Rainfall Generation Research Articles

Development of a stochastic rainfall generator to yield unprecedented rainfall events

Unprecedented rainfall events are defined as rainfall events that have extremely high magnitude with low probability of occurrence. Although such events have occurred in the recent past, there might be no historical record of unprecedented rainfall. While significant efforts have been made to understand the major drivers of unprecedented events, the use of statistical modelling to estimate the likelihood of unprecedented events has gained little to no attention. The current study proposes a ‘serial’ type stochastic rainfall generator (SRG) to simulate daily rainfall with an additional focus on unprecedented events. The study introduces resampling of model parameters to yield the non-stationarity in simulated rainfall distribution. The proposed approach is applied to simulate daily rainfall over twenty-six grid points in Southeast Texas, which experienced Hurricane Harvey in 2017. Towards this, the study obtains rainfall accumulation data during Harvey from eight weather stations during. We found that the proposed approach can successfully generate unprecedented events if two power law tail tuning parameters are adjusted with respect to future warming levels. Rainfall magnitudes correspond to 50, 100, 250 and 500-year return periods simulated by the proposed model are found to be substantially higher than historical (i.e., observed) return period. Furthermore, the study found that the estimated return periods of Hurricane Harvey like rainfall is almost the same as the return period estimated by a former study of Emmanuel (2017). The potential benefit of the proposed approach in hydrological risk assessment has also been discussed.

CON-SST-RAIN: Continuous Stochastic Space–Time Rainfall generation based on Markov chains and transposition of weather radar data

In this study, we present CON-SST-RAIN, a novel stochastic space–time rainfall generator specialized for model-based urban drainage design and planning. CON-SST-RAIN is based on Markov Chains for sequences of dry/rainy days and uses stochastic storm transposition (SST) to generate realistic rainfall fields from weather radar data. CON-SST-RAIN generates continuous areal rainfall time series at arbitrary lengths. We propose a method for updating the Markov Chains by each passing year to better incorporate low-frequency variation in inter-annual rainfall values. The performance of CON-SST-RAIN is tested against multi-year records from rain gauges at both point and catchment scales. We find that updating the Markov Chains has a significant impact on the inter-annual variation of rainfall, but has little effect on mean annual/seasonal precipitation and dry/wet spell lengths. CON-SST-RAIN shows good preservation of extreme rain rates (including sub-hourly values) compared to observed rain gauge data and the original SST framework.

A Nonstationary Stochastic Rainfall Generator Conditioned on Global Climate Models for Design Flood Analyses in the Mississippi and Other Large River Basins

AbstractExisting stochastic rainfall generators (SRGs) are typically limited to relatively small domains due to spatial stationarity assumptions, hindering their usefulness for flood studies in large basins. This study proposes StormLab, an SRG that simulates precipitation events at 6‐hr and 0.03° resolution in the Mississippi River Basin (MRB). The model focuses on winter and spring storms caused by water vapor transport from the Gulf of Mexico—the key flood‐generating storm type in the basin. The model generates anisotropic spatiotemporal noise fields that replicate local precipitation structures from observed data. The noise is transformed into precipitation through parametric distributions conditioned on large‐scale atmospheric fields from a climate model, reflecting spatial and temporal nonstationarity. StormLab can produce multiple realizations that reflect the uncertainty in fine‐scale precipitation arising from a specific large‐scale atmospheric environment. Model parameters were fitted monthly from December–May, based on storms identified from 1979 to 2021 ERA5 reanalysis data and Analysis of Record for Calibration (AORC) precipitation. StormLab then generated 1,000 synthetic years of precipitation events based on 10 CESM2 ensemble simulations. Empirical return levels of simulated annual maxima agree well with AORC data and show an overall increase in 1‐ to 500‐year events in the future period (2022–2050). To our knowledge, this is the first SRG simulating nonstationary, anisotropic high‐resolution precipitation over continental‐scale river basins, demonstrating the value of conditioning such stochastic models on large‐scale atmospheric variables. StormLab provides a wide range of extreme precipitation scenarios for design floods in the MRB and can be further extended to other large river basins.

