By continuing to browse this site you agree to us using cookies as described in About Cookies
Wiley Online Library is migrating to a new platform powered by Atypon, the leading provider of scholarly publishing platforms. The new Wiley Online Library will be migrated over the weekend of March 17 & 18. You should not experience any issues or loss of access during this time. For more information, please visit our migration page: http://www.wileyactual.com/WOLMigration/
Significant increases in precipitation have been observed in The Netherlands over the last century. At the same time persistent spatial variations are apparent. The objective of this study is to analyse and explain these spatial patterns, focussing on changes in means and extremes for the period 1951–2009. To investigate different possibilities for the causes of spatial variations, a distinction was made between six regions based on mean precipitation, soil type and elevation, and four zones at different distances to the coast. Spatial maxima in mean precipitation inland and over elevated areas are mainly formed in winter and spring, while maxima along the coast are generated in autumn. Daily precipitation maxima are found in the central West coast and over elevated areas. Upward trends in daily precipitation are highest from February to April and lowest from July to September. The strongest and most significant increases are found along the coast. For several seasonal and climatological periods diverging behaviour between coastal and inland zones is observed. We find that distance to the coast gives a more consistent picture for the seasonal precipitation changes than a classification based on surface characteristics. Therefore, from the investigated surface factors, we consider sea surface temperature to have the largest influence on precipitation in The Netherlands.
Assessment and estimation of weather and climate extremes contributes to our understanding of the changing climate (e.g. Klein Tank et al., 2009) and studies analysing trends and especially extremes are numerous. For precipitation some recent European studies (Anagnostopoulou and Tolika, 2012; Burauskaite-Harju et al., 2012; Karagiannidis et al., 2012) find opposing trends in extreme precipitation. This is no surprise because, trends can be even of opposing directions for different regions within a country (del Rio et al., 2011; Lupikasza et al., 2011) and obviously within Europe as a whole (e.g. Klein Tank and Konnen, 2003). Furthermore, trends are often dependent on the time period examined (Turco and Llasat, 2011; Brunetti et al., 2012) and care has to be taken when results based on different periods are compared.
The most recent precipitation trend analysis for The Netherlands has been performed by Buishand et al. (2013). They determined annual precipitation amounts, precipitation amounts in winter and summer halves of the year, number of days per year with a precipitation amount greater than 20 and 30 mm, and the 5-d annual maximum precipitation. All six indices show increasing trends, the strongest of which were found in winter precipitation and the number of days with more than 20 and 30 mm rain. The mean exceedance frequency of the 30 mm threshold for the wettest parts of the country was found to be about twice as large as that for the driest part and shows a relatively strong increase from the beginning of the 1980s. Overall, the largest changes were found in the coastal area.
Especially for extremes, the achieved results are strongly dependent on the chosen index (Ustrnul et al., 2012). Ustrnul et al. show the advantages and disadvantages of different methods used to identify extremes in temperature and precipitation and their influence on long-term variability and trends. They found the method of percentiles to be the most suitable for spatial analysis. Percentiles split a set of ordered data into hundredths, so 90% of the data should fall below the 90th percentile. Percentile values are site-specific because they sample the same part of the distribution at each station while the value over threshold approach can favour sites at specific locations where events with high precipitation amounts occur more frequently.
This study will extend the work of Buishand et al. (2013), who investigated threshold exceedance frequencies, with an analysis of percentiles and focus on regional differences in precipitation changes for The Netherlands. Our objective is to analyse and explain persistent spatial patterns in relation to several factors, which will be described in more detail below. Soil type, sea surface temperature (SST), topography, and urbanization have been investigated. We choose these, because (1) soil moisture has a large impact on the energy partitioning between latent (moisture) and sensible heat fluxes at the land surface (Seneviratne et al., 2010) and therefore has the potential to influence precipitation; (2) a large fraction of The Netherlands is located along the relatively shallow North Sea and the influence of SST on precipitation has been investigated and proven for The Netherlands (Lenderink et al., 2009); (3) even though elevation differences in The Netherlands are small, they have been shown to affect local precipitation (ter Maat et al., 2012); and (4) the influence of urbanization on precipitation is receiving more and more attention (Shepherd, 2005; Trusilova et al., 2009) and might be significant.
