Rainfall-induced shallow landslides represent a critical natural hazard in mountainous regions, with their frequency controlled by hydrological processes. Climate change is expected to alter both precipitation patterns and soil moisture dynamics but quantifying these impacts on landslide susceptibility remains challenging.In this study, we integrate physically-based stability thresholds with distributed hydrological modeling to assess future landslide hazard evolution under multiple climate scenarios. The study is conducted for a small basin (~28 km2) located in the north-eastern Friuli Venezia Giulia (Italy).Spatially explicit Critical Soil Moisture (CSM) and Critical Wetness Index (CWI) thresholds at 50 m resolution were derived in a previous effort for multiple failure depths (0.75 to 2.00 m) by inverting the infinite slope stability analysis. The thresholds represent hydrological conditions at which slope failure may initiate through either unsaturated zone processes or groundwater table rise. These thresholds were coupled with a calibrated distributed and physically-based hydrological model, the Triangulated Irregular Network‐based real‐time integrated basin simulator (tRIBS), which simulates hourly soil moisture and groundwater dynamics, to assess the occurrence of failure over 100-year periods for three synthetically generated climate scenarios: current conditions, moderate emissions (RCP4.5, 2050), and high emissions (RCP8.5, 2050). The synthetic series of meteorological variables, and particularly precipitation, were generated by combining the AWE-GEN (Advanced WEather GENerator) model with a procedure to correct the distribution of extreme events.We quantify exceedance frequencies, i.e., the proportion of time during which CSM and CWI thresholds are exceeded, as a measure of temporal exposure to landslide-conducive conditions. Results reveal that, under RCP4.5, exceedance frequencies decrease by up to 14.6% (CWI) and 10.9% (CSM), due to a reduction in annual precipitation despite an increase in mean intensity per event. In contrast, RCP8.5 shows bidirectional patterns, with maximum increases reaching 5.1% (CWI) and 3.6% (CSM), indicating that precipitation intensification begins to overcome the reduction in annual precipitation. Critically, climate impacts amplify with failure depth; the 2.00 m failure depth exhibits changes in magnitude up to three times greater than those at 0.75 m, suggesting that deeper failures become disproportionately more sensitive to climate change.This research received funding from European Union NextGenerationEU – National Recovery and Resilience Plan (PNRR), Mission 4, Component 2, Investment 1.1 -PRIN 2022 – 2022ZC2522 - CUP G53D23001400006.

Abstract. Rainfall-induced shallow landslides are among the most widespread natural hazards in mountainous regions, where intense precipitation, steep topography, and subsurface hydrological processes interact to trigger slope failures. Physically based approaches commonly derive rainfall-triggering thresholds using the framework proposed by Montgomery and Dietrich (1994), which defines instability conditions as a function of groundwater table position. However, this formulation neglects the stabilizing contribution of matric suction in unsaturated soils, potentially limiting its applicability. This study introduces a complementary metric, the Critical Soil Moisture (CSM), which, together with the classical Critical Wetness Index (CWI), provides a continuous hydro‑mechanical description of stability across the full range of hillslope moisture states. The methodology is applied to the 28.6 km² Pontaiba basin in the Carnic Alps (northeastern Italy), a region characterized by steep terrain, high precipitation, and documented shallow landslides. Spatially distributed analyses based on topographic, soil, and landslide inventory data are combined with sensitivity analyses and an ensemble calibration procedure using Receiver Operating Characteristic (ROC) metrics to constrain uncertain parameters. Results delineate three stability regimes, unconditionally stable terrain, groundwater-controlled instability (CWI), and moisture-controlled instability (CSM), and identify slope-dependent hydrological thresholds that can support landslide early warning by focusing on state variables (groundwater, soil moisture) rather than rainfall alone.

