Climate change projections have been incorporated into the upper Raba River SWAT model as variant scenarios with the use of the delta change method. This commonly applied method is based on so-called monthly change rates, obtained from the climate change models with respect to the original historical climate data (reference period) [29, 30].
In the current study two sets of projections, using precipitation and temperature data, have been used. The first set included scenarios for the single synoptic station (WMO 12566, 19°47'42''E, 50°04'40''N) (point approach, P), and has been applied in the previous modeling studies for this area [28, 31]. The second set has been prepared for the purpose of this study and includes E-OBS gridded data on a spatial resolution of 0.1° regular longitude-latitude grid for the area of the upper Raba River catchment (areal approach, A). In both approaches, described in detail (Supplementary S2.2.), 14 different future climate GCM-RCM model pairs were chosen as the best reflecting the annual cycle of both precipitation and air temperature (Supplementary Table S4). In the current study two historical periods were taken into account: 10-year (2006–2015) and 30-year (1981–2010). Moreover, the climate projections were calculated for two time horizons: near (2026–2035) and far future (2046–2055), and for two Representative Concentration Pathways: RCP 4.5 and RCP 8.5 to follow the original scenarios developed to facilitate climate change adaptation in Poland [32].
Previous studies indicated that precipitation and its projected changes have a decisive impact on the initiation and transport of soil particles both in the land and river bed phase [e.g. 33–35]. Therefore, although both precipitation and temperature change projections were implemented in the suspended sediment load simulations, only the results of precipitation scenarios were analyzed. Moreover, to elucidate the impact of rainfall amount on load analyses, two outermost sub-ensembles, dry and wet, have been distinguished based on the already selected GCM-RCM model pairs (Supplementary Table S5). Finally, 20 variant scenarios have been analyzed in the current study (Fig. 2). Predicted precipitation changes for all variant scenarios have been presented in Supplementary S2.3.).
Pitfall No. 1: Choice of the precipitation data source
Lack of ready-to-use climate change projections for catchments of interest, and needless to say, the lack of appropriate knowledge among the majority of environmental modelers to prepare climate scenarios themselves, forces the latter to use locally available sources of such projections. We have tested projections built on the data from a single station (point approach) which was dictated simply by the availability of projections with monthly resolution, prepared and published nationwide [32]. Although the selected station was localised outside of the modeled catchment, the distance to the model simulation cross-section and elevation difference were in between the World Meteorological Organization's (WMO) guidelines [36]. However, taking into consideration that representativeness of such data quickly decreases in mountainous areas, constituting the upstream part of the studied catchment, the areal approach has been also taken into consideration for comparative purposes. In this approach the diverse topography of the studied catchment, causing large differences in the distribution of rainfall [37, 38] was reflected more accurately.
Here, both scenario approaches (point and areal) suggested a future increase in yearly sediment loads compared to the baseline scenario (approx. 2,700 t/y on average), however, larger increases are expected for areal (up to 4,277 t/y) than for point scenarios (up to 2,833 t/y) (Supplementary Table S6). As for monthly load distribution (Fig. 3; Supplementary S2., S3.) both approaches indicated that an unusual increase of loads delivered to the reservoir can be expected in April, by over 1700 t/m, and almost 1900 t/m on average for point and areal scenarios, respectively. The second peak of the sediment loads delivery shall be expected in the summer months (June-July), reflecting precipitation variability in this area with distinctly higher rainfall in these months [39, 40]. Although the monthly trend is similar for both approaches, the higher summer loads for areal scenarios should be observed, resulting from higher precipitation changes predicted for the entire catchment area than for the single meteorological station adopted originally. Therefore, especially in mountainous areas, the areal approach should be adopted as better reflecting a catchment’s characteristics and used whenever possible.
Pitfall No. 2: Choice of the reference periods
Lack of explanation for Aprils’ (more than 1 April) extreme loads in this catchment, also observed in other areas where climate change projections were adopted from the very same source [32, 41, 42], prompted an interest in the impact of the reference period on the sediment simulations. Typically, 30 years are used as a climatic reference period in line with WMO’s recommendations [43]. Since such a period is long enough to filter out interannual variations in climate parameters or anomalies, it is yet short enough to show climate trends. However, many studies to date show that these guidelines are followed in quite a limited way in predicting future changes in catchment ecosystems. The limited access to quality controlled and homogenized historical data in some areas frequently forces environmental modelers to shorten the reference period down to 10–20 years [e.g. 44–49].
Here, comparison of sediment loads in the areal approach based on the 30-year long reference period (1981–2010) and the 10-year period, selected to cover the same period as used in the point approach (2006–2015), interestingly did not show significant differences in average yearly loads. In both sets of simulations sediment loads delivered to the reservoir were higher by over 4,000 t/y from the baseline scenario values, but differed from each other only by approx. 2% (Supplementary Table S6). However, the monthly distribution of sediment loads turned out to be distinctly different for simulations based on both reference periods (Fig. 4) with no Aprils’ extremum for the longer reference period.
