Abstract

The lower Neretva river basin includes a fertile valley at the estuary into the Adriatic Sea, where intense agricultural production occurs, and the higher terrain where drinking water resources exist. To provide input for the further assessment of crop-yield production and hydrological risks, climate and climate change were analysed using the Opuzen station air temperature and total precipitation data for the 1961–2015 period. Both historical and future climates (2021–2050) were assessed based on simulations of three regional climate models (RCMs). The RCMs were forced by the observed concentrations of greenhouse gases (GHGs) from 1951 to 2000, and the IPCC A1B scenario of the GHG emissions was applied from 2001 onwards. The models were compared with the observations, and two bias adjustment methods were evaluated. The results generally showed a significant increase in the mean annual and seasonal temperature and a weak decreasing trend in annual and seasonal precipitation. Projections revealed a predominant increase in the mean temperature by the mid-21st century for all three RCMs (between 0.5 and 3.5 °C). The precipitation changed by between −60 and +60% throughout the year for the different models, although the changes generally were not statistically significant.

INTRODUCTION

The observed global and regional warming has been unequivocal since the middle of the 20th century (IPCC 2014) and is present along the eastern Adriatic coast (e.g. Branković et al. 2013). Climate change shows a tendency to drier conditions for the Adriatic (e.g. Cindrić et al. 2010; Patarčić et al. 2014), which is a common feature for the broader European part of the Mediterranean (e.g. Giorgi & Lionello 2008). The fact that precipitation change is regionally highly variable increases the need for a more accurate regional and local precipitation change analysis to improve the assessment of climate change impacts.

In the present paper, we studied the lower Neretva river catchment, which includes the fertile valley at the estuary into the Adriatic Sea. Because of the typical Mediterranean climatic conditions and the abundance of water, agriculture is well developed in the Neretva river valley in the lowest part of the catchment and has a great potential for the intensive production of vegetables and Mediterranean fruits (Romić et al. 2008). The agricultural and water supply systems in the catchment are greatly influenced by the meteorological and hydrological conditions. The impact that climate change has on plant production and hydrological systems at a sub-catchment scale can be seen in the changes in the length of the yearly growth period, potential evapotranspiration, soil moisture, streamflow, groundwater recharge and the magnitude/frequency of agricultural and fluvial floods and droughts. Such agricultural and hydrological impacts of climate can be assessed by forcing crop and hydrological models with high-resolution regional climate model (RCM) simulations. In this study, we performed a climate analysis for the Neretva river valley that included 2-m air temperature and total precipitation variability and time trends in historical (present) and future climates. The obtained results can further be used as meteorological inputs for the projections of crop yield models and hydrological water balance estimates relevant to the projections of drinking water supplies or irrigation needs over the selected region. Moreover, the analysis may serve as an example of a climate study for local and national strategies to adapt to climate change.

The paper is organised as follows: the methodology incorporating the local climate and the characteristics of climate models is introduced in the following section, followed by the results and finally the main conclusions.

METHODOLOGY

Local climate analysis

General climate characteristics and the observed climate variability and trends were analysed from the available DHMZ (Meteorological and Hydrological Service of Croatia (Državni hidrometeorološki zavod; DHMZ)) climatological data in the lower Neretva river catchment. All DHMZ data were continuously monitored and a routine quality control was applied. The air temperature and precipitation time series from the Opuzen meteorological station (altitude h = 3 m a.s.l., φ = 43° 1′ 3″, λ = 17° 33′ 31″) at the mouth of the Neretva river in the southern Croatian Adriatic coastal region were used for the analysis (Figure 1).

Figure 1

Orography height (m) with the location of the Neretva river and the Opuzen meteorological station. The locations of the lakes and the river network were based on the GSSHG (Global Self-consistent, Hierarchical, High-resolution Geography Database) and WDBII (CIA World Data Bank II) datasets (Wessel & Smith 1996).

Figure 1

Orography height (m) with the location of the Neretva river and the Opuzen meteorological station. The locations of the lakes and the river network were based on the GSSHG (Global Self-consistent, Hierarchical, High-resolution Geography Database) and WDBII (CIA World Data Bank II) datasets (Wessel & Smith 1996).

The local climate characteristics were described using the corresponding monthly, seasonal and annual averages over the reference climate period of 1961–1990. Seasonality was described in terms of annual cycles of the mean monthly air temperature and precipitation values, and their interannual variability was described by the standard deviations of monthly means and coefficients of variation for precipitation. The discussion of extremes in annual and seasonal air temperature and precipitation averages was based on percentiles calculated from the empirical cumulative distribution function (CDF).

Trends in seasonal and annual mean air temperature and precipitation amounts were calculated for the longer period of 1961–2015. Trends were estimated by Kendall's tau method (or Sen's slope; Sen 1968). The trend values are expressed as decadal values for both variables. Additionally, trends in precipitation amounts are given as the percentage of corresponding seasonal and annual means from the 1961–1990 reference period. The statistical significance of trends was estimated by the non-parametric Mann–Kendall test (Gilbert 1987).

Regional climate models

The assessments of historical (present) and future (2021–2050) climates were based on the results of the numerical simulations of the three RCMs that were analysed for the CC-WaterS project (Koeck 2012). These models participated in the EU FP6 ENSEMBLES project (van der Linden & Mitchell 2009), with downscaling simulations at a 25-km horizontal resolution over Europe. In this study, the model data were analysed for those model grid cells that were closest to the location of the Opuzen meteorological station (representing the lower Neretva river catchment).

