationship between Some Large-Scale Atmospheric Parameters and Rainfall over Southeast Asia: A Comparison with Features over India, Theo. And Appl. Cli., 59, 1-11.
Krishna Kumar, K., M. K. Soman and K. Rupa Kumar, 1995: Seasonal Forecasting of Indian Summer Monsoon Rainfall: A Review, Weather, 50(12), 449-467.
Krishnamurthy, V. and B. N. Goswami, 2000: Indian Monsoon-ENSO Relationship on Interdecadal Timescale, Journal of Climate, 13, 579-595.
Kumar, K. Krishna, Balaji Rajagopalan and Mark A. Cane, 1999: On the Weakening Relationship between Indian Monsoon and ENSO, Science, 284, 2156-2159.
Lall, U. and A. Shama, 1996: A Nearest Neighbor Bootstrap for Resampling Hydrologic Time Series, Water Res. Res., 32(3), 679-693.
Lau, K.-M. and H. T. Wu, 2001: Principle Modes of Rainfall-SST Variability of the Asian Summer Monsoon: A Reassessment of the Monsoon-ENSO Relationship, Journal of Climate, 14, 2880-2895.
Lau, Ngar-Cheung and Mary Jo Nath, 2000: Impact of ENSO on the Variability of the Asian-Australian Monsoon as Simulated in GCM Experiments, Journal of Climate, 13, 4287-4309.
Loader, C. Local Regression Likelihood. New York: Springer, 1999.
Local Climatology Sub-Division, The Climate of Thailand, Climatology Division, Thailand Meteorological Department, August 2000.
Lui, Z., J. B. Valdes and D. Entekhabi, 1998: Merging and Error Analysis of Regional Hydrometeorological Anomaly Forecasts Conditioned on Climate Precursors, Water Res. Res., 34(8), 1959-1969.
Mann, Michael E. and Jonathan M. Lees, 1996: Robust Estimation of Background Noise and Signal Detection in Climate Time Series, Climatic Change, 33, 409-445.
Ministry of Finance. Export Value Statistics Classify by Commodity Group. [Online] Available , September 2003.
Owosina, A. Methods for Assessing the Space and Time Variability of Ground Water Data. Thesis (M.S.), Utah: Utah State University 1992.
Piechota, T. C., F. H. S. Chiew, J. A. Dracup and T. A. McMahon, 2001: Development of an Exceedence Probability Streamflow Forecast, ASCE J. of Hydrologic Eng., 6(1), 20-28.
Prairie, J. R. Long-term Salinity Prediction with Uncertainty Analysis: Application for Colorado River above Glenwood Springs, M.S. Thesis. Colorado: University of Colorado at Boulder, 2002.
Prairie, J., B. Rajagopalan, T. Fulp and E. Zagona, 2003: A modified K-NN Model for Generating Stochastic Natural Streamflows, (submitted to) Water Resources Research.
Ragagopalan, B. and U. Lall, 1998: Locally Weighted Polynimial Estimation of Spatial Precipitation, J. of Geographic In formation and Decision Analysis, 2(3), 48-57.
Rajagopalan, B. and U. Lall, 1999: A Nearest Neighbor Bootstrap Resampling Scheme for Resampling Daily Precipitation and other Weather Variables, Water Resources Research, 35(10), 3089-3101.
Rajagopalan, B., U. Lall and S. Zebiak, 2002: Optimal Categorical Climate Forecasts through Multiple GCM Ensemble Combination and Regularization, Monthly Weather Review, 130, 1792-1811.
Rasmusson, E. M. and T. Carpenter, 1983: The Relationship between Eastern Equatorial Pacific Sea Surface temperature and Rainfall over India and Sri Lanka, Mon. Wea. Rev., 111, 517-527.
Reynolds, R. W. and T. M. Smith, 1994: Improved Global Sea Surface Temperature Analyses Using Optimum Interpolation, Journal of Climate, 7, 929-948.
Reynolds, Richard W., Nick A. Rayner, Thomas M. Smith, Diane C. Stokes and Wanqiu Wang, 2002: An Improved In Situ and Satellite SST Analysis for Climate, Journal of Climate, 15, 1609-1625.
Rupakuma, K. and G. B. Pant. Climate of South Asia. John Wiley and Sons, 1997.
Saji, N. H., B. N. Goswami, P. N.Vinayachandran and T. Yamagata, 1999: A Dipole Mode in the Tropical Indian Ocean, Nature, 401, 360-363.
Salas, J. D., 1985: Analysis and Modeling of Hydrologic Time Series, Handbook of Hydrology, New York: McGraw-Hill, 19.1-19.72.
Shukla, J. and D. A. Paolino, 1983: The Southern Oscillation and Long-Range Forecasting of Summer Monsoon Rainfall over Indian, Mon. Wea. Rev., 111, 1830-1837.
Thailand Meteorological Department. Flood. [Online] Available . go.th/ risk1/flood.htm, September 2003.