Only the detailed analysis of monthly delta change rates for both sets of precipitation projections allowed us to finally understand the mechanism of April's anomaly. The average precipitation for the month of April in the period of 2006–2015 was lower by almost 42% when compared to 1981–2010. Therefore, the delta change rates for this month also notably differed between both periods, reaching 90% and 35% in the near and far future projections, respectively. Which resulted in a significant increase in precipitation for the simulations based on the short term (2006–2015) reference period. Such an impact of low and high delta changes for the catchment SWAT modeled parameters has been observed before [50], and here resulted in an approx. 1,185 t/m difference of sediment loads introduced into the reservoir in April.
It should also be noted that for precipitation sums, unlike temperature, no statistically significant trends related to climate change are observed in Central Europe [37]. Which is also noticeable when monthly precipitation is compared decade by decade for the catchment area (Supplementary Fig.S5). Therefore, selection of short reference periods when implementing model variant scenarios without their careful examination, may lead to random over- or underestimation of sediment loads in individual months.
Pitfall No. 3: Choice of the model ensemble
The decisive impact of precipitation on sediment loads prompted in turn an interest in rainfall performance in the GCM-RCM model pairs applied in the current study. Generally, the selection of the final model ensemble for this particular area was performed based on the skill to simulate present and near-past climate, or the ability to represent the same connection pattern that drives the climate of the studied region [51, 52]. Climate models with fundamental errors (e.g., unrealistically represented processes) should be disqualified as they cannot be improved by statistical postprocessing [24]. Although the bias adjustment (BA) procedure is dedicated to amend the raw model output, it has been proved that if applied without considering the underlying processes it may introduce even large artifacts and constitute itself a significant source of uncertainty [24, 53].
For environmental studies moisture content seems to be crucial, nevertheless, it is only occasionally observed in terms of the models’ moisture variability [52, 54–58], and its impact on environmental parameters is rarely studied. Here, we have used a probabilistic method to identify wet and dry models and subsequently to estimate the range of possible extreme values in sediment loads. As expected, variant scenarios based on the extracted wet and dry sub-ensembles translated into highs and lows of the expected future yearly sediment load delivery into the reservoir. Simulations performed with use of both sub-ensembles revealed that the difference in total sediment loads could exceed 3,000 t/y (Supplementary Table S6). What is more important, our results also show that use of only dry and wet models significantly changes the pattern of sediment monthly loads (Fig. 5). For dry sub-ensembles the peak of sediment delivery is expected in May-June, while for wet sub-ensembles the high load delivery period is shifted to early spring. Moreover, this peak can last from March to August, which results from projected precipitation changes (Supplementary Fig.S4). Moreover, it should be noted that the extreme sediment loads in the dry sub-ensemble may exceed the wet sub-ensemble projections in a given month. Therefore, a composition of model ensemble, in terms of wet and dry projections, is of particular value for environmental modelers dealing with pollutants transported via particles eroded from the catchment. Here, if only wet models in the ensemble were taken into consideration, the associated peak of sediment loads would be displayed earlier and coincided with the pre-vegetation period of increased fertilization in the analysed catchment. This would mean an increased risk in reservoir contamination with biogenic compounds, and consequently, a higher risk of eutrophication.
Integration steps
It is generally accepted that current changes in the climate system and those expected in the future will increasingly have significant impacts on ecosystems. Since adaptation plays a key role in reducing risks and vulnerability from these changes, the need for reliable assessments of their impacts on the environment is drastically growing. The results of climate, and consequently of catchment models, are increasingly becoming the basis for shaping water policy and management for the upcoming decades. It is therefore not surprising that the pressure on climate scientists, whose work plays a key role here, intensifies extremely. Both types of modeling, climatological and environmental, are performed by specialists expertly, but unfortunately not jointly.
As pointed out above, the abundance of climate change projections and lack of clear information on model representativeness for particular regions and applications on one side overwhelms their potential recipients and users. And while additionally combined with a lack of specialist knowledge and/or sufficient data sources on the other, leads to a gross over- or underestimation of modeled environmental parameters. Moreover, this situation leads to confusion and deepening distrust towards climate change forecasts, and consequently to ignoring or disregarding warings on future scenarios.
Presented here, three selected pitfalls help to illustrate how the choice of meteorological data, reference period, and model ensemble can affect sediment load estimations. For our study venue (Carpathian catchment delivering suspended sediment to the dammed reservoir) calculated differences in average future loads could reach up to even 6,000 tons of sediment per year. However, large differences in monthly loads (up to 2,000 tons) are also visible in selected months. In point of fact, all precipitation variability driven parameters and processes in the catchment (e.g. runoff, flow, and flooding) can be subjected to such discrepancies depending on the climate change scenario choice.
Our example and personal experience show that awareness of what the work of both groups of modelers requires is minimal and there is a lack of communication prevailing. It is now time for better integration between climatologists and environmental modelers and to focus attention on joint projects, workshops, conferences, and publications. These activities should concentrate mainly on mid-scale cooperation to elucidate regional climate and environmental peculiarities, and to introduce future climate multi-model specific impact ensembles in order to facilitate an informed choice of available climate information. We propose as well development of the concept of rigorous science, already existing in climatology, taking into account biases in climate model simulations [24], choice of meteorological data, reference period, and not least model ensemble.