• No results found

Rainfall estimation using both radar and rain gauges

Another complementary approach to rainfall estimation with radar is the combination of radar QPE with external rainfall measurements, mainly from rain gauges. The ap- proaches taken include using rain gauges to adjust radar data at a given temporal scale (Cole and Moore, 2008; Joss and Lee, 1995; Collier, 1986, for example), the merging of the data sources together using geospatial techniques such as Kriging (Jewell and Gaussiat, 2015; Goudenhoofdt and Delobbe, 2009, for example) and recent developments in the generation of stochastic ensembles which represent the combined error structure of the two data sources (Wu et al., 2015; Germann et al., 2009; Llort et al., 2008, for example). Each of these methods utilises rain gauge measurements of rainfall as do many validations of radar QPE, and it is important to recognise that rain gauges have their own sources of error and operate at a different spatial and temporal scale to weather radars.

2.5.1 Rain gauge measurements of rainfall

Rain gauges measure the amount of rainfall falling in a small area (typically in the order of 200 cm2) and vary in complexity from manually read devices to weighing principle gauges

Chapter 2. Weather radar for hydrology 31

with complex filtering electronics. Many studies have been undertaken to determine the accuracy of these gauges with common sources of error being evaporation, wetting loss, wind under catch, splash and false readings during high intensity rainfall (Beven, 2011; Lanza and Vuerich, 2009; Sieck et al., 2007; Sevruk, 1982). While many of these errors have corrections or mitigation techniques rain gauge measurements of rainfall are still subject to a degree of uncertainty which is highly dependant on the type of gauge used and its location. (Sevruk et al., 2009; Duchon and Essenberg, 2001; Humphrey et al., 1997).

Another aspect of uncertainty when comparing or merging rain gauges and radar esti- mates of rainfall is the resolution differences between the two observations. Rain gauges are an accumulated point measurement while radar is an instantaneous measurement over a wider area. Studies have shown that the effect of these differences on comparisons are most apparent when comparing data on short time scales in small areas and that both spatial and temporal integration of the data reduces the random errors in the comparison (Villarini and Krajewski, 2008; Ciach, 2003; Habib et al., 2001).

Despite these errors rain gauges are routinely used as the primary measurement of rain- fall both for validation of other datasets (radar QPE, satellite QPE, numerical weather prediction) and for hydrological modelling. The next section covers there usage in con- junction with radar QPE.

2.5.2 Combination with radar

The simplest application of rain gauges in relation to radar is for validation of new rainfall estimation methods (Diederich et al., 2015b; Jiang et al., 2012; Biggs and Atkinson, 2011, for example) however they have also been used to adjust radar measurements to account for differences between radar and rain gauges accumulations. Radar rainfall estimates are often adjusted to remove any bias between them and rain gauges over a given time period (Collier, 1996, 1986). For example the UKMO adjust radar rainfall estimates such that 24 hour accumulations are unbiased in relation to 24 hour rain gauge accumulations (Harrison et al., 2000; Golding, 1998).

More recent adjustment techniques are actually spatial merging of the two data sources, typically using a variant of Kriging (Jewell and Gaussiat, 2015; Goudenhoofdt and De- lobbe, 2009, for example). These techniques account more for the spatial variation of the biases between gauge and radar while maintaining the spatial resolution of the radar data. A new variation of the Kriging with external drift technique has just become op- erational in the UKMO radar system based on the work of Jewell and Gaussiat (2015) (Katie Norman, personal communication, July 2016).

Stochastic rainfall ensembles derived from radar are another relatively new addition to hydro–meteorology (Ciach et al., 2007; Germann et al., 2006a). The main advantage of these statistical ensembles are their computational and experimental ease of implemen- tation, as all the radar’s error sources are lumped into one residual error derived through comparison with the rain gauges. One limitation is that the residual error is also in- fluenced by the errors in ground observations, another being that the method does not directly identify the cause of the errors be they beam blockage, attenuation or parametri- sation errors. Each of the ensemble techniques contains a stochastic perturbation which is modulated by residual error distributions obtained by comparison to ground observations (Villarini and Krajewski, 2010b; Germann et al., 2009, for example). This perturbation can then be used to generate multiple realisations of the rainfall, the distribution of which has the same statistical structure as the radar gauge comparisons. One emerging use of these ensembles is the quantification of overall uncertainty in the hydrological modelling process, allowing comparison with other sources of uncertainty including other inputs, model structure and seasonal parametrisations (Liechti et al., 2013; Quintero et al., 2012; Schröter et al., 2011; Zappa et al., 2011).

Each of these merging and ensemble techniques will provide more accurate rainfall esti- mates if the radar QPEs used within them is the best possible estimate available, and dual polarisation provides the ability to improve these estimates as discussed earlier in this Chapter. The primary focus of the following research is to ensure this accuracy for rainfall estimates from the NCAS mobile radar, prior to their application in any future stochastic studies of hydrological modelling of river flow.

Chapter 3

Instrumentation and data

acquisition

The data used in this study comes from the National Centre for Atmospheric Science’s dual polarisation Doppler mobile X-band radar, primarily obtained during the Convec- tive Precipitation (COPE) field experiment in the summer of 2013 (3.2). This data is supported by additional measurements from the field campaign (3.2.2) and also from the United Kingdom’s operational monitoring network (3.4). Additional data has been analysed obtained during testing deployments of the radar at the Burn field site (York- shire, UK) and also during theICE-D(Cape Verde) andSESAR(Braunschweig, Germany) projects (3.3).

3.1

The NCAS dual polarisation Doppler mobile X-band

radar

The primary data source for this study is a dual polarisation Doppler mobile X-band radar operated by NCAS. The radar is a Meteor 50DX manufactured by Selex ES GmbH, with a custom fitted 2.4 m antenna which precludes the use of a radome (Fig. 3.1). The radar operates at a frequency of 9.375 GHz (a wavelength of 3.2 cm) at a peak transmission power of 83 kW allowing operational ranges in excess of 150 km. The addition of a larger antenna reduces the half power beam width to 0.98 degrees and the radar is capable of

scanning from -1 to 90 degrees in elevation (unlike many operational radars) at a speed of up to 36◦s−1. The radar operates in hybrid transmission mode, simultaneously transmit- ting in horizontal and vertical polarisations, and can also transmit in single polarisation (H or V) while still receiving in both channels. Many of the specific scan parameters, including the azimuth and range gate spacing, the pulse repetition frequency and the an- tenna rotation speed are fully customisable depending on the research requirements and as such are specific to each deployment. The following section (3.2) covers the specifics of the radar’s deployment during COPE, along with a more general introduction to the field campaign.

Figure 3.1: The modified Meteor 50DX operated by NCAS deployed at the Burn

testing site, 30-01-2013. Notice the absence of a radome due to the oversized parabolic antenna fitted to the radar.