The RCMs used were the Aladin (Bubnova et al. 1995), Promes (Castro et al. 1993) and RegCM3 models (Pal et al. 2007). The RCMs were forced by the observed concentrations of the greenhouse gases (GHGs) from 1951 to 2000, and the Intergovernmental Panel on Climate Change (IPCC) A1B scenario of the GHGs emissions was applied from 2001 onwards. The initial and boundary data for each RCM were provided by different global climate models (GCMs): the ECHAM5 GCM dataset was used to force the RegCM3 RCM, the Aladin RCM was forced by the Arpege GCM, and the Promes RCM was forced by the HadCM3Q GCM (van der Linden & Mitchell 2009). The choice of these three RCM-GCM pairs covered the typical range of the climate change responses from the entire ENSEMBLES set of simulations (Koeck 2012). For the historical (present) climate, models were compared with the local DHMZ observations and with the E-OBS gridded temperature and precipitation data based on the contributing stations of varying density across Europe (Haylock et al. 2008). The following two abbreviations were used in the study:

  • (1)

    RCMcorr: this type of simulation was based on the raw RCM output bias adjusted using the E-OBS data (see, e.g., Déqué (2007) and Formayer & Haas (2010) for the description of the bias correction/adjustment methodology). The RCMcorr data are available from the CC-WaterS database at http://climdat.boku.ac.at/opendap/ccwaters.

  • (2)

    RCMcorr_adj: this type of time series was based on a further adjustment of RCMcorr due to the differences between the E-OBS data and local DHMZ observations; it was based on applying the additive (for both air temperature and total precipitation) and multiplicative adjustments (for total precipitation) of RCMcorr monthly values based on the differences/quotients between RCMcorr and DHMZ observations over the 1961–1990 (P0) period. The adjustment differences were computed for RCMcorr from the DHMZ values for each month i (i = 1,…,12) in the P0 period as adj(i)=RCMcorr(i)-DHMZ(i), where RCMcorr(i) and DHMZ(i) are monthly temperature averages (or precipitation amounts) from the RCM and observations, respectively. These adjustment differences were then applied on the RCMcorr time series for the entire analysed period (1951–2050) to obtain RCMcorr_adj in the following manner: RCMcorr_adj(i,j)=RCMcorr(i,j)-adj(i); i = 1,…12; j = 1951, …, 2050. In a similar manner, multiplicative adjustment for the total precipitation was based on mul(i)=DHMZ(i)/RCMcorr(i), which led to RCMcorr_adj2(i,j)=RCMcorr(i,j) · mul(i).

Due to the experimental nature of the RCMs, several limitations should be emphasised:

  • (i) The spatial resolution of the RCM simulations applied in this study was 25 km. At this resolution, the main orographic features and the land-sea boundary of the Croatian coast were reasonably well resolved. However, local characteristics for the specific station or catchment were not fully resolved at the same resolution. Both the physics and the horizontal resolution of the limited models contributed to systematic errors (biases), which motivated the use of the bias-adjustment step before, for example, forcing the crop-yield impact model with the output from the RCM.

  • (ii) All of the RCMs in this study were forced by historical (observed) concentrations of the GHGs for the period 1951–2000. However, the IPCC A1B scenario was applied from 2001 onwards, and only one assumption of the GHG concentration was thus evaluated. This must be considered when evaluating the amplitude of projected climate changes (e.g. the higher GHG emission scenarios are usually associated with higher temperature increases). However, since our focus is on the future 2021–2050 period, the choice of the GHG concentration scenario is less important compared with the later parts of the 21st century.

  • (iii) The three RCM models used here accounted for only some of the possible modelling uncertainties. The use of the multi-model ensemble approach in climate projection studies is a common approach to avoid projection dependence on specific model assumptions. However, users often require a minimum number of input time series for impact studies to simplify data management and analysis. In this case, the use of three different RCMs, each forced by a different GCM, could address part of these issues.

  • (iv) In the analysed RCM simulations of the reference climate, the RCMs did not reproduce the actual variability observed in the real climate system. Since the RCMs were forced at the boundaries by different GCMs (each of which had its own internal variability, for example the sequence of warm and cold years over Europe), the RCMs simulated different variabilities, for example, their own sequence of warm and cold years (or dry and wet years). The specific values indicated in the time series presented in this study do not signify a specific prediction for a specific year. As an alternative approach, RCMs are often forced by the atmospheric reanalyses, where large parts of the inter-annual variability can be reproduced. However, for the future model projections, only GCM projections can provide relevant boundary data. Models can be compared with observations and with each other in terms of the reference and projected mean climate and overall variability. However, model simulations of the future climate should be interpreted as projections of the possible state(s) of the climate system that are sensitive to the applied initial and boundary conditions, GHG scenarios and model internal configurations. These projections are expected to represent future trends and changes over longer time periods as realistically as possible but not the ‘forecasts’ for some specific future year.

A detailed discussion on the modelling limitations emphasised here can be found in Hawkins & Sutton (2009) and Jacob et al. (2014).

RESULTS AND DISCUSSION

Climate characteristics

The climate characteristics of the lower Neretva river basin were determined by the mid-latitude air circulation modified by the influence of the sea, which reaches deep into the mainland through the valley of the Neretva river. To a lesser extent, they are governed by altitude, relief configuration, soil type and other factors. The lower Neretva river basin is under the influence of the subtropical high-pressure zone during summer, with dry and warm weather. The Adriatic Sea, with its large thermal capacity, moderates air temperature extremes: it has a cooling effect in summer and reduces the cold in winter. In contrast to the Neretva river valley, which is primarily used for agriculture, there are also water springs in the basin that may be included in systems for drinking water supply. The karstic soil type contributes to warming in the summer. During the cold part of the year, the area is within the zone of the main western winds that dominate mid-latitudes, with a constant change between low- and high-pressure systems (Zaninović et al. 2008).

