Outcome of environmental change from historical sediment discharge in a Mediterranean fluvial basin, 1500–2019 CE

The Sediment Discharge Model (SEDIM_ARB)—equation (1)—summarises, through mean areal rainfall erosivity, winter precipitation amount, monthly storm-severity index values (MSSI), and woody vegetation inputs, the relationship between temporal patterns of climate and SD, consistent with a sample (1954–1984 CE) of annual observation data (see Material and Methods in the Supplement (available online at stacks.iop.org/ERC/3/071002/mmedia)). For the calibration period 1954–1984 CE, we obtained the coefficients A = 5.162628·10⁻⁴ MJ mm⁻¹ hm⁻² h⁻¹ yr⁻¹, α = 1.00 Mg MJ⁻¹ mm⁻¹ h, β = 0.10 Mg MJ⁻¹ mm⁻¹ h, φ = 2.80 Mg km⁻¹ mm⁻¹ yr⁻¹ and σ = 1.00 Mg km⁻¹ mm⁻¹ yr⁻¹ in Eqs. (1), (2) and (3). There is no indication of marginal serial correlation in the residuals (Durbin-Watson statistic p > 0.05) and a significant relationship exists between observed and predicted values according to the ANOVA test (p < 0.05). The data-points are aligned within 95% prediction bounds, with negligible differences from the theoretical 1:1 line, except for one data-point (not shown). Taking out the 1955 CE outlier of 140 kg s⁻¹ (∼3% of the dataset of 31 years), not considered for model calibration, we obtained an annual mean value of 42.8 ± 27.4 kg s⁻¹ over the study period, which is close to the mean of actual SD data (42.8 ± 29.2 kg s⁻¹). The regression line has an intercept a = −0.06 (±3.43) kg s⁻¹ and slope b∼1.00 (±0.07), near the optimum values (a = 0 kg s⁻¹ and b = 1). The Nash–Sutcliffe efficiency value equal to 0.89 indicates limited uncertainty in model estimates. The R² statistic equal to 0.88 means that the model explains 88% of the variability in actual SD data (figure 3(a)). The mean absolute error (MAE), equal to 7.5 kg s⁻¹, is lower than the standard deviation of the residuals (10.0 kg s⁻¹). In addition, MAPE (mean absolute percent error) equal to 25% and Kling-Gupta Efficiency (KGE) equal to 0.92 indicate satisfactory model performance and efficiency. F-ratios were highly significant for each independent variable, the highest p-value being 2.26·10^–5 for the term related to the BGE component.

Zoom In Zoom Out Reset image size

Figure 3(b) indicates the normal approximation of model residuals (normality test p > 0.05). The distribution of the percentiles of modelled SD (figure 3(c), orange curve) approaches the distributional shape of the observed erosivity, indicating a satisfactory prediction over the range of erosivity values (including low and high values). Figure 3(d) presents an indirect validation of the model, obtained by comparing the predicted SD with the observed soil sediment volume at the outlet of the ARB over the period 1951–2010 CE. This independent validation (extending over a longer time span than the calibration period) shows that the model follows fairly well (r = 0.79) the temporal pattern of erosional sediment processes occurring in the ARB (figure 3(d)).

We know from historical documentary sources that the ARB has always been subject to seasonal erosional changes due to its markedly torrential morphological nature. During intense or long-duration storms, the river has often produced marked fluctuations in runoff and floods and, in turn, sediment delivery. Sometimes, sediment was produced in large quantity, depending on the hydroclimatic forcing and the landscape structures that resisted to storm forcing (see Material and Methods in the Supplement).

