Professional Documents
Culture Documents
Corigliano Et Al ICOLD2023Goteborg
Corigliano Et Al ICOLD2023Goteborg
ABSTRACT: A particularly intense meteorological event hit some Piedmont basins in the first
days of October 2020, with generally very heavy effects on the territory. In the Gesso valley,
however, the effects were minor due to the storage action of volumes of water carried out by the
reservoirs in the area. The article describes in the first part how the event was managed with the
Chiotas and Piastra reservoirs, and how it was possible to reduce the effects of the weather event
on the territory. Against these environmental benefits obtained, the pumping and the retention of
water volumes have however led to a greater damage of the machines due to the suspended ma-
terial. Moreover, the basins had to withstand a solid load that was not released downstream, and
which therefore contributed to the silting up of the basins. The rapidity of the water level increase
in the Chiotas reservoir required careful monitoring of the dam behaviour both during and after
the event. The second part of the article describes the analysis of some representative parameters
of the response of the Chiotas dam. The analysis is carried out in two steps, the first using algo-
rithms to identify anomalies to highlight any deviations from the historical series. If an anomaly
is identified, a detailed analysis is carried out to analyse the causes of the deviation from the
expected behaviour. The analysis of the representative parameters of the dam response has shown
that the rapid variation of the water level can in some cases produce a deviation from the expected
behaviour in ordinary conditions, which however can always be explained by introducing appro-
priate variables in the statistical models. In conclusion, even during the rapid change in the reser-
voir level occurred during the October 2020 event, the dam responded with no evidence of irreg-
ularities.
INTERNAL
1 INTRODUCTION
On 2-3 October 2020, a heavy perturbation caused exceptionally high precipitation over the Cu-
neo region, favoured by a thermal zero positioned above 3,000 m, which limited the snowfall to
the highest areas.
This was an intense event triggered by cold polar air and warm-humid currents from the Atlan-
tic as well as the Mediterranean, which from a meteorological point of view is very rare at our
latitudes, also considering the season in which it occurred.
The effects on the territory of this meteorological event were very heavy in some areas, with
major damage recorded in the Vermenagna valley. In the nearby Gesso valley, on the contrary,
thanks to the presence of a number of hydroelectric dams, the situation was quite different, as the
reservoirs gave a significant contribution to store part of the rainfall volumes.
From a pluviometric point of view, the event was characterised by absolutely exceptional pre-
cipitation heights. The precipitation values are extremely important: in the upper Tanaro valley,
the Limone Piemonte (CN) station recorded approximately 600 mm, almost all on the day of 2
October; extreme values were also recorded at the station in the municipality of Garessio (CN)
with over 400 mm. These values represent, at station level, more than 50% of the average annual
precipitation. In the stations most affected by the rainfall, the estimated return periods for duration
from 1 to 24 hours exceed 200 years, as verified in the case of Limone Piemonte (CN) and Chiotas
dam (CN), confirming the exceptionality of the event.
In the Gesso Valley, the event of 2-3 October caused violent torrential activity along the upper
sections of the streams, indicatively above an altitude of 1100/1300 m, while downstream there
were no particular situations of fluvial dynamics (ARPAP, 2020). Along the Gesso della Barra
there was an erosion on the left bank that removed the abutment of the bridge used to access the
Vallone di Monte Colombo. The abundant solid transport consisting of earth, detritus and plant
material that developed mainly along the Vallone della Barra and secondarily along the limestone
Vallone della Rovina flowed into the Piastra artificial reservoir (ARPAP, 2020).
One of the beneficial effects of the presence of dams and reservoirs in the territory is certainly
their lamination capacity, i.e. the reduction of peak flows to the benefit of protecting downstream
territories. This is a very site-specific capacity that depends on the type of reservoir, its volume,
the type of pre-existing discharges and the size of the upstream basin.
