Land use and damaging hydrological events temporal changes in the Sarno River basin: potential for green technologies mitigation by remote sensing analysis

Nowadays over half of the world's population is living in cities and by 2050 it is estimated that roughly 6.4 billion people, almost the planet's current population, will live in a city (United-Nations 2014). Urban growth has left cities facing a number of problems and they will inevitably increase in the near future unless best management practices are adopted to mitigate those effects. As cities have grown and more development has occurred, the natural landscape has been replaced by roads, buildings, housing developments, and parking lots. Changes in land use have been followed by an increase in impervious surfaces, propagating into hydro-geological and hydraulic risk changes (Konrad 2003).

In addition to the land use changes, critical infrastructures are also becoming more sensitive to weather and climate extremes, which include a multiplicity of hazardous events falling into a wider category of the Multiple Damaging Hydrological Events (MDHEs), severe stormy periods during which floods, landslides, lightning, windstorms, hail, or storm surges can harm people (Petrucci et al. 2018). According to several works, the main cause of major MDHE effects in urban areas around the world is the increasing impervious surfaces. Huang et al. (2018) found out that in Guangzhou, China, the rapid increase in flood events was influenced by the change in the land cover. This was also affirmed by Paprotny et al. (2018), who argued that the upward trend of the MDHEs in Europe is linked to the rapid urbanization of the growing regions. Similar studies in different Italian regions also identified a relationship between increasing imperviousness and increasing occurrences of damaging hydrological events (Apollonio et al. 2016; Sofia et al. 2017; Bentivenga et al. 2020; Recanatesi & Petroselli 2020). However, contrasting messages come from the same geographical areas probably because of the urban context specificities. Indeed, Esposito et al. (2018) in the Pozzuoli area (Campania region) ascribed the increasing occurrence of MDHEs to the variation of the rainfall regime and not to the increasing percentage of sealed areas.

Unfortunately, conventional urban water management practices are not fully prepared to deal with this problem. Design and planning methodologies based on stationary scenarios are no longer reliable (Longobardi et al. 2016). Action should be taken to prepare and to adapt our cities to changing environmental conditions, in the medium-long temporal horizon.

In light of this, innovative approaches in sustainable urban development and water management, such as the SuDS (Sustainable Drainage Systems), appear as a set of multi-benefit technologies, based on the use of green and blue technologies, able to increase the urban areas’ resilience to environmental changes (Sartor et al. 2018). SuDS are able to control runoff by reducing its quantity and improving water quality (Voyde et al. 2010). Among SuDS, green roofs (GRs) have proven to be a very effective measure if used in response to water quality and quantity issues at the city scale (Morgan et al. 2013). They allow for a reduction in runoff volume and peak flow and an increase in delay time, consequently, its application at a large scale by retrofitting the existing traditional roofs helps to significantly reduce the hydrological risk (Masseroni & Cislaghi 2016).

The effectiveness of a widespread application of GRs on stormwater generation has been tested, over time, by several authors using a number of hydrological models such as the SWMM (Storm Water Management Model), L-THIA-LID 2.1 model, Curve Number (CN) method, Visualizing Ecosystem and Land Management Assessments (VELMA v2019-07-22) model. Among these, the SWMM of the Environmental Protection Agency (EPA) has resulted in very accurate and precise reproduction of the hydrological behavior of GRs applied at a large scale (Versini et al. 2016). On this point, it is necessary to consider that the hydrological performances of GRs at the catchment scale strongly depend on the roof's surface suitable for GR retrofit.

The possibility to use satellite images to detect the areas with the potential for GR retrofit has been investigated during the past decades. Karteris et al. (2016) used high spatial resolution orthoimages from a Leica ADS40 Airborne digital sensor in the Mediterranean city of Thessaloniki, Northern Greece, and quantified the benefits of the large-scale application of GR technology in terms of energy consumption and rainwater retention. Pappalardo et al. (2017), in Southern Italy, derived the land use by Google satellite images, investigating the role of the SuDS in urban runoff control. Zhou et al. (2019) used panchromatic and multispectral images of the ZY-3 satellite to identify the building roofs suitable for greening retrofitting within the city of Beijing.