Thomson, D. J., 1982: Spectrum Estimation and Harmonic Analysis, IEEE Proc., 70, 1055-1096.
Vranes, Kevin. Indian Ocean Dipole Mode Index. [Online] Available . ldeo.columbia.edu/~kvranes/research/DMI/, March 2002.
Walpole, Ronald E. and Raymond H. Myers. Probability and Statistics for Engineers and Scientists. 3rd ed. New York: Macmillan Publishing, 1985.
Wang, Bin and Zhen Fan, 1999: Choice of South Asian Summer Monsoon Indices, Bulletin of the American Meteorological Society, 4, 629-638.
Wang, Bin, Renguang Wu and K.-M. Lau, 2001: Interannual Variability of the Asian Summer Monsoon: Contrasts between the Indian and the Western North Pacific-East Asian Monsoons, Journal of Climate, 14, 4073-4089.
Wilks, D. D., 1998: Multi-Site Generalizations of a Daily Stochastic Weather Generatation Model, Journal of Hydrology, 210, 178-191.
Wilks, D. S. and R. L. Wilby, 1999: The Weather Generation Game: A Review of Stochastic Weather Models, Progress in Physical Geography, 23, 329-357.
Wilks, D. S. Statistical Methods in the Atmospheric Science: An Introduction. San Diego: Academic Press, 1995
Xie, P. and P. A. Arkin, 1997: Global Precipitation: A 17-year Monthly Analysis based on Gauge Observations, Satellite Estimates, and Numerical Outputs, Bull. Amer. Meteor. Soc., 78, 2539-1558.
Xu, Jianjun and Johnny C. L. Chan, 2001: The Role of the Asian-Australian Monsoon System in the Onset Time of El Nino Events, Journal of Climate, 14, 418-433.
Yates, D., S. Gangopadhyay, B. Rajagopalan and K. Strzepek, 2003: A Nearest Neighbor Bootstrap Technique for Generating Regional Climate Scenarios for Integrated Assessments, Water Resources Research, 39(7), 1199.
Yevjevich, V. M., 1972: Stochastic Processes in Hydrology, Water Res. Publi., Colorado: Fort Collins.
-----------------------
[1] Ministry of Finance. Export Value Statistics Classify by Commodity Group. [Online] Available , September 2003.
[2] Electricity Generating of Thailand. Welcome to Electricity Generating Authority of Thailand. [Online] Available , September 2003.
[3] Thailand Meteorological Department. Flood. [Online] Available , September 2003.
[4] Department of Agricultural Extension. Flood 2002. [Online] Available , September 2003.
[5] Local Climatology Sub-Division. The Climate of Thailand. Climatology Division, Thailand Meteorological Department, August 2000.
[6] Local Climatology Sub-Division. The Climate of Thailand. Climatology Division, Thailand Meteorological Department, August 2000.
[7] Local Climatology Sub-Division. The Climate of Thailand. Climatology Division, Thailand Meteorological Department, August 2000.
[8] Local Climatology Sub-Division. The Climate of Thailand. Climatology Division, Thailand Meteorological Department, August 2000.
[9] GAME-T. Long Term Discharge Observation by RID. [Online] Available , March 2002.
[10] Rainy season (ASO) discussed in section 2.3
[11] Streamflow season (SON) discussed in section 2.3
[12] Climate Diagnostic Center. CDC Interactive Plotting and Analysis Pages. [Online] Available , January 2002.
[13] Climate Diagnostic Center. Create a monthly/seasonal mean time series from the Dataset. [Online] Available , March 2002.
[14] Intergovernmental Panel on Climate Change. Trends and Variability in Key Climate Variables. [Online] Available , November 2003.
[15] Climate Diagnostic Center. CDC Interactive Plotting and Analysis Pages. [Online] Available , January 2002.
[16] Climate Diagnostic Center. CDC Interactive Plotting and Analysis Pages. [Online] Available , January 2002.
[17] Climate Predictor Center. Cold and Warm Episodes by Season. [Online] Available .
[18] Japan Marine Science and Technology Center. Indian Ocean Dipole (IDO) Home Page. [Online] Available , July 2002.
[19] Climate Diagnostic Center. CDC Interactive Plotting and Analysis Pages. [Online] Available , January 2002.
-----------------------
Interdecadal and interannual variability of Thailand hydroclimate
Precipitation
Streamflow
Temperature
Identify Predictors
Tropical SSTs
SLPs
SATs
Ensemble Forecast
Categorical probability model
Modified K-nn model
Implication to water resources management and planning
r = 0.36
r = 0.23
r = 0.25
r = 0.37
................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related searches
- antonyms and synonyms list of words pdf
- positive and negative effects of technology
- tongue and quill letter of appreciation
- positive and negative effects of video games
- positive and negative impacts of technology
- singular and plural forms of nouns
- determine and display product of numbers pseudocode
- hebrew and greek dictionary of the bible
- aerobics and fitness association of america
- positive and negative effects of social media
- positive and negative effects of globalization
- positive and negative effects of the internet