Pyraingen: A python package for constrained continuous rainfall generation

Continuous rainfall is often required for flood estimation and water resources assessment. However, stochastically generated continuous rainfall records are typically inconsistent with intensity-frequency-durations (IFDs) used in design, and do not simulate the rainfall behaviour we can expect in a future warmer climate. Here, we present the python package pyraingen to generate ensembles of continuous subdaily rainfalls at a single location that preserve IFD relationships. Rainfall generation can also be conditioned on a climate covariate, in this case temperature, to produce rainfall timeseries reflective of a future climate. We present an implementation at five climatologically unique sites across Australia and demonstrate its ability to produce rainfall timeseries that reflect both present day and future changes in rainfall persistence, temporal patterns, means, and extremes. This package will enable studies using continuous simulation to assess the impacts of climate change using the current guidance on changes to both IFDs and mean annual rainfalls.

Impacts of temporal/spatial rainfall heterogeneities on peak runoff distribution and intensities for an urban river basin of south China

AbstractIn this research, a modeling approach of rainfall generator coupled with high resolution rainfall products were proposed to generate designed rainfall events under multiple spatial and temporal distributions, which was then employed to analyze the impacts of spatial and temporal rainfall heterogeneities on peak runoff for watersheds. Three scenarios were developed under multiple degrees of impermeable underlying surface areas within an urban watershed in south China. Detailed runoff processes were analyzed through the adoption of a distributed hydrological model (GSSHA). A covariance analysis method combined with rainfall spatio‐temporal heterogeneity characteristic were used to quantify heterogeneity effects on peak runoff. Results indicated that coupling short period (2008–2016) remotely rainfall data and RainyDay results could successfully reproduce designed rainfall events, spatio‐temporal heterogeneity of rainfall contributed significantly to the peak runoff, which was greater than those by rainfall duration and capacity, and the increase in impermeable underlying surface enhanced the complexities of the effects. Over each rainfall duration with increasing rainfall return period, the indicator of rainfall peak coefficient (RWD) would decrease and then increase. Regarding the total rainfall center (tg), 25 mm/h threshold rainfall spatial coverage (A25) decreased with increasing imperviousness, 1‐h maximum rainfall (Rmax) surged with increasing imperviousness at rainfall duration of 2 and 24 h. Innovations of this research lied in: combination of a rainfall generator model based on a stochastic storm transposition technique and remote‐sensing rainfall data to generate designed rainfall events, a rainfall spatial and temporal heterogeneities index system was developed to reveal how the changing characteristics of rainfall distribution and the impacts on peak runoff, and in‐depth analysis of the impacts on runoff peak under multiple urban development scenarios for increasing capability in flood control/prevention.

Possibility of using the STORAGE rainfall generator model in the flood analyses in urban areas

In this investigation, we evaluated the applicability of the Stochastic Rainfall Generator (STORAGE) as a data source for deriving design hydrographs in urban catchments. This assessment involved a comparison with design rainfall calculated using Intensity-Duration-Frequency (IDF) curves derived from observed time-series data. The resulting design rainfall values from both methods were incorporated into a hydrodynamic model of the storm sewer network. To simulate peak discharge and flood areas, the Storm Water Management Model (SWMM) program was employed in conjunction with SCALGO. Our findings indicate that design rainfall values obtained from the STORAGE model exceeded those derived from the observed time-series, with a more pronounced difference for shorter rainfall durations. Simulations further revealed that peak runoff disparities between the two approaches were most evident at a 0.10 probability of exceedance compared to a 0.01 probability. Hydrodynamic simulations demonstrated that the flooding volume induced by design rainfall based on the STORAGE model surpassed that resulting from observed rainfall. Across all events, both the flooding volume and area from STORAGE were consistently greater than those derived from IDF curves. The integration of the SWMM model with the SCALGO application introduced a novel functionality for dynamic visualization of flooding, offering valuable insights for effective flood management in urban areas.

Performance of low impact development on peak flow reduction in an urban system

AbstractThis study proposes an approach to evaluate the efficiency of low impact development (LID) in reducing urban runoff using a rainfall generator to disaggregate daily rainfall into sub‐hourly rainfall data, which are used as input of a hydrological model at the urban watershed scale. Twelve scenarios are analyzed combining four percentages of impervious area retrofitted with LIDs (25%, 50%, 75% and 100%), and three LID combinations of green roofs (GRs) and rain gardens (RGs). The rainfall generator Rainsim V.3 is used to generate 500 years of rainfall data with a 15‐min time step to analyze the performance of LIDs in the long‐term with the LID module of the Soil and Water Assessment Tool hydrological model. An urban watershed of 3 km2 located in Florence (Italy) is selected as a case study. Results show the performances of GRs and RG on peak flow reduction, highlighting a maximum flow reduction of single facilities ranging between 15% and 60% that can improve in case of their combination. The hydrological performances of LID combinations are very sensitive to the intensity of rainfall events, as well as percentages of area treated underlining the importance of simulating multiple scenarios of intervention to determine the most efficient combination of LIDs for a given case study and support their proper design from a urban water hydrology perspective.