The Sarno River basin, in southern Italy, represents an emblematic case. It is a peri-urban basin highly prone to flooding events, and it is the most polluted river in Europe and one of the 10 most polluted rivers in the world (Baldantoni et al. 2018). In previous works, the correlation between the increasing number of MDHEs, during the last two decades, and the climate change in the Sarno River basin was investigated with negative results (Califano et al. 2015; Longobardi et al. 2016). Until now, only traditional engineering measures have been designed and realized to reduce the risk related to the occurrence of flooding events, including water detention basins and drainage channels (Di Vito et al. 2019). These strategies, which intended to modulate peak discharge and delay, were not significantly effective and the above-mentioned measures did not take into account the qualitative-related issues affecting the area.

With reference to the Sarno River basin, the presented research has a twofold objective. On one side, to analyze in more detail the potential cause-and-effect relationship between the increase in MDHE occurrences and the changes in the land cover. On the other side, to investigate, by a hydrological simulation approach, the impact of green infrastructure solutions as a proposal for stormwater management, conscious of the benefit they could furthermore address in terms of water quality issues in a seriously endangered environment. The link between the two objectives is represented by the use of satellite imageries which were used to both assess the changes in the land cover, over the period a raise in MDHEs was observed, and to quantify the potential for GR retrofit. Different from what was overall considered in the literature, synthetic-aperture radar (SAR) imageries, from ERS-1 and COSMO-SkyMed missions, were used for the purpose. Compared to the multispectral imagery, SAR allows for a lower computational complexity, not strictly required for the purposes of the present investigation. Multispectral images return an accurate mapping of land use and land cover features but spectral indices (normalized difference built-up index, the built-up index, normalized difference impervious surfaces index) were used to identify impervious surfaces that require higher computational effort levels (Su et al. 2022). The exploitation of SAR sensors allows for a quick and reliable extraction of urban agglomerates, thanks to the possibility to exploit the information carried by the phase of the radar signal. Its stability is a peculiar characteristic of the built-up feature (Franceschetti & Lanari 1999). This procedure has not been thoroughly investigated until recently (Liang et al. 2022; Wu et al. 2022) and allows the identification of most of the impervious surfaces without the need to consider spectral information.

From the climate point of view, the Sarno River basin is characterized by a warm-summer Mediterranean climate with warm, dry summers, and cool mild winters and with most of the annual precipitation falling in the autumn-winter seasons. The area presents an average annual air temperature of about 17 °C and average annual rainfall of approximately 1,200 mm. The rain gauge monitoring network includes 25 stations for which rainfall data observations are available from 1990 to the present at the sub-daily scale. In Figure 1(b), the available rain gauges (red circles) and their area of influence resulting from the Thiessen polygons method have been shown.

The basin is a flood-prone area which, in time, has been the cause of several hazardous hydrological events including flash floods and landslides (Califano et al. 2015; Longobardi et al. 2016). The vulnerability associated with urban flooding is considerably high since the basin is populated by more than 1 million people with an average population density of about 1,800 inhabitants per km² and a peak density of up to 2,200 inhabitants per km², which make it among the most densely populated and urbanized areas in Italy. The peak density is reached along the coastal belt and the mouth of the Sarno River (Figure 1(c)) and it is lower in the rest of the plain, which is used for intensive agriculture of vineyards, chestnuts, etc.

The Sarno River basin is a flood-prone area that has been affected over time by several damaging hydrological events. Longobardi et al. (2016) investigated the occurrence of MDHEs in a relevant peri-urban sub-catchment, the Solofrana watershed (Figure 1(a)), showing a temporal increase in their frequencies. For the period between 1951 and 2014, a total of 45 events causing seriously damaged urban infrastructures and loss of human lives were recorded. Temporal changes in MDHE occurrences over this period of time were investigated in the light of a climate change perspective with no significant results (Longobardi et al. 2016). With the aim to further investigate and address the possible motivation besides the increase in MDHE occurrences in this geographical region, in the current research, the collection of the events was extended to the whole catchment. The collected information includes:

• affected areas;

• the date of occurrence; and

• the source of information.

Only MDHEs that caused damage to structures such as bridges, levees, and roads and/or economic and social impact on properties and/or damage to land and loss of natural resources and in extreme circumstances, losses of life and injury to people, were taken into consideration. The date of occurrence of these events is available with daily accuracy. Papers, books, articles from local newspapers, event databases from state civil protection agencies, and scientific and technical reports were used as the source of information about the MHDEs. Rainfall data were collected from the regional rain gauge network (Figure 1(b)) in order to characterize the properties of the triggering rainfall events. Unfortunately, streamflow data for the Sarno River basin are only available starting from 2001, thus it was not possible to have a further characterization of flow rates.

