HS7.7 | Hydrometeorologic stochastics: from theoretical advancements in extremes, scales and probabilities to applications in industry
Hydrometeorologic stochastics: from theoretical advancements in extremes, scales and probabilities to applications in industry
Convener: Jose Luis Salinas Illarena | Co-conveners: Carlotta Scudeler, Gaby GründemannECSECS, Stergios EmmanouilECSECS, Bora ShehuECSECS
Posters on site
| Attendance Wed, 17 Apr, 10:45–12:30 (CEST) | Display Wed, 17 Apr, 08:30–12:30
Hall A
Wed, 10:45
Over the last decades, a significant body of empirical and theoretical work has revealed the departure of statistical properties of hydrometeorological processes from the classical statistical prototype, as well as the scaling behaviour of their variables in general, and extremes in particular, in either state, space and/or time. Extremes and more generally the statistics of hydrometeorologic processes are the key input for hydrological applications, e.g. in natural catastrophe modelling. An example of this is the estimation of design rainfall. Beside the estimation of the absolute rainfall amount related to a certain return period, the intra-event rainfall distribution, its spatial extension and the rainfall intensities at neighbouring stations can be required depending on the intended application and thus the spatial and temporal scales of interest should be determined. Another good example are the large scale connections between hydrometeorologic extremes and climatic oscillations such as NAO or ENSO, and how these correlations can evolve in a changing climate. These are only two examples among numerous hydrologic applications.
On the one hand, the estimation of the hydrometeorological extremes and their probability distribution, the identification and incorporation of supporting information to improve these estimates, and their hydrologic application over a wide range of scales remain open challenges. On the other hand, hydrometeorologists had never access to so much computer power and data, including novel AI approaches, to face these open challenges.
This session welcomes, but is not limited to submissions on the following topics:
- Coupling stochastic approaches with deterministic hydrometeorological predictions, in order to better represent predictive uncertainty
- Development of robust statistics under non-stationary conditions for design purposes
- Development of parsimonious representations of probability distributions of hydrometeorological extremes over a wide range of spatial and temporal scales in risk analysis and hazard prediction
- Improvements for reliable estimation of extremes with high return periods under consideration of upper or lower limits due to physical constraints
- Linking underlying physics and hydroclimatic indices with stochastics of hydrometeorologic extremes
- Exploration of supporting data sets for additional stochastic information as well as the use of novel AI and machine learning approaches

Posters on site: Wed, 17 Apr, 10:45–12:30 | Hall A

Display time: Wed, 17 Apr 08:30–Wed, 17 Apr 12:30
Chairpersons: Carlotta Scudeler, Jose Luis Salinas Illarena, Bora Shehu
Francesco Marra, Marika Koukoula, Antonio Canale, and Nadav Peleg

We present a new statistical method for estimating extreme sub-hourly precipitation return levels that explicitly hinges on our physical understanding of the processes. The TENAX (TEmperature-dependent Non-Asymptotic statistical model for eXtreme return levels) model is based on two modules: (i) a magnitude module describes precipitation intensities using temperature as a covariate. It includes all the information about thermodynamics and local dynamics of the processes at a given temperature; (ii) a temperature module accounts for the distribution of daily temperature during rainfall events. Using the total probability theorem, the two modules are linked to provide a physics-based estimate of the marginal distribution of the precipitation intensities. Return levels are then estimated using a non-asymptotic method. Assuming that the physics of convection remains unchanged in the future (i.e., no change in the magnitude module) and that convection remains the dominant process, the TENAX model enables to project future sub-hourly precipitation return levels only based on the projected changes in daily temperature during rainy days. We will discuss the theory behind TENAX and show it can reproduce return levels with the same accuracy as more parsimonious non-asymptotic methods. We will additionally show that the model reproduces known properties of the extreme precipitation-temperature scaling relation for which it was not explicitly designed. Last, in hindcast, we will demonstrate that TENAX trained on observations of precipitation and temperature can well reproduce “future” unseen return levels only based on projections of daily temperatures. As projections of daily temperature from climate models are more readily available and accurate than those of sub-hourly extreme precipitation, TENAX could allow one to derive future sub-hourly return levels in any location globally where observations of past sub-hourly precipitation and daily near-surface air temperature are available.

How to cite: Marra, F., Koukoula, M., Canale, A., and Peleg, N.: A physics-based statistical model to predict sub-hourly extreme precipitation intensification based on temperature shifts, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-3337, https://doi.org/10.5194/egusphere-egu24-3337, 2024.

Artemis Venardos and Carlotta Scudeler

Insurers use catastrophe models to assess the risk related to catastrophic events such as floods and windstorms and, in turn, to inform their pricing, manage their risk accumulation, and allocate their capital for regulatory purposes. One of the major challenges when it comes to catastrophe modelling is to ensure that any likelihood of adverse scenarios correlating across different perils is appropriately captured. When this correlation of risk is underestimated, it can lead to weaker financial protection. Insurers’ European catastrophe models generally assume windstorms and inland floods to be independent perils, while recent storm case studies such as Storm Kyrill in 2007 and Storm Desmond in 2015 have suggested that these two hazards occur simultaneously in the same weather systems, increasing joined risk.  The purpose of this study is to quantify the correlation in the UK, by analysing historical wind speed and precipitation extremes. It employed an event-based approach that utilises the Copernicus C3S Footprint dataset, as well as E-OBS daily gridded precipitation data, to investigate whether the windstorms between 1979 to 2021 correlate with precipitation extremes. The Storm Severity Index (SSI), accompanied by a newly developed Precipitation Severity Index (PSI), were used to assess the severity of each windstorm, and inform the subsequent correlation analysis. Among the major findings, it is shown that 1) wind and precipitation extremes exhibit a probability of simultaneous occurrence; 2) the highest SSI (category 1 and 2) windstorms are the most correlated to precipitation extremes; 3) rainfall events with the highest PSI most likely follow a clustering series of storms, leading to prolonged and heavy rainfall (e.g. storm Ciara, 2020 and storm Desmond, 2015); and 4) the SSI severity category of each storm might play a role in whether wind and precipitation extremes occur in the same location or not. The key takeaway of this study suggests the importance of incorporating the correlation between these two hazards (wind and precipitation that leads to floods) into insurers' catastrophe models. Further analysis will look at river flow data and wind and flood losses.  

How to cite: Venardos, A. and Scudeler, C.: Windstorm and Flood correlation in UK , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-22052, https://doi.org/10.5194/egusphere-egu24-22052, 2024.

Jaya Bhatt and Vemavarapu Venkata Srinivas

Flood estimates corresponding to probable maximum precipitation (PMP) are desirable for planning, designing and risk assessment of large hydraulic structures, such as spillways of large dams, whose failure may have catastrophic consequences on ecology, economy, and the environment. In practice, PMP estimates are obtained using hydrometeorological or/and statistical methods as per the recommendations of the World Meteorological Organization. In regions where data of hydrometeorological variables (e.g., precipitation, temperature, relative humidity) are limited or unavailable, practitioners often resort to various statistical methods which require only precipitation records to estimate PMP. Among statistical methods, the Hershfield method is widely used when records from several sites in a region are available whereas conventional probabilistic approach is preferred for at-site analysis. However, arriving at reliable PMP estimates at data-sparse locations is still a challenge. Thus, there is a growing need to improve the existing statistical methods and develop/explore alternate methods. Against this backdrop, this study proposes a new variant of Bethlahmy method, which is a non-parametric method, to facilitate the estimation of PMP at locations with sparse records of extreme precipitation.

The proposed Bethlahmy variant involves (i) mapping of datapoints and corresponding ranks of target site’s annual maximum precipitation series to a non-dimensional space (NDS), (ii) using the mapped information to arrive at a surrogate estimate for PMP in the NDS, and (iii) mapping the surrogate estimate to PMP in the original space. The efficacy of the proposed Bethlahmy variant over various existing statistical techniques is demonstrated through Monte Carlo Simulation experiments and a case study on 37,872 stations from a global precipitation database. The existing techniques include the original Bethlahmy and Hershfield methods, conventional probabilistic approach, and relevant variant(s). Results revealed that the proposed Bethlahmy variant outperforms other methods/variants across samples varying in size and extreme precipitation characteristics, making it a promising statistical alternative for PMP estimation.

How to cite: Bhatt, J. and Srinivas, V. V.: Estimating reliable Probable Maximum Precipitation at data-sparse locations , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-244, https://doi.org/10.5194/egusphere-egu24-244, 2024.