Air temperature

The annual cycle of air temperature monthly averages over the lower Neretva river catchment has maritime characteristics, with autumn being warmer than spring by 1.8 °C on average (Table 1). The winters are mild, with an average air temperature of 7.4 °C, and the summers are moderately warm (23.8 °C). The standard deviation of mean monthly air temperature ranges between 0.8 (July) and 1.9 °C (February), which indicates a small interannual variability (Figure 2(b)). July is the month that is least likely to change its thermal character and February is the most unstable month.

Table 1

Basic statistics (mean, standard deviation, maximum and minimum) for annual and seasonal mean air temperature and the percentiles from the corresponding empirical distributions for the reference period 1961–1990 for the Opuzen meteorological station

  DJF MAM JJA SON Year 
Mean (°C) 7.4 14.5 23.8 16.3 15.5 
Std. dev (°C) 1.0 0.9 0.5 0.9 0.4 
Max (°C) 9.1 15.9 24.9 18.0 16.2 
Min (°C) 5.4 12.3 22.6 13.9 14.6 
p1 5.4 12.3 22.6 13.9 14.6 
p2 5.5 12.4 22.6 13.9 14.6 
p5 5.7 12.9 22.8 14.4 14.8 
p10 5.8 13.3 23.1 15.1 14.8 
p90 8.5 15.7 24.5 17.4 16.1 
p95 8.8 15.8 24.7 17.8 16.2 
p98 9.1 15.9 24.9 18.0 16.2 
p99 9.1 15.9 24.9 18.0 16.2 
  DJF MAM JJA SON Year 
Mean (°C) 7.4 14.5 23.8 16.3 15.5 
Std. dev (°C) 1.0 0.9 0.5 0.9 0.4 
Max (°C) 9.1 15.9 24.9 18.0 16.2 
Min (°C) 5.4 12.3 22.6 13.9 14.6 
p1 5.4 12.3 22.6 13.9 14.6 
p2 5.5 12.4 22.6 13.9 14.6 
p5 5.7 12.9 22.8 14.4 14.8 
p10 5.8 13.3 23.1 15.1 14.8 
p90 8.5 15.7 24.5 17.4 16.1 
p95 8.8 15.8 24.7 17.8 16.2 
p98 9.1 15.9 24.9 18.0 16.2 
p99 9.1 15.9 24.9 18.0 16.2 
Figure 2

Annual cycle of (a) mean monthly air temperature, (b) its standard deviation, (c) cumulative distribution of mean annual air temperature, (d) mean, maximum and minimum monthly precipitation amounts, (e) standard deviation and coefficient of variation of monthly precipitation and (f) cumulative distribution of annual precipitation amounts for the period 1961–1990 for the Opuzen meteorological station.

Figure 2

Annual cycle of (a) mean monthly air temperature, (b) its standard deviation, (c) cumulative distribution of mean annual air temperature, (d) mean, maximum and minimum monthly precipitation amounts, (e) standard deviation and coefficient of variation of monthly precipitation and (f) cumulative distribution of annual precipitation amounts for the period 1961–1990 for the Opuzen meteorological station.

The empirical cumulative distribution of the mean annual air temperature for Opuzen is shown in Figure 2(c). The percentiles that determine extreme values of annual and seasonal mean temperatures are given in Table 1.

Precipitation

The lower Neretva river catchment has a maritime annual cycle (Figure 2(d)). During the cold half-year (October–March) this area receives more precipitation than in the warm half-year (up to 70% of the annual total on average). The monthly amounts were above 150 mm from October to December, with the maximum in November (180 mm). The lowest monthly precipitation amounts occurred in the warm period of the year (April–September).

In some years, a considerable deviation from the average precipitation conditions by months was observed. The coefficient of variation indicated such a high interannual variation in the mean monthly precipitation amounts (Figure 2(e)). These coefficients were higher than 50% for each month, with the maximum in July.

The CDF of annual precipitation is shown in Figure 2(f). The percentiles that determine extreme values are given for annual and seasonal precipitation in Table 2. For all observed return levels on the right tail of the distribution (90–99th percentile), the highest seasonal values of precipitation amounts can be found for winter and autumn. In contrast, extremely dry seasons are those with precipitation amounts lower than the second percentile. This value was the lowest for summer (21 mm) and the highest for autumn (182 mm). On a monthly scale, there may have been no precipitation at all from July to October and in January, whereas for other months, the second percentile values ranged between 11 and 34 mm (not shown).

Table 2

Basic statistics (mean, standard deviation, coefficient of variation, maximum and minimum) for annual and seasonal precipitation amounts and the percentiles from the corresponding empirical distributions for the reference period 1961–1990 for the Opuzen meteorological station

  DJF MAM JJA SON Year 
Mean (mm) 416.3 281.9 170.4 436.2 1308.5 
Std. dev (mm) 184.0 100.5 82.1 156.2 223.4 
cv (mm) 0.45 0.36 0.49 0.36 0.17 
Max (mm) 851.4 549.7 342.1 739.5 1734.6 
Min (mm) 69.5 95.0 33.8 173.0 710.3 
p1 69.5 95.0 33.8 173.0 710.3 
p2 72.6 97.3 35.4 174.7 725.9 
p5 100.1 117.8 49.5 189.5 866.5 
p10 128.9 159.0 61.3 227.6 1047.0 
p90 659.3 417.6 296.3 667.1 1636.5 
p95 768.7 492.7 331.1 694.3 1661.0 
p98 843.1 544.0 341.0 735.0 1727.2 
p99 851.4 549.7 342.1 739.5 1734.6 
  DJF MAM JJA SON Year 
Mean (mm) 416.3 281.9 170.4 436.2 1308.5 
Std. dev (mm) 184.0 100.5 82.1 156.2 223.4 
cv (mm) 0.45 0.36 0.49 0.36 0.17 
Max (mm) 851.4 549.7 342.1 739.5 1734.6 
Min (mm) 69.5 95.0 33.8 173.0 710.3 
p1 69.5 95.0 33.8 173.0 710.3 
p2 72.6 97.3 35.4 174.7 725.9 
p5 100.1 117.8 49.5 189.5 866.5 
p10 128.9 159.0 61.3 227.6 1047.0 
p90 659.3 417.6 296.3 667.1 1636.5 
p95 768.7 492.7 331.1 694.3 1661.0 
p98 843.1 544.0 341.0 735.0 1727.2 
p99 851.4 549.7 342.1 739.5 1734.6 

Observed trends

Time series of the mean seasonal and annual air temperature and precipitation amounts, with the associated trend lines and their mean values from the reference period 1961–1990, are shown in Figures 3 and 4, respectively.