Soil type and soil moisture are directly related to each other through the differences in water holding capacity and pore size. Although soil moisture-precipitation feedbacks are usually found at large (i.e. continental or global) scales, the influence on smaller scales should not be dismissed. Even within The Netherlands large differences in Bowen ratio or evaporative fraction can occur. This is shown by a quick comparison of two Fluxnet sites: Loobos, a pine forest on sand in the East region (see Figure 1 for the location of the regions), and Cabauw, a grass field on a peat/clay soil in the West region. Observed mean Bowen ratios computed from Fluxnet towers (available online: www.fluxdata.org) are 2.0 for Loobos and 0.6 for Cabauw in spring (MAM) and are 1.3 for Loobos and 0.4 for Cabauw in summer (JJA). Seneviratne et al. (2010) have written an extensive review on large scale soil moisture–climate interactions that also focusses on soil moisture-precipitation feedbacks. They conclude that because of the direct impact of precipitation on soil moisture, links of causality between soil moisture and precipitation are difficult to establish. Two opposing feedback mechanisms that enhance convective activity and increase precipitation have been proposed (Findell and Eltahir, 2003). On one hand, dry soils could trigger convection because the higher sensible heat flux increases the strength of the updraft, this combined with high moisture availability at the top of the atmospheric boundary layer enhances cloud formation (Westra et al., 2012). On the other hand, wet soils could increase convection through the build-up of a comparatively shallow and moist boundary layer and a large(r) net radiative energy flux (Schär et al., 1999) so convective updraft saturates at relatively low levels. The feedback mechanism over wet soils is mainly found in modelling studies and might be an artefact of model parameterizations at low resolutions. Hohenegger et al. (2009), for example, found significant differences between simulated soil moisture-precipitation feedback at 25 and 2.2-km resolution. A recent global observational analysis showed afternoon precipitation falls preferentially over dry soils, in contrast with the precipitation simulated by global climate models (GCMs) (Taylor et al., 2012). In addition, Orlowsky and Seneviratne (2010) warned that apparent soil moisture-precipitation feedbacks for GCM simulations can often as well or even better be attributed to the influence of global SST fields.
There are several studies on influence of SST on large scale precipitation (e.g. Benestad and Melsom, 2002; Kjellström and Ruosteenoja, 2007) and a few at smaller scales (Messager et al., 2004; Persson et al., 2005; Jung et al., 2010). For The Netherlands the influence of SST on precipitation has been found significant and particularly strong in the coastal area less than 30–50 km from the coastline (Lenderink et al., 2009). The adjacent North Sea is a shallow coastal sea (20–200 m deep) and therefore cools or warms relatively fast (on time scales of weeks to months) dependent on the atmospheric conditions and shortwave radiation. Van Oldenborgh et al. (2009) found that the North Sea temperatures follow the temperature rise of The Netherlands, which is higher than the global mean trend. Some of the observed increases in precipitation along the coast are likely related to the land-sea temperature contrast. Temperatures over sea are higher from August onward and enhance coastal precipitation until December when the atmosphere is on average too stable for frequent convective showers to occur (Attema and Lenderink, 2013). From April to July coastal precipitation is suppressed by relatively cold sea water temperatures.
Topography plays an important role in the distribution of precipitation, even in Northwestern Europe where gradients are relatively small (e.g. Osborn et al., 2000; Gilles et al., 2006). The Netherlands is a very flat country with maximum elevation of about 325 m in the South of the country and just over 100 m on the Veluwe. The Veluwe is a densely forested and elevated area located somewhat east of the middle of the country. Nonetheless even the limited topography of the Veluwe has been shown to influence the local precipitation climate (ter Maat et al., 2012). Ter Maat et al. showed that the difference in precipitation between the Veluwe and its surroundings, with a maximum of 14.5%, is highest in winter. Their analyses suggest that the precipitation maximum can only be explained by a combination of topography and land-use.
Both modelling and observational studies show increases in precipitation downwind of or near large urban areas (e.g. Comarazamy et al., 2010; Shepherd et al., 2010; Li et al., 2011; Mitra et al., 2012). These increases seem to be caused by a combination of physical and chemical processes. Urban areas have a larger heat-storage capacity, Bowen ratio, and an increased surface roughness in comparison with rural areas (Oke, 1982). These differences affect the surface energy budget and planetary boundary layer, which in turn impact the regional climate in and around urban areas. Furthermore, Russell and Hughes (2012) found that increases in precipitation are significantly and negatively correlated with NOx emissions. Other simulations with a microphysics cloud model show that heightened urban aerosol concentration in combination with the low-level updraft induced by the urban heat island leads to enhanced precipitation (Han et al., 2012). Urbanization in The Netherlands has strongly increased in the last century and its influence on observed precipitation increases cannot be ruled out. Therefore we integrate urbanization into some of the analyses in this study.