Yichao Xu, Zhiqiang Jiang, Yanpeng Dai, and Zhijin Li

The Twenty-Four Solar Terms are an ancient and unique Chinese contribution, developed by agrarian laborers for the convenience of agricultural scheduling. Historically, each Solar Term has exhibited a significant correlation with climate change. Our study revisits and validates the Solar Terms’ sensitivity to regular precipitation metrics over the past half-century. We assessed the reliability of this temporal framework by comparing 30-day cumulative precipitation and the number of effective rainfall days around typical Solar Terms, along with the correlation of China’s Standardized Precipitation Index (SPI) with the Heihe-Tengchong Line, known as the Hu Line, during various Solar Terms. The research further investigates the correspondence between multiple scenarios of future extreme precipitation events, including Event-based Extreme Precipitation (EEP), and the Solar Terms. This study focuses on identifying which Solar Terms have historically been, and are likely to continue being, prone to extreme precipitation, as well as their spatial distribution patterns across China. The result indicates that under four different socio-economic pathways in future scenarios, over 55.2% of the regions in China will witness a concentration of extreme precipitation events during the Minor Heat and Major Heat Solar Terms. These events are predominantly expected to occur in the Qinghai-Tibet region, averaging over 10 instances every five years that exceed the 95th percentile for extreme rainfall, showing an increasing risk trend over time. This study not only enhances the cultural depth of our research but also fosters a profounder understanding of the cyclical patterns of extreme precipitation and the related flood risks it entails, offering a novel perspective for guiding flood prevention efforts and the study of extreme precipitation patterns.

How to cite: Xu, Y., Jiang, Z., Dai, Y., and Li, Z.: Chinese Wisdom: Investigating Future Patterns of Extreme Precipitation Based on the Twenty-Four Solar Terms, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4277, https://doi.org/10.5194/egusphere-egu24-4277, 2024.

Jannis Hoch, Izzy Probyn, Francesco Marra, Chris Lucas, James Savage, Oliver Wing, Chris Sampson, and Nans Addor

Intensity–duration–frequency (IDF) curves are representations of the probability that a given rainfall intensity over a given duration [GU1] will be exceeded [GU2] within a given period. To construct IDF curves, rainfall observations are required, ideally at the sub-daily temporal resolution. Unfortunately, such measurements are available only for a few locations world-wide. This poses a major challenge for simulations of global pluvial flood hazard and risk which require information of intensity, duration, and probability as boundary conditions.

As an alternative to global IDF curves created from remotely sensed rainfall, we here propose a bottom-up approach which departs at the gauge level and employs machine-learning for regionalizing information on IDF curves from gauged to ungauged areas.

To that end, we use available quality-controlled sub-daily precipitation data from the GSDR data set to derive Simplified Metastatistical Extreme Value (SMEV) parameters at around 10,000 locations world-wide. After combining these parameters with globally available data of precipitation drivers, a random forest regression model is applied. Results indicate that some SMEV parameters can be better regionalized than others. With globally available SMEV parameters, it is possible to obtain rainfall intensity for any combination of duration and frequency.

We then evaluated these IDF maps against analytical intensities derived at the GSDR stations directly. Results show overall good agreement, yet the tails of the distributions are not entirely represented in our simulated intensities.  Additionally, we benchmarked our intensity maps against similar datasets such as PPDIST and GPEX. Last, we assessed practical implications by comparing flood maps created with the various datasets used as pluvial boundary condition. While there are fundamental differences in how each of the datasets is derived, our analysis indicates overall similar spatial patterns and distributions of rainfall intensities.

While such data-driven approaches clearly depend on the quality and quantity of available sub-daily rainfall observations, our proposed bottom-up approach seems to be able to scale local data to global data applicable in both flood risk research and practice.

How to cite: Hoch, J., Probyn, I., Marra, F., Lucas, C., Savage, J., Wing, O., Sampson, C., and Addor, N.: A bottom-up regionalization approach for extreme rainfall events, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5682, https://doi.org/10.5194/egusphere-egu24-5682, 2024.