Artificial Neural Network Technique for Annual Rainfall Generation Applied to Three Selected Sites in Kurdistan Region, Iraq

Artificial Neural Network Technique for Annual Rainfall Generation Applied to Three Selected Sites in Kurdistan Region, Iraq

Process‐Based Quantification of the Role of Wildfire in Shaping Flood Frequency

AbstractModerate to high severity wildfire can abruptly alter watershed properties and enhance extreme hydrologic responses such as debris flows and floods. The compounding effects of wildfire on flood hazard, represented here via flood frequency analysis (FFA; e.g., 100‐year flood) are of growing importance. Standard statistical FFA approaches are ill‐suited to examining this issue because wildfire‐affected flood peak observations are limited in number and violate the assumption of independent and identically distributed events. Here, we developed a process‐based FFA framework that integrates a stochastic rainfall generator, wildfire simulation, inverse modeling, and a physics‐based hydrological model to directly simulate the impacts of wildfire on FFA. We applied this framework in the upper Arroyo Seco watershed in Southern California, which experienced Moderate to high burn during the 2009 Station Fire. An FFA analysis, performed with simulated peak flows from the first year since fire demonstrates the 100‐year flood can be three times larger than simulations that only consider peak flows in non‐fire‐affected years. On the other hand, coupling process‐based FFA with stochastically simulated wildfire events and watershed's time‐varying hydrologic recovery yields “fire continuum FFA”, a concept introduced here for the first time. Fire continuum FFA accounts for multiple wildfires within very long synthetic time series. Variability in upper tail flood peaks is substantially higher in fire continuum results as compared with pre‐wildfire FFA. This result highlights the importance of wildfire inter‐arrival time and post‐wildfire recovery processes, both of which are expected to change as a result of climatic change and evolving fire management strategies.

Role of updraft in dry‐season torrential rainfall in Greater Jakarta, Indonesia

AbstractCoastal urban areas in tropical regions, such as Greater Jakarta (GJ) in Indonesia, are susceptible to flood hazards from torrential rainfall. Efforts to understand the convective mechanisms leading to this type of rainfall have been carried out extensively, especially for events occurring in the wet season. However, hydrometeorological risks persist even in the dry season, when the rainfall is more closely correlated with the diurnal wind circulation. Among the aspects of convective activities and this diurnal circulation, the details of the updraft remain the least studied and open for discussion. We investigate the role of the updraft in urban torrential rainfall over GJ, using data from a C‐band Doppler radar and reanalysis dataset during the dry season of 2018–2019. Employing back‐trajectory calculations, we locate updraft origins corresponding to the peak of daytime inland rainfall. Most updraft origins are localized at an NW–SE‐oriented front nearly parallel to the coastline on the low‐lying plain, suggesting a region of atmospheric destabilization favorable for convective activity. The dry‐season torrential rainfall over GJ may involve convective activity from this front, that from earlier updrafts originating above the densely populated area, and further enhancement coming from orographic contributions. Our findings suggest the important role of updrafts in the rainfall generation, providing insights into the convective mechanism over GJ.

Evaluation of a Tilt-Based Monitoring System for Rainfall-Induced Landslides: Development and Physical Modelling

Landslides in northern India are a frequently occurring risk during the rainy season resulting in human, animal, and property losses as well as obstructing transportation facilities. Usually, numerical and analytical approaches are applied to predicting and monitoring landslides, but the unpredictable nature of rainfall-induced landslides limits these methods. Sensor-based monitoring is an accurate and reliable method, and it also collects accurate and site-specific required data for further investigation with a numerical and analytical approach. This study developed a low-cost tilt-based rainfall-induced landslide monitoring system using the economical and precise MEMS sensor to record displacement and volumetric water content. A self-developed direct shear-based testing setup was used to check the system’s operational performance. A physical slope model was also prepared to test the monitoring system in real scenarios. A debris failure occurred at Kotrupi village in the Mandi district of Himachal Pradesh, India, which was chosen for the modelling to investigate the failure mechanism. A rainfall generator was developed to simulate the rainfall, equipped with a flow sensor for better simulation and data recording. The tilt angle records the deviation in terms of angle with a least count of 0.01 degrees, and the moisture content was recorded in terms of percentage with a least count of 1. The results show that the developed system is working properly and is very effective in monitoring the rainfall-induced landslide as it monitors the gradual and sudden movement effectively. This study explains the mechanism behind the landslide, and it can be helpful in monitoring the slope to enable the implementation of preventative actions that will mitigate its impact.