The temporal distribution of the collected MDHEs and the possible existence of any temporal trends or patterns in the occurrence of these events were examined.

In particular, the existence of linear trends in the number of MDHE records was inspected using statistical parametric (Pearson test) and non-parametric tests (Mann Kendall test and Sen's test).

The null hypothesis (no trend) H₀: ρ = 0 is tested against the alternative hypothesis H₁: ρ ≠ 0 at a level of significance α. The null hypothesis is rejected when the t value is greater in absolute value than the critical value t_α/2.

The previous tests show the trend existence and direction but, using Sen's test, it is possible to estimate the trend magnitude.

S > 0 indicates an upward trend in the time series and vice versa.

Next, M₁ = (N′ − C_α)/2 and M₂ = (N′ + C_α)/2 are computed. The lower and upper limits of C_α, Q_min and Q_max are the M₁th and the (M₂ + 1)th of the N′ ordered slope estimates. The slope S is statistically different from zero if the two limits Q_max and Q_min have the same sign.

The processing and elaboration of SAR images acquired over the Sarno River basin were performed in order to track the urban development of the area and investigate the changes in the extension of the impervious areas during the last decades. Due to the availability of radar imagery, the period considered for the present investigation covers about 20 years from 1995 to 2016. Indeed, the first SAR images covering the study area have been provided by the ERS-1 sensor that was launched in 1995 by the European Space Agency (ESA) but ended its operations in 2000, so, in order to integrate these data, the satellite images provided by the COSMO-SkyMed mission and released by the Italian Space Agency (ASI) were used. The mission launched four identical satellites between 2007 and 2010 and all are still operating. Provided the spatial resolution of the remote sensing data, once the percentage of the total impervious area within the basin is identified by the SAR detection analysis, the use of building construction indexes will allow the areal extension of buildings over the whole sealed area to be quantified, which is essential to investigate the hydrological benefits associated with a widespread GR retrofitting of the existing traditional roofs. In this context, the land use scenario detected in 1995 will be referred to as the pre-development scenario, whereas the 2016 land use scenario will be referred to as the post-development scenario.

The first step was the data download. Two sets of archive satellite images were used. The first one consisted of six images acquired between March and December 1995 by the ERS-1 satellite. The second dataset was acquired by the COSMO-SkyMed mission between May and December 2016, and consisted of seven images. The main characteristics of the two aforementioned sensors are summarized in Table 1. Details about the processed datasets are provided in Table 2. These two datasets provided, respectively, the two Sarno River basin impervious area scenarios under analysis.

The pre-processing block includes data coregistration and the estimation of the interferometric coherence. Coregistration is the process of alignment of images with respect to one of them selected as the reference (Franceschetti & Lanari 1999). In our case, reference (or master) images for the ERS1 and COSMO-SkyMed time series were acquired, respectively, on March 1995 and May 2016.

The coherence image for all interferometric pairs for the years 1995 and 2016 was produced using as the reference image, the master image of the previous step. The coherence ranges between 0 and 1. The higher the coherence, the higher the phase stability of the target. Built-up areas are typically stable with respect to the phase, so they are expected to exhibit high values of the interferometric coherence (Franceschetti & Lanari 1999), even when images are acquired with a long-time baseline. On the other side, natural land cover is expected to exhibit low coherence values, even when images are acquired at short time scale. This is due to the intrinsic variability of features like crops or bare soils. The feature extraction block includes the temporal mean of the coherence maps grouped per year. In other words, all the coherence maps relevant to images acquired in the same year are stacked together and processed for temporal average. This is useful to enhance the presence of man-made stable targets at the expense of natural scene features.

This procedure returned two cluster-based threshold images relating to the years 1995 and 2016 from which it was possible to quantify, for each year, the amount of pervious and impervious areas.

The SWMM model was used to assess the hydrological response of the Sarno River basin, both in the pre-development (1995) and post-development (2016) scenarios, showing the hydrological changes related to the changes in land cover. Furthermore, the SWMM model was used to assess the hydrological performance of a GR retrofitting stormwater management plan, which, conscious of the importance of SuDS areal extension and investment costs, takes into account different scenarios of GRs retrofitting, considering different scenarios, corresponding to 5, 30, and 100% of traditional roofs conversion to GRs.

The RUNOFF block is used to produce the runoff hydrograph from sub-catchments. Indeed, the algorithm allows us to generate surface runoff in response to precipitation. The method employs the surface water budget approach as shown in Figure 3 with a single inflow represented by the precipitation and several discharges including the runoff ‘Q’, the infiltration ‘f’, and the evapotranspiration ‘ET’.