Sonja Jankowfsky, Mohammad Sharifian, Edom Moges, Ludovico Nicotina, Shuangcai Li, and Arno Hilberts

Running inundation on a stochastic event set with thousands of events can be quite time consuming, especially if physically based methods such as the shallow water equations are used. In order to optimize runtime and to keep a high spatial resolution, event footprints are often reconstructed using return period maps. However, this means that design storm events need to be constructed for each return period which should ideally be representative for different event durations.

Here, we compare the alternating block method, a popular design storm model, to actual event hyetographs from a selection of storm events in Florida. The hyetographs are input to a 10m grid-based Green-and-Ampt infiltration model which is coupled to a two-dimensional shallow-water inundation model. The difference between the alternating block method and the actual event hyetograph is measured based on the flood extent and depth.

How to cite: Jankowfsky, S., Sharifian, M., Moges, E., Nicotina, L., Li, S., and Hilberts, A.: How appropriate is the alternating block method to represent flooding from extreme precipitation events?, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6064, https://doi.org/10.5194/egusphere-egu24-6064, 2024.

Abubakar Haruna, Juliette Blanchet, and Anne-Catherine Favre

Accurately quantifying the impact of climate change on past and future precipitation, especially in terms of extreme precipitation events, remains a significant challenge. Furthermore, the scarcity and high variability of extreme events make this task particularly daunting. Current approaches based on extreme value theory (EVT),  relying on annual maxima series or exceedances above large thresholds, are limited in their efficiency as they only consider a small fraction of the available data. Additionally, these methods do not model the bulk of the distribution, which has applications in areas such as water resources management, urban water supplies, and hydropower. To address these limitations, Naveau et al. (2016) proposed the Extended Generalized Pareto distribution (EGPD), which models the entire non-zero precipitation range while remaining consistent with EVT in both lower and upper tails. While the EGPD has seen wide applications in modeling precipitation in various regions, its application has predominantly been within a stationary framework (e.g. Haruna et al.,2022, 2023). This study explores the potential of a non-stationary version of the EGPD to jointly model trends in both the bulk of the precipitation distribution and in the extremes. The non-stationarity is accommodated by allowing the EGPD parameters to be parametric functions of relevant explanatory variables. The proposed model is then applied to precipitation datasets in Switzerland, a region where long term warming of twice the global average has already been experienced.

  • Haruna, A., Blanchet, J., and Favre, A.-C. (2023). Modeling Intensity-Duration-Frequency Curves for the Whole Range of Non-Zero Precipitation: A Comparison of Models. Water Resources Research, 59(6):e2022WR033362.
  • Naveau, P., Huser, R., Ribereau, P., and Hannart, A. (2016). Modeling jointly low, moderate, and heavy rainfall intensities without a threshold selection. Water Resources Research, 52(4):2753–2769
  • Haruna, A., Blanchet, J., and Favre, A.-C. (2022). Performance-based comparison of regionalization methods to improve the at-site estimates of daily precipitation. Hydrology and Earth System Sciences, 26(10):2797–2811

How to cite: Haruna, A., Blanchet, J., and Favre, A.-C.: Non-stationary extended generalized Pareto distribution  for joint assessment of trends in the bulk and extreme precipitation, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-12133, https://doi.org/10.5194/egusphere-egu24-12133, 2024.

Felix Fauer and Henning Rust

Extreme precipitation and flooding events in middle Europe lead to high death tolls and huge existential and financial losses. Evaluating how the probability of such events changes with respect to climate change can help in preventing casualties and reducing impact consequences. We create Intensity-Duration-Frequency (IDF) curves which describe the major statistical characteristics of extreme precipitation events (return level, return period, time scale). They provide information on the probability of exceedance of certain precipitation intensities for a range of durations and can help to visualize how extreme the event for different durations is. We modeled the underlying distribution of block maxima with the Generalized Extreme Value (GEV) distribution. Maxima from different durations are used and enable a model that can evaluate different time scales. All durations are modeled in one single model in order to prevent quantile-crossing and to assure that estimated quantiles are consistent.

Large-scale information is included by letting the GEV parameters depend on variables like NAO, temperature and blocking. We found an increase in probability of extreme precipitation with year as proxy for climate change and temperature, while the effect of the other variables depends on the season. Since it is easier to project average values than extremes, we use the relations between average large-scale covariates and extreme precipitation to create future IDF-relations based on climate projections. Furthermore, we plan to add a spatial component to the model that enables the usage of data from several neighboring stations in one model and interpolate to ungauged sites. This will be the basis for investigating how gridded data sets can be used to complement the station-based approach. One focus will lie on the challenge of dependence between neighboring grid points.