Our objective is to analyse and explain persistent spatial patterns and changes herein in relation to surface forcings. To achieve this, we divide the country into regions based on surface characteristics and zones at different distances to the coast. Topography is taken into account in the regions based on surface characteristics. The regions and zones will be introduced in Section 'Data and methods'. Section 'Spatial differences and trends' identifies persistent spatial patterns and adds percentiles to the analyses of Buishand et al. (2013). In this section we also analyse mean and extreme annual and seasonal precipitation amounts and annual trends. The regional differences in monthly mean and extreme precipitation changes are compared in Section 'Regional trend analysis' where we also compare annual precipitation differences over two climatological periods. This article concludes with a discussion and summary of results in Sections 'Discussion' and 'Conclusions', respectively.
2. Data and methods
Daily precipitation data for the period 1951–2009 are obtained from the KNMI manual rain gauge network. The data used in this study has been homogenized as part of the study by Buishand et al. (2013) with the automated homogenization procedure of Menne and Williams (2005). About 37% of the initial 377 precipitation series were designated as inhomogeneous and were corrected with the homogenization procedure. We use the same dataset as Buishand et al. (2013) for consistency, therefore the years 2010–2012 are not included. Only stations that had no more than 2% of daily precipitation observations missing are used. Missing data were supplemented with data from the nearest station at a maximum distance of 50 km.
Spatial differences for the precipitation changes presented in Section 'Regional trend analysis' are computed for regions based on mean precipitation, elevation and soil type, and zones at different distances to the coast. The Netherlands has four major soil types, sand, clay, loam, and peat. Clay and loam are mainly present along the seacoast and the (former) courses of the major rivers, Rhine and Meuse. Peat has been excavated in major parts of the country, but is still present in the North and West, while sandy soils dominate the east and south of the country. Combining the soil characteristics with average precipitation data led to the creation of five different regions. The sixth ‘region’, referred to as ‘high’, is based on topography and encompasses the Veluwe area and five stations in the southeast of The Netherlands that lie above an elevation of 80 m. The main characteristics of these regions are given in Table 1 and their spatial extent is given in Figure 1. The use of soil characteristics as such was not possible in the regression analysis with mean precipitation trends (Section 'Regional trend analysis'), therefore we used soil moisture capacities for each of the different soil classes based on Wösten et al. (2001).
Table 1. Characteristics of the six regions that were investigated in this study. Mean precipitation is calculated over the entire investigated 59-year period
Main soil type
Number of stations
Mean precipitation (mm)
Mean height (m)
Sand and loam
Clay and peat
Sand and peat
The degree of urbanization is calculated as the fraction of urban area within a 5 km radius around each station using the European Corine land cover database (EEA, 2002) for the year 2000. Land cover categories classified as urban are: discontinuous urban fabric, industrial and commercial units, road and rail network, port areas, airports, mineral extraction sites, dump sites, construction sites, green urban areas, and sport and leisure facilities.
To create the coastal distance zones, the shortest distance to a coastline enclosing the northern islands and south-western peninsula is calculated for each station. The zones are defined such that each contains about a quarter of the total number of precipitation stations available in this study. This results in four zones at 0–25, 25–50, 50–100 and 100–200 km from the coast. The distance zones are depicted in Figure 2.
Percentile values in this study are calculated by sorting the precipitation measurements over wet days (>1 mm) and taking the respective value. Trends are based on a simple linear regression on time and their degree of significance is assessed using the related P-values computed with ordinary least squares fitting. We take a 0.05 significance level. The results are found statistically significant if the P-value is less than the significance level. Percentual changes are calculated by subtracting the last value of the linear trend analysis from the first and dividing by the first, i.e. the difference between 1951 and 2009 expressed as percentage. For the regions and zones the data of all the encompassed stations is combined before the percentile values and trends are computed. A similar approach was taken for the countrywide changes of percentile values given in Table 2.