Figure 4 shows the estimated time-series of SD with some related environmental changes, dating back to 1500 CE. The long-term fluctuations of the erosional activity determined in a lacustrine area of central Italy (42° 12' N, 12° 57' E) are in general agreement (figure 4(a)) with our reconstructed time-series of SD. In particular, until the middle of the 17th century, the SD time-series of the ARB had frequent low-erosion storm, with ∼110 kg s⁻¹ occurring once every two years (figure 4(d)). SD values of ∼370 kg s⁻¹ are present only once every 20–30 years (i.e., in the years 1515, 1532, 1547, 1571, 1589, and 1646 CE). The maximum value, 1905 kg s⁻¹, occurred in the year 1758 CE. Historical documents show that in the 14th and 15th centuries, the Arno riverbed underwent an unsafe uplift due to sediments that continued to settle and that the presence of bridges, locks and barriers ended up constantly favouring [33]. Thus, since the end of the 17th century SD has often become livelier, with annual values that are more frequently >300 kg s⁻¹ (about once every two years), while the values once expected every 20–30 years have increased strongly to 608 kg s⁻¹. The values of 1423, 1905, 873 and 868 kg s⁻¹ (which occurred in 1688, 1758, 1805, and 1868 CE, respectively) are considerably high, as they are SD estimates for the entire basin area. This means that even higher SD values may have occurred widely within the basin, leading to continuous discharge of sediments until the 19th century. According to Rossini (1855 [34], p. 27): Miglior ventura non dovette correre in altri tempi la valle superiore ed inferiore dell'Arno, nè pi ù prospera fu concessa alle nostre terre e paesi da quell'epoca (1761) fino a noi, benchè di continuo fossero praticati nell'Arno dei ripari considerevoli, poichè per la copia delle acque sempre crescente [...] vengono facilitate spesso le erosioni ed i trabocchi ('The upper and lower valley of the Arno did not have to run better in other times, nor more prosperous was it granted to our lands and villages from that time (1761) up to us, although considerable shelters were continuously practiced in the Arno, as for the ever-increasing copiousness of the waters [...] erosions and overflows are often facilitated'). Even if SD values have not increased in the ARB during the 20th century (figure 4(d)), gross erosion (GE), which represents soil movement in the basin, has in recent years reached remarkably high rates (figure 4(c)) of 2163, 1849, 1795 and 2150 Mg km⁻² yr⁻¹. These values, estimated in 1951, 1960, 2010 and 2014 CE, respectively, are comparable to those of the Little Ice Age (LIA; ∼1300–1850 CE), and exceed the soil loss tolerance of ∼500–1000 (Mg km⁻² yr⁻¹) [35, 36].

Zoom In Zoom Out Reset image size

Significant change-points in the SD time-series were detected in 1740 (SNHT-double shift), in the central part of the LIA, and in 1903 (SNHT-double shift and Mann-Whitney-Pettitt test), at the onset of modern warming (red arrows in figure 4(d)). The second change-point in the estimated SD time-series mainly coincides with a low erosive forcing during a landscape restoration, confirmed by the reforestation of the basin in the decades after 1903 CE. An earlier change-point, in the year 1687 CE, was detected in the GE time-series (figure 4(c), orange curve) with the SNHT-double shift (not shown). In fact, more frequent and severe hydrological events occurred in Italy at the turning point of the 17th and 18th centuries [27], approximately during the so-called Maunder Minimum, which began c. 1645 CE and lasted until c. 1715 CE, a period characterised by extremely rare sunspots and mostly below-average temperatures even for the LIA [37]. After GE values peaked at the end of the 17th century, there has been a continuous decline until the end of the 19th century. Then, GE increased again significantly throughout the 20th century (Mann-Kendall test p < 0.01), in a similar way as it had done in the central part of the LIA.

The difference between gross and net erosion (B = GE–NE) is a sink term representing the fraction of GE stored in the basin during the re-sedimentation process. The term B is derived from the sediment delivery ratio (SDR), also known as the connectivity ratio [38], which describes the extent to which eroded sediments are stored within the basin: B = (1– SDR)·GE [39]. In this way, the amount of sediment that reaches a stream over the amount of eroded soil characterises the efficiency of the slope-to-channel transfer, depending on the transport capacity, which in turn is modified by the effects of the slope shape and drainage pattern. The average long-term estimate of B for the ARB is 690 Mg km⁻² yr⁻¹, with SDR = 0.24. This estimate is higher than 44 to 235 Mg km⁻² yr⁻¹ reported for Alpine basins [40], with comparable values of the SDR (from 0.27 to 0.48). This indicates that high amounts of soil are mobilized in the ARB (on average, GE = 905 Mg km⁻² yr⁻¹) due to erosion in mountain ranges (which only lie along the northwestern margin of the basin; figure 2(c)) but a relatively high fraction is retained in the Marshy areas of the basin, which are large reservoirs for the collection of flood water and sediments.