This study proposes an advanced hydrologic/landslide modeling application to assess the spatial distribution of rainfall-induced landslides for a sub-basin in central Puerto Rico. The framework implements a stability component into a spatially distributed physically-based hydrological model coupled to a model of plant physiology. Puerto Rico is an ideal study site to assess the performance of landslide modeling efforts due to the availability of thousands of catalogued landslides triggered by Hurricane Maria (HMA) during September 19-22, 2017. The main objective of the study is to simulate the observed landslide events forcing a coupled eco-hydrological-stability model, the tRIBS-VEGGIE-Landslide, with weather data of HMA. The tRIBS-VEGGIE-Landslide model has the advantage of accounting for the vegetation dynamics that affect the soil moisture patterns at an hourly scale and for the soil-water characteristic curve and the saturated shear strength parameters (cohesion and friction angle) to assess the factor of safety (FS) in space and time, using an infinite slope model. The modeling application focuses on two small sub-basins of the Rio Saliente watershed, each smaller than 1 km2. The small study area allows for the use of a 5m DEM resolution topography, which has been derived from a 1m resolution LiDAR measurements. Since many radar and ground stations were destroyed during the hurricane, the hourly time series of the HMA event has been reconstructed by using the NCEP (National Centers for Environmental Prediction) – Environmental Modeling Center (EMC) gridded Stage IV data, produced by NOAA National Weather Service. The precipitation data resulted in a maximum hourly intensity of 64.52 mm/hr, maximum daily intensity of 294.56 mm/day, and rainfall total of 332.15 mm, consistent with other daily reconstructions. Preliminary results demonstrate the importance of the spatial computational mesh and accurate characterization of soil parameters, which play an essential role in simulating landslides with mechanistic models.

The Friuli Venezia Giulia (FVG) region, northeastern Italy, records the heaviest precipitation annual totals of the country. The region counts on a dense ground-station network constituted by 2 main rain-gauges networks, whose sampling frequency has been progressively increased from 60 up to 1 min step. In this work, we propose a comprehensive analysis of the available dataset of continuous series at high temporal resolution (i.e. 60, 30, 5 and 1 min) to verify whether trends in very short rainfalls are underway. We adopt the quantile regression (QR) method which allows to detect changes in the tails of the rainfall distributions and to screen the whole rainfall time series. At this aim, we first introduce a method to check and correct the continuous series by removing the suspicious outliers, based on references values. Significant increasing trends at 5 % of significant level have been detected on some of the analysed stations.

Initiation of rainfall-induced landslides is intricately linked to hydrological conditions, mainly soil water content (SWC), which directly reflects precipitation intensity and patterns. Initiation may occur only on areas that are susceptible to the movement, i.e., the so-called conditionally stable areas. Existing methods delineate unconditionally and conditionally stable areas in “partially saturated” soils based on topography, mechanical properties, and a steady state wetness index (WI) or depth of groundwater level.This study presents a methodology that delineates conditionally stable areas under fully unsaturated soil water conditions, i.e., in the absence of groundwater. In particular, the methodology identifies (i) the ‘partially-saturated’ conditionally stable areas previously mentioned in terms of groundwater level or positive pressure head, and (ii) an ‘unsaturated’ conditionally stable areas, assessed in terms of SWC or negative pressure head. This is obtained computing the factor of safety (FoS) by using two equations of the infinite slope model, which account for both saturated and unsaturated soil conditions. The region delineation ultimately depends on the spatial heterogeneity of topographic and hydro-mechanical properties of the terrain. Finally, for the conditionally stable areas, both ‘partially saturated’ and ‘unsaturated,’ we derive critical maps of landslide initiation, either in terms of SWC or pressure head, respectively. In order to provide efficient and easy-to-interpret maps, the methodology generates Homogeneous Soil Units (HSUs) where each unit is represented by a unique combination of slope and hydro-mechanical properties of the terrain. A unique critical value of SWC or pressure head will result for each HSU at a given hypothetical failure surface, i.e., soil depth.We apply the methodology over the Friuli Venezia Giulia region, Italy, and central Puerto Rico, where thousands of shallow landslides were triggered by Hurricane Maria in September 2017.This research received funding from European Union NextGenerationEU – National Recovery and Resilience Plan (PNRR), Mission 4, Component 2, Investiment 1.1 -PRIN 2022 – 2022ZC2522 - CUP G53D23001400006.

Prolonged droughts and water scarcity have become more frequent in recent years, exacerbating the problem of artificial reservoir management in the Mediterranean area. This study proposes a methodology that combines a Nonlinear AutoRegressive network with eXogenous input (NARX) data-driven model with seasonal forecast (SF) data, with the aim to predict the water volume stored in reservoirs at a mid-term scale, as requested by the local authority. The methodology is applied to four Sicilian reservoirs that experienced water scarcity in the recent past. SFs produced at the European Centre for Medium-Range Weather Forecasting are used to force the NARX models. Also, the reservoirs are in a typical data-scarce environment, where very few or no measurements at all are available. The results show that the NARXs have the capability to reproduce the volumes stored in the considered reservoirs for the investigated period up to four months in advance. The performance of the modelling system strictly depends on: (i) the quality of climate forecasts and (ii) the strength of the autocorrelation for the water volumes.

This presentation provides a compilation of results from an eco-hydrological-landslide model: tRIBS-VEGGIE-Landslide. This effort couples a well-established distributed basin simulator with a slope stability model that considers topography, soil moisture and vegetation. The research explores the impact of soil parameter uncertainty, of vegetation and roots, and of numerical resolution on predictions of rainfall triggered landslides. The Luquillo Mountains on the northeast of Puerto Rico are used as a case study. The uncertainty associated to the soil hydrological and geotechnical parameters is considered by implementing the First- Order Second Moment (FOSM) probabilistic method. The model produces maps and times series of two main representative quantities of probability (hereinafter p.) of failure: (i) the p. of plane of failure at a given soil depth (PrPFi), which indicates dynamically the most probable depth of failure, based on soil moisture dynamics and soil parameters; (ii) the p. of failure of the column, PrFC, that can be generated anywhere within the column. Figure 1 shows the time series of the two probabilities evaluated for the rainfall event reported in Figure 1a. The most probable failure surfaces occurred at depths between 300 and 1000 mm, indicating high probability of shallow landslides. The hydrological and mechanical effects of roots on slope stability were investigated assessing the role of two different vegetation species, shrubs and trees, in comparison to a case without vegetation cover. The model estimates the additional root reinforcement exerted by roots, in addition to the apparent cohesion due to soil suction under unsaturated soil. Figure 2 shows representative results in terms of Factor of Safety (FS) obtained with or without additional root cohesion (∆FSr). The stabilizing effect of the hydrological process is more effective in the case of trees. The resulting drier conditions, due to transpiration, over the most critical depths led to an increase of FS. Finally, to evaluate the influence of terrain resolution on the hydro-geomorphological processes involved in slope stability, we run tRIBS-VEGGIE-Landslide by using five grid-DEM resolutions of the case study basin, i.e., 10, 20, 30, 50, and 70 m (Arnone et al., 2021). Since the model implements Triangulated Irregular Network (TIN), a grid-DEM to TIN algorithm is involved. Using irregular meshes reduces the loss of accuracy with coarser resolutions in the derived slope distribution, in comparison to that estimated using the original grid-DEM. Additionally, from the hydrological perspective, the ultimate impact of resolution on slope stability is significant mostly when lateral water exchanges are allowed within the model framework. REFERENCES Arnone, E., Francipane, A., Dialynas, Y.G., Noto, & Bras, R.L. (2021). Implications of terrain resolution on modeling rainfall-triggered landslides using a TIN-based model. Environ. Model. Softw., 141, 105067. Arnone, E., Dialynas, Y.G., Noto, L.V., & Bras, R.L. (2016a). Accounting for soils parameter uncertainty in a physically based and distributed approach for rainfall-triggered landslides. Hydrol. Process., 30, 927–944. Arnone, E., Caracciolo, D., Noto, L.V., Preti, F., & Bras, R.L. (2016b). Modeling the hydrological and mechanical effect of roots on shallow landslides. Water Resour. Res., 52 (11), 8590-8612.