Table 2. Results of the trend analysis for the period 1951–2009. Changes in mean precipitation and the 90th, 95th, and 99th percentiles. Columns 3–6 give the changes relative to the 1951–1980 period
However, determining whether surface characteristics exert significant influence on precipitation is more complicated. As introduced before, we define a set of regions based on surface characteristics and coastal distance zones. The changes in these regions and zones are compared with a random sample of stations to give an indication of the local variability. For the number of stations we use 50, which is about the average number of stations in the regions. Accordingly we resample 50 stations randomly out of all stations available and compute the same statistics from this set of stations as was done for the regions and zones. We repeat this bootstrap procedure 1000 times, and the 5th and 95th percentiles are presented. Thus we compare the statistics results of the regions and zones to random regions assuming spatially uncorrelated data (in short, random region). For most statistics, however, spatial correlation exists. This can be either due to the scale of the weather systems that are involved, or due to surface forcing. We are interested in the role of the surface, but this cannot be separated statistically from the role of the implicit scale of the weather systems. As a complication, the regions and zones defined have different shapes, with more circular shapes for the regions based on soil type and more elongated shapes for the coastal distance zones. As a result spatial correlation due to the scale of the weather systems is not necessarily the same in both sets and is likely larger in the regions based on soil type as these are generally spatially more confined areas. This effect is very difficult to incorporate in the bootstrap resampling and therefore we only compare to the random regions. This is by no means an absolute measure of significance, but merely a rough measure of the spread which would be expected if there is no spatial correlation in the data. Finally, we note that bootstrap procedures have been used with spatially correlated data (e.g. Douglas et al., 2000), for e.g. by resampling fields at the same time. Such a procedure answers a different question (significance of trends) than the one we aim to answer (how different trends in regions are from each other).
3. Spatial differences and trends
Here we discuss the main features of precipitation in The Netherlands from 1951 to 2009. In general, The Netherlands has seen an increase in both mean and extreme precipitation. Despite the fact that precipitation has a large inter annual variability, mean annual precipitation has increased almost 16% over the 59 years investigated. Figure 3 shows the annual mean precipitation and the climatological means for the standard WMO 30-year period (WMO, 1989). The 1980–2009 climatological values for mean precipitation are strikingly different from the others. This is consistent for all of the investigated indices of extremes, such as the annual number of days with precipitation greater than 20 and 30 mm and the 90th, 95th, and 99th percentiles.
Despite the small size of The Netherlands, persistent spatial variations in precipitation can be observed. The whole country ranges 312 km in the north–south direction and 264 km in the east–west direction. Relatively wet areas are present in the middle of the West coast, over the Veluwe area and at a few stations in the South. These areas receive the largest mean precipitation, around 900 mm in contrast to the country average of 800 mm, as well as the highest daily amounts. The values of the 95th and 99th percentile for these areas are on average 16 and 28 mm, respectively which is about 20% more than the countrywide values of roughly 13 and 24 mm. We will first focus on extreme events and compare the spatial distribution of the number of days on which a specific threshold is exceeded to the spatial distribution of percentile values for extreme events (see Figure 4). The 20 mm threshold (panel a) is approximately exceeded five times per year over the Veluwe as well as over the West coast. However, on average the amount of precipitation is higher along the West coast, which is reflected in higher values of the 99th percentile (panel b).
We also investigated the percentage of precipitation that falls above a number of fixed thresholds and above different percentiles, the results can be seen in panel (c) (a threshold of 20 mm) and D (the 95th percentile) of Figure 4. The highest values for the percentage of precipitation above the percentile thresholds (of approximately 45, 30 and 10% for the 90th, 95th, and 99th percentile, respectively) are found at a few stations in the north(-east) of the country. The percentage of precipitation over the 20 and 30 mm thresholds in contrary shows maxima over the Veluwe area and near the south-west coast. Taking an average, we can say that more than 10% of whole year precipitation falls on days with extreme precipitation, i.e. days with more than 20 mm rain and/or above the 90th percentile.