During the 2-3 October event, the Entracque power station began operating in pumping mode,
transferring part of the precipitation water volumes into the large Chiotas reservoir, with the effect
of generating an important lamination effect of the flood wave that had no significant effect down-
stream of the Piastra dam. The rapid filling of the Chiotas reservoir required careful monitoring
of the behaviour of the dam during and after the event.
The first part of the article illustrates the hydroelectric system constituted by the Chiotas, Pias-
tra and Rovina reservoirs and the Entracque hydroelectric power plant, and how they contributed
to limit the damage on the downstream territories due to this exceptional event. While the second
part analyses some representative monitoring parameters of the Chiotas dam behaviour, looking
for possible deviations from the historical series and, in the presence of an anomaly, a detailed
analysis is conducted to identify the causes of the deviation from the expected behaviour.
The Entracque hydroelectric power station discharges into the Piastra reservoir. This power sta-
tion connects the Piastra reservoir with the Chiotas and Rovina reservoirs by means of reversible
groups. In particular, the Chiotas reservoir is connected through eight groups whose maximum
flow rate is about 16 m3/s each; the same groups can operate in pumping mode from Piastra to
Chiotas with a maximum flow rate of about 15 m3/s each. On the other hand, Lake Rovina is
connected to the Piastra reservoir through a single group, whose maximum flow rate is 27.0 m3/s
during generation and approximately 19.0 m3/s during pumping. In the whole, the maximum total
flow that can be derived from the two reservoirs of Chiotas and Rovina into Lake Piastra is ap-
proximately 155 m3/s, while the maximum flow that can be pumped from Piastra upstream is
approximately 140 m3/s.
The Chiotas dam was built between 1970 and 1978 on the Bucera river and is a double-curva-
ture concrete arch-gravity type, with a height of 120 m and a crest length of approximately 230
m. The thicknesses at the dam shutters vary from 5 m at the crest to approximately 37.50 m at the
base (Figure 2). The direct catchment area has a surface of approximately 11.6 km2 and a total
reservoir volume of approximately 30 million cubic metres.
The Chiotas dam is equipped with a complex system of instruments to control the behaviour of
the structure that allow the monitoring of hydro-meteorological parameters, plano-altimetric dis-
placements of both the dam and the embankments, displacements of the imposed section, defor-
mations of the structure, temperatures of the wall mass, pressures at the rock-foundation contact,
piezometric levels and weirs for the measurement of leakages.
INTERNAL
Figure 3. Levels and flows at Chiotas during the 2-3 October 2020 event. Reconstruction using the balance
equation.
At the conclusion of the event, in the afternoon of 3 October, the Chiotas reservoir reached a
level of 1954.03 m a.s.l., corresponding to a total increase in volume of approximately 7x106 m3.
The temporal trend of the levels at Chiotas and of the natural tributary and discharged flows are
shown in the graph in Figure 3. For the sake of simplicity, only the trend of the levels of the
Chiotas reservoir is shown, since this is the largest of the three reservoirs and the one, therefore,
whose contribution in terms of flood wave lamination is greater. An analysis of Figure 3 shows
that the maximum inflow to the Chiotas reservoir was approximately 70m3/s, with a basin area of
11.6 km2. The specific areal contribution can therefore be quantified as approximately 6 m3/s/km2.
The maximum contribution of the entire basin closed at Piastra, on the other hand, was approxi-
mately 250 m3/s with a catchment area of 88.3 km2, corresponding to a contribution of 2.83
m3/s/km2. The areal contribution of the overall Piastra basin was obviously lower because the
intense central core of precipitation fell over the upper part of the basin, as well as in the nearby
French basin.
Symposium “Management for Safe Dams” - 91st Annual ICOLD Meeting – Gothenburg 13-14 June 2023
The maximum total flow withdrawn from the Piastra dam through the turbines of the Andonno
power station and through the opening of the outlets, however, was never more than 60 m3/s and
this was decisive in containing the maximum flows downstream of Piastra and after the conflu-
ence with the Gesso.