We present a procedure to predict peak flows in mountain and ungauged basins, by addressing two challenges: fine temporal resolution required to capture intense storms; scarcity of streamflow measurements needed to calibrate hydrological models. The study area is the Fella River basin at Pontebba and its upstream Uque sub-basin, in the northeastern Italian Julian Alps. A non-stationary hydraulic model is combined with field measurements to derive rating curves at the downstream outlet which lacks discharge data. Five-minute rainfall series are exploited to predict hydrographs through a semi-distributed hydrological model, at continuous and event scales. The hydrological modelling is verified on an ungauged basin at the inner upstream outlet. Results indicated that the predictability of intense events is improved when a single storm is evaluated compared to a continuous hydrograph; event-based hydrographs at the upstream outlet are well reproduced by the model calibrated with downstream data, thus allowing the use of the parameters in an ungauged basin.

The Friuli Venezia Giulia (FVG) region, located in the northeast of Italy, is characterised by frequent heavy precipitations that recurrently trigger debris flow phenomena. On August 2003, an intense rainfall concentrated in the north-eastern Julian Alps of FVG produced several floods and debris flow events, widespread on the entire basin of the Fella river watershed, with great economic damage and some casualties. In the light of this, forecasting tools for the debris-flow analysis are useful with a view to a territorial planning. The general aim of our research is to develop a hydro-morphodynamical framework to study debris flow phenomena, which includes the hydrological modelling of the rainfall triggering event, the estimate of the solid-liquid discharge of the debris-flow and the hydraulic modelling of its propagation. While previous works have accomplished the hydrological analysis, in the present study we focus on the evaluation of the solid-liquid discharge and the simulation of its propagation down the slope till its stop. Specifically, we considered a sub-basin of the Fella river watershed, the Uque at Ugovizza, and, in particular, a sub-area of the basin from which the debris flow that swept the village of Ugovizza in 2003 came off. The resulting liquid discharge obtained from the previous hydrological analysis was the input data to derive the solid-liquid discharge of the debris flow, which was assessed by using a formulation proposed in literature. In order to study the propagation of the debris flow, we first identified a rheology model suitable to represent this kind of events. This was then implemented into an in-house numerical model, which integrates the bidimensional shallow water equations by means of finite volume techniques. Furthermore, an appropriate runout criterion was also assessed, so that the final stages of the phenomenon can be represented. The first results of the application of the developed hydro-morphodynamic framework to this case study are presented and discussed.

Leonardo's rule (Lrule) applied to below-ground systems defines a simple topological scheme that describes how the branches of root architectures develop within the soil. The approach does not consider the soil-climate-root interactions. From another hand, eco-hydrological approaches exploit physically-based formulations to derive the dynamic evolution of root profile based on soil and climate characteristics. In homogenous soil and simplified hydrological conditions, analytical solutions can be derived, as demonstrated by Laio's model, who proposed a simple exponential formulation to derive the Root Area (AR) profile. Apart from Laio's model, more generalized functions, i.e. derived by two and three parameters gamma distribution or others, can be efficiently used to derive the AR profile.This communication proposes a combination of the Lrule and eco-hydrological approaches to derive the AR profile, at given soil and climate conditions, allowing to identify a physical and theoretical meaning of the Lrule's parameters. A comprehensive root dataset from field measurements carried out in the region of Tuscany (Italy) is used. Results demonstrate that values of Lrule's parameters derived throughout the proposed mathematical relationships tend to constant values in case of exponential function, which is valid for homogenous soils. Moreover, in a realistic vegetated soil, where top-soil is different than deep-soil, functions derived from a two and three parameters gamma distribution may reproduce better root data observations.