How to cite: Fauer, F. and Rust, H.: Large scale influence on extreme precipitation, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15037, https://doi.org/10.5194/egusphere-egu24-15037, 2024.

Jose Salinas, Rahul Sojitra, and Sonja Jankowfsky

Extreme rainfall events have distinct spatio-temporal characteristics that are usually determined by climatic conditions and dominant weather types at the site of interest. When trying to simulate synthetic rainfall events, this spatio-temporal structure needs to be reproduced in order to achieve suitable design attributes (e.g. design of a drainage network, or flood protection at different spatial scales).

The study are of this analysis is the state of Florida, in the USA. Tropical Cyclone induced rainfall extremes tend to be more catastrophic in this region, therefore we focused only on historical TC events from the last 20 years. We analysed the rainfall Return Period composition at different temporal and spatial scales for observed gridded precipitation, and explored the consequences that these spatio-temporal characteristics have for design rainfall applications.

How to cite: Salinas, J., Sojitra, R., and Jankowfsky, S.: Analysis of the Return Periods of Tropical Cyclone Rainfall events across temporal and spatial scales in the state of Florida (USA), EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-22043, https://doi.org/10.5194/egusphere-egu24-22043, 2024.

Mauricio Zambrano-Bigiarini, Cristóbal Soto, and Violeta Tolorza

In this work, we computed IDF curves for the climatically and topographically diverse Chilean territory (17-56ºS) using both stationary and non-stationary approaches based on three state-of-the-art gridded datasets: i) the Integrated Multi-Satellite Retrievals for GPM (IMERGv06B), ii) the fifth generation ECMWF ERA5 reanalysis, and iii) ERA5-Land, the high-resolution reanalysis from ECMWF. The three gridded datasets were used to compute the annual maximum intensities for 12 durations (1, 2, 4, 6, 8, 10, 12, 18, 24, 48, and 72 h) and six return periods (T=2, 5, 10, 25, 50, and 100 years), for the period 2001-2021, using a common spatial resolution of 0.10° and hourly temporal resolution. Data from 161 quality-checked hourly rain gauges are used to calculate IDF curves that serve as a benchmark for the curves derived from the gridded datasets.

First, we calculated a bias correction factor for the annual maximum intensities for each duration, based on the comparison of the gridded value with the point one. Second, the previous correction factors were applied to the annual maximum intensities derived from each gridded dataset, using splines over the entire spatial domain. Third, the bias-corrected annual maximum intensities were calculated for each duration and return period using the Gumbel probability distribution, assuming stationary conditions. Then, the same annual maximum intensities were calculated for each duration and return period using a non-stationary Gumbel probability distribution with a varying mean. Fifth, we used the non-parametric Mann-Kendall test to assess the existence of trends in annual maximum intensities. Finally, 41 (1981-2021) and 21 (2001-2021) years of hourly precipitation data from ERA5 and ERA5-Land were used to test the effects of data length on the resulting stationary and non-stationary IDF curves.

Our results reveal how the bias of annual maximum intensities changes in space, with generally smaller biases for longer periods. In addition, minor differences were found between the maximum annual intensities calculated using the stationary and non-stationary approaches for 2001-2021. Unexpectedly, we found some significant decreasing trends (p-value < 0.05) in Central-Southern Chile (32-43ºS) for the annual maximum intensities derived from ERA5 and ERA5-Land, while these trends were more localised and divergent (increasing and decreasing) for IMERG. Finally, there were only minor differences between the annual maximum intensities derived from ERA5 and ERA5-Land when using 21 years compared to 41 years of hourly records, for both the stationary and non-stationary approaches.

We gratefully acknowledge the financial support of ANID-Fondecyt Regular 1212071,  ANID-PCI NSFC 190018, and ANID-Fondecyt Iniciación 111908064.

How to cite: Zambrano-Bigiarini, M., Soto, C., and Tolorza, V.: Spatially-distributed Intensity-Duration-Frequency (IDF) curves for Chile using sub-daily gridded datasets, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-21043, https://doi.org/10.5194/egusphere-egu24-21043, 2024.