The number of days with precipitation varies throughout the year from around 50% in spring and summer to 60% in autumn and winter in The Netherlands, but the maximum daily totals are much higher in summer when precipitation has a more convective character. In spring and winter daily maxima are on average 25 mm, while summer and autumn daily totals can be more than twice as high. The seasonal variations in mean precipitation are quite large as well (see Figure 5). Spring (MAM) is the driest season with little spatial variation because of typical (high pressure) circulation patterns (Buishand and Velds, 1980). Generally the Veluwe area receives somewhat more and the coast somewhat less precipitation than the rest of the country. Summer (JJA) precipitation has a similar gradient along the coast, this is likely caused by a suppression of precipitation by the relatively cold sea water temperature in spring and early summer. Precipitation in autumn (SON) is the highest and has a clear opposing coastal gradient with highest amounts along the coast and minima further inland. The contrast between coastal and inland precipitation has become larger in recent times presumably because the increasing SST feeds convective systems. In winter (DJF) amounts are generally low again and the most outstanding feature is the maximum over the Veluwe area. Annual mean precipitation is consequently dominated by summer and autumn.
A simple trend analysis through station-wise linear regression of daily precipitation shows a positive change over all stations in The Netherlands (Figure 6). The increases are largest near the coast and smallest in the Southeast. Station-wise assessment of the related P-values shows that over the entire country nearly 90% of the observed changes are significant at the 0.05 level. Nearly all changes in daily precipitation at stations along the coast are significant, while some inland stations are not. We note that we do not take spatial correlation between neighbouring stations into account in estimating the significance. The changes are largest in spring and winter (can be seen in Figure 8 which will be discussed in the next section) and smallest in summer on average, although the difference between coastal and inland changes are largest in summer. The largest increases, on average up to 35%, are found in the months February to April. The changes in the following months are successively smaller until hardly any change can be detected in July. After September the changes increase again up to 20% from October to January. Although early spring and winter changes are the largest, the spatial distribution in these seasons does not change. Hence, the overall spatial pattern in precipitation changes is dominated by early summer and late autumn.
Urban stations, i.e. stations with an urban fraction greater than or equal to 25%, are indicated in Figure 6 with diamonds. Almost half the cities designated as urban lie in the central West coast, where the percentile values and threshold exceedances are higher than in the rest of the country. Mean precipitation changes in these stations are relatively high, however, urban stations in the rest of the country have similar changes as nearby nonurban stations.
Last we explore the changes in percentile values. On average these changes are close to the change in mean precipitation (see Table 2, column 2, all significant at the 0.05 level). This is an indication that the shape of the distribution of precipitation has not altered much throughout time, as extreme events still occur in about the same proportions. Columns 3–5 of Table 2 give the changes in percentile values relative to the 1951–1980 period. It shows that all indices increase throughout time, but the largest changes are found in the 1980–2009 period. The overall picture is that of increasing mean precipitation as well as percentile values and the percentage of precipitation above the percentile values.
4. Regional trend analysis
We now investigate spatial precipitation trends and how differences between the regions and coastal zones can be explained. Simple linear regression analyses between the change in mean precipitation and the investigated factors: soil moisture, distance to the coast, elevation and degree of urbanization show little correlation for soil moisture (capacity) and urbanization. Both topography and distance to the coast have a negative relation with the observed precipitation changes, although the relation for distance to the coast is stronger. Topography and distance to the coast themselves are also correlated as stations with low elevation lie close to the coast and elevated stations mainly lie in the southeast and on the Veluwe.
Figure 7 shows the negative relation between the station-wise increase in precipitation (i.e. the difference between 1951 and 2009 expressed as percentage) and distance to the coast. The regression between precipitation change and distance to the coast is approximately 8% per 100 km in the 59 years that were investigated. The changes are given in percentages rather than millimetres to be able to compare stations with different annual precipitation amounts equally. The correlation coefficient (r2) is about 0.4. On average, precipitation increases are greatest near the coast and decrease further inland. For stations that change less than 9%, the changes turn out not to be significant. Urban stations are present throughout the entire scatter and there is no coherent clustering, which, if present, would suggest the influence of the degree of urbanization on precipitation amounts and as a result on trends.
As mentioned in the previous sections, we defined a set of regions based on surface characteristics and zones at different distances to the coast (see Figures 1 and 2). The remainder of this article will focus on the differences when comparing these with each other. Figure 8 provides two panels for each statistic, the left gives the changes for regions based on surface characteristics, the right for coastal distance zones. Panels (a) and (b) give the change in mean precipitation for each month of the year. The most distinctive feature is the order of the regions from April to June and its near reversal from July to September. In panel (b) it is easy to see that the strongest increases in these months are along the coast and the lowest further inland. The largest differences between the regions can be found from April to June, when the change in mean precipitation is about 20% higher along the coast than further inland.
