Winter daily precipitation variability over the East Anglian region of Great Britain and its relationship with river flow


Correspondence to: Dr I. D. Phillips, School of Geography, Earth and Environmental Sciences, The University of Birmingham, Edgbaston, Birmingham B15 2TT, UK. E-mail:


This paper investigates winter (DJF) precipitation variability over East Anglia at the daily and 2 d timescales using rain gauge data from 134 sites. The first three principal components (PCs) explain 80.2% (82.0%) of the variance in daily (2 d totals). Precipitation displays higher spatial coherence in winter, with the first three PCs of summer daily totals (Neal and Phillips, 2009, International Journal of Climatology 29: 1661–1679) explaining only 72.3% of the variance. The winter PCs for both 24- and 48-h timescales display precipitation maxima in South Essex (PC 1), NE Norfolk (PC 2), and NW Cambridgeshire (PC 3). The rainfall maximum over South Essex is more important in winter than in summer. Lamb Weather Types and synoptic charts are used to explain the causes of the PCs. A precipitation regionalization is also produced. Discharge for 11 rivers is then modelled as a linear function of the rainfall patterns (as represented by the PC scores). With the exception of the River Babingley, whose discharge is not significantly related to the PCs, these models explain on average 26.5% of the variance. For most rivers, there are significant relationships between discharge and precipitation that fell up to 4 d ago on average. In general, East Anglian river flows show persistence over 6–8 d.

1. Introduction

This paper considers the geographic variability of winter precipitation (DJF) over East Anglia at the daily and 2 d timescales. River flow variability in the region is then linked to the main patterns of rainfall variability.

This work builds upon an earlier study of summer rainfall variability over the same region by Neal and Phillips (2009). This study identified the most important patterns of precipitation variability by applying principal component (PC) analysis (PCA) and cluster analysis (CA) to gauge data from 136 sites, but did not relate these patterns to river discharge. In our paper, the winter was chosen for two reasons: (1) to assess whether the summer rainfall patterns identified by Neal and Phillips (2009) are stable to a change of season and (2) 75% of floods in East Anglia occur in the winter half-year (Mokrech et al., 2008), which means that identifying rainfall–runoff associations during this season have the most practical utility in the fields of infrastructure and flood prediction. Furthermore, there are concerns that more intense winter rainfall (Maraun et al., 2008) and rising sea levels will make East Anglia particularly vulnerable to a warmer climate. Mokrech et al. (2008) predicted the area at risk of flooding in East Anglia in 2050 by using different scenarios of temperature increase, sea-level rise, and changes in winter and summer precipitation. Studying winter rainfall in East Anglia also has applications to agriculture. East Anglia is the main cereal growing area in Britain, with more than one-quarter of England's wheat and barley being grown in the region. Important crops like peas, wheat, and beans are all sown in the winter, with crop growth being critically dependent on the soil's moisture content. In addition, more than 50% of Britain's entire sugar beet crop and almost one-third of England's potato crop are harvested in East Anglia during the early winter.

Climate is the first-order control of river discharge, with flow being influenced by the local seasonal cycle of precipitation and evaporative losses (Dettinger and Diaz, 2000). Laizé and Hannah (2010) analysed seasonal climate–river flow associations in 104 basins in Great Britain, with CA being used to group similar basins. They found that flow in the higher elevation and more impermeable basins in western and northern Britain had the strongest relationships with precipitation. For instance, during the winter, rainfall accounted for 74% of the variance in the discharge of seven rivers in central and northern Scotland compared to only 33–54% for those regions covering East Anglia. Potential evaporation did not emerge as a significant predictor of winter flows in East Anglia, but a modest improvement was achieved by including the soil moisture deficit in the regression models. During the winter, a stronger North Atlantic Oscillation (NAO) leads to stronger rainfall–runoff associations and more contiguous river flow regions across Britain than in summer. However, Laizé and Hannah (2010) did not investigate daily rainfall–runoff dependence in East Anglia, a research gap that is addressed in this paper. The dependence of discharge on precipitation has also been investigated in the Danube Basin in Europe (Rimbu et al., 2002), in eastern China (Chen et al., 2009), and in the Nile Basin in Africa (Conway and Hulme, 1993). These international studies often use atmospheric circulation indices [e.g. NAO index and Southern Oscillation Index (SOI)] and sea-surface temperatures in conjunction with precipitation. However, studies in Britain (Phillips et al., 2003; Laizé and Hannah, 2010) have shown that atmospheric circulation predictors are generally inferior to regional climate predictors (precipitation and temperature).

Hydrologists use rainfall–runoff models to predict flow from precipitation. In 1851, Mulvaney modelled peak discharge from the basin's area and average rainfall intensity. Mulvaney's equation—the first use of the rational method—also includes a scaling parameter (C) to reflect the fact that not all of the rainfall becomes discharge (Beven, 2000). The rational method has been superseded by distributed models that consider the different routes (e.g. overland flow, throughflow, and groundwater flow) by which precipitation reaches the river channel in different areas of the catchment. Antecedent conditions are important in predicting runoff, and accordingly hydrologists attempt to identify the component of the discharge (the baseflow) that would have occurred regardless of recent rainfall.

Rainfall–runoff associations are modified by catchment characteristics (a second-order control). A distinction can be drawn between catchments fed by groundwater and those fed by surface water, the latter exhibiting the greater climate sensitivity (Phillips et al., 2003; Bower et al., 2004). With East Anglia dominated by a chalk aquifer and dry conditions, these catchment controls (e.g. permeability) will have proportionally greater influence in this region than in northern and western Britain (Laizé and Hannah, 2010).

The objectives of the paper are as follows:

  • (i)to document the main patterns of winter rainfall variability in East Anglia at the daily and 2 d timescales;
  • (ii)to identify the driving mechanisms responsible for these patterns in terms of the atmospheric circulation; and
  • (iii)to explore the relationship between the rainfall patterns and river discharge at selected gauging stations at the daily timescale.

The paper is structured as follows. Details of the region's climate and hydrology are provided in Section '2. Study Area'. The datasets and methods used are described in Section '3. Data and Methods'. Results are presented in Section '4. Results' and discussed in Section '5. Discussion' before conclusions are drawn in Section '6. Conclusion'.