Figure 3

Time series of mean seasonal and annual air temperatures with associated trend lines for the period 1961–2015 for the meteorological station Opuzen. Corresponding decadal trend values (in °C/10 yrs) are given on each panel, and the significant trends at the 5% level are indicated by *.

Figure 3

Time series of mean seasonal and annual air temperatures with associated trend lines for the period 1961–2015 for the meteorological station Opuzen. Corresponding decadal trend values (in °C/10 yrs) are given on each panel, and the significant trends at the 5% level are indicated by *.

Figure 4

Time series of seasonal and annual precipitation amounts with the associated trend lines for the period 1961–2015 for the Opuzen meteorological station. Corresponding decadal trend values (in mm/10 yrs and %/10 yrs) are given in each panel, and the significant trends at the 5% level are indicated by *.

Figure 4

Time series of seasonal and annual precipitation amounts with the associated trend lines for the period 1961–2015 for the Opuzen meteorological station. Corresponding decadal trend values (in mm/10 yrs and %/10 yrs) are given in each panel, and the significant trends at the 5% level are indicated by *.

During the period 1961–2015, mean annual air temperature anomalies were mainly positive. In the most recent 20 years, the air temperature trend was amplified. The consequence of such temperature fluctuations is that nine of the ten warmest years in the observed 55-year period were recorded after the year 2000. This amplification in the temperature trend(s) is in accordance with observed regional and global warming (IPCC 2014). The annual trend reveals a statistically significant increase in the mean air temperature of 0.2 °C/10 yrs since 1961, according to the Opuzen data (Figure 3). The annual temperature increase was predominantly because of a significant increase in the mean air temperatures of the summer (0.4 °C/10 yrs) and spring (0.2 °C/10 yrs). The detected temperature trends agree with the observed temperature trends over the Croatian Adriatic found by Branković et al. (2013).

The trends in precipitation amounts revealed drying in the annual (–2.1%/10 yrs) and seasonal amounts, except MAM, although they were not statistically significant (Figure 4). The main contribution to annual drying primarily came from the reduction in summer precipitation totals (–6.4%/10 yrs). These results are consistent with the observed trends from the middle to the southern Adriatic zone (Branković et al. 2013).

The combined influence of observed meteorological parameters, air temperature and precipitation affects the water balance components. The detected increase in air temperature in spring and summer causes an increase in evapotranspiration. When associated with seasonal trends in precipitation, decreasing trends in summer, autumn and winter precipitation, and a weak positive trend in spring precipitation, the precipitation deficit was expected to increase in the warm season. At the same time, runoff and the filling of aquifers in autumn and winter could be reduced because of the negative precipitation trends in these seasons and could impact the water supply.

RCM simulations

Bias-corrected (or bias-adjusted) simulations (RCMcorr) of the three RCMs, DHMZ observations and E-OBS data in the reference period 1961–1990 for the Opuzen location were compared and are shown in Figure 5. The E-OBS annual cycle shows a lower air temperature compared with the DHMZ observations. Due to the nature of the bias correction applied to RCMs, RCMcorr was colder than the DHMZ observations, and all RCMcorr graphs overlap with that of E-OBS (Figure 5(a)). For precipitation, an underestimation of the DHMZ observations by E-OBS and RCMs is seen, and it was particularly large in the colder part of the year (Figure 5(d)).

Figure 5

Opuzen station: measured and modelled annual cycles of (a) mean monthly temperature, (d) mean monthly precipitation amount, (b) mean monthly temperature standard deviation, (e) coefficient of variation of monthly precipitation amount; time series of (g) mean annual temperature and (h) annual precipitation amount; empirical CDFs of (c) mean annual temperature and (f) annual precipitation amount. Model time series are RCMcorr. The analysis period is 1951–2000 in panels (g) and (h) and P0 (1961–1990) in all other panels.

Figure 5

Opuzen station: measured and modelled annual cycles of (a) mean monthly temperature, (d) mean monthly precipitation amount, (b) mean monthly temperature standard deviation, (e) coefficient of variation of monthly precipitation amount; time series of (g) mean annual temperature and (h) annual precipitation amount; empirical CDFs of (c) mean annual temperature and (f) annual precipitation amount. Model time series are RCMcorr. The analysis period is 1951–2000 in panels (g) and (h) and P0 (1961–1990) in all other panels.

For the air temperature standard deviation and total precipitation coefficient of variation, some differences between RCMcorr and the two observational datasets are present (Figure 5(b) and 5(e)). Whereas Aladin RCMcorr overestimated the air temperature variability in the summer and at the end of the year, bias-corrected RegCM3 and Promes reproduced the interannual variability of both temperature and precipitation reasonably well. However, the models were not successful in capturing the July maximum in the total precipitation coefficient of variation found in the DHMZ observations (Figure 5(e)). This maximum was also underestimated in the E-OBS data.