Analysing the graph in Figure 3, it can be seen that the maximum contribution in terms of
lamination by the three above-mentioned reservoirs is represented by the curve “Incremental flow
per volume variation” and can be quantified as approximately 200 m3/s. The corresponding water
volume filled the three reservoirs during the event and subsequently returned to the riverbed after
the event had passed.
A careful inspection of the machinery immediately highlighted the severe worn caused by the
sediments and the consequent need for extraordinary, not foreseen, maintenance to restore the
safety of the machinery as well as its efficiency in order to allow its immediate resumption of
operation.
In addition, the decision to fill in the water volumes rich in suspension, which otherwise could
have caused serious problems for the territories downstream, led to the accumulation of consid-
erable amounts of sediment, resulting in significant silting, especially of the Piastra reservoir.
maximum and minimum value of the parameter considered in the training period, in this way the
result is dimensionless and constitutes an indicator of the goodness of the fit between the model
and the measure.
The overall response of the dam to the rapid increase of the reservoir was generally good, for
example the seepage measurement at the toe of the dam were limited (with peaks of 0.15 l/s) and
the pendulums at the crest had a response consistent with the statistical model, several parameters
has been analysed, but for sake of brevity cannot be described in detail in the paper. In the fol-
lowing the analysis is concentrated on the response of a long-base extensometer in the foundation
(code ER11KLEV) and of a piezometer placed at rock-foundation contact (code PZ13KPRL),
both placed in the central block of the dam. Both parameters highlighted some anomalies in the
first step of analysis, and they are characterized by a very high Pearson correlation coefficient
with the level of the reservoir (code GI26KINV), of 0.97 and 0.99 respectively. For this reason
the statistical models for anomalies detection were obtained by neglecting the seasonal contribu-
tion, because the response is mostly governed by the reservoir level.
In the central block the long-base extensometers in the foundation are arranged along three
alignments, two alignments are vertical and placed respectively in the upstream (ER2xx group)
and downstream (ER3xx group) part of the foundation, furthermore there is an inclined alignment
at the upstream part (ER1xx group), see Figure 4. On each alignment there are three measurement
stations at different depths. The ER11KLEI parameter measures the displacement at the most
shallow measurement position along the inclined extensometer.
Figure 4. Transversal section of the Chiotas dam: location of the instrumentation analysed.
The statistical model in the training period, not showed in the paper, reproduces the measure-
ments well, as confirmed by the r2 value close to unity (i.e. 0.948) and the small NRMSE value
equal to 0.062. Observing the test period (Figure 5) it can be seen that in correspondence with the
rapid change in the reservoir level, between 2 and 3 October 2020, there is an evident deviation
between the measurement and the statistical model (i.e. r2=0.85, NRMSE =0.099). The difference
between the statistical model and the measurement can be interpreted as an anomaly in the be-
haviour with respect to normal operating conditions occurred in the past. As a result of the rapid
reservoir level increase, the statistical model predicts an increment of displacement in the
Symposium “Management for Safe Dams” - 91st Annual ICOLD Meeting – Gothenburg 13-14 June 2023
Figure 5. Test period for the parameter ER11KLEI: Reservoir level (GI26KINV) (high); Comparison be-
tween measured parameter and model prediction for anomaly detection (low). The growth of negative val-
ues indicates a downstream displacement.
To consider this aspect, the statistical model was modified by introducing the temperature of
the upstream face (code G27KTAM) as a further cause parameter. In the training period, not
showed in the paper, the fitting of the statistical model is only slightly better than in the previous
case, as confirmed by the analysis of the good of fit parameters (r2=0.97 and NRMSE=0.047).
Index of the fact that under normal operating conditions, the variation of the temperature of the
upstream face does not affect the long-base extensometer.