2. Study Area

2.1. Climate

East Anglia is comprised of four counties: Cambridgeshire, Norfolk, Suffolk, and Essex. The region is mostly flat, with some parts of the Fenland district of Cambridgeshire and Lincolnshire being nearly 5 m below sea level. The highest area is the East Anglian Heights, which are located to the southeast of Cambridge (maximum elevation = 244 m). Its flat topography and easterly location result in East Anglia being the driest region of the British Isles, with some parts of Cambridgeshire and South Essex recording < 600 mm of rainfall per annum (Glenn, 1987). North Norfolk is the wettest area (Stone, 1983) because it is furthest north and is more exposed to onshore flows from the North Sea that have a longer fetch compared to Suffolk and Essex. North Norfolk also contains the comparatively high ground of the Cromer Ridge.

With limited potential for relief rainfall, most precipitation in East Anglia is caused by depressions or convection. Winter rainfall tends to be frontal in nature (hence lower intensities) compared to the more convectively driven summer regime (hence higher intensities). Rainfall is more evenly distributed between the four seasons than is the case in western Britain where the winter is the wettest season. Across East Anglia, there are about 30 rain days (totals > 1 mm) in winter (DJF) compared to fewer than 25 d in summer (JJA) (Met Office, 2009).

Neal and Phillips (2009) identified five main patterns (as represented by the PCs) of summer rainfall variability over East Anglia at the daily timescale. The most dominant pattern (PC 1) is a gradient between wetter conditions in NE Norfolk and drier conditions in Cambridgeshire and Essex, which is favoured by a cyclonic north to northeast airflow. PC 2 is almost as dominant, and shows a west to east decline in rainfall totals that is typically associated with W-SW flows. In PC 3, the highest totals are observed in Essex, which tends to occur when areas of low pressure follow a southerly track along the English Channel and the southern North Sea. PC 4's wettest area occurs furthest inland in SE Cambridgeshire and West Suffolk, with convectional rainfall being triggered or accentuated by the East Anglian Heights. PC 5 shows the rainfall distribution under a cyclonic N-NW flow, whereby showers generated over The Wash penetrate inland and produce the most rainfall in West Norfolk around King's Lynn and Thetford.

Due to its proximity to mainland Europe, East Anglia has one of the most continental climates in the British Isles (Sumner, 1988). Continental winds from the east have little time to be modified during their short passage over the southern North Sea. An easterly airstream needs to be slow moving to enable it to acquire sufficient moisture from the North Sea before it can be truly classified as being maritime in nature (Bonacina, 1949). During the winter, the region's long coastline has a warming effect on coastal areas (Joshi et al., 2007). The increased temperature is likely to increase evaporation and hence condensation. Simpson (1994) notes how rainfall can be generated in a sea-breeze convergence zone due to onshore winds hitting the land in different directions. This is caused by the curvature of the coastline, which consequently causes turbulent air to rise.

2.2. Hydrology

There are over 500 km of rivers in East Anglia. The region's flat topography ensures the need for an efficient drainage system. Many man-made waterways have been constructed, such as the New Bedford River in the Cambridgeshire Fens, to cope with drainage problems (Hills, 2003). The Fens were fully drained by the end of the 18th Century to prevent prime agricultural land from flooding. In addition to river flooding, storm surges down the North Sea can result in coastal flooding (Svensson and Jones, 2002), such as seen in 1953.

East Anglian runoff (effective rainfall) averages < 25% of the rainfall (Centre for Ecology and Hydrology, 2009). Unsurprisingly, time series of river discharge mirror rainfall series considerably. However, rivers in East Anglia tend to have longer lag times between the rainfall peak and the discharge peak than is seen in other regions of Britain. This is because the area's permeable geology and low gradients do not favour a rapid run-off response. In general, discharge is higher in winter and lower in summer, with inland rivers exhibiting more seasonal variability than those nearer the coast. Hiscock et al. (2001) investigated long-term changes in the discharge of three rivers in Norfolk and whether any changes were driven by rainfall. Over the period 1964–1992, rainfall–runoff modelling showed that the relationship between rainfall and discharge had remained essentially unchanged both on an annual and seasonal basis. They found no evidence of significant changes to the seasonal distribution of river flow.

3. Data and Methods

Daily rainfall totals (millimetres) were obtained from the British Atmospheric Data Centre (BADC; an online repository of British Meteorological Office datasets) for the ten winters (DJF) from 1999–2000 to 2008–2009. This is better than the 5 years used in Neal and Phillips' (2009) study of summer rainfall over East Anglia. There were 173 rain gauges in East Anglia with daily winter records over this period. There were not enough sites to replicate this study at the hourly time resolution. After removing stations with more than 10% missing data, 138 remained. All high rainfall totals (>100 mm) were scrutinized to see whether they were feasible: for example, a total of 198.1 mm at one station was removed because neighbouring sites recorded only 1.3 mm. Four sites were also removed because they were highly correlated (r > 0.95) and very close to other gauges (hence multicollinear), and so would have added little new information about rainfall's variability at the regional scale. This resulted in a data matrix containing daily totals at 134 sites (Figure 1) on 334 d (44 756 data entries). This is a gauge density of 1 per 121.1 km2, which roughly equates to an average of one gauge for every 11 km × 11 km grid square. Of the 134 sites, only four are above 100 m (station numbers 99, 104, 107, and 125 in the East Anglian Heights); 27 gauges are ⩽10 m above sea level. Whilst this network is the best available for the 10 years, the distribution of gauges is far from even. For instance, there is a paucity of gauges in NE Cambridgeshire and West Norfolk near Wisbech and Downham Market, Central and East Suffolk (removing site 78 would create an even larger gap), and Central Essex around Great Dunmow and High Easter. The Suffolk coastline around Southwold is also poorly represented.

Figure 1.