The time series of the mean annual air temperature and annual precipitation amounts in the 1951–2000 period (i.e. the model historical period) confirmed the previously discussed relationship among the five datasets: the E-OBS and RCMcorr air temperature and annual precipitation were lower than DHMZ observations (Figure 5(g) and 5(h)). The corresponding differences during the P0 (1961–1990) period were 1 °C for temperature and ranged from 568 to 580 mm for precipitation. The empirical CDFs of the annual data in the P0 (1961–1990) period again showed a constant shift of the E-OBS and RCMcorr air temperature and precipitation amounts from DHMZ observations (Figure 5(c) and 5(f)). Additionally, the Promes model tended to simulate higher air temperatures than did the other two models in the CDFs associated with the highest mean annual air temperatures in the P0 period.

The adjustment differences between RCMcorr and DHMZ in P0 were negative throughout the year (Figure 6); they ranged from −1.8 in the summer to −0.2 °C in December. The bias-corrected models, RCMcorr, were thus colder than the DHMZ observations, and the differences were statistically significant for all three models for most of the year, except in winter (Figure 6(a)). For precipitation, the adjustment differences (quotients) for all three models were negative (larger than 1.0). The differences were statistically significant for most of the year, except from May to July (Figure 6(b) and 6(c)). The magnitudes of the adjustment differences (and quotients) were similar in all three models, which was expected since the adjustment was applied on the already-corrected time series (RCMcorr) and not on the raw RCM output.

Figure 6

Adjustment differences for (a) mean monthly temperature and (b) mean monthly precipitation amounts and adjustment quotients for (c) mean monthly precipitation amounts for the Opuzen station. Differences and quotients were based on the 1961–1990 period. The availability of DHMZ observations in this period was 100%. Statistically significant differences according to the Wilcoxon–Mann–Whitney non-parametric rank-sum test at the 5% significance level are marked by the filled circles.

Figure 6

Adjustment differences for (a) mean monthly temperature and (b) mean monthly precipitation amounts and adjustment quotients for (c) mean monthly precipitation amounts for the Opuzen station. Differences and quotients were based on the 1961–1990 period. The availability of DHMZ observations in this period was 100%. Statistically significant differences according to the Wilcoxon–Mann–Whitney non-parametric rank-sum test at the 5% significance level are marked by the filled circles.

The impact of the adjustment procedure (for the case of RCMcorr_adj) can be seen as a shift in the model annual cycles and time series towards the DHMZ observations (Figure 7(a), 7(d), 7(g) and 7(h)). For precipitation, the model-adjusted annual cycles fully overlapped with observations; however, the precipitation coefficient of variation was reduced in all three models, which may be interpreted as deterioration in the RCMcorr_adj data. In contrast, the RCMcorr_adj2 preserved the precipitation coefficient of variation (dashed lines in Figure 7(e)) and was equal to the corresponding RCMcorr values.

Figure 7

Same as Figure 5 but for the adjusted model data RCMcorr_adj (full lines) and RCMcorr_adj2 (dashed lines).

Figure 7

Same as Figure 5 but for the adjusted model data RCMcorr_adj (full lines) and RCMcorr_adj2 (dashed lines).

For the projected climate changes of the RCMcorr (and RCMcorr_adj) data, an increase in the mean air temperature by the mid-21st century, i.e. in P1 (2021–2050) relative to P0 (1961–1990), was simulated by all three RCMs (RCMcorr; Figure 8(a)). The projected warming in the P1 period ranges between 0.5 and nearly 3.5 °C and, in most cases, it is statistically significant. The Promes model tended to simulate a larger temperature increase than the other two models. In contrast, the amplitude of the projected precipitation change varies greatly throughout the year from one model to another (between −60 and +60% in P1 relative to P0), although the variations were almost insignificant (Figure 8(b)). Since we estimated relative precipitation changes, all three types of the bias-adjusted timeseries (i.e. RCMcorr, RCMcorr_adj and RCMcorr_adj2) gave the same estimates of precipitation change in the 21st century projections.

Figure 8

P1 vs. P0 change for (a) monthly mean temperature (in °C); (b) relative monthly precipitation change (in %); (c) empirical CDFs of mean annual temperature in P0 and P1; (d) same as (c) but for annual precipitation amount. Time periods are: P0 1961–1990 and P1 2021–2050. Statistically significant differences in (a) and (b) according to the Wilcoxon–Mann–Whitney nonparametric rank-sum test at the 5% significance level are marked by solid circles. Statistically significant differences according to the Kolmogorov–Smirnov test at the 5% significance level between CDFs in the two periods for each model in panels (c) and (d) are marked by solid circles. Model time series are RCMcorr (panels (a) and (b) are identical for RCMcorr_adj and RCMcorr_adj2, by definition).

Figure 8

P1 vs. P0 change for (a) monthly mean temperature (in °C); (b) relative monthly precipitation change (in %); (c) empirical CDFs of mean annual temperature in P0 and P1; (d) same as (c) but for annual precipitation amount. Time periods are: P0 1961–1990 and P1 2021–2050. Statistically significant differences in (a) and (b) according to the Wilcoxon–Mann–Whitney nonparametric rank-sum test at the 5% significance level are marked by solid circles. Statistically significant differences according to the Kolmogorov–Smirnov test at the 5% significance level between CDFs in the two periods for each model in panels (c) and (d) are marked by solid circles. Model time series are RCMcorr (panels (a) and (b) are identical for RCMcorr_adj and RCMcorr_adj2, by definition).

The warming signal is also present in the empirical CDFs of the mean annual temperature (Figure 8(c)); CDFs in the P1 and P0 periods for all three models are significantly different according to the Kolmogorov–Smirnov test at the 5% significance level. The corresponding CDFs for the annual precipitation are not significantly different (Figure 8(d)).