Figure 6 (low) shows the comparison between the model and the measurement in the test pe-
riod, as can be seen the deviation highlighted in the anomaly detection model (Figure 5 low) is
substantially reduced (i.e. r2=0.935, NRMSE=0.065), the measurements are 95% of the time
within the prediction range of the statistical model, confirming that the increase of the upstream
face temperature is the cause in the anomaly in the long-base extensometer.
INTERNAL
Figure 6. Test period for the parameter ER11KLEI: Reservoir level (GI26KINV) and temperature of the
upstream face (GI27KTAM) (high); Comparison between measured parameter and model prediction for
detailed analysis (low). The growth of negative values indicates a downstream displacement.
The analysed piezometer (code PZ13KPRL) is placed below the pulvinus in the upstream area
of the central block at a level of 1852.9 m a.s.l. The statistical model was created considering the
reservoir level (code GI26KINV) as a cause parameter with a polynomial expansion degree equal
to 4. In relation to the sampling frequency of the data, which consider one datum per day, the
model is well constrained, and the degree of polynomial expansion considered does not generate
"overfitting" issues. The model reproduces the measurements well, the r2 value is close to unity
(i.e. 0.98) and the NRMSE value is rather small (i.e.0.032). Even the piezometric level, like the
long-base extensometer in the foundation, is sensitive to the rapid variation of the reservoir level,
and in correspondence with the rapid water level increase at the beginning of October, the meas-
urement has a peak that the model for anomalies identification is unable to capture. The statistical
model tends to overestimate the response because it predicts a higher piezometric level value than
the one actually measured (Figure 7 low).
Figure 7. Test period for the parameter PZ13KPRL: Reservoir level (GI26KINV) (high); Comparison be-
tween measured parameter and model prediction for anomaly detection (low).
Symposium “Management for Safe Dams” - 91st Annual ICOLD Meeting – Gothenburg 13-14 June 2023
Analysing the behaviour of the long-base extensometer, it was observed that the response of
the dam foundation is conditioned by the temperature variation of the water which was at a higher
temperature than the air. This temperature variation induced a reduction in the displacement of
the long-base extensometer in the foundation, with a consequent variation in the induced state of
stress in the underlying rock mass. To consider this aspect, it is not possible to introduce in the
statistical model of the piezometer the upstream face temperature as a further cause parameter,
because it is not physically attributable to the variation of the piezometric level. Because there are
no representative measures of the variation of the state of stress or of the hydraulic conductivity
of the rock mass to be introduced in the statistical model, it was decided to use as a representative
parameter of this phenomenon directly the displacement of the long-base extensometer in the
foundation (code ER11KLEI), which it was introduced into the statistical model as an additional
cause parameter. The comparison between the measure and the model in the training period is
slightly better than in the previous case (i.e. r2=0.99, NRMSE=0.025). Index of the fact that under
normal operating conditions this parameter does not affect the response of the piezometric level.
Figure 8 (low) shows the comparison between the model and the measurement in the test period,
as can be seen, the deviation highlighted in the model for anomaly detection (Figure 7 low) is
substantially reduced, as confirmed by the improvement of all good of fitness parameters (r2=0.98,
NRMSE=0.036).
Figure 8. Test period for the parameter PZ13KPRL: Reservoir level (GI26KINV) and long-base extensom-
eter ta foundation (ER11KLEI) (high); Comparison between measured parameter and model prediction for
detailed analysis (low).
5 CONCLUSIONS
The meteorological event that occurred between 2nd and 3rd October 2020 heavily affected in
Piedmont the basins of the Cuneo province, on the border with France. In Italy, in particular, the
damage caused by the waters on the Vermenagna basin had a wide media coverage. Far more
limited damage was instead recorded in the nearby Gesso basin, where there is a hydroelectric
system consisting of three reservoirs for hydroelectric use and a hydro power plant with pumping
station (Entracque powerplant), managed by Enel Green Power. The hydroelectric system
promptly intervened in pumping on the basis of the observations of rain and weather forecasts,
allowing considerable volumes of water to be filled in the basins and released into the riverbed
when the flood was finished. This made it possible to create a strong lamination of the flood wave
which, downstream of the dam of Piastra, had no effect.