Location of rain gauges and places mentioned in the text. Height of each station (metres) is shown in parentheses

The main patterns of rainfall variability at the daily and 2 d timescales were identified using PCA. The logic of repeating the PCA at the 48-h timescale is that it should provide greater insight into the effect of precipitation on flooding. Probable maximum precipitation is calculated using 2 d totals and this is often used to estimate probable maximum flood (PMF) (Rakhecha and Soman, 1994). The dataset contains sufficient days (the 48-h dataset contains 327 rows) because it is commonly stated that 300 cases is a good sample size for PCA (Comrey and Lee, 1992; Tabachnick and Fidell, 2001). It is worthwhile performing PCA because the Kaiser–Meyer–Olkin measure of sampling adequacy (which varies from 0 to 1) for the daily totals is 0.984, which shows that there are distinct sub-groups of stations that would be represented by the PCs. Furthermore, Bartlett's test of sphericity confirmed strong inter-station correlations between the daily totals (χ2 = 88 073, p < 0.001). After substituting 0.01 mm for days with no rainfall (as log10 of zero is undefined), the rainfall totals were all logged before performing PCA to reduce the degree of positive skew and make the arithmetic mean more representative of the distribution's central tendency. PCA of a single meteorological variable (in this case rainfall) can be performed in S and T modes. S mode was chosen because this identifies common variance amongst the sites rather than the days. The PC axes were rotated orthogonally using the VARIMAX algorithm and the eigenvalue-one rule was used to determine the number of PCs to retain. For S mode PCA, the component loadings show the association between each rain gauge and the PC. Each day is assigned a PC score, with the rainfall pattern represented by the PC being best developed on days with high positive PC scores. On days with high negative scores, the inverse geographic pattern should be evident. Jolliffe (2002) and Field (2005.) provide more details about PCA and the procedural issues discussed above.

The Lamb Weather Type (LWT) (Lamb, 1972.) was obtained for all days with high positive and negative PC scores (thresholds of ± 2 and ± 1.5 were used depending on the PC) to discover whether a particular rainfall pattern is disproportionately caused by one airflow pattern. It is important to acknowledge that the LWTs are indicative of the weather situation over the British Isles as a whole and may not always be representative of the circulation over East Anglia. As the rain gauges are emptied at 0900 GMT, the previous day's LWT at 1200 GMT was always used as an indicator of that day's atmospheric circulation. Clearly, it would have made no sense to relate the rainfall totals to the atmospheric circulation 3 h after all the gauges had been emptied. Daily synoptic charts were also studied to understand the driving mechanisms responsible for each rainfall pattern.

Finally, CA (Davis 1986) was used to partition East Anglia into regions that show similar variations in daily rainfall over time. In other words, the stations in one cluster should all tend to be wet or dry on the same day, and show variations over time that are different from the stations in the other clusters. The component loadings of PCs with eigenvalues > 1 were entered into Ward's agglomerative hierarchical clustering method. The squared Euclidean distance was used as the similarity measure.

Mean daily river discharge values (Q) were abstracted from the National River Flow Archive (NRFA) on the Centre for Ecology and Hydrology's (CEH) website. There were 11 gauging stations with available data in East Anglia (Figure 2 and Table 1). When the analyses were performed in summer 2009, river discharge data were only available for the seven winters up until 2005/2006. Only three of these basins are classified by the NRFA as ‘benchmark catchments’ with near-natural regimes; thus, human influences have a marked influence on river flow in the other eight catchments. In this paper, discharge will be compared only to the 1 d precipitation patterns identified by PCA. It would have been more difficult to isolate a clear relationship between daily Q and the 48-h rainfall patterns because of the mismatch in the temporal resolutions. The dependence of discharge on rainfall was quantified using multiple linear regression (MLR) (Field, 2005.) and cross-correlation functions (CCF). Both techniques will enable differences to be identified in the driving mechanisms that link precipitation to discharge. First, stepwise regression equations were developed to predict river discharge from the best combination of rainfall patterns. The input variables into the models were the PC scores of those PCs with eigenvalues > 1. Second, CCFs were calculated between discharge and rainfall (the average of the five rain gauges closest to the gauging station was used) to quantify lead and lag times between the two variables. A separate CCF was derived for each winter. This is because this study uses winter data only, which means that the time series is not continuous. Clearly, the last day of February cannot be treated as the day before 1 December. Like the standard Pearson correlation coefficient, CCF assumes that the variables are normally distributed. Hence, the highly skewed daily rainfall totals (millimetres) and discharge values were logged to the base ten before calculating the CCF. For MLR, the PC scores are already normally distributed with a mean of zero and so do not require transformation.

Figure 2.

Location of river gauging stations

Table 1. Details of the rivers used in this study
River and gauging station nameLength of river (km)Altitude of station (m)Catchment size (km2)Benchmark catchment?EGIRPCatchment rock type
  1. The maximum altitude of the rivers varies from 65 to 145 m. Human influences in each catchment are shown by crosses: E, runoff increased by effluent returns; G, runoff influenced by groundwater abstraction and/or recharge; I, runoff reduced by industrial and/or agricultural abstraction; R, regulation from surface water and/or ground water; P, runoff reduced by public water supply abstraction, N/A, not available.

(1) Babingley at Castle Rising19.64.547.7NoXXX80% chalk, 20% boulder clay
(2) Bure at Ingworth80.012.2164.7NoXX55% boulder clay, 45% sand/gravel
(3) Stringside at WhitebridgeN/A2.598.8YesXX80% chalk, 20% boulder clay
(4) Little Ouse at Abbey Heath60.07.2688.5NoXXX80% chalk, 20% boulder clay
(5) Waveney at Needham MillN/A16.5370.0NoXX10% chalk, 90% boulder clay
(6) Kym at Meagre FarmN/A17.2137.5YesXX100% clay (very flashy by Anglian standards)
(7) Cam at Dernford64.014.7198.0NoXXX50% chalk, 50% boulder clay
(8) Gipping at Stowmarket22.125.1128.9NoXXX5% clay and chalk, 95% boulder clay
(9) Stour at Langham76.06.4578.0NoXXXX25% chalk (north), 75% London clay
(10) Colne at Lexden11.08.2238.2YesXXX70% London clay, 30% sand/gravel
(11) Chelmer at Springfield20.023.1190.3NoXXX75% London clay, 25% boulder clay

4. Results

4.1. Patterns of rainfall variability

When the 1- and 2 d totals were subjected to PCA, there were six PCs with eigenvalues > 1 (Table 2). The first three PCs collectively account for more than 80% of the variance and so will be considered in greater detail.