The Julian Alps, located in the region of Friuli Venezia Giulia (FVG, Northeastern Italy), record the heaviest precipitation annual totals in the country. Due to the complex orography and several other prone factors, effects of both prolonged and extreme precipitation can be particularly damaging in this area, causing debris flow, flash floods, avalanches. A proper planning of protection against natural hazards then requires the understanding of possible modification in rainfall characteristics. Since the mountain watersheds of the Alpine area are characterized by a very short time of concentration and hydrological response, extreme events are of particular interest, and rainfall analyses at sub-daily scale could not be appropriate.The region counts on a dense ground-station network which is managed by the regional Civil Protection Agency, constituted by 2 main rain-gauges networks, based on CAE and Micros-SIAP technology, respectively; this last is co-managed by the OSMER-ARPA (OSservatorio MEteorologico Regionale-Agenzia Regionale per la Protezione dell’Ambiente) FVG. The networks count a total of about 200 rain-gauges; for some stations, data at 5-minute resolution are available since the 1996 (CAE network), whereas Micros-SIAP works continuously and at high resolution since the early 2000s. Over the last two decades, the temporal resolution of stations has been progressively increased up to 1-minute step.In this work, we propose a comprehensive analysis of the available dataset at high temporal resolution (i.e. 30 min, 5 min and 1 min) in order to verify whether trends in very short rainfall duration are underway. At this aim, we first analyzed the continuous time series of data recorded by a sample of rain-gauges by the two networks. A preliminary analysis aims at verifying the consistency of the dataset at the higher resolutions. Statistical trends are then assessed by comparing two methods, i.e., the classical Mann-Kendall and the quantile regression at different thresholds and durations. The quantile regression method, which is increasingly used in hydrology, allows to detect changes in the tails of the rainfall distributions and to screen the whole rainfall time series, differently than the traditional methods that require a subset of data (e.g., the rainfall annual maxima).

The northeastern area of Italy, and specifically of Friuli Venezia Giulia region (FVG), is characterized by the heaviest precipitation annual totals in the country. Effects of both prolonged and extreme precipitation can be particularly damaging in this area, causing debris flow, flash floods, avalanches. Due to the very short times of concentration and hydrological response of the mountain watersheds of the analyzed area, extreme and short events are of particular interest. The region has a dense ground-station network which is managed by the regional Civil Protection Agency, constituted by 2 main rain-gauges networks, based on CAE and Micros-SIAP technology, respectively; this last is co-managed by the OSMER-ARPA (OSservatorio MEteorologico Regionale-Agenzia Regionale per la Protezione dell’Ambiente) FVG. The networks count a total of about 200 rain-gauges; for some stations, data at 5-minute resolution are available since the 1996 (CAE network), whereas Micros-SIAP works continuously and at high resolution since the early 2000s. Over the last two decades, the temporal resolution of stations has been progressively increased up to 1-minute step. This work presents a comprehensive analysis of the available dataset at high temporal resolution (i.e. 30 min, 5 min and 1 min) to verify whether trends in very short rainfall duration are underway. The continuous time series of data recorded by a sample of rain-gauges by the two networks are first analyzed. A preliminary analysis aims at verifying the consistency of the dataset at the higher resolutions. Statistical trends are then assessed by comparing two methods, i.e., the classical Mann-Kendall and the quantile regression at different thresholds and durations. Differently than the traditional methods that require a subset of data (e.g., the rainfall annual maxima), the quantile regression method allows to detect changes in the tails of the rainfall distributions and to screen the whole rainfall time series.