ET rates can be stated as a single constant value, a set of monthly average values, a user-defined time series of values, values computed from the daily temperatures contained in an external file, and daily values read from an external file. In the present study, it has been neglected during the calibration procedure carried out at an event scale while it has been set as a constant value during the simulations of the GR conversion scenarios performed considering a longer time scale. It has been estimated according to the Blaney–Criddle model (Blaney & Criddle 1950).

Provided the urban development spatial features, while the land cover change analysis and the temporal investigation of MDHE occurrences refer to the whole Sarno River basin, the hydrological modeling is only applied to the downstream area of the Sarno River basin (Figure 1(a)), hereinafter called Sarno Mouth sub-basin or investigated urban area.

The inflows from the peri-urban area located in the upstream part of the Sarno River basin (Solofrana and Cavaiola sub-catchments in Figure 1(a)) were not considered because a previous study demonstrated that the hydrological connection between the peri-urban and the urban system (Sarno Mouth sub-catchment in Figure 1(a)) can be neglected (Mobilia 2018).

The average surface slope of the catchments was obtained using the slope analysis of the Digital Elevation Model in QGIS while the width of the overland flow path was given, according to the SWMM user's manual, by the catchment area divided by the average maximum overland flow length. After the GR placement, the sub-catchment width has required adjustment to compensate the amount of the initial watershed area replaced by the green technology. The original sub-catchment width has been modified taking into account the width of the installed GRs intended as the total length along the edge of the roof where runoff is collected. More details about the experimental GR dimensions are provided in Table 3.

These data were used to define the CN. The CN is tabulated in the literature and, for open spaces with grass cover on 75% or more of the area and hydrological soil group (HSG) B, it assumes the value of 61, whereas for grass cover on 50–75% of the area and B-HSG, it equals 69. The B group includes soils with a moderate infiltration rate attributable to the texture of silt loam or loam which is the dominant soil texture of the Sarno Plain (Adamo et al. 2014). In Table 4, more details about the input parameters of the sub-catchments, conduits and nodes are available.

The SWMM is equipped with an LID tool. It has been specifically designed for modeling the storm water management performance of various types of LID practices, including GRs, bioretention cells, permeable pavements, etc.

According to several studies (Burszta-Adamiak & Mrowiec 2013; Cipolla et al. 2016; Limos et al. 2018), the bioretention cell LID module has been selected in this study to mimic the GR hydrological behavior. In detail, this LID type consists of a number of horizontal layers including the surface layer, the soil layer, and the storage substrate layer with only vertical movement of water within and between them. Beyond the literature studies, this particular model fits the building practices of the experimental GRs used to calibrate the model, as later explained.

The input parameters required by the bioretention module and used in the present study refer to an experimental GR and are listed in Table 4.

In detail, the values of the thickness and the slope of the layers are geometrical characteristics of the system while the roughness of the surface layer and the void ratio of the storage layer were sourced from the user's manual of the SWMM which draws on the literature data. Concerning the soil layer, the value of the porosity was provided by the material supplier while the values of the other parameters were derived by Rawls et al. (1983) who suggested average values of the infiltration parameters for different kinds of soil. With reference to the drain parameters, the drain offset is a geometrical characteristic of the system, while the user's manual suggests values for the flow exponent when the drain acts like an orifice, and for the drain coefficient estimated to be 60,000 times the ratio of the total slot area to the LID area.

The experimental roof is located within the campus of the University of Salerno, in Fisciano, Southern Italy. It was selected as a test bed since it falls within the study area (Figure 1(a)). It is of extensive type and has a surface of 2.5 m² (2.5 m × 1 m; width × length) and a depth of 15 cm. The support layer is made up of a mix of peat and zeolite while the storage layer is made up of expanded clay. The vegetation layer hosts succulent plants called Mesembryanthemum. Further details about the GR test bed are available at Longobardi et al. (2019).

The SWMM hydrological modeling required a spin-up period of 72 h in order to give the model the time to reach a state of equilibrium after being started from initial conditions. The soil moisture content of the LID module before each simulated event was fixed as the initial condition. As suggested by Cipolla et al. 2016, with the aim of reducing the dependency of the model on the degree of saturation of the LID, the value of this parameter of the day before the rainfall event was set to 100% so that it could be estimated by the model as a function of the weather data.

