3.1. Hydrological Conditions during the Period of Study
Many storms reached the coast during the measurement period. Using the definition of storm surge described in Section 2.4
(based on sea levels of at least 0.45 m above mean sea level), a total of 61 storms affected the cliffs in Bansin, compared to 43 in Międzyzdroje and 62 in Wicie. The distribution of surges was highly uneven, as shown in Figure 3
. The most intense period lasted from late November 2016 to mid-January 2017. Around 10 surges were distinguished during that period, with water levels exceeding 1.4 m above average at all locations on 4–5 January 2017. This water level corresponded to an event with a return period of 15–20 years [44
]. The maximum water level of 1.55 m was observed at the Koserow tide gauge close to the Bansin cliffs. Conversely, the waves reached their maximum height throughout late 2016, culminating on 7 December 2016.
Another period of stormy weather lasted from mid-October 2017 to early January 2018, during which around 20 surges affected the coast. However, neither the water levels nor the wave heights were as extreme as those during the 2016–2017 storm season. The most intense storm in the 2017–2018 storm season occurred around 29–30 October 2017 during which the water levels slightly exceeded 1 m above mean in all study areas. Considering the stricter definition of storm surge presented by Wiśniewski and Wolski [44
], i.e., the exceedance of water levels of 0.6 m above mean, the first half of the study period had three times more storms than the long-term average of about four per year, including a very unusual occurrence of a storm surge in June; the second half of the study period was close to an average year.
3.2. Descriptive Analysis of Cliff Erosion
During the monitoring period, the sediment budget was definitely negative with a total loss of 49,330 m3. Erosion was most significant on the cliffs (over 58,000 m3), while a positive budget was observed on the beaches, with a value slightly exceeding 9000 m3. This positive balance shows that not all of the cliffs’ material was swept into the sea, but some of it remained on the beaches.
Erosion and sedimentation were unevenly distributed in time and space (Figure 4
). At the beginning of the 2016–2017 storm season, erosion was principally visible on the beach (over 80% of total erosion volume in Bansin and Międzyzdroje). As the successive lowering of beach proceeded, the proportions changed, and the cliff erosion started to dominate, reaching over 85% of the total erosion volume. Due to the very narrow beach, the Wicie area suffered cliff-dominated erosion of more than 90% of the total loss in this coast section. In fact, the sediment budget was obviously negative both for the beach and cliff during the winter season. The maximum negative volume of eroded material measured between the third and fourth topographic campaigns was also the highest during the monitoring period. Erosion volume on the beach varied at different test sites, reaching from 627 to 2191 and 2566 m3
for Międzyzdroje, Bansin, and Wicie beaches, respectively. However, the first group of severe storms affected the cliff face much stronger than the beach, exceeding the maximum volumes of 6000, 12,000, and 18,000 m3
for Międzyzdroje, Bansin, and Wicie cliffs, respectively. Notwithstanding the clear erosion dominance across the whole study area during the 2016–2017 storm season, the retreat of the cliff top was relatively small compared to changes of the 1 m contour line and the cliff base line. While the cliff top retreated by a maximum of 11 m in Wicie, the average change on all areas was less than 1 m, and the median was only 0.03 m. The maximum changes of shoreline and cliff base lines were similar, reaching around 11 m. However, the average change of shoreline and cliff base lines of 2.5 and 1.3 m, respectively, as well as medians of 1.7 and 0.15 m, respectively, suggested more even distribution.
The period between storm seasons contained higher variability in both the time and space distributions, even though the total volumes were much lower. Furthermore, the compilation of the next five surveys revealed both accumulation and erosional patterns with a rather modest positive overall sediment budget (1800 m3). Before the 2017 winter season approached, the dominant processes were much weaker, but cliff erosion still occurred along with the overall recovery of beach height and length. The volume values between surveys fluctuated from –2870 to 9280 m3 and –3520 to 3683 m3, respectively, for beach and cliff. However, the negative values for the beach and the positive for the cliffs were a consequence of landslide processes that pushed the cliff base line in the seaward direction rather than significant erosion or deposition episodes.
The second period of stormy weather as well as the following spring season (2017–2018) revealed strong similarities to the corresponding earlier periods. This observation was supported by a comparison of data from the last four topographic surveys. Erosion was still principally visible on the cliffs, though the water levels and wave heights were not as extreme as those during the 2016–2017 storm season. The much weaker waves were not able to clean all the debris, and in some of the investigated areas, the cliff base line migrated seawards, and the volume values presented an inverse pattern to what was observed during the first storm season. The after-storm period was again characterized by beach recovery processes.