Extreme precipitation follows roughly the same pattern throughout the year as mean precipitation. Panels (c) and (d) give the change in the 95th percentile value. It is hard to find a coherent picture in panel (c) and not easy in panel (d) either, because both the most inland and the coastal zones behave differently from the rest. Sometimes the coastal zone has the largest (smallest) trend and sometimes the most inland zone. From August to November the change in the most inland zone is up to 15% higher than the coastal zones.
Panels (e) and (f) show the change in the value of the 99th percentile calculated over all days, thus not only over wet days as before. The regions and zones are somewhat easier to distinguish in these graphs than in the last. Panel (e) shows a very large spread, of about 30%, between the regions in the months December and January. This feature cannot be seen in panel (f) where the changes differ only about 10%. The months June, July, and August show very unalike behaviour for the most coastal zone, with changes up to 15% higher than the others. Conversely, the trend of the most inland zone is also larger than the two in between zones in the months June and July.
Last, the differences between the regions are investigated for two different climatological periods that split the dataset in half. So Figure 9 shows the differences in yearly mean precipitation between the 1980 and 2009 and 1951 and 1980 periods. The difference for each of the regions is plotted against the percentile, given as a fraction between 0 and 1. Thus the most extreme values recorded have a cumulative probability of 1 and the x-axis gives the probability that x or lower occurs. Every station in a region is individually averaged for each year in the two periods. For every region the yearly precipitation sums of each station and each year in the considered 30-year period are then ordered to get their corresponding percentiles and compared with the other period.
On average precipitation has increased about 10% when the earlier period is compared with the latter. For probabilities higher than 0.8 and lower than 0.1 contrasting behaviour for the coastal and inland zones can be seen. For high probability values, the coastal zones, up to 50 km from the coast, increased up to 5% more than average, while precipitation in some inland zones even decreased. For low probability values, however, inland zones have become wetter than the regions near the coast. Especially for low probabilities the differences are more distinct when comparing coastal distance zones to one another than comparing the regions based on surface characteristics.
In comparing the two sets of surface characteristic regions and coastal distance zones we showed bootstrap confidence intervals. Although we find a lot of consistency in trends and between zones, a large proportion of the data points lie outside the bootstrap confidence intervals in Figures 8 and 9. Some differences probably arise because the bootstrap sample is based on 50 random stations, while the regions range in amount of stations from 36 to 66 with exception of the high region which has only 15 stations. However, the shape of the regions and spatial correlation between the stations likely plays a much larger role. Spatial correlation decreases the effective sample size in a dataset and not taking it into account results in an overestimation of trend significance. Several studies report significant impact on trends if spatial correlation would not have been taken into account (e.g. Douglas et al., 2000; Adamowski and Bougadis, 2003). Our station-wise trend assessment of the significance of mean precipitation is for this reason much higher than the significance reported by Buishand et al. (2013) who assessed the entire field significance. In other situations correlation has been dealt with by using a regional or seasonal Kendall trend test (Helsel and Frans, 2006). For our regional trend analyses these methods are less appropriate because we are not so much interested in the field significance of a trend, but more in how trends are affected by the surface. The bootstrap confidence intervals currently plotted in the figures should give a lower boundary of the expected spread assuming spatially uncorrelated data.
Increases in both mean and extreme precipitation indices were found. Both yearly and seasonal precipitations have increased over the investigated 59-year period, but the largest changes are generally found in the last 30-year climatological period. This is because the increase in precipitation is not gradual and consistent, but dry and wet periods can be distinguished and the inclusion of the dry period peaking in 1976 into all of the other climatological periods cause the later period to be strikingly different. The North Atlantic Oscillation (NAO) has important links with precipitation over Europe (Hurrell, 1995) and most of the temporal changes between wet and dryer periods in The Netherlands are strongly influenced by changes in the atmospheric circulation (Lenderink et al., 2009; Haren et al., 2013). The overall rise in precipitation causes the thresholds for the annual number of days with precipitation greater than 20 and 30 mm to be exceeded more frequently. Although similar spatial results are found for the analyses of exceedance frequencies and intensity changes (i.e. percentile values), the changes of respectively 44 and 53% for the annual number of days with precipitation greater than 20 and 30 mm reported by Buishand et al. (2013) appear much higher than our changes in percentiles. The results are comparable, however, because a 10% increase in intensity decreases the return time by roughly a factor two on average.