A fundamental role was played by the Entracque powerplant which started operating in pump-
ing mode, pouring a flow rate up to approximately 100 m3/s towards the large Chiotas reservoir.
Overall, considering the storage capacity of the three reservoirs of Chiotas, Rovina and Piastra, it
INTERNAL
was possible to reduce the flow in the riverbed up to over 200 m3/s, to the full advantage of the
downstream territories.
The paper also describes the analysis of the response of the Chiotas dam following the rapid
increase in the reservoir level. Significant parameters representative of the dam behaviour were
analysed through statistical models using a two-step approach. First, anomalies with respect to
past behaviour are sought, and in the presence of an anomaly, the analysis is deepened to identify
the causes. The term anomaly does not necessarily indicate a critical aspect, but a deviation from
the expected behaviour of a model generated on the basis of past measurements.
These algorithms for anomaly detection were developed and tested as part of the platform
called "Dam Behavior" used by Enel Green Power to control the behaviour of dams. They are
characterized by the fact that they do not require interaction with the user for the model genera-
tion, because are generated automatically, and use a reduced number of input parameters that fit
the parameter to be modelled. In this way they can be used for a massive analysis of parameters
of different types.
From the analysis of the various measures, it can be observed that despite the exceptional nature
of the October 2020 event, the dam's response did not highlight any critical behaviour. The algo-
rithm used for anomaly detection has identified, in correspondence with the period of rapid res-
ervoir variation, for two parameters (i.e. a long-base extensometer in the foundation and a pie-
zometer placed in contact with the rock), a deviation from the expected behaviour. The detailed
analysis to understand the reasons for this deviation showed that the pumping of a high volume
of water from the Piastra reservoir produced in the lower part of the dam a minor decrease in the
average temperature of the dam face, due to the effect of the highest reservoir level. This temper-
ature variation had an opposite effect in terms of movement of the long-base extensometer in the
foundation with respect to that resulting from the increase in the reservoir level. As a consequence
of this, the part of the rock mass underwent a change in the stress state which conditioned the
response of the piezometer.
The analysis of the parameters representative of the dam response has highlighted that the rapid
variation of the reservoir level can in some cases produce a deviation from the expected behaviour
in ordinary conditions, which can be explained by introducing the appropriate variables in the
statistical models. Despite the rapidity of variation of the reservoir level that occurred during the
event of October 2020, the dam responded without evidence of irregularities.
REFERENCES
AA.VV., 2013. Atlante dei bacini imbriferi Piemontesi, Renerfor.
ARPAP, 2020. Eventi alluvionali in Piemonte, “Rapporto evento 02 Ottobre 2020”, a cura del Dipartimento
Rischi Naturali e Ambientali, Torino 14 ottobre, rev 1.1
Basano S., Bernardi F., Bonafè A., Fornari F. (2021). L’evento del 2-3 ottobre 2020 nel bacino del Gesso
dal punto di vista del gestore idroelettrico. L’Acqua (4).
Corigliano M., Torasso D., Sciolla M., Scuppa I., Fornari F., Lezzi R., Moscariello M., Benanti A., Bernardi
F. (2022). La risposta della diga di Chiotas al rapido invaso per la laminazione della piena conseguente
l’evento meteo del 2-3 ottobre 2020. L’Acqua (5).
Enel (1980). Impianti Idroelettrici del Gesso DIGHE DEL CHIOTAS E DI COLLE LAURA.
Hyndman R.J., Athanasopoulos G. (2018). Forecasting: Principles and Practice. OTEXTS.
www.arpa.piemonte.it/rischinaturali/tematismi/acqua/risorsa-idrica/H-Q-giornaliere.html