Four rainfall–runoff events were selected for the calibration of the RUNOFF module. They belong to the historical MDHEs database previously described. Being all hazardous events, they were chosen according to the following selection criteria:

• The number of affected areas;

• The rainfall return period; and

• The rainfall intensity (5-min interval)

A single event can affect more than one municipality, and in this case, it probably has a major impact, so, in the present study, the rainfall events which triggered urban MDHEs in several municipalities were considered.

The 10-year return period is generally selected as the design standard for most urban drainage systems and when it is exceeded, the systems are overloaded and damaging hydrological events can occur. In light of this, only the rainfall events with a return period higher than 10 years were selected for the analysis.

Many authors have asserted that rainfall intensities are of prime importance for flood generation (e.g. Pitlick 1994); indeed, for higher rainfall intensities, higher hydrological responses are expected (Habonimana 2014). In this research, the triggering rainfall events which occurred in at least three locations of the basin with an intensity higher than 15 mm/h (heavy events) were chosen.

For the reasons related to the modeling of the LID tool, the selected rainfall events were divided into convective and stratiform categories based on their rainfall shape profile (Terranova & Iaquinta 2011). They were separately modeled since it was shown by Mobilia & Longobardi (2020) that SWMM LID tool is affected by the behavioral effects of events with different shapes of rainfall profiles. Two different calibration phases were performed for the two classes of events. Their relevant properties are illustrated in Table 6.

The calibration of the LID module was carried out using rainfall/runoff events collected at an experimental roof located within the investigated river basin, at the university campus of Salerno. The experimental extensive GR is fully equipped with a rain gauge for recording the precipitation and a digital scale for weighing the runoff volume since 2017. The rain gauge is part of a weather station, Watchdog 2000 Series (Model 2550) while the digital scale is a LAUMAS AC60 Kg and above it, a tank is placed where the rainwater is collected. The measurements are taken at 5 min time step and aggregated at the desired work scale (Longobardi et al. 2019).

According to a number of sensitivity analyses from several experimental studies in this area of research, the suction head Ψ is the most sensitive parameter (Baek et al. 2020; Paithankar & Taji 2020; Lisenbee et al. 2022). For example, in the case of the RUNOFF module, stratiform and convective events were clustered. A complete list of 29 rainfall events recorded during the period 2017–2019, used for the LID calibration phase, and their characteristics such as rainfall intensity, duration, cumulative, and retention coefficient (RC) are shown in Table 7.

The calibration procedures returned the best parameters for LID and RUNOFF modules to be used to run SWMM for subsequent analysis. In detail, the average value of the suction heads obtained by the calibration of the convective events was used for the simulation of events with this shape of rainfall profile and the same applies to the stratiform events. On the other side, the calibrated values of d_p and n for each of the four selected events were used for the reproduction of the same events in the modeled scenarios.

The land use cover corresponding to 2016 (post development) was set as the baseline scenario and differences with pre-development (1995) and greening technologies were considered. Thus, R_V,0, P_F,0, F_S,0, D_T,0 are, respectively, the runoff volume (m³), the peak flow (m), the number of flooded sections (–) and the delay time (hours) referred to the post-development scenario, while R_V,1, P_F,1, F_S,0, D_T,1 are the same parameters but referred both to the different greening scenarios and to the pre-development scenario.

The database of the MDHEs occurred between 1951 and 2016 within the Sarno River basin, and the related information such as the areas hit by the events, the date of occurrence and the documentary resources is accessible in Table S1 (see Supplementary Materials).

As SAR images and consequently land uses change is only available from 1995 to 2016, a statistical analysis was performed for trend detection in this specific period in order to assess the significance of the temporal change in MDHE occurrences (Figure 6). The Mann–Kendall test, Pearson test and Sen's Slope estimator test were applied for detecting the existence of a possible monotonic trend direction and the magnitude of change at 5% significance level. Each of the three tests rejects the null hypothesis, that is the presence of a statistically significant trend is confirmed (Table 8). In addition, the positive value of the test-statistics suggests an upward trend direction, as evident in Figure 6.

Various sources of uncertainty might affect the objectivity and the outcome of the statistical analysis. The urban growth is inevitably associated with an increase in the urban infrastructure exposed to risk such as the use of an increasing number of media in recent years may have led to a physiological increase in the records of flood event occurrences compared to the past. It is however extremely difficult to identify them, but in a general context in which society's perception is an effective increase in the risk of urban flooding, these limitations in objectivity seem altogether acceptable.