All three bias-corrected models simulated statistically significant increasing linear trends (and Sen's slope) in the mean annual temperature for the period 1951–2050: 0.19 °C/10 yrs in RegCM, 0.27 °C/10 yrs in Aladin and 0.31 °C/10 yrs in Promes (Table 3). For the period corresponding to the available DHMZ observations (1961–2015), all three models agreed with the observations in the sign of the trends, and the simulated trend slopes were close to the observed slope (i.e. 0.2 °C/10 yrs in DHMZ observations for Opuzen; Figure 3), except for RegCM3. The trends in the mean seasonal temperature were highest for the summer season and when using the Promes model; in most cases, they were statistically significant (Table 3).

Table 3

Decadal trends (linear trend and Sen's slope (in brackets)) of seasonal and annual means of air temperature according to the three RCMs (based on RCMcorr and RCMcorr_adj), for the Opuzen station

  Period RegCM3 Aladin Promes 
MAM 1961–2015 0.04 (0.01) 0.30 (0.30) 0.34 (0.36) 
1951–2050 0.18 (0.17) 0.24 (0.23) 0.31 (0.31) 
JJA 1961–2015 0.15 (0.13) 0.40 (0.37) 0.14 (0.14) 
1951–2050 0.21 (0.22) 0.37 (0.36) 0.39 (0.38) 
SON 1961–2015 0.19 (0.22) 0.30 (0.31) 0.06 (0.05) 
1951–2050 0.20 (0.19) 0.28 (0.29) 0.27 (0.28) 
DJF 1961–2015 0.05 (0.05) 0.15 (0.15) 0.31 (0.32) 
1951–2050 0.15 (0.16) 0.19 (0.19) 0.27 (0.27) 
Annual 1961–2015 0.10 (0.07) 0.28 (0.29) 0.20 (0.20) 
1951–2050 0.19 (0.19) 0.27 (0.27) 0.31 (0.30) 
  Period RegCM3 Aladin Promes 
MAM 1961–2015 0.04 (0.01) 0.30 (0.30) 0.34 (0.36) 
1951–2050 0.18 (0.17) 0.24 (0.23) 0.31 (0.31) 
JJA 1961–2015 0.15 (0.13) 0.40 (0.37) 0.14 (0.14) 
1951–2050 0.21 (0.22) 0.37 (0.36) 0.39 (0.38) 
SON 1961–2015 0.19 (0.22) 0.30 (0.31) 0.06 (0.05) 
1951–2050 0.20 (0.19) 0.28 (0.29) 0.27 (0.28) 
DJF 1961–2015 0.05 (0.05) 0.15 (0.15) 0.31 (0.32) 
1951–2050 0.15 (0.16) 0.19 (0.19) 0.27 (0.27) 
Annual 1961–2015 0.10 (0.07) 0.28 (0.29) 0.20 (0.20) 
1951–2050 0.19 (0.19) 0.27 (0.27) 0.31 (0.30) 

Statistically significant trends at the 5% level according to Mann–Kendall test are in bold. Units are °C/10 yrs.

For annual precipitation in the period 1951–2050, RegCM3 and Aladin simulated an increasing trend, whereas Promes simulated a decreasing trend (Table 4); however, these trends are not statistically significant. For the period 1961–2015, when the DHMZ annual precipitation amounts showed a non-significant decreasing trend (i.e. −27 mm/10 yrs; Figure 4), only Aladin simulated a (non-significant) decreasing trend (–5.87 mm/10 yrs). The climate change signal of the ENSEMBLES simulations was weak for the first part of the 21st century, and significant trends in seasonal precipitation consistent in all three RCMs were not found (Table 4). As for the air temperature trends, the additive adjustment of RCMcorr to the l DHMZ observations (i.e. change from RCMcorr to RCMcorr_adj) shifted only the annual and seasonal time series; no changes in the trends' amplitude or sign were present. However, although multiplicative adjustment of the precipitation amount (or any other flux variable) is a standard approach in practice (for example, it ensures no negative precipitation amounts in the adjusted time series), it introduced additional uncertainty into the climate projections by changing the trend amplitude (and, in some cases, even the sign of the trend estimate, e.g. RegCM3, MAM, 1951–2050). Although it is not the topic of this study, even more advanced bias correction/adjustment methods (e.g. applied in deriving RCMcorr from the raw RCM simulations and based on the E-OBS observations) may suffer from the same problem. In general, all bias correction/adjustment methods are only statistical approaches needed because of the existence of systematic errors in the climate models and because of the methodological limitations when comparing point observations (e.g. DHMZ local observations) and gridded (E-OBS and climate models) products that represent the mean values over the areas defined by the grid spatial resolution.

Table 4

Decadal trends (linear trend and Sen's slope (in brackets)) of seasonal and annual means of the total precipitation amount according to the three RCMs (based on RCMcorr, RCMcorr_adj and RCMcorr_adj2) for the Opuzen station