The effect of plant roots on hillslope stabilization against rainfall-induced shallow landslides depends on the mutual interaction between biotechnical characteristics of the root system (i.e., root length, root tensile strength, root area, root diameter profile) with the soil root-zone and the hydrological processes therein. Describing adequately the root architecture of a plant species is useful when root strength models, such as the Root Bundle Model (RBM), are applied to assess the ultimate root reinforcement. This study describes the preliminary results of the calibration of an existing Root Topological Model (RTM) combined with a RBM model to estimate the additional roots shear resistance of vegetation typical of a subtropical climate. Specifically, the dataset of the root system of four Hong Kong native species of shrubs (Rhodomyrtus tomentosa and Melastoma sanguineum) and trees (Schefflera heptaphylla and Reevesia thyrsoidea) has been used. The dataset includes the measurements relative to both the root architecture, i.e., root diameter classes and number of roots as function of depth, and the root resistance, i.e. root tensile strengths for each diameter classes, which were obtained from laboratory test. The present application allows for calibrating and exploiting the potentiality of the framework RTM-RBM in a climatic environment different from the Mediterranean one analyzed so far for its development, thus testing the response and the flexibility of the modeling framework. The availability of such a tool could enhance, for example, the assessment of the most suitable plant species to be adopted for the slope stabilization in different soil and/or climatic conditions.

The Northeastern Italy and the therein Friuli Venezia Giulia (FVG) region are frequently hit by heavy and prolonged precipitations, which cause frequent debris flow and diffused shallow landslides. In this study we focus on a mountain sub-basin of the Fella river watershed, the Uque at Ugovizza, located in the northeastern Julian Alps of the FVG, where a disruptive rainfall-triggered debris flow occurred in 2003. The work aims at pursuing two main targets: (i) implementing a rainfall-runoff and hydro-morphodynamical framework for the analysis of debris flow initiated by intense heavy precipitation; ii) exploiting, for the first time, the flexibility of the GEOframe-NewAge semi-distributed hydrological model simulating high temporal resolution simulations (5-minutes) rainfall-runoff events. The GEOframe-NewAge is an open-source component-based modeling framework, which simulates the entire hydrological cycle of the study area, including the snow melting, the soil water storage and the runoff production and routing in the river network; the model is suitable for the rainfall-runoff event scale simulations in Alpine environment with scarce measurements. Specifically, we describe the results of the calibration and validation procedures applied to four selected intense events occurred in the period 2009-2019. Meteorological data at 5 minutes-step are used to rainfall-runoff modeling, whereas streamflow at 30 minutes is used for the model calibration and validation. Preliminary results show that the models is able to capture the temporal and spatial dynamic of extremes short events, providing satisfying Nash and Sutcliffe coefficient values.

This study employs a distributed eco-hydrological-landslide model, the tRIBS-VEGGIE-Landslide, to evaluate the influence of terrain resolution on the hydro-geomorphological processes involved in slope stability analysis. The model implements a Triangulated Irregular Network (TIN) to describe the topography starting from a grid-DEM. Five grid-DEM resolutions of the case study basin, i.e., 10, 20, 30 and 70 m, are used to derive the corresponding TINs. The results show that using irregular meshes reduces the loss of accuracy with coarser resolutions in the derived slope distribution in comparison to slope distributions estimated from the original grid-based DEM. From a hydrological perspective, the impact of resolution on soil moisture patterns and on slope stability is significant mostly when lateral water exchanges are allowed. The degrading of resolution leads to a reduction of the predicted unstable areas, with respect to the highest resolution case, from about 15% (20 m) to more than 40% (70 m).

In this paper, we discuss the drought-alert decision support system (DA-DSS), a solution developed by the Small Medium Enterprise Amigo s.r. l., (Rome, Italy) and MaP Ltd., (Athens, Greece). within the Horizon 2020 cross-sectoral project "cross-climate". DA-DSS is a prototype of the WebGIS application aimed to support the water utilities (WUs) in the management of drinking water in a changing climate. DA-DSS combines climate data, infrastructures geospatial data, and visualization tools. Climate data exploit the seasonal forecasts provided by the EU Copernicus services in order to obtain short and mid-term DA, up to a six months' time range. This information is integrated with geospatial data with advanced geographic information system technology, and finally implemented into a WebGIS application. Two small and medium WUs, located in Greece and Italy respectively, served as test cases for the system development in a co-design approach, based on their existing decision chains. The prototype system was then released in the municipality of Zakynthos, Greece. The investigation of the user's needs suggested that both the WUs are interested in predictions on 3 months-cumulated precipitation, 3 months in advance, from November to March. The test of the prototype in November 2017 demonstrated that the uncertainty, assessed in terms of false rate, is low for the short-term prediction (3 months).

Root topological models are schematic representations of the root structure based on a defined topology graph theory. In the context of hillslope stabilization modeling against rainfall-induced shallow landslides, the root topological models may be used in combination with root strength models assessment, such as the Root Bundle Model (RBM), to estimate the ultimate root reinforcement. The effect of plant roots on slope stabilization is determined by the interaction between soil and the hydrological processes (within the root zone) and the biotechnical characteristics of the root system, such as root length, root density, root tensile strength, root area, root diameter profile and the total number of roots. Describing adequately the root architecture of a plant species is useful, for example, to evaluate how the root structure may change in different soil and/or climatological conditions and, ultimately, as an example, to assess the most suitable plant species to be adopted. This study exploits the potentiality of a root topological model based on Leonardo’s rule in describing root architectures of (i) different species (and tree individual) at given growth conditions, (ii) same species at different environmental conditions, e.g., exposure to light, water and nutrient availability. The former is supported by field campaign measurements from Tuscany region, the latter are reproduced starting from a reference case and imposing growth assumptions. Next, the information of the root system, in terms of root length, density, root diameter profile, total number of roots, are used to estimate, through a RBM approach, the additional root tensile force, deriving it from the force-deformation theory of linear elasticity in a rigorous framework aimed to derive the additional shear resistance from the Mohr-Coulomb’s failure plane. The preliminary results demonstrated the capability of the root topological model of reproducing different types of root system; additional data are required to further validate the model, with regard to the growth conditions simulation. Similarly, laboratory test of root strength would allow to quantify the improvement derived from the rigorous method adopted to estimate the additional root strength.

Wanner air has the potential to hold more water vapour and, therefore, to provide more water to rainfall events. Studying the relationship between rainfall and temperature represents an emerging issue in hydrology and meteorology, since it can be considered fundamental for evaluating the effects of global warming on future precipitation. Various approaches have been tested across different parts of the world, in many cases observing an intensification of extreme precipitation at higher temperatures consistent with the well-known thermodynamic Clausius-Clapeyron relation (CC-scaling rate of 6-7% degrees C-1). However, at different locations for hourly time-scales, the temperature-extreme rainfall scaling can be higher (super-CC) or lower (sub-CC). This study contributes to the understanding of the scaling relationship between extreme rainfall and temperature under climate conditions characteristic of Mediterranean semi-arid regions, rarely explored in the past. The role of different factors, such as rainfall characteristics and climatic seasonality, modelling framework and rainfall accumulation period are investigated through an application to Sicily (Italy). In particular, the suitability of different types of regression models used to interpret the relationship between hourly and sub-hourly extreme rainfall and surface temperature is explored. We find overall a sub-CC scaling for most of the island of Sicily. However, the rainfall-temperature scaling relationship is not constant over the temperature range and may be dependent on the season. The different results obtained highlight the importance of modelling choices for analyses in regions characterized by semi-arid climates. More specifically, we observe increasing scaling rates for decreasing rainfall accumulation periods, and significant sensitivity of scaling rates to the selected extreme rainfall quantile. Our novel use of piecewise and locally-weighted scatter plot smoothing regression-based approaches allow the accurate characterization of the temperature dependence of extreme rainfall in Sicily. This identifies a peak-like structure for the drier season, not detected by the simple application of the commonly-used exponential regression based approach.