Merging of radar and rain gauge data considering the rainfall characteristics

다국어 초록 (Multilingual Abstract)

Improving the quality of radar rain rate data is important to maximize the application of data. Generally, high-quality rainfall information is used as a key input for hydrologic analysis. Quantitative precipitation estimation (QPE) has been performed to minimize the various errors of radar rain rate data. Recently, merging of the radar data and rain gauge data is important as it is expected to get higher-quality rain rate field. However, the quality of the radar rain rate data still does not meet the expected level. To make better quality of rain rate field, this study evaluated the effect of rainfall characteristics in the application of a data merging technique. Specifically, this study focused on (1) analysis of zero measurements and distribution function of rainfall data on the correlation coefficient, (2) evaluation of rainfall characteristics on the simple Kriging, and (3) evaluation of rainfall characteristics on the merging of radar and rain gauge rain rate data.
First, the effect of zero measurements and distribution function on the correlation coefficient was analyzed in the application of the Kriging method. Generally, the correlation coefficient is found without considering the zero measurements of data and with the assumption that data follow the Gaussian distribution. However, rain rate data is generally positively skewed, also showed a strong spatial and temporal intermittency. These characteristics change the correlation coefficient which also affect the shape of the variogram. As a result, the shape of the variogram decides the covariance and finally the Kriging weighted values.
Second, the effect of rainfall intermittency and log-normality on the simple Kriging was evaluated, in the estimation of rainfall spatial coverage. In this study, the artificial data and radar and rain gauge rain rate data were applied to the simple Kriging. The results showed that the zero values has the tendency to decrease the sample variance but to increase the correlation coefficient among data. When the data do not follow the Gaussian distribution, the correlation length could be longer when taking the natural logarithm to the original data. It was confirmed that the data intermittency and data log-normality should be considered to derive a proper variogram. Overall, it was found that the consideration of the data intermittency and data log-normality can improve the simple Kriging result. Especially, the effect of considering the data intermittency was found very significant. However, it was also found that several abnormally high values can be generated or the area of no rain can be decreased due to the longer correlation length.
Third, the effect of rainfall intermittency and log-normality on the merging of radar and rain gauge rain rate data using the co-kriging was evaluated. As a result, for both variogram and cross-variogram, the correlation length was found to be longer, but sill height was smaller in the cases with consideration for data characteristics. The longest correlation length was derived when considering both the data intermittency and log-normality. Additionally, the co-kriging result can be better in case of considering the data characteristics. Especially, the data log-normality was found to be higher effect on the spatial coverage and quality of the rain rate field than the data intermittency. Furthermore, when considering the data characteristics, the mean of the merged rain rate field became more or less the same as that of the ground data.

목차 (Table of Contents)

• 1. Introduction 1
• 1.1 Background of Research 2
• 1.2 Review of the Previous Studies 7
• 1.2.1 Studies on Quantitative Precipitation Estimation 7
• 1.2.2 Studies on Kriging 11
• 1.3 Objectives and Organization of the Dissertation 17
• 1.3.1 Study Purpose 17
• 1.3.2 Contents and Scope of Dissertation 20
• 2. Radar Rainfall Estimation 22
• 2.1 Radar Rainfall Observation 23
• 2.1.1 Rainfall Observation Using Single Polarization Radar 23
• 2.1.2 Rainfall Observation Using Dual Polarization Radar 24
• 2.2 Radar Polarimetric Parameters 26
• 2.2.1 Polarimetric Parameters for Single Polarization Radar 26
• 2.2.2 Polarimetric Parameters for Dual Polarization Radar 28
• 2.3 Radar Rain Rate Equation 32
• 2.3.1 Radar Rain Rate Equation for Single Polarization Radar 32
• 2.3.2 Radar Rain Rate Equation for Dual Polarization Radar 34
• 2.4 Bias Correction for Radar Rain Rate Using the G/R Ratio 39
• 3. Kriging and Correlation Coefficient According to Zero Measurements and Distribution Function of Rainfall Data 41
• 3.1 Theory of Kriging 42
• 3.1.1 Variogram 42
• 3.1.2 Kriging 48
• 3.2 Correlation Coefficient of Normally Distributed and Intermittent Data 55
• 3.3 Correlation Coefficient of Log-normally Distributed and Intermittent Data 60
• 4. Evaluation of Rainfall Characteristics on the Simple Kriging 65
• 4.1 Introduction 66
• 4.2 Analysis of Simple Kriging Using Artificial Data 70
• 4.2.1 Preparation of Artificial Data 70
• 4.2.2 Variograms of Artificial Data 72
• 4.2.3 Application Results of Simple Kriging to Artificial Data 76
• 4.3 Analysis of Simple Kriging Using Rain Rate Data 82
• 4.3.1 Data 82
• 4.3.2 Variograms of Radar Rain Rate Data 86
• 4.3.3 Application Results of Simple Kriging to Radar Rain Rate Data 89
• 4.4 Conclusions 97
• 5. Evaluation of Rainfall Characteristics on the Merging of Radar and Rain Gauge Rain Rate 99
• 5.1 Introduction 100
• 5.2 Analysis of co-kriging Using Radar and Rain Gauge Rain Rate Data 105
• 5.2.1 Data 105
• 5.2.2 Determination of Variograms and Cross-variorams 113
• 5.2.3 Merging of Radar and Rain Gauge Rain Rate Data Using co-kriging 122
• 5.3 Conclusions 129
• 6. Summary and Discussions 131
• References 136