  Period RCMcorr/RCMcorr_adj
 
RCMcorr_adj2
 
RegCM3 Aladin Promes RegCM3 Aladin Promes 
MAM 1961–2015 –5.14 (–8.18) –3.99 (–4.4) –4.19 (–4.43) –8.7 (–12.31) –4.91 (–8.33) –7.66 (–8.15) 
1951–2050 –0.27 (–0.71) 0.9 (1.16) 3.56 (3.88) 0.13 ( − 0.46) 1.81 (1.97) 6.01 (5.98) 
JJA 1961–2015 1.14 (0.5) –1.31 (–1.18) 6.7 (6.51) 2.01 (1.43) –3.64 (–1.71) 14.08 (14.7) 
1951–2050 1.75 (1.47) –0.57 (–0.47) –1.93 (–1.79) 3.98 (2.98) –0.83 (–1.29) –2.44 (–2.33) 
SON 1961–2015 8.76 (5.55) 10.5 (11.66) 0.86 (3.7) 16.24 (12.21) 21.39 (23.84) –1.21 (2.4) 
1951–2050 –3.29 (–3.48) 6.92 (6.87) –0.48 (0.91) –6.16 (–6.66) 13.65 (13.14) –0.33 (2.1) 
DJF 1961–2015 8.29 (14.08) –11.43 (–9.76) 4.27 (4.2) 13.56 (24.03) –20.77 (–18.08) 9.85 (10.24) 
1951–2050 3.48 (3.7) –6.15 (–6.66) –2.99 (–3.3) 5.48 (5.72) –11.11 (–11.99) –5.7 (–6.16) 
Annual 1961–2015 15.64 (18.93) –5.87 (–8.92) 9.05 (11.2) 28.24 (33.85) –7.21 (–10.04) 17.4 (18.01) 
1951–2050 2.22 (1.14) 0.28 (–1.13) –1.61 (–0.06) 4.38 (2.55) 2.05 (–3.71) –2.14 (–0.27) 
  Period RCMcorr/RCMcorr_adj
 
RCMcorr_adj2
 
RegCM3 Aladin Promes RegCM3 Aladin Promes 
MAM 1961–2015 –5.14 (–8.18) –3.99 (–4.4) –4.19 (–4.43) –8.7 (–12.31) –4.91 (–8.33) –7.66 (–8.15) 
1951–2050 –0.27 (–0.71) 0.9 (1.16) 3.56 (3.88) 0.13 ( − 0.46) 1.81 (1.97) 6.01 (5.98) 
JJA 1961–2015 1.14 (0.5) –1.31 (–1.18) 6.7 (6.51) 2.01 (1.43) –3.64 (–1.71) 14.08 (14.7) 
1951–2050 1.75 (1.47) –0.57 (–0.47) –1.93 (–1.79) 3.98 (2.98) –0.83 (–1.29) –2.44 (–2.33) 
SON 1961–2015 8.76 (5.55) 10.5 (11.66) 0.86 (3.7) 16.24 (12.21) 21.39 (23.84) –1.21 (2.4) 
1951–2050 –3.29 (–3.48) 6.92 (6.87) –0.48 (0.91) –6.16 (–6.66) 13.65 (13.14) –0.33 (2.1) 
DJF 1961–2015 8.29 (14.08) –11.43 (–9.76) 4.27 (4.2) 13.56 (24.03) –20.77 (–18.08) 9.85 (10.24) 
1951–2050 3.48 (3.7) –6.15 (–6.66) –2.99 (–3.3) 5.48 (5.72) –11.11 (–11.99) –5.7 (–6.16) 
Annual 1961–2015 15.64 (18.93) –5.87 (–8.92) 9.05 (11.2) 28.24 (33.85) –7.21 (–10.04) 17.4 (18.01) 
1951–2050 2.22 (1.14) 0.28 (–1.13) –1.61 (–0.06) 4.38 (2.55) 2.05 (–3.71) –2.14 (–0.27) 

Statistically significant trends at the 5% level according to Mann–Kendall test are in bold. Units are mm/10 yrs.

CONCLUSIONS

In this paper, the present climate characteristics of the lower Neretva river valley were presented based on the time series of air temperature and precipitation amounts for the Opuzen station. An increase in air temperature during spring and summer was detected, whereas a decreasing tendency was observed in precipitation amounts during summer, fall and winter seasons, as was a slight increase in spring. These results imply the possibility of an increase in the evapotranspiration rates. At the same time, runoff and the filling of aquifers in autumn and winter could have been reduced because of the negative precipitation trends in those seasons. The obtained results for the present climate also served as a basis for the validation of the simulated climate using ENSEMBLES/CC-WaterS RCMs. The results showed that the E-OBS dataset and the RCMs had a lower air temperature than did the DHMZ observations for the location of the Opuzen station. For precipitation, an underestimation of DHMZ observation by E-OBS and RCMs was seen and was particularly large in the colder part of the year. The empirical CDFs of the annual data in the historical period again showed a constant shift of the E-OBS and RCMcorr air temperature and precipitation amounts from corresponding observations for the Opuzen station.

The three bias-corrected RCMs simulated statistically significant increasing trends in the mean annual temperatures for the period 1951–2050 of between 0.19 °C/10 yrs and 0.31 °C/10 yrs. For the period corresponding to the available DHMZ observations (1961–2015), all three RCMs agreed with observations in the sign of the trends, and the simulated trend slopes were close to the observed slope except for the RegCM3 model.

RegCM3 and Aladin simulated a non-significantly increasing trend in annual precipitation for the period 1951–2050, whereas Promes simulated a non-significantly decreasing trend. For the period 1961–2015, when the DHMZ annual precipitation amounts showed a non-significantly decreasing trend, all models simulated a trend of the opposite sign. The comparison of two adjustment methods (additive and multiplicative correction) generally showed minor impacts on seasonal and annual time-scales.

The results of this study should be revisited using new generations of the RCMs and the emission scenarios; however, no major differences from this study are expected for the near-future projections.

ACKNOWLEDGEMENTS

This study has been partly supported by the EU project ‘DRINK-ADRIA’ (Programme IPA ADRIATIC CBC 2007-2013) and by the Croatian Science Foundation (HRZZ) under the project 2831 (CARE). We thank the three anonymous reviewers for their valuable suggestions.