Application of gene expression programming for seasonal rainfall forecasting in Western Australia using potential climate indices

AbstractThis study presents the development of rainfall forecast models using potential climate indices for the Kimberley region of Western Australia, using 100 years of rainfall and climate indices data for four rainfall stations. Three different modeling techniques: multiple linear regression (MLR), autoregressive moving average with exogenous input (ARIMAX), and gene-expression programming (GEP) were applied to develop prediction models. Preliminary analysis suggests that Western Tropical Indian Ocean (WTIO) and Southern Oscillation Index (SOI) have significant impacts on summer rainfall generation for the region. Developed models’ performances were evaluated using Pearson correlation coefficient ($$r$$ r ), root mean square error ($$RMSE$$ RMSE ), mean absolute error $$(MAE)$$ ( M A E ) , Nash–Sutcliffe efficiency $$(NSE)$$ ( N S E ) , and refined Willmot index of agreement ($${d}_{r}$$ d r ). It is found that the GEP model exclusively outperforms the other two alternatives. In the calibration period, the GEP model resulted in a Pearson correlation coefficient (r) values ranging from 0.76 to 0.85, which are significantly higher than that achieved from MLR (0.32 to 0.44) and ARIMAX (0.53 to 0.83) models, while for the validation period, the correlation values for the models ranged from 0.74 to 0.87 for GEP, 0.35 to 0.51 for MLR and 0.59 to 0.77 for ARIMAX models. Considering other statistical error statistics it can be concluded that the GEP model is the best representative seasonal rainfall forecasting model for the region.

Influences of Urbanization on an Afternoon Heavy Rainfall Event over the Yangtze River Delta Region

Abstract An afternoon heavy rainfall event occurred over coastal Nantong, an area of 70–100 km downstream from the Shanghai–Suzhou–Wuxi–Changzhou city belt over the Yangtze River Delta, under the influences of weak southwesterly monsoonal flows and lake/sea breezes on 26 July 2018. An observational analysis shows the emergence of pronounced urban heat island (UHI) effects along the city belt during the late morning hours. A series of nested-grid cloud-permitting model simulations with the finest grid spacing of 1 km are performed to examine the impacts of urbanization on convection initiation (CI) and the subsequent heavy rainfall event. Results reveal the generation of lake breezes and warm anomalies in the planetary boundary layer along the city belt and low-level convergence, thereby inducing upward motion for CI. The southwesterly flows of the monsoonal warm–moist air, enhanced by the UHI effects along the city belt, allow the development of convective cells along the city belt, some of which merge with convective clusters during their downstream propagation, and the ultimate generation of several distinct heavy rainfall centers by local convective clusters over coastal Nantong where atmospheric columns are more moist and potentially unstable under the influences of sea breezes. Sensitivity simulations show small contribution of Nantong’s UHI effects to the heavy rainfall event. The above findings help elucidate how the UHI effects could assist the CI in a weak-gradient environment, and explain why urbanization can contribute to increased downwind mean and extreme precipitation under the influences of favorable regional forcing conditions. Significance Statement The urban heat island (UHI) effects tend to produce more rainfall on its downwind side than that on the other sides, but alone could hardly account for the generation of heavy rainfall. This study examines the influences of the UHI effects associated with a city belt over the Yangtze River Delta on generating an afternoon heavy rainfall event over coastal Nantong that is 70–100 km downwind from the city belt. Results show (i) the downwind advection of the UHI effects; (ii) the initiation of convective storms along the city belt, and their subsequent downstream propagation, leading to the generation of heavy rainfall over Nantong; and (iii) an important role of sea breezes in generating the heavy rainfall event.

Hypothetical Ground Radar-Like Rain Rate Generation of Geostationary Weather Satellite Using Data-to-Data Translation

This study proposes a deep-learning-based data-to-data (D2D) translation framework to simulate a radar-like retrieval of rainfall rates using the advantages of spatial coverage and temporal resolution of geostationary (GEO) satellite observation. The D2D method comprises normalization and denormalization in pre- and post-processing and an adversarial learning structure for an inter-domain conversion between physical values of data such as albedo and brightness temperature (BT) unlike the image-to-image translation using digital number values in image data. The GEO-KOMPSAT-2A (GK2A) and radar hybrid surface rainfall (HSR) datasets over the Korean Peninsula from September 2019 to September 2021 were used as the source and target domains for training and testing the D2D model. The constructed D2D model for ground radar-like rainfall generation was validated using the ground radar-observed rainfall data and compared to the GK2A rainfall rate (RR), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Cloud Classification System (PERSIANN-CCS), and Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG) rainfall products. The D2D model exhibited excellent performance for various rain types in the study area compared to the GK2A RR, PERSIANN-CCS, and IMERG data. Consequently, the D2D model can provide valuable and accurate radar-like rainfall intensity and distribution data with a high temporal resolution and complementary rainfall information over lands and oceans without radar observation.

Implications of modeling seasonal differences in the extremal dependence of rainfall maxima

For modeling extreme rainfall, the widely used Brown–Resnick max-stable model extends the concept of the variogram to suit block maxima, allowing the explicit modeling of the extremal dependence shown by the spatial data. This extremal dependence stems from the geometrical characteristics of the observed rainfall, which is associated with different meteorological processes and is usually considered to be constant when designing the model for a study. However, depending on the region, this dependence can change throughout the year, as the prevailing meteorological conditions that drive the rainfall generation process change with the season. Therefore, this study analyzes the impact of the seasonal change in extremal dependence for the modeling of annual block maxima in the Berlin-Brandenburg region. For this study, two seasons were considered as proxies for different dominant meteorological conditions: summer for convective rainfall and winter for frontal/stratiform rainfall. Using maxima from both seasons, we compared the skill of a linear model with spatial covariates (that assumed spatial independence) with the skill of a Brown–Resnick max-stable model. This comparison showed a considerable difference between seasons, with the isotropic Brown–Resnick model showing considerable loss of skill for the winter maxima. We conclude that the assumptions commonly made when using the Brown–Resnick model are appropriate for modeling summer (i.e., convective) events, but further work should be done for modeling other types of precipitation regimes.

Combined basin-scale and decentralized flood risk assessment: a methodological approach for preliminary flood risk assessment

In this study, we focused on a combined basin-scale and decentralized (localized) approach (municipalities) to riverine flood risk mapping and assessment in the Gidra River Basin (Slovakia). The methodology applies direct spatial and probabilistic relation among the physical processes of riverine flooding with a given vulnerability. Flood hazard mapping at basin scale was based on ungauged conditions. First, we calculated the design peak discharges using the rainfall generation model and rainfall-runoff model, in particular, the Stochastic Rainfall Generator (STORAGE) and the Continuous Simulation Model for Small and Ungauged Basins (COSMO4SUB). After that, we used the obtained flood scenarios for Hydrologic Engineering Center´s River Analysis System (HEC-RAS) hydraulic modelling. The modelled flood extent, flow depth, and velocity were considered as hazard indicators. The results were expressed in terms of flood hazard (FHI), vulnerability (FVI), and risk index (FRI): the highest values were obtained for municipalities located in the northern-central part of the investigated area. The proposed methodology can constitute a conceptual basis for updating the Preliminary Flood Risk Assessment (PFRA) in Slovakia.

Statistical characterization of rainfall fields based upon a 12-year high-resolution radar archive of Belgium

Spatial and temporal variability of rainfall input plays a critical role on the performance of the urban runoff simulations. For this reason, urban hydrological application demands accurate products of high-resolution and high-quality rainfall estimates, which use and understanding become critical for more robust models.In this study, the spatial and temporal characteristics of rainfall in Belgium are analyzed at fine scales. This is done by constructing a conceptual rain storm model using historical long-term radar data archive. This application involved the use of a state-of-the-art tracking algorithm, developed by Muñoz et al. (2018), to associate the spatial and temporal variations in rain storms and cells This algorithm enables better handling of high-resolution details and is capable of extracting rainfall objects/entities across various scales. Through tracking and conceptualizing these storm entities, spatial and temporal rainfall properties, such as geometric features, intensities, velocity, rainfall types or spatial patterns, can be extracted and analyzed. To accomplish this, twelve years of continuous high-resolution radar data are utilized in this work. The resulting analysis of the spatial and temporal characteristics of rainfall radar data is intended to set the ground for the future applications. More specifically, it can be used to refine and improve the existing long-term spatial rainfall generators, such as that the one implemented by Willems' in 2001.

Effect of indigenous and scientific forecasts on pastoralists’ climate change perceptions in the Rwenzori region, Western Uganda

ABSTRACT This study investigated the effect of climate information on the climate change perceptions of pastoralists in the Rwenzori region, Western Uganda. Data on socio-economic characteristics, climate risk perceptions, use of indigenous knowledge-based forecasts (IFs) and scientific knowledge-based forecasts (SFs) were collected from 269 pastoralists using household surveys. The findings indicated that pastoralists using IFs only were more likely to perceive flood increase. Pastoralists using both IF and SF were more likely to perceive temperature increase, flood increase and drought increase. Apart from climate forecasts, other factors of significance included gender, education level, social capital, herd size, farm size, access to hired labour, herd mobility, agricultural extension access, access to government programmes on climate change and dissemination channels of climate change information. Pastoralists tend to put more attention on climate risks that are less likely to happen and less attention on events that are more likely to happen. Cognitive factors like education level can overweight small probabilities and underweight large probabilities of climate risk perceptions. Stakeholder engagement in rain-gauge management for on-farm rainfall generation and co-production of climate services can enhance climate change perceptions of pastoralists resulting in a reduction in vulnerability to climate change. Increasing the rain-gauge network in pastoral areas could promote more scientific perceptions of climate change.

Comparison of rainfall generators with regionalisation for the estimation of rainfall erosivity at ungauged sites

Abstract. Rainfall erosivity values are required for soil erosion prediction. To calculate the mean annual rainfall erosivity (R), long-term high-resolution observed rainfall data are required, which are often not available. To overcome the issue of limited data availability in space and time, four methods were employed and evaluated: direct regionalisation of R, regionalisation of 5 min rainfall, disaggregation of daily rainfall into 5 min time steps, and a regionalised stochastic rainfall model. The impact of station density is considered for each of the methods. The study is carried out using 159 recording and 150 non-recording (daily) rainfall stations in and around the federal state of Lower Saxony, Germany. In addition, the minimum record length necessary to adequately estimate R was investigated. Results show that the direct regionalisation of mean annual erosivity is best in terms of both relative bias and relative root mean square error (RMSE), followed by the regionalisation of the 5 min rainfall data, which yields better results than the rainfall generation models, namely an alternating renewal model (ARM) and a multiplicative cascade model. However, a key advantage of using regionalised rainfall models is the ability to generate time series that can be used for the estimation of the erosive event characteristics. This is not possible if regionalising only R. Using the stochastic ARM, it was assessed that more than 60 years of data are needed in most cases to reach a stable estimate of annual rainfall erosivity. Moreover, the temporal resolution of measuring devices was found to have a significant effect on R, with coarser temporal resolution leading to a higher relative bias.

Synthetic rainfall data generator development through decentralised model training

Recent heavy rainfall-induced flood events, for example in Germany, Australia and USA, have highlighted the relevance of countermeasures in saving human lives and preventing property damage. Newly introduced ML-based flood forecasting methods rely on high-intensity synthetic rainfall events due to the sparsity of their real counterpart. Such synthetic data instances can be produced by precipitation generators trained in an adversarial setting on historical rainfall data. Capturing processes for rainfall data are often highly distributed, with multiple radar stations contributing to a centralised data set. However, data centralisation entails challenges regarding data-stream logistics, data locality, and memory overhead. Distributed Analytics (DA) aims to overcome these challenges through decentralised model training by bringing the algorithm to the data instead of vice versa. In this work, we propose a feasibility study evaluating the applicability of DA on hydrological data. As example of use, we choose the decentralised training of rainfall data generators. We introduce a rainfall generator training procedure relying on Generative Adversarial Networks (GANs) and evaluate two DA algorithms: Federated Learning (FL) and Cyclic Institutional Incremental Learning (CIIL). We compare the resulting training outcomes with the centralised model training (CL) approach and find CIIL performed similarly to CL but less stable, while FL outperformed CL by 7.5%. We conclude that the proven feasibility of FL in our simulated distributed setting lays the groundwork for utilising this approach in realistic environments of grander scale while overcoming potential privacy concerns or logistical challenges in the setting of centralised analytics.