At a time when the extension of pastures and an extensive cultivation of ploughed areas at the expense of forest was planned (figure 4(b)), a strong erosive forcing likely caused the incessant expulsion of tree stands and strong rainwater runoff that washed the fertile soil in the plains, causing the land to rise in the river valley, as had happened in the 14th and 16th centuries [41, 42]. This is confirmed by the forestry legislation of Tuscany during the time of the Medici family (1434–1737 CE), in which (our translation) 'renovations were all aimed at avoiding the washout of the Apennine slopes and the transport of earth debris downstream, causes of the progressive rise of the bed and mouth of the Arno and the potential swamping of the plains along the course of the river' (rinnovazioni tesero tutte ad evitare il dilavamento delle pendici appenniniche ed il trasporto a valle dei detriti terrosi, cause del progressivo innalzamento del letto e della foce dell'Arno e del potenziale impaludamento delle pianure lungo il corso del fiume; Belli, 1998 [43], p. 127).

The origin of this instability has not only been climatic, but has often been favoured by repeated and excessive exploitation of forest resources in the high hills and mountain areas to meet the growing demand for firewood, but has sometimes turned into real deforestation due also to the growing demand for agricultural and grazing areas [44], as shown by historical representations of the landscape around Florence. It is precisely during the 18th century that the ARB became most vulnerable, being subject to agro-forestry activities and deforestation [41, 45]. It is not a coincidence that the climate in the central and final parts of the LIA has provided some damaging scenarios, with storms of unprecedented strength. In particular, the most striking group of erosive events occurred between the end of the 17th century and the end of the 18th century, when the ARB experienced frequent high erosivity values of ∼2000 (MJ mm hm⁻² h⁻¹ yr⁻¹) [46]. Also recently this climatic phase has been identified as erosive, characterised by heavy runoff in the southern Carpathians [47] and flooding at most European sites [17].

Subsequently, the water of the river cleared again when turbidity levels dropped suddenly during the 20th century and have remained low until present. This was mainly due to a major change in land management policy after about 1850 CE, which promoted the reforestation of large mountain areas, the stabilization of mountain slopes and river banks, and the construction of dams and other retention systems [48]. However, sustained GE rates have also been maintained in the recent period. This is evident from figure 4(c) (blue curve), which shows a significant increase of extreme erosion values after ∼1900 CE. This is consistent with an intensification of the geo-hydrological hazard over the past five decades for the inner hilly areas of central Italy [49].

The hazard increase associated with water soil erosion in the basin indicates an interruption of high GE rates, together with a decrease in NE rates and, consequently, a decrease in SD values. During recent decades, for instance, there has been in the ARB a decrease in total precipitation with a significant decrease in the frequency of winter rainy days, while there has been an increase in extreme precipitation over short (one to three hours) time-scales [50]. If this was reported for the ARB, there has essentially been an absence of trends in the intensity of extreme events in the whole of Tuscany [51]. This reflects the data presented here, which show a decrease in long-distance solid transport associated with long precipitation events, and a substantial levelling, or weak rise, of erosion rates associated with short-distance soil movement in the basin, typical of heavy but short-duration rainfall. According to Brandolini et al [52], the degradation of slopes in Tuscany can be linked to the intensive management of cultivated lands and the increase in hazardous rainfall, which favour the acceleration of erosion processes. We highlight that to restore the sustainability of intensively managed land, the rural parts of the ARB should plan for the incorporation of hedges and trees in agricultural areas, thereby improving the hydrological and structural characteristics of the landscape (and limiting soil erosion in sub-basins) so as to increase soil fertility and, together with it, agricultural productivity.