Description of data

The UTRCA maintains and operates 20 tipping bucket (TB) rain gauges in the UTRb. Each tip of the bucket is based on rainfall accumulations of either 0.20 mm or 0.25 mm. The majority of the rain gauges record at an hourly resolution with a few gauges recording at resolutions of 15 minutes. The rain gauge network distribution follows no uniform pattern. Due to ease of installation and maintenance rain gauges were installed to be co-located with stream gauges. Therefore, as seen in Fig. 2, there is an increase in density of the rain gauge network around major urban centres, which are primarily located in the southern portion of the watershed (in particular around the City of London). The poor uniformity of the rain gauge network influences the errors observed during the spatial interpolation of the point data. Following Looper and Vieux (2012) the rain gauges undergo quality control before being implemented in the various merging methods, where the bias between the rain gauge and the radar accumulation at the gauge location is calculated for each rain gauge. The rain gauge is removed as an outlier if the bias is outside two standard deviations of the mean bias.

3.2.2 Radar data

Radar data are provided for the Exeter radar station (see Fig. 2) by EC through the Canadian Meteorological Centre (CMC). The radar data are not yet widely available and are provided as part of a collaborative research effort. The radar data are part of EC experiment number 28 of the experimental Canadian Precipitation Analysis (CaPA) system (version 2.4). CaPA is a rainfall estimation program that combines rainfall forecasts, rain gauges and radar to produce six hour estimates of rainfall accumulations on a 10 x 10 km grid across North America. Radar data are processed and corrected using the Unified Radar Processing (URP) software. Before being issued for the study the radar product undergoes substantial correction, including correction for: attenuation, clutter removal, beam blocking and anomalous propagation. The data are provided in a constant altitude plan position indicator (CAPPI) view set in Cartesian coordinates at an altitude of 1.5 km. The radar data have a spatial radial resolution of one km by one degree and a

temporal resolution of one hour. The Exeter radar station has a Doppler range of 120 km covering the entire extent of the UTRb. The data are processed and georeferenced before being applied using ArcGIS (version 10.2). Additional details on the radar tower are displayed in Table 2.

Figure 2: Location of rain gauges and radar station in the Upper Thames River basin

Not all rain gauges in the basin are able to be used in this study. A woodlot located in close proximity to the radar station generates a shadow zone as illustrated in Fig. 3, extending out from the radar station and resulting in a region of unknown radar rainfall. Two rain gauges located within the shadow zone cannot be directly compared to the radar rainfall estimates, and are therefore omitted from this study.

For the remainder of this thesis the term “raw radar” is used to describe the corrected EC radar product unadjusted by ground based rain gauges. This allows for distinction between unadjusted and adjusted radar data.

Table 2: Characteristics of Exeter radar station

Radar station Exeter

SiteID WSO

Location Exeter, Southern Ontario

Latitude 43.3703

Longitude -81.3842

Ground Height 303 masl

Measurement cycle 10 min

Frequency band C (5.6 cm)

Doppler mode Yes

3.2.3 Stream flow data

The UTRCA in conjunction with Environment Canada (EC) measures stream flows at 23 locations within the UTRb, as displayed in Fig. 4. The stream gauges (SG) are maintained and operated by EC through the Water Survey of Canada. These stations measure water levels along main channels within the basin, largely in close proximity to damage centres. The measured water levels are converted to flow rates (in cubic meters per second) using EC calibrated rating curves (Lane 1999). Flow rates are determined at an hourly resolution.

Figure 4: Stream gauge locations in the Upper Thames River basin

3.3 Gauge-radar merging methods

Four merging methods were selected for the analysis in Chapters 4 and 5. These methods applied were selected based on their prominence in literature, their widespread

operational use in other regions and their ability to be implemented in near-real time in the UTRb (Gjertsen et al. 2004; Goudenhoofdt and Delobbe 2009; Berne and Krajewski 2013). For this analysis, a mean field adjustment method (MFB), two spatially-dependent adjustment methods (BSA and LB) and a geostatistical merging method (CM) are evaluated. See Chapter 2 for the comprehensive review and description of these merging methods.

The direct comparison entails comparing the point rain gauge value with the radar pixel located directly above the rain gauge location. For the BSA and LB correction methods rain gauges recording less than 2.5 mm are not used, as minor differences between the observed accumulations can produce excessively large or small calibration factors (Brandes 1975), leading to an erroneous correction field. The LB adjustment method relies on ordinary kriging to distribute the correction factors generated at each rain gauge location. Empirical Bayesian Kriging (EBK) was also explored as an alternative to ordinary kriging, however, EBK was not found to improve the accuracy of the results. In this study a simple spherical variogram is used and the data is assumed to be isotropic. Other variograms explored included circular, exponential, and Gaussian, however, no substantial improvements were observed in accuracy. The parameters of the ordinary kriging spherical model are based on the models recently established by the UTRCA to develop rainfall fields from their rain gauge network. No transformation of the data is conducted and, therefore, the assumption of normality is not upheld.

Chapter 4