ALBERT

Note: Dissertation, Gottfried Wilhelm Leibniz Universität Hannover, 2017 , Contents 1 Introduction 1.1 Motivation, research questions and overview 1.1.1 Rainfall estimation at high spatial and temporal resolution 1.1.2 Precipitation estimation with cars 1.1.3 Motion estimation from in-situ sensor data 1.2 Outline 2 Basics 2.1 Precipitation 2.1.1 Resolution, accuracy and precision of precipitation measurements 2.1.2 In-situ point measurements of precipitation by rain gauges 2.1.3 Weather radar 2.2 Wireless Sensor Networks 2.2.1 Modeling sensor networks 2.2.2 Sensor network algorithms and protocols 2.3 Statistics 2.3.1 Basics and notation 2.3.2 Regression 2.3.3 Stochastic processes 2.3.4 Stochastic filtering and the Kalman filter 2.3.5 Geostatistics 2.4 Interpolation methods 2.4.1 Inverse-Distance-Weighted 2.4.2 Ordinary kriging 2.4.3 Regression kriging 2.4.4 Cross-validation for performance assessment 2.5 Optical flow 2.5.1 Optical flow intensity conservation 2.5.2 Gradient-based optical flow 2.5.3 Probabilistic optical flow 3 Related Work 3.1 Quantitative precipitation estimation from rain gauges, weather radar and other data sources 3.1.1 Precipitation estimation with weather radar 3.1.2 Precipitation estimation by interpolation of rain gauges measurements 3.1.3 Geostatistical merging of radar and rain gauge data 3.1.4 Motion-based methods used in nowcasting 3.1.5 New data sources for precipitation estimation 3.2 Decentralized estimation with geosensor networks 3.2.1 Estimation of spatio-temporal field properties with GSN 3.2.2 Object-tracking with GSN 4 Methodology for precipitation intensity estimation at 1-min resolution 4.1 Time-window approach for estimation 4.1.1 Estimation of field motion 4.1.2 Weather radar upsampling 4.1.3 Variogram estimation 4.2 Estimation methods 4.2.1 Spatial rain gauge interpolation methods 4.2.2 Space-time symmetric rain gauge interpolation method 4.2.3 Space-time asymmetric rain gauge interpolation methods 4.2.4 Radar-rain gauge merging methods 4.2.5 Estimation methods solely based on radar 4.3 Summary 5 Methodology for precipitation intensity estimation with car sensors 5.1 Car sensors 5.1.1 Wiper Frequency Sensor 5.1.2 Xanonex optical sensor 5.1.3 Other sensors investigated 5.1.4 Experimental setup and preprocessing 5.2 Theoretical considerations for the calibration of the W-R relationship in the field . 5.3 Dependency between car speed, windscreen angle and sensor readings 5.3.1 Manually-operated windscreen wipers 5.3.2 Automatically-operated windscreen wipers 5.3.3 Xanonex optical sensor 5.4 Summary 6 Methodology for motion estimation with a geosensor network 6.1 Algorithm overview 6.2 Network and field model 6.3 Gradient constraint estimation in the network 6.3.1 Gradient constraint estimation from irregular data 6.3.2 Requirements on node stationarity and sampling synchronicity 6.3.3 Estimation of partial derivative error 6.3.4 Gradient constraint selection and derivation of gradient constraint error 6.4 Temporal coherence: Kalman filter for recursive motion estimation 6.4.1 Estimation of process noise Q 6.4.2 Estimation of measurement noise R 6.4.3 Difference to common Kalman filtering problems 6.5 Algorithm protocol 6.6 Algorithm complexity 6.6.1 Communication complexity 6.6.2 Load balance 6.6.3 Computational complexity of partial derivative estimation 6.6.4 Computational complexity of motion estimation 6.7 Summary 7 Results 7.1 Precipitation intensity estimation at 1-min resolution 7.1.1 Study area and data basis 7.1.2 Performance assessment via cross-validation 7.1.3 Exploratory and visual data analysis 7.1.4 Radar estimation and rain gauge cross-validation results 7.1.5 Summary 7.2 Precipitation intensity estimation with cars 7.2.1 Study area and data basis 7.2.2 Selection of the reference method 7.2.3 Manually-operated windscreen wipers 7.2.4 Automatically-operated windscreen wipers 7.2.5 Xanonex optical rain sensor 7.2.6 Results of experiments on the VW rain track 7.2.7 Summary 7.3 Motion estimation with a geosensor network 7.3.1 Study Area, sensor network and deployment strategies 7.3.2 Error measures 7.3.3 Setting the filter parameters 7.3.4 Results - simulated field 7.3.5 Results - radar field 7.3.6 Summary 8 Summary and discussion of the research hypotheses 8.1 Discussion of research hypotheses 1 and 2: 1-min precipitation intensity estimation 8.2 Discussion of research hypothesis 3: precipitation estimation with cars 8.3 Discussion of research hypothesis 4: decentralized motion estimation 8.4 Outlook 9 Appendix 9.1 Discussion on the 'frozen field' distance function 9.2 Executable Kalman filter equations for the motion estimation algorithm 9.3 Controllability and Observability of the Kalman filter for motion estimation List of Figures List of Tables References