Results are furthermore supported and confirmed by previous research which investigated the variation of the frequency of flash floods in urban areas, during time periods similar to the one considered in this study. In detail, within the same region, Esposito et al. (2018) observed an increase in flooding events between 2000 and 2014 in the Pozzuoli area. They ascribed this increasing trend to the variation in the rainfall regime since urbanization has increased only for 0.7% of the watershed area. In a geographically close region, Bentivenga et al. (2020) showed a recent increase in the frequency of flooding events in the Ionian Belt of Basilicata Region confirming the role played by both climate changes and human activities. Apollonio et al. (2016), found a good correlation between flooding areas and land use changes from 1984 to 2011 in the Cervaro river basin in Apulia Region. In the Lazio region, Recanatesi & Petroselli (2020) investigated the relationship between land cover change and flood risk during the period from 1954 to 2018, showing how the flood risk changes due to an increase in urbanization. In the Veneto Region, Sofia et al. (2017) coupled the increase in floods from 1900 to 2010 to both climate and land use change.

Other works, worldwide, confirmed the upward trend of urban flood events. For instance, according to Huang et al. (2018), Guangzhou, China experienced a rapid increase in flood events during the period of 2009–2015 influenced by both change in precipitation and impervious surface. In Europe, the amount flooding in the affected areas has seen a rise since 1870 and among the main causes, the changes in the distribution of land cover/use can be included (Paprotny et al. 2018).

As shown in Table 1, images acquired by the two sensors have slightly different pixel size. This is not expected to influence significantly the segmentation, especially over dense urban areas. Misclassification of small agglomerates and/or isolated structures can occur using ERS-1 images. However, the impact of these areas on the calculation of the percentage of soil sealing, which is the parameter of interest for hydrological modeling, is expected to be negligible (see Table 9). This analysis was performed at the sub-catchment scale, in order to identify the areas of the basin which, in time, were most affected by the land use change.

The percentage of soil imperviousness heavily increases from 1995 to 2016 for each sub-basin. The built-up area in the whole basin changes from 7.4 to 12.38% in 2016, doubling its original value in 20 years. The basin with the lowest increment of impervious area is the Calvagnola with an increase of about 30% between 1995 and 2016, while the catchment which experiences the highest variation of the urbanized surfaces, around 75%, is the Sarno Mouth basin. These findings, on one side, reinforce the choice to focus the hydrological analysis on the downstream part of the Sarno River basin, where the increase of built-up area is considerable. On the other side, they also support the idea of the increase in flooding risk due to land cover changes. As reported in the introduction section, for what concerns the Sarno river basin, in previous research, the temporal variability of the precipitation over the same time-horizon was studied (Califano et al. 2015). The findings excluded both the existence of a significant variability and the increase in storm events severity. Despite the discussed limitations of the statistical empirical analysis and considering the presented results in terms of land use changes, it is reasonable to assume that the cause of the recent rise of MDHEs is most likely related to the increase in the impervious surface. This result would suggest that a valid solution to reduce the frequency of occurrence of damaging events could be found in sustainable landscape planning, by green technologies implementation.

In order to return accurate simulations of the GR hydrological response at the basin scale, the LID and RUNOFF modules of the SWMM were calibrated.

For what concerns the event scale calibration of the RUNOFF module, hourly rainfall and streamflow data occurring on 31.10.2012, 21.01.2014, 20.11.2013, and 13.09.2012 were used (see Table 6). The calibration parameters are the impervious depression storage coefficient d_p and channel Manning's roughness coefficient n and their input values were iteratively changed in order to reach the maximum value of the NSE index for each event. The optimal values of the calibration parameters and the corresponding NSE are shown in Table 10.

With the aim of interpreting the findings related to the GR effectiveness in mitigating the runoff at the basin scale, it is useful to observe that the retention capacities of the two groups of events substantially differ in quantitative terms: the convective events present an average RC of about 60% which is higher than that related to the stratiform events which approaches 45% (Figure 9(c) and Table 7). With the average rainfall intensities being similar for the two groups, the lower retention capacity of the stratiform group is probably due to the fact that the average duration of the rainfall events belonging to this category (504 min) is higher than in the case of a convective group (297 min), confirming the importance of more rainfall properties beyond the relevant severity (Longobardi et al. 2019; Mobilia & Longobardi 2020).

Besides the NSE index, which describes the model goodness-of-fit at the event scale, it is possible to observe how well the simulations match the measured data for both types of events. Indeed, the model, besides the runoff volume, is able to well reproduce the temporal pattern of the events too.