REFERENCES

REFERENCES
Castro
,
M.
,
Fernández
,
C.
,
Gaertner
,
M. A.
1993
Description of a meso-scale atmospheric numerical model
. In:
Mathematic, Climate and Environment
(
Díaz
,
J. I.
&
Lions
,
J. L.
, eds).
Masson
,
SA, Paris
, pp.
230
253
.
Cindrić
,
K.
,
Pasarić
,
Z.
&
Gajić-Čapka
,
M.
2010
Spatial and temporal analysis of dry spells in Croatia
.
Theor. Appl. Climatol.
102
,
171
184
.
Formayer
,
H.
&
Haas
,
P.
2010
Correction of RegCM3 model output data using a rank matching approach applied on various meteorological parameters
. In:
Deliverable D3.2 RCM Output Localization Methods
. (
accessed 17 November 2016
).
Gilbert
,
R. O.
1987
Statistical Methods for Environment Pollution Monitoring
.
Wiley
,
New York
.
Giorgi
,
F.
&
Lionello
,
P.
2008
Climate change projections for the Mediterranean region
.
Glob. Planet. Change
63
,
90
104
.
Hawkins
,
E.
&
Sutton
,
R.
2009
The potential to narrow uncertainty in regional climate predictions
.
Bull. Am. Meteorol. Soc.
90
,
1095
1107
.
Haylock
,
M. R.
,
Hofstra
,
N.
,
Klein Tank
,
A. M. G.
,
Klok
,
E. J.
,
Jones
,
P. D.
&
New
,
M.
2008
A European daily high-resolution gridded dataset of surface temperature and precipitation for 1950–2006
.
J. Geophys. Res.
113
,
D20119
.
IPCC
2014
Climate Change 2014: Synthesis Report
. In:
Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
(
Core Writing Team
,
Pachauri
,
R. K.
&
Meyer
,
L. A.
, eds).
IPCC
,
Geneva
,
Switzerland
.
Jacob
,
D.
,
Petersen
,
J.
,
Eggert
,
B.
,
Alias
,
A.
,
Christensen
,
O. B.
,
Bouwer
,
L. M.
,
Braun
,
A.
,
Colette
,
A.
,
Deque
,
M.
,
Georgievski
,
G.
,
Georgopoulou
,
E.
,
Gobiet
,
A.
,
Menut
,
L.
,
Nikulin
,
G.
,
Haensler
,
A.
,
Hempelmann
,
N.
,
Jones
,
C.
,
Keuler
,
K.
,
Kovats
,
S.
,
Kroner
,
N.
,
Kotlarski
,
S.
,
Kriegsmann
,
A.
,
Martin
,
E.
,
van Meijgaard
,
E.
,
Moseley
,
C.
,
Pfeifer
,
S.
,
Preuschmann
,
S.
,
Radermacher
,
C.
,
Radtke
,
K.
,
Rechid
,
D.
,
Rounsevell
,
M.
,
Samuelsson
,
P.
,
Somot
,
S.
,
Soussana
,
J. F.
,
Teichmann
,
C.
,
Valentini
,
R.
,
Vautard
,
R.
,
Weber
,
B.
&
Yiou
,
P.
2014
EURO-CORDEX: new high-resolution climate change projections for European impact research
.
Reg. Environ. Change
14
(
2
),
563
578
.
Koeck
,
R.
, (ed.).
2012
CC-WaterS Monograph: Climate Change and Impacts on Water Supply
.
Vienna Water
,
Vienna, VA
,
USA
, p.
215
.
Pal
,
J. S.
,
Giorgi
,
F.
,
Bi
,
X. Q.
,
Elguindi
,
N.
,
Solmon
,
F.
,
Gao
,
X. J.
,
Rauscher
,
S. A.
,
Francisco
,
R.
,
Zakey
,
A.
,
Winter
,
J.
,
Ashfaq
,
M.
,
Syed
,
F. S.
,
Bell
,
J. L.
,
Diffenbaugh
,
N. S.
,
Karmacharya
,
J.
,
Konare
,
A.
,
Martinez
,
D.
,
da Rocha
,
R. P.
,
Sloan
,
L. C.
&
Steiner
,
A. L.
2007
Regional climate modeling for the developing world: the ICTP RegCM3 and RegCNET
.
Bull. Am. Meteorol. Soc.
88
,
1395
1409
.
Patarčić
,
M.
,
Gajić-Čapka
,
M.
,
Cindrić
,
K.
&
Branković
,
Č.
2014
Recent and near-future changes in precipitation-extremes indices over the Croatian Adriatic coast
.
Clim. Res.
61
,
157
176
.
Romić
,
D.
,
Zovko
,
M.
,
Romić
,
M.
,
Ondrašek
,
G.
&
Salopek
,
Z.
2008
Quality aspects of the surface water used for irrigation in the Neretva Delta (Croatia)
.
J. Water Land Dev.
12
,
59
70
.
van der Linden
,
P.
&
Mitchell
,
J. F. B.
2009
ENSEMBLES: Climate Change and its Impacts: Summary of Research and Results From the ENSEMBLES Project
.
Met Office Hadley Centre
,
Exeter
,
UK
.
Zaninović
,
K.
,
Gajić-Čapka
,
M.
,
Perčec Tadić
,
M.
,
Vučetić
,
M.
,
Milković
,
J.
,
Bajić
,
A.
,
Cindrić
,
K.
,
Cvitan
,
L.
,
Katušin
,
Z.
,
Kaučić
,
D.
,
Likso
,
T.
,
Lončar
,
E.
,
Lončar
,
Ž.
,
Mihajlović
,
D.
,
Pandžić
,
K.
,
Patarčić
,
M.
,
Srnec
,
L.
&
Vučetić
,
V.
2008
Klimatski Atlas Hrvatske / Climate Atlas of Croatia 1961–1990, 1971–2000
.
Državni hidrometeorološki zavod / Meteorological and Hydrological Service
,
Zagreb
.
Wessel
,
P.
&
Smith
,
W. H. F.
1996
A global self-consistent, hierarchical, high-resolution shoreline database
.
J. Geophys. Res.
101
,
8741
8743
.