Trends in precipitation patterns were analysed with respect to soil type, distance to the coast, topography and urbanization. Urban stations were found to have similar changes as nearby non-urban stations, so no distinctive influence was found. The fact that especially modelling studies do find an impact of urban areas could be due to the short time span that they look at or the location of the urban effect. The effect of urban areas is expected downwind which is not defined in our case as we look at yearly or seasonal averages. Topography has been found influential before, for example, over the Veluwe (ter Maat et al., 2012) where average yearly precipitation sums of up to 100 mm higher than the rest of the country have been observed. We find the increases in this region, however, are among the smallest during the period investigated, both for mean and extreme precipitation. So, over time the higher precipitation amounts over the Veluwe area have become less distinct. For soil type and precipitation changes no direct relation was found and regions based on soil type show less consistent behaviour than the coastal distance zones.
Coastal precipitation increases reported here are consistent with the results of Lenderink et al. (2009), who investigated the influence of SST on precipitation. They showed that, apart from the spring months, the coastal area has consistently become wetter compared with the inland area since the 1950s. This wetting of the coastal area compared with the inland is confirmed by this study. Some inconsistent behaviour was found for the months July and August when the increase in mean precipitation is higher for both inland regions compared with the coastal area. Lenderink et al. found that the influence of SST is particularly strongly less than 30–50 km from the coastline. This finding is strengthened by the current study, although we show that for extreme summer precipitation the influence seems to be confined to the first 25 km. The differences between the first and second half of the investigated period show further diverging behaviour between the stations less than 50 km from the coast and those further away. The influence of SST on coastal precipitation in The Netherlands might be larger than in other countries because the temperature of the relatively shallow North Sea has increased more than the global mean trend.
The consistent increases in precipitation in The Netherlands are quite unusual in comparison to other European trends which are highly variable, sometimes giving opposing trends for different regions and/or time periods (e.g. del Rio et al., 2011; Lupikasza et al., 2011; Karagiannidis et al., 2012). A common conclusion from other studies is that precipitation amounts have large interannual variability, on both annual and seasonal scales. We confirm this and also find the period examined can have a large effect on the results. For example, the standard 30-year climatologies are not appropriate to compare with each other either, as in- or exclusion of a relatively wet or dry period can cause large differences between the periods. A running trend analysis, such as the one utilized by Brunetti et al. (2012) and Turco and Llasat (2011) might be the most appropriate to investigate changes throughout time. For extreme precipitation analysis several other techniques are available, such as those referred to and used in (Ntegeka and Willems, 2008).
We have shown that the spatial patterns found in The Netherlands, although persistent, are not time invariant. Future precipitation patterns in the Dutch climate scenarios (van den Hurk et al., 2006) are the same as in the present day climate. We suspect this assumption is not valid, because regional differences in past trends are apparent. Future modelling work at smaller temporal and/or spatial resolution might be able to find local influence of urbanization and the land surface that we have not because of our relatively coarse approach.
In this study spatial patterns and seasonal precipitation trends were investigated with homogenized daily precipitation data from 240 stations in The Netherlands for the period 1951–2009. Homogenization of the data and a first analysis was conducted by Buishand et al. (2013). This study has extended their results regarding persistent (seasonal) spatial patterns and regional differences in precipitation changes.
Overall, both mean and extreme precipitations have increased on seasonal and annual scales over the last 59 years and especially in the last 30 years. Persistent patterns in mean annual precipitation are related to a combination of seasonal variations and the spatial distribution of extreme precipitation. Linear regression shows a positive change in daily precipitation for the entire country with the largest changes along the West coast. Both annual and seasonal changes show increases in precipitation, but the observed spatial patterns are sensitive to seasonality and the investigated time frame. Observed changes are largest in spring and winter, while absolute amounts peak in summer and autumn.
We find that distance to the coast explains more of the variance in the dataset than the other investigated factors, being soil type, topography, and urbanization. Zones based on distance to the coast give a more consistent picture for precipitation changes over time than regions based on surface characteristics do. The zones up to 50 km from the coast show a number of different characteristics than the inland zones, especially regarding extreme precipitation on both seasonal and climatological timescales.
This study was supported by the Dutch research program Knowledge for Climate. We would like to thank the two anonymous reviewers for their constructive comments. Furthermore, we thank Geert Jan van Oldenborgh for making the precipitation data available through the KNMI Climate Explorer and Theo Brandsma for useful comments on an earlier version of this manuscript.