3.3. Statistical Analysis of Cliff Erosion
The statistical analysis was performed using the BN presented in Figure 5
. The final model, constructed following the procedure explained in Section 2.4
., included five cliff erosion indicators explained by two morphological factors, eleven hydrological factors, and two meteorological factors. The morphological factors were additionally explained by two hydrological factors and one meteorological factor.
The shoreline is the most dynamic component of the coastline; therefore, its changes (Shore) have the highest number of explanatory variables. The highest correlation was observed with the 95th percentile of wind speed (WindSpeed_95), which gave a slightly higher correlation than the wave height indicators. A likely explanation for this relationship is that wind is more dynamic than offshore waves containing significant inertia and hence is a better predictor of the small wind-driven waves that contribute to shoreline retreat. The second factor influencing shoreline retreat was the width of the beach (Width) before the occurrence of erosion. Wider beaches have more material to be eroded, resulting in larger shoreline retreat. The beach width was influenced by both the maximum wave height (P_WaveHeight_Max), which resulted in shorter beaches, and the average temperature (P_Temp_Avg), which is an indicator of the time of the year, as beaches tend to be shorter during the autumn and winter storm season than during the warmer spring or summer. Other factors contributing to shoreline retreat were the 95th percentile of water levels (WaterLevel_95), average wave direction during storm surges (WaveDirect_Storm), and average wave peak period (WavePeakPer_Avg), all of which resulted in higher and longer waves attacking the shoreline, resulting in erosion.
Beach volume balance (Beach) was highly correlated (0.72) with shoreline retreat, which incorporated the influence of several factors. The average water level during storms (WaterLevel_Storm) further contributed to beach erosion, as higher baseline sea levels allowed waves to reach further onto the beach, while the 95th percentile of significant wave height (WaveHeight_95) indicated the importance of high waves in beach erosion.
Cliff foot retreat (Foot) showed a relatively low correlation (0.24) with beach volume balance, as more complex mechanisms were observed: material from cliff erosion could be deposited on the beach, which would result in a weak dependency between beach and cliff erosion. However, some of the waves eroding the beach still cut into the cliff. Specifically, waves that were both particularly high and long contributed to cliff foot retreat, as revealed by the wave power (WavePower_95) indicator, which was proportional to the product of significant wave height and the mean wave period. Additionally, the cliff was more prone to erosion if more vertical than inclined, as shown by the cliff slope (Slope) variable. The cliff slope showed the highest correlation with the average mean wave period in the preceding period (P_WaveMeanPer_Avg), where stormy periods resulted in lower cliff slopes due to erosion.
The cliff volume balance (Cliff) depended primarily on waves undercutting the cliff, resulting in the eventual collapse of the cliff. Erosion was further increased by very high waves, as shown by the accumulated excess energy (Storm_AEE) indicator. The accumulated excess energy indicator represented the energy of waves above a 2 m threshold (including sea level), which was close to the average elevation of cliff foots in the study area; hence, this indicator counted only the waves that actually eroded the cliff. Two other variables correlated with the cliff volume balance were the maximum mean wave period (WaveMeanPer_Max), which indicated the occurrence of very long waves, and the maximum water level (WaterLevel_Max), as the high baseline sea level increased the number of waves that could reach the cliff.
Finally, erosion of the cliff could also result in retreat of the cliff top (Top). This erosion indicator was the least dynamic and depended mostly on factors already included in previous erosion indicators. Some correlation existed with the total precipitation recorded during storms (Prec_Storm), as rainfall could weaken the structure of the cliff, making it more susceptible to collapse. Other factors showed only a small conditional correlation; the largest was for the maximum wave height (WaveHeight_Max), which indicated the occurrence of extreme waves having the biggest impact on the cliff.
The model was validated by analyzing the correlation between predicted and observed changes in the variables of interest (Table 2
). This was carried out for different choices of input sample, thus analyzing how transferable is the model between locations. The small sample size resulted in a non-negligible variation of results between different model runs; therefore, the results shown are averages of 100 model runs per each variant of location or sample source. A split-sample validation (using half of the data as input sample, and the other half to run the model) showed only marginally lower performance than using the same data for both purposes. Of the three study sites, data from the Bansin cliff is the most transferable. For individual variables, the highest correlation between modeled and observed data is for beach volume balance, followed by shoreline retreat and beach width (correlations of 0.4-0.6). Correlations for cliff foot and volume balance are in the 0.3–0.4 range, and lower for the cliff top, which was the least dynamic part of the cliff in the timeframe of the study.