The calibration process returns values of Ψ ranging from 3 to 300 mm (the last is an outlier in the series of estimated parameters) for convective events and between 6 and 100 mm for the stratiform ones with average values, respectively, of 49 and 39.8 mm. The boxplots in Figure 9(b) show that the empirical distribution of Ψ values is actually wider for convective than for the stratiform events which means higher uncertainty in the selection of the most appropriate value of the calibration parameter for the first type of event.

Indeed, the value of the suction head for stratiform and convective events is never higher than 100 and 140 mm, respectively, with an outlier represented by the event with Ψ = 300 mm. These values are in line with the maximum value suggested by Rossman & Huber (2016) for the LID module which was set at 4 inches. The large variability of the values assumed by Ψ can be explained by the choice of using only this single variable as the calibration parameter. If from one side, the system has been forced to be governed by only one parameter, from the other side, this choice avoids an overparameterization and reduces the complexity of the retention process schematization that would otherwise have to take into account multiple parameters.

The results of SAR images elaboration suggest that the increase in the number of MDHEs within the studied basin is potentially related to the rapid expansion of the built-up area, therefore the GR retrofitting of existing roofs, with the consequent reduction of impervious surfaces and runoff production, could represent a valid solution for the flood risk management in the investigated catchment. In light of this, the hydrological behavior of large-scale GRs retrofitting scenarios were furthermore compared to the pre-development scenarios of 1995 and post-development scenarios of 2016. The specific modeling was considered for the four events used for the calibration of the RUNOFF module.

The comparison allows us, on one hand, to investigate how the variation of impervious areas between 1995 and 2016 affects the runoff production and, on the other hand, if the implementation of GRs at a wide scale could effectively help to restore the hydrological behavior of the system in the pre-development scenario. For the greening scenarios, a reduction of the impervious area taken by traditional roofs in 2016 is planned. In detail, it is assumed to replace 5, 30, and 100% of the traditional rooftops with GRs. Since the SAR technology allows us to identify the total paved areas without distinguishing among concrete roofs, walkways, sidewalks, and streets, the percentage of surface hosting traditional covers, suggested by SNPA (2019), was used in this study. According to this source, in Southern Italy, building area corresponds to 35% of the total impervious surface after 2005 and to 26% in the previous years.

For comparison in quantitative terms, the evaluation of the four assessment indices (Equations (15)–(18)) was carried out (Table 11). The reduction of the runoff volume detected after the GR conversion of the existing traditional roofs ranges from a maximum value of 49.7%, for the scenario with a conversion of 100%, to a minimum value of 16.8% for the scenario with a retrofit of 5%. In general, ΔR_V is higher for the scenarios with a higher green requalification than for the ones with a lower implementation of green technologies; indeed, the average runoff reduction reaches average values across all rainfall events of about 41, 27, and 22%, respectively, for 100, 30, and 5% GR retrofit scenarios. The 1995-scenario, where the percentage of imperviousness is nearly half the 2016-scenario (10.53 against 18.42%), is the best performing scenario with an average runoff mitigation of about 47%. The 100%-scenario can only partially achieve this behavior. This finding shows the negative impact of sealing within the basin in terms of runoff production, which cannot even be compensated by a complete green conversion of the existing traditional roofs. With reference to the shape of the rainfall profile of the events, the largest reduction of the runoff volume occurs for the stratiform events characterized by lower average intensity (about 19 mm/h) and longer duration (between 16 and 58 h). For these events (21.01.2014 and 31.10.2012), the average ΔR_V across all the scenarios is, also in this case, around 40% against the 29% of the convective events (20.11.2013 and 13.09.2012) characterized by higher average intensities (respectively, of 34 and 72 mm/h) and shorter duration (between 12 and 14 h). Similar considerations can be extended to the reduction of the peak flow. ΔP_F ranges between 40.8 and 0%. In detail, for the stratiform events, its average value is around 29% while for the convective ones, it approaches 13%. It is confirmed that the reduction is higher for the 1995-scenario and 100% GR retrofit scenario with average values, respectively, of 29 and 25% and values lower than 15% on average in the other cases. The reduction of the flooded sections is significant both for stratiform and convective events and reaches maximum and minimum values of 100 and of 25%, respectively. Here too, on average, the best performing scenarios are the ones corresponding to the 1995 scenario and to the 100% GR conversion scenario. Finally, the increase in delay time is more evident for the events occurring on 13.09.2012 and 20.11.2013. For these events, the delay time ranges between 1 and 3 h. For the other two events, the time of peak discharge is delayed at most for 1 h.

A due consideration is that the suitability for GR retrofit of traditional roofs depends on a number of criteria, beyond the potential retrofitting surface, including the slope and the orientation of the roof, the number of stories of the building and the number of site boundaries (Wilkinson & Reed 2009). The percentage of buildings that meet all the required attributes and are physically suitable for GR adaptation is never higher than 20% (Wilkinson & Reed 2009; Karteris et al. 2016; Mobilia & Longobardi 2021). In light of this, the really feasible scenarios are the ones corresponding to 5 and 30% of GR conversion. The 100% GR conversion scenario was analyzed in order to perform an exhaustive analysis, also taking into account ideal conditions.

Based on these considerations, the condition occurring in 1995 can never be fully restored. Indeed, this scenario is partially equaled only by a complete (100%) green conversion of the existing traditional roofs but this condition is clearly not feasible. This finding confirms the negative effect of imperviousness on the stormwater production. An attempt to restore the hydrological behavior of the system before the massive increase of paved areas in 2016 could be made using GR infrastructures in combination with other types of LID practices (bioretention cells, permeable pavements).

The results of this study are consistent with previously published research which confirms that the diffuse installation of GRs allows a reduction of both runoff volume and flow peak, which are variables with the rainfall intensity and the duration of the event. In detail, the performances are better for events with a smaller magnitude than for the ones with larger magnitude. The scientific literature also confirms that the reduction of runoff volume is higher for the 100% GR conversion scenarios and smaller for a lower percentage of green requalification. Consistent is also the data around the volume reduction which ranges from 35 to 50% for conversion of 100% and goes from 30 to 51% for the peak flow reduction (Palla et al. 2008; Qin et al. 2013; Masseroni & Cislaghi 2016; Ercolani et al. 2018).

With the support of an archive collection of damaging events that occurred within the study area between 1951 and 2016, the present study has shown that in the Sarno basin, the MDHE occurrences have been characterized by an overall upward trend. Contextually, an increase in pervious surfaces has been detected by the use and analysis of SAR images by the Cosmo-SkyMed constellation, comparing SAR images referring to 1995 (pre-development scenario) and 2016 (post-development scenario). These results have brought to light a potential correlation between the two phenomena (MDHE occurrence and soil sealing), hence the possibility to act in a sustainable landscape planning context, with implementation of GRs at the basin scale to mitigate the effects related to excessive runoff production. Three GRs retrofitting scenarios (corresponding to retrofitting 5, 30, 100% of roofs areal extension as detected by SAR images elaboration) have been analyzed using the SWMM and compared to the pre- and post-development scenarios using performance indices. The findings suggest that soil sealing has a negative impact on catchment hydrology, indeed, in the post-development scenario, the catchment is expected to see increased surface runoff. On the other side, simulation results point out that the extensive use of GRs, even on only 5% of all roofs of the catchment, allows us to mitigate the negative effects of the stormwater excess. In this case, the reduction of runoff is recorded at most as 32%, whereas 31% reduction of the peak flow can be reached. The hydrological performances of the 100% greening scenario, with a maximum percentage of runoff and peak flow reduction, respectively, of 50 and 40%, appears excellent but, however good, it is not able to totally restore the hydrological pre-development conditions of the watershed. In addition, it should be considered that the spatial potential for GR retrofitting can hardly reach a percentage of 100% due to the building attributes required for the conversion.

The present work confirms that GRs are an effective stormwater mitigation strategy for the selected basin and suggests that a massive green restoration planning for the existing roofs of the area can improve the environmental benefits from a hydrologic point of view. The possibility to combine GR technologies with other SuDS practices within the study basin, with the aim to mimic a pre-development hydrological condition, can be explored in future works. The present investigation can be considered paradigmatic for any degraded and vulnerable urban context interested in a high frequency of MDHEs and where an extreme land use change occurred rapidly with visible results in just a few decades.

The authors would like to thank anonymous reviewers for their helpful comments and suggestions which resulted in an improved manuscript version. This study was carried out using CSK Products ASI (Italian Space Agency), delivered under an ASI licence to be used in the framework of ‘COSMO-SkyMed Open Call for Science Initiative’ (Proposal ID 452 – ‘Use of high-resolution SAR images for monitoring urban dynamics’-Principal investigator: Antonia Longobardi).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.