Long-term Flood Research Articles

Enhancing Long-Term Flood Forecasting with SageFormer: A Cascaded Dimensionality Reduction Approach Based on Satellite-Derived Data

Floods, increasingly exacerbated by climate change, are among the most destructive natural disasters globally, necessitating advancements in long-term forecasting to improve risk management. Traditional models struggle with the complex dependencies of hydroclimatic variables and environmental conditions, thus limiting their reliability. This study introduces a novel framework for enhancing flood forecasting accuracy by integrating geo-spatiotemporal analyses, cascading dimensionality reduction, and SageFormer-based multi-step-ahead predictions. The framework efficiently processes satellite-derived data, addressing the curse of dimensionality and focusing on critical long-range spatiotemporal dependencies. SageFormer captures inter- and intra-dependencies within a compressed feature space, making it particularly effective for long-term forecasting. Performance evaluations against LSTM, Transformer, and Informer across three data fusion scenarios reveal substantial improvements in forecasting accuracy, especially in data-scarce basins. The integration of hydroclimate data with attention-based networks and dimensionality reduction demonstrates significant advancements over traditional approaches. The proposed framework combines cascading dimensionality reduction with advanced deep learning, enhancing both interpretability and precision in capturing complex dependencies. By offering a straightforward and reliable approach, this study advances remote sensing applications in hydrological modeling, providing a robust tool for mitigating the impacts of hydroclimatic extremes.

Decision-making under flood predictions: A risk perception study of coastal real estate.

Flood models, while representing our best knowledge of a natural phenomenon, are continually evolving. Their predictions, albeit undeniably important for flood risk management, contain considerable uncertainties related to model structure, parameterization, and input data. With multiple sources of flood predictions becoming increasingly available through online flood maps, the uncertainties in these predictions present considerable risks related to property devaluation. Such risks stem from real estate decisions, measured by location preferences and willingness-to-pay to buy and rent properties, based on access to various sources of flood predictions. Here, we evaluate the influence of coastal flood predictions on real estate decision-making in the United Kingdom by adopting an interdisciplinary approach, involving flood modeling, novel experimental willingness-to-pay real estate surveys of UK residents in response to flood predictions, statistical modeling, and geospatial analysis. Our main findings show that access to multiple sources of flood predictions dominates real estate decisions relative to preferences for location aesthetics, reflecting a shift in demand toward risk averse locations. We also find that people do not consider flood prediction uncertainty in their real estate decisions, possibly due to an inability to perceive such uncertainty. These results are robust under a repeated experimental survey using an open access long-term flood risk map. We, therefore, recommend getting flood models "right" but recognize that this is a contentious issue because it implies having an error-free model, which is practically impossible. Hence, to reduce real estate risks, we advocate for a greater emphasis on effectively communicating flood model predictions and their uncertainties to non-experts.

Optimizing nature-based solutions for urban flood risk mitigation: A multi-objective genetic algorithm approach in Gdańsk, Poland.

Nature-based Solutions (NbS) have emerged as a sustainable approach to managing flood risks by enhancing natural water retention and reducing surface runoff in urban areas. As climate change and rapid urbanization exacerbate flood hazards, optimizing the spatial deployment of NbS is crucial for improving urban resilience and mitigating flood impacts. This study presents a comprehensive optimization framework for the spatial allocation of fourteen different NbS types aimed at mitigating urban flood risks in Gdańsk, Poland. Leveraging a genetic algorithm alongside the Urban Flood Risk Mitigation (UFRM) model of the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) software suite, we identified areas of the city most vulnerable to pluvial flooding and optimized NbS placement to maximize water retention and reduce runoff. The optimization process balanced multiple objectives, including minimizing implementation costs and maximizing water retention capacity, ensuring that the solutions are both economically feasible and environmentally effective. Three distinct scenarios were proposed: a cost-effective solution, a high-retention solution, and a balanced solution, offering urban planners a range of strategies to address flood risks based on their specific priorities and constraints. Results demonstrated considerable variations in water retention effectiveness across different NbS configurations, with denser urban regions showing the most significant improvements from targeted interventions. The optimized placement of NbS resulted in estimated total water retention improvement of approximately 15.5% for the best solution considered. These findings provide valuable insights for integrating NbS into urban flood management strategies, enhancing citywide resilience, sustainability, and long-term flood mitigation.