Table 2. The eigenvalue and explained variance of the most important PCs of the daily and 2 d rainfall totals
PCEigenvaluePercentage explained varianceCumulative percentage explained variance
Daily totals
PC 140.129.929.9
PC 238.628.858.7
PC 328.921.580.2
PC 44.93.783.9
Two-day totals
PC 144.333.133.1
PC 240.730.363.4
PC 324.918.682.0

4.1.1. Daily totals

PC 1 explains 29.9% of the variance in winter precipitation variability. PC 1's spatial eigenvector (Figure 3(a)) shows a clear gradient between wetter conditions in southern Essex and drier conditions in NW Norfolk. The five stations with the highest PC 1 loadings (101, 115, 124, 127, and 134: Figure 1) are found in South Essex and all averaged at least 4.6 mm across all days with high PC 1 scores (≥1.5). By contrast, the five stations in NW Norfolk (46, 48, 50. 58, and 61) with the lowest PC 1 loadings averaged < 1.6 mm across all days with high PC 1 scores. From examining the LWT codes (Table 3) for days with high PC 1 scores (≥1.5), 19 of the 25 d had either a directional westerly (LWT code = 16) or experienced pure cyclonic conditions (LWT code = 20). On most days with high PC 1 scores, low pressure is positioned directly over Britain or just to the north. An example of two consecutive days with high PC 1 scores is 3 and 4 January 2001, which saw several areas of low pressure and troughs move from west to east over Britain. The parent low was positioned off the west coast of Scotland. The fact that a daughter depression developed over southern Britain on the parent low's trailing cold front is likely to explain why Essex was the wettest county on these 2 d.

Figure 3.

The first three PCs of daily rainfall over East Anglia: (a) PC 1, (b) PC 2, and (c) PC 3

Table 3. The percentage frequency of occurrence of each LWT for days with high PC scores (≥1.5 for PCs 1, 2, and 3; ≥ 2 for PCs 4, 5, and 6) obtained from the PCA of the daily rainfall totals
  PC 1PC 2PC 3PC 4PC 5PC 6
  1. For simplicity, the anticyclonic and cyclonic types were categorized into two larger groups. A directional weather type is defined as a situation when neither high nor low pressure is more dominant over the British Isles on that particular day. The most frequently occurring LWTs for each PC are shown in bold font.

0–8Anticyclonic (all wind directions)842165008
20–28Cyclonic (all wind directions)362007388
11Directional (NE)4260707
12Directional (East)040007
13Directional (SE)0000120
14Directional (South)4040127
15Directional (SW)40200130
16Directional (West)400320021
17Directional (NW)4024362521
18Directional (North)0400014
− 1Unclassified044007

PC 2 explains nearly as much variance as PC 1 (28.8% vs 29.9%). Under PC 2's pattern (Figure 3(b)), NE Norfolk receives the highest totals and western Cambridgeshire and southern Essex record the lowest. Across the 19 d with high PC 2 scores (≥1.5), precipitation at the five stations with the highest (lowest) PC 2 loadings averaged 4.5 mm (0.3 mm). PC 2's pattern tends to occur in northeasterly airflows with anticyclonic vorticity; 68% of days with high PC 2 scores had LWTs that were anticyclonic, basic northeasterly, or a combination of the two (Table 3). An example of two consecutive days with high PC 2 scores is 19 and 20 February 2005. During these days, the wind veers from a northwesterly to a northerly, as low-pressure areas move south over the Norwegian Sea and North Sea and the Azores High extends a ridge northwards to Iceland. This pressure distribution results in an N-NE flow over East Anglia, with NE Norfolk being most exposed to the onshore wind.

PC 3 (explained variance = 21.5%) shows a clear gradient between higher rainfall totals in NW Cambridgeshire; and lower totals in East Norfolk, and along the coast of Suffolk and Essex (Figure 3(c)). Across the 25 d with high PC 3 scores (≥1.5), rainfall averaged 3.5 mm (1.1 mm) at the five stations with the highest (lowest) PC 3 loadings. PC 3's pattern was best developed on 19 January 2003. At the five stations with the highest PC 3 loadings, mean rainfall over NW Cambridgeshire around Peterborough was 15.8 mm on that day compared to only 5.7 mm at the five stations with the lowest loadings (⩽0.3) along the east coast near Lowestoft. Nineteen of these 25 d were confined to three LWTs: directional SW, W, and NW (Table 3). The remaining 6 d saw mainly pure anticyclonic conditions (LWT code = 0) with occasional light southerly winds (LWT code = 14). An example of a day with a high PC 3 score is 27 December 1999, which saw strong northwesterlies over the eastern Atlantic and the British Isles. In this flow, a shallow low positioned over the Atlantic (45°N, 40°W) at 1200 GMT on 26 December tracked rapidly eastwards and deepened (984 hPa) to be located off the coast of Brittany in northern France by 1200 GMT on 27 December. This active zonal regime was seen on most days with high PC 3 scores. In this situation, western Cambridgeshire is likely to be wetter than eastern parts of East Anglia because areas of low pressure generally become less intense as they track from west to east over Britain. The airmass will also dry out as it passes over the land.

The main properties of PC 4, PC 5, and PC 6 are summarized in Table 4. These three PCs are much less important than PCs 1–3 (Table 2), with even PC 4 accounting for < 4 % of the variance. Whereas PCs 1–3 show clear trend surfaces, the component loading maps for PCs 4–6 (Figure 4) all exhibit concentric zone patterns in which there are localized rainfall maxima in different parts of East Anglia.

Figure 4.

PCs of daily rainfall over East Anglia: (a) PC 4, (b) PC 5, and (c) PC 6

Table 4. The main properties of PCs 4–6 of daily winter rainfall over East Anglia
 Wettest areaRainfall average (high PC loadings)Rainfall average (low PC loadings)Atmospheric circulation
  1. The rainfall averages (millimetres) were calculated using the five stations with the highest PC loadings and the five stations with the lowest PC loadings across all days with high PC scores (≥2). Also, refer to eigenvalues and explained variance statistics in Table 2 and the LWTs in Table 3.

PC 4East Anglian Heights (near Newmarket and Bury St. Edmunds)2.40.6 (South Essex)86% of days were directional NW and anticyclonic LWT Unstable air flow pattern Low pressure close to East Anglia triggers localized shower activity in inland areas
PC 5On the border between SE Suffolk and NE Essex (Felixstowe, Ipswich, and Colchester area). Also, a wetter corridor up through Suffolk and western Norfolk to The Wash1.20.4 (NE Norfolk and East Cambridgeshire)63% of days were directional NW and cyclonic NW with low pressure over Scandinavia. Agrees with the fact that wetter conditions were aligned in an NW-SE corridor Generally from the SE on other days
PC 6NW Norfolk near The Wash (King's Lynn and Hunstanton)1.50.3 (East Suffolk)Very variable, although a slight majority (8 of 14 d) had directional W, NW, and N NW Norfolk is most exposed to onshore wind. A combination of showery activity generated over The Wash and fronts weakening as they move SE across East Anglia because of high pressure to the south of Britain

4.1.2. Two-day totals

The first six PCs explained 88.3% of variance in the 2 d totals (Table 2). The component loading maps for PCs 1–4 of the 2 d totals (not shown) are almost identical to the daily patterns. Hence, the four main modes of precipitation variability do not change significantly when the timescale is changed from 24 to 48 h. There are, nonetheless, subtle differences between some of the daily and 2 d maps. For PC 1, higher component loadings occur around The Wash at the 2 d timescale. This suggests that this area is less likely to be dry when subjected to PC 1's synoptic forcing mechanisms over 48 h. For PC 2, the wettest region (NE Norfolk) is more extensive at the 2 d timescale. PCs 5 and 6 swap around between the two timescales, with higher totals around The Wash (PC 6 of the daily totals) now being more likely over 2 d than higher totals on the border between SE Suffolk and NE Essex (PC 5 of the daily totals).

4.2. Rainfall regions

The regions of coherent precipitation variability derived by CA of daily totals are shown in Figure 5. In order to assess the relative importance of the PCs in producing rainfall in each region, regional precipitation averages were calculated on the 2 d with the highest scores on each PC (Figure 6). Region one (West and Central Cambridgeshire) is wettest under PC 3's pattern. The unstable westerly flows and Atlantic depressions that are dominant under PC 3 provide this area with wetter conditions due to its location in the far west of East Anglia. Neal and Phillips (2009) also found the most westerly cluster in summer to be wettest under the advance of eastward moving low-pressure systems. Region two is landlocked and includes parts of all four counties; a similar region was seen on the summer rainfall classification. Region three (East Cambridge and West Norfolk) receives comparatively high amounts of rainfall across many of the PCs. The highest rainfall in this region occurs on the day with the second highest PC 3 score (1.46 mm) and the highest PC 6 score (1.44 mm). This is logical because 32 and 24% of days with high PC 3 scores (≥1.5) experienced westerly and northwesterly winds, respectively (Table 4). The location of region three could indicate that high levels of precipitation are caused by these airflows gaining moisture from The Wash. This is also seen in the localized shower activity regime described by PC 6. Region four (Central and Northern Norfolk) is almost identical to Neal and Phillips' (2009) summer region four, with onshore northeasterlies (e.g. 20 February 2005) producing the most rainfall (winter PC 2 and summer PC 1). This flow pattern also produces the most rainfall in region five (SE Norfolk and NE Suffolk), which is also exposed to onshore winds from the North Sea. The only day when rainfall in region six (Central and East Essex and SE Suffolk) exceeded 5 mm was 9 February 2001, which was the day with the highest PC 1 score (5.66 mm). This is logical because PC 1 shows a precipitation maximum in southern Essex. Region six is comparable to region seven of Neal and Phillips' (2009) summer classification.

Figure 5.

Daily precipitation regions

Figure 6.

Arithmetic mean daily rainfall (millimetres) in each region on the days with the highest and second highest score on each PC

4.3. River flow's dependence on rainfall

The 95th percentile of winter daily discharge was calculated at each gauging station. For eight of the 11 sites, the 95th percentile was exceeded most frequently in the winter of 2001–2002, which was the wettest of the seven winters considered in this study. This preliminary finding demonstrates a clear association between rainfall and discharge, which is worth exploring further in stepwise multiple regression models whereby the discharge is expressed as a linear function of the rainfall patterns (Table 5). For each river, CCFs were calculated between discharge and rainfall (Table 6) for each winter in turn (Section '3. Data and Methods'). The autocorrelation function (ACF) of each river's discharge was also calculated to determine the persistence of the flow (Table 7). The river flow results are analysed in Section '5. Discussion'.

Table 5. Stepwise MLR models of river discharge from the daily rainfall patterns
RiverFr2 (%)Regression equation
BabingleyNo model
Bure39.028.3Log10Q = 0.23 + 0.05PC 2 + 0.03PC 1 + 0.03PC 3
Stringside9.28.5Log10Q = − 0.01 + 0.04PC 1 + 0.03PC 3 + 0.03PC 2
Little Ouse18.19.3Log10Q = 0.83 + 0.06PC 2
Waveney30.834.4Log10Q = 0.53 + 0.13PC 1 + 0.12PC 2 + 0.08PC 3 + 0.04PC 5 − 0.03PC 6
Kym39.434.7Log10Q = − 0.15 + 0.2PC 1 + 0.18PC 3 − 0.09PC 4 + 0.09PC 2
Cam14.813.0Log10Q = 0.18 + 0.09PC 1 + 0.05PC 3 + 0.03PC 2
Gipping51.846.8Log10Q = 0.06 + 0.17PC 1 + 0.12PC 2 + 0.11PC 3 + 0.04PC 5 − 0.04PC 4
Stour28.527.8Log10Q = 0.73 + 0.13PC 1 + 0.07PC 2 + 0.07PC 3 − 0.04PC 4
Colne35.632.5Log10Q = 0.28 + 0.15PC 1 + 0.08PC 3 + 0.06PC 2 − 0.05PC 4
Chelmer30.629.2Log10Q = 0.24 + 0.14PC 1 + 0.08PC 3 + 0.05PC 2 − 0.04PC 4
Table 6. CCFs between daily discharge and rainfall
Discharge station number and river nameLag numbers by which the coefficients for each winter were significant
  1. A negative (positive) lag indicates that discharge lags (leads) rainfall. Only those lags with statistically significant correlations are listed, with the strongest correlation shown in bold font. With the exception of the River Babingley, all statistically significant (p < 0.05) correlations were positive. The sign of the Babingley correlations in each year is shown.

(1) Babingley3, 4, 5, 6, 7 (negative)0, − 1, − 2, − 3, − 4 (positive)1, 0, − 1, − 2, − 3, − 4, − 5, − 6, − 7 (positive)Not sig.1, 0, − 1, − 2 (positive)2, 3, 4, 5, 6, 7 (negative)4, 3, 2, 1, 0, − 1, − 2, − 3, − 4, − 5, − 6 (positive)
(2) Bure1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 42, 1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 22, 1, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 4, − 5
(3) Stringside0, − 11, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 31, 0, − 1, − 2, − 30, − 12, 1, 0, − 1, − 2, − 3, − 4
(4) Little Ouse0, − 1, − 2, − 3Not significantNot significant1, 0, − 1, − 2, − 3, − 4, − 50, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 4, − 5, − 6, − 70, − 1, − 2, − 3, − 4, − 5, − 6
(5) Waveney1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 43, 2, 1, 0, 1, − 2, − 3, − 4, − 51, 0, 1, − 2, − 3, − 41, 0, 1, − 2, − 3, − 4, − 5, − 60, 3
(6) Kym0, − 11, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 4, − 52, 1, 0, − 1, − 2, − 31, 0, − 1, − 2, − 30, − 1, − 2, − 30, − 1, 2, − 3, − 4, − 5
(7) Cam0, − 11, 0, − 1, − 2, − 3, − 40, − 1, − 51, 0, − 1, − 2, − 31, 0, − 1, − 2, − 30, − 1, − 2, − 30, − 3, − 4
(8) Gipping1, 0, − 1, − 2, − 31, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 42, 1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3,1, 0, − 1, − 2, − 3, − 4, − 52, 1, 0, − 1, − 2, − 3, − 4
(9) Stour0, − 11, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 4, − 52, 1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 40, − 1, − 2, − 30, − 1
(10) Colne0, − 1, − 21, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 40, − 1, − 2, − 30, − 1, − 2, − 3, − 4
(11) Chelmer0, − 1, − 22, 1, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 3, − 4, − 51, 0, − 1, − 2, − 3, − 41, 0, − 1, − 2, − 30, − 1, − 21, 0, − 1, − 2, − 3, − 4
Table 7. Number of statistically significant lags for each river's daily discharge time series
Discharge station number and river nameNumber of days discharge can be predicted into the future for each winterMean
  1. The maximum number of lags was set at 16. This is because as the lag time increases the sample size used to calculate the autocorrelation coefficient decreases.

(1) Babingley111111161651212
(2) Bure134584847
(3) Stringside8410716758
(4) Little Ouse61016869
(5) Waveney546810726
(6) Kym4412714357
(7) Cam3447151558
(8) Gipping565710666
(9) Stour454564116
(10) Colne74968356
(11) Chelmer441048566

5. Discussion

5.1. Rainfall patterns

The variance explained by the first six PCs of daily winter rainfall totals for East Anglia (86.0%) is almost identical to the percentage explained (86.3%) for the seven statistically significant PCs identified for summer rainfall totals in the same region (Neal and Phillips, 2009). These are both marginally higher percentages than the 83.4% explained by the first six PCs of winter rainfall in Southwest England (Phillips and Denning, 2007). Similar synoptic situations were observed between many of the PCs outlined in this paper and those discovered by Neal and Phillips (2009).

For winter daily totals, PC 1 shows a gradient between wetter conditions in southern Essex and drier conditions in NW Norfolk. This mode of variability is more important in the winter because Neal and Phillips (2009) found this to be only the third most important pattern (PC 3) during the summer. Winter PC 1 is usually generated by maritime airmasses, with 40% of days with high PC scores (≥1.5) having a directional westerly flow (Table 3). The first PC of precipitation over Wales (Bonell and Sumner, 1992) and Southwest England (Phillips and Denning, 2007) is also explained by the same active zonal regime.

With the wettest conditions in NE Norfolk, winter PC 2 is like summer's PC 1 pattern (Neal and Phillips, 2009). Winter PC 2 is generated disproportionately by onshore NE winds (Table 3), which given the order of the PCs must be a less important generator of rainfall in winter than in summer. The increased exposure of North Norfolk to winds from the North Sea was noted by Stone (1983).

For winter PC 3, rainfall totals show a west to east decline in rainfall, with the wettest conditions being in NW Cambridgeshire and the driest conditions along the east coast. PC 3 typically occurs under a westerly wind, with the air becoming drier as it passes over land. This geographic pattern is the same as PC 2 of summer rainfall (Neal and Phillips, 2009), which means this pattern is less important in winter than in summer. This is surprising because eastward-moving depressions are the main cause of winter precipitation as opposed to convection in summer. On the other hand, winter PCs 1 and 3 are both maritime PCs, and tend to occur under westerly flows. Despite this apparent similarity, the two winter PCs have different rainfall distributions because the depression track is further north in PC 3.

Winter PC 4 is almost identical to summer PC 4 (Neal and Phillips, 2009), with the highest rainfall totals occurring in SE Cambridgeshire and West Suffolk near Bury St. Edmunds and Newmarket. This precipitation maximum coincides with the East Anglian Heights, with high PC 4 loading stations all having relatively high altitudes (42–76 m) compared to East Anglia as a whole. PC 4 mostly occurs under anticyclonic situations with light W-NW airflows (Table 3). The fact that rainfall totals vary markedly over short distances in this area on days with high PC 4 scores (e.g. 2.9–12.0 mm on 6 February 2004) suggests this component is showing localized shower activity over the East Anglian Heights. PC 4 explains less variance in winter (3.7%) than in summer (7.0%), which is consistent with PC 4 being partly a convectively driven pattern. The sea-breeze convergence effect would also be weaker in the winter.

Winter PC 5 explains 1.1% of the variance, and combines elements of summer PCs 6 and 7 (Neal and Phillips, 2009). Unlike PCs 1–4, rainfall totals do not show a clear trend surface; thus, considerable mesoscale variability must be superimposed on PC 5. Nonetheless, a slightly wetter NW-SE corridor from The Wash to South Suffolk can be identified, which can be explained by NW winds being relatively frequent on days with high PC 5 scores (Table 3).

In winter PC 6, the rainfall is concentrated in NW Norfolk around The Wash. This is the same as Neal and Phillips' (2009) PC 5 summer pattern. In general, this pattern is most likely to occur in a W-NW airstream with showers developing over The Wash in the unstable onshore airstream and moving inland to produce slightly higher rainfall totals in the area between King's Lynn and Thetford.

The first three PCs derived from the 48-h totals are almost identical to the daily PCs. Given that PMF is often estimated from 2 d totals, this implies that flooding during the winter is most likely to occur in South Essex (PC 1), NE Norfolk (PC 2), and NW Cambridgeshire (PC 3). It must be stressed, however, that high rainfall does not necessarily relate directly to flood risk potential. It is merely one factor because the areas of the region with the lowest elevation (e.g. the Norfolk/Suffolk Broads, the Fens) are vulnerable to coastal inundation as well.

5.2. Relationship between rainfall and discharge

5.2.1. The dependence of river flow on the rainfall patterns

The first PC of rainfall is the best predictor of discharge for 8 of the 11 rivers (Table 5). The first three PCs appear in far more regression models (PC 1 = 10 times, PC 2 = 9, and PC 3 = 9) than PCs 4–6 (PC 4 = 5 times, PC 5 = 2 times, and PC 6 = 1 time).

The Babingley is a short (19.6 km) river in West Norfolk near The Wash with a boggy, flat catchment. The river's discharge is not significantly related to any of the PCs (Table 5), which suggests that the lag effect caused by the chalk aquifer has a much greater influence on discharge than the main concurrent patterns of precipitation variability.

The Bure's discharge is most strongly associated with PC 2, and then PC 1 and PC 3. It is logical that PC 2 is the best predictor because this PC produces the most rainfall in NE Norfolk where this river is located. Hiscock et al. (2001) found that the relationship between climate and the Bure's flow was stable between 1964 and 1992, which suggests that the river's association with PCs 1, 2, and 3 is likely to have remained unchanged over a longer period of time.

The Stringside, a tributary of the Ouse near the border of Cambridgeshire and Norfolk, is significantly associated with the first three PCs. However, the explained variance is low (8.5%) because the dominance of chalk in the basin's geology (80%) markedly weakens the association between concurrent rainfall and discharge. The Little Ouse is another chalk-dominated catchment and so it is logical that its regression model explains little of the variance (9.3%). PC 2 is the only significant predictor of the Little Ouse's discharge.

The Waveney and Gipping catchments in East Suffolk are mainly boulder clay and are the only two rivers whose discharge is related to five of the PCs. PC 1, PC 2, and PC 3 (selected in that order) are the best predictors of the two rivers' discharge. The explained variance is thus high for both rivers (Waveney = 34.4% and Gipping = 46.8%); indeed, the Gipping has the highest r2 of the 11 rivers.

The flow of the Kym (a tributary of the Bedford Ouse) is significantly associated with the first four PCs (r2 = 34.7%), with PCs 1 and 3 being the best predictors. PC 2 only appears as the fourth variable in the model because the river is in western Cambridgeshire far away from the NE Norfolk precipitation maximum represented by PC 2.

The Cam's discharge is significantly related with PCs 1, 3, and 2 in that order. However, the explained variance is low (r2 = 13.0%) because the catchment is 50% chalk, which acts to weaken the instantaneous rainfall–runoff association.

The Stour, Colne, and Chelmer are all located in eastern Essex, and have regression models with the same predictor variables selected in the same order (PC 1, PC 2, PC 3, and PC 4). It is logical that PC 1 appears first because this PC has a rainfall maximum in southern Essex. The discharge of the three rivers is negatively related to PC 4, which indicates that high rainfall over the East Anglian Heights is associated with lower discharge in these rivers.

Apart from the geology, there are other reasons why the regression models explain < 50% of the variance in discharge. Primarily, the models do not consider antecedent conditions and human influences.

The relationship between rainfall and runoff is usually accepted to be nonlinear because more runoff will be generated when the catchment is wet before the event (Beven, 2000). However, the rainfall–runoff models derived in this paper treat discharge solely as a linear function of the concurrent precipitation patterns. Snowfall leads to a delay in the runoff response, but this is unlikely to be a significant factor over this study period. This is because the seven winters from 1999/2000 to 2005/2006 saw very little snowfall. For example, at Lowestoft on the Suffolk coast, there were four or fewer days per month with snow lying on the ground at 0900 GMT; the town's mean maximum temperature was also above the 1961–1990 monthly average in 19 of 21 months. If the study period had included several snowy winters, then it is likely that temperature would have to be used alongside precipitation as a predictor of river discharge.

Other reasons for the models' low explanatory power include spatial and temporal mismatches between the rainfall and discharge data. In this study, discharge was related to the rainfall average of the five nearest gauges rather than all the gauges upstream of the site; and the daily discharge mean was related to a 24-h precipitation total that was offset by 9 h from the calendar day (i.e. 0900 GMT to 0900 GMT on the following day). With East Anglian catchments being small (<600 km2) by global standards, better fitting regression models might have been obtained if sub-daily data had been used. It is possible that the discharge peak occurs so quickly in the smaller catchments that it is not adequately modelled using only daily values.

With only three of the basins being benchmark catchments, human influences will markedly modulate and weaken the rainfall–runoff response. In all catchments, discharge is reduced because of abstractions for industrial and agricultural uses (Table 1). Runoff is increased by groundwater abstraction and/or recharge in 8 of the 11 basins; effluent returns have a significant effect on discharge in 5 basins. Public water supply abstraction is only important for the three rivers (Stour, Colne, and Chelmer) that have predominantly London Clay catchments (Table 1).

5.2.2. Leads and lags between rainfall and discharge

For seven of the rivers (Bure, Stringside, Waveney, Gipping, Stour, Colne, and Chelmer), today's discharge is most strongly related to precipitation that fell 4 d ago (Table 6). In other words, a rainfall peak is most likely to produce a discharge peak 4 d later. The Cam and Kym have comparatively shorter lag times, indicating that their discharge responds more quickly to rainfall: for instance, in the flashy Kym catchment (90% clay), today's rainfall is the best indicator of discharge in four of the seven winters.

The Babingley's CCF varies between winters making it difficult to generalize. For example, the river's discharge lagged 3 d behind the rainfall in the winter of 2000–2001 (a logical finding) but was illogically four (seven) days ahead of it in the winters of 1999–2000 (2003–2004). Furthermore, in the winters 1999–2000 and 2003–2004, the cross correlations are negative, which is counter-intuitive because high discharge would not usually be associated with low rainfall. The discharge of the Babingley is affected by effluent returns, groundwater abstraction and/or recharge, and abstractions for industrial and/or agricultural uses (Table 1). The fact that the Babingley's discharge could not be predicted from the PC scores (Table 5) further suggests that the river's discharge shows almost no predictable relationship with rainfall.

The highest CCF observed across all rivers and lead/lag times was for the Little Ouse in the winter of 2004–2005 (r = 0.6). In this winter, discharge could be predicted up to 7 d in advance from the rainfall. In general, statistically significant correlations extend to a greater number of lags for the Little Ouse than for other rivers. Thus, the Little Ouse with its 80% chalk catchment can be said to have the most steady flow and relationship with precipitation (as a driver of discharge) than any other East Anglian river.

5.2.3. Persistence of discharge

ACFs show that discharge (logarithm to the base ten) persistence is typically between 6 and 8 d (Table 7). The River Babingley has a persistence of almost 2 weeks (12 d), indicating that its discharge shows less variability at the daily timescale than the other rivers. The Little Ouse's discharge has the second longest persistence time (9 d on average). This indicates that today's discharge can be estimated from its value 9 d ago. The rivers in Essex and southern Suffolk (Gipping, Stour, Colne, and Chelmer) all have a mean persistence of 6 d. The rivers Bure, Stringside, Kym, and Cam all show slightly longer persistence times (7 or 8 d). The persistence times for all the East Anglian rivers (around 1 week) are much higher than for rivers in northern and western Britain (Ward and Robinson, 2000). This is explained by East Anglia's easterly location within the British Isles, together with its subdued topography, which results in the region being less prone to high rainfall totals and intensities in the winter. Consequently, river flows fluctuate less from day to day in East Anglia compared to rivers in western and northern Britain.

It should be remembered that this paper has considered precipitation–discharge relationships over seven winters only. It would be desirable to replicate these analyses over a longer period (>30 years) to ensure that extremes in precipitation and discharge are adequately captured. However, even if a longer period was used, it would not resolve the potential issue of non-stationarity in rainfall–runoff associations. Whilst Hiscock et al. (2001) found the relationship between climate and the Bure's flow to be stable between 1964 and 1992, rainfall–runoff associations are not always stationary over time, whereby one unit of precipitation input produces the same discharge output. Causes of non-stationarity include changes in catchment land use (e.g. level of urbanization) and installing field drainage.

6. Conclusion

Winter precipitation variability over East Anglia was explained by six PCs for 24- and 48-h totals. The first three PCs cumulatively explain 80.2% (82.0%) of the variance in daily (2 d totals), and display precipitation maxima in South Essex (PC 1), NE Norfolk (PC 2), and NW Cambridgeshire (PC 3). Whilst the same three most important geographic patterns were identified in winter and summer (Neal and Phillips, 2009), their relative importance differs between the two seasons. In particular, the rainfall maximum in South Essex explains more of the daily variance in winter (29.9%) than in summer (20.9%). For the winter, the three most important patterns of variability remain unchanged when the timescale is changed from 24 to 48 h. Stepwise regression models, in which river flow is expressed as a linear function of the rainfall patterns, were able to explain 24.0% of the variance in discharge when averaged across the 11 rivers. The Babingley's discharge cannot be modelled from the rainfall patterns nor does it show a predictable relationship with antecedent rainfall. For most rivers, there are significant relationships between discharge and precipitation that fell up to 4 d ago on average. In general, East Anglian river flows show persistence over 6–8 d.

Whilst it is instructive to compare the relative importance of the PC patterns between winter and summer, it should be remembered that the winter PCs were derived over 10 years (2000–2009) compared to only 5 years (1994–1998) for Neal and Phillips' (2009) summer PCs. Nonetheless, rainfall displays higher spatial coherence in the winter, with the first three PCs of the daily totals explaining 80.2% of the variance compared to 72.3% in the summer. The period for the winter was chosen to make the study as contemporary as possible, whilst ensuring that the PCs generalize better to the population than the summer PCs. Despite all available Meteorological Office rain gauges being used, their distribution was not perfect, which may have led to minor errors in the mapping of the PCs.

The research presented in this paper could be expanded in the following ways. First, as daily and 2 d totals are not always reliable indicators of precipitation intensity, hourly totals from 28 sites in East Anglia could be investigated. Second, a T-mode PCA could be employed to group together those days with similar spatial rainfall distributions. Third, precipitation variability in the transition seasons of spring and autumn could be investigated. Last, temperature could be entered into the regression models alongside the rainfall patterns to explain more of the variance in the discharge.

This paper has successfully identified the main patterns of winter rainfall variability over East Anglia and has shown clear linkages to river discharge.


The authors would like to thank the British Atmospheric Data Centre (BADC) and the Centre for Ecology and Hydrology (CEH) for providing us with access to the precipitation and river flow data, respectively. We would like to thank Mrs Anne Ankcorn, Drawing Office, School of Geography, University of Birmingham for all the figures.