Effects of War-Related Human Activities on Microalgae and Macrophytes in Freshwater Ecosystems: A Case Study of the Irpin River Basin, Ukraine

Throughout the world, river basins are directly or indirectly affected by human activities, reducing local and global biodiversity and preventing the ecosystem from properly functioning. Our research focused on the Irpin River basin (Ukraine), whose water bodies have experienced various impacts due to human activities, including the unexpected extremes caused by military operations in the catchment area: long-term flooding, disturbance of free flow, significant water level fluctuations, etc. The study hypothesized that the primary factors determining the structural and spatial distribution of quantitative indicators of microalgae and aquatic macrophytes are the result of various hydromorphological changes, that lead to changes in the physical and chemical parameters of the aquatic environment. Very high values of chlorophyll a in the water column (59–106 µg · L−1), an increase in the abundance (number of cells) and biomass of algae (due to the predominance of certain groups in the transformed sections), as well as saprobic index were recorded in the sections of the Irpin River basin that underwent significant hydromorphological changes. Our results revealed a strong correlation between phytoplanktonic (in the water column) chlorophyll a levels and water temperature (r = 0.76, p < 0.001), as well as organic phosphorus and polyphosphate concentrations (r = 0.61, p < 0.01). ANOVA and Monte Carlo permutation tests in a Canonical Correspondence Analysis (CCA) showed that the abundance of different divisions of phytoplankton and phytobenthos were significantly and similarly related to several environmental variables. We observed a positive correlation between the number of cyanobacteria and the concentration of ammonium nitrogen, nitrites, and phosphorus compounds. An increase in dissolved organic matter in the water can explain the increase in the biomass of Dinoflagellata and Euglenophyta. Species richness and the cover values of the macrophytes also clearly reflected changes in vegetation activity in sections of the Irpin River caused by hydromorphological changes. The results indicated that long-term flooding had the most negative impact on macrophyte communities. At some sites, the impact was so severe that the number of macrophyte species was very low. The total number of macrophyte species showed a significant negative correlation with total suspended solids (r = −0.51, p < 0.05) and phytoplankton chlorophyll a concentration (r = −0.73, p < 0.001). Our results provide a scientific basis for predicting changes in riverine microalgal and aquatic macrophyte communities due to extreme hydrological events.

Improving Long-Term Flood Forecasting Accuracy Using Ensemble Deep Learning Models and an Attention Mechanism

Improving Long-Term Flood Forecasting Accuracy Using Ensemble Deep Learning Models and an Attention Mechanism

Assessment of socio-economic strategies for managing regional flood risk in an urban coastal catchment

With the cascading flooding events over the past decades, several structural and non-structural measures have been effectively implemented for improved flood management. However, the uneven impacts of the flooding events have highlighted that socio-economic inequality amongst the community needs to be addressed along with hazards to devise long term and sustainable flood management strategies. This study aims to evaluate the impact of socio-economic adaptive scenarios over an urban coastal catchment for reduction in flood risk. Flood risk is estimated utilizing socio-economic aspects (vulnerability) and flood inundation (hazard), for the highly flood-prone coastal city of Mumbai, India. A multivariate approach is used for assessing socio-economic vulnerability, starting with Principal Component Analysis followed by Data Envelopment Analysis, for various socio-economic scenarios, i.e., by reducing percentage of certain feasible socio-economic sensitivity indicators, to identify the key indicators, which contribute to significant change in vulnerability. The flood hazard is assessed through a comprehensive hydrodynamic flood modelling approach to quantify the flood inundation depth and extent. The flood risk is, thereafter, evaluated by combining socio-economic vulnerability and flood hazard using bivariate risk mapping approach. The results indicate that, with the decrease in alteration percentage of certain feasible socio-economic sensitivity indicators, the overall vulnerability and flood risk may be reduced. The areas under very high and high vulnerability classes reduce gradually for all scenarios as the percentage reduction for various indicators is varied from 5 % to 15 %, total illiteracy being the most significant indicator. This study marks the first effort to discern socio-economic vulnerability indicators that have the potential to significantly diminish flood risk, thereby constituting a viable long-term flood risk mitigation strategy. This novel framework is generic and applicable to any flood prone catchments, to devise appropriate long-term flood management measures.

Cognitive limits of perceived flood risk on residential property values

Examining cognitive limits in flood risk perception for residential property values, we analyse the Richmond housing market in the state of New South Wales, Australia. Using micro-level home sales data, our study reveals that the market has integrated long-term flood risk into property values. A notable 10.8 % price discount is observed for properties within 1–100 Annual Exceedance Probability (AEP) flood zone, 4.4 % for those in an AEP 500 zone, with no discounts for AEP 1000 flood zone properties. Comparisons of 2019 and 2023 flood maps and property's Time-on-Market (TOM) affirm that people's cognitive limits constrain to the AEP 500 level.

Comparison of Nitrous Oxide Consumption of Paddy Soils Developed from Three Parent Materials in Subtropical China

Paddy soils developed from various parent materials are widely distributed in the subtropical region in China and have a non-negligible but unclear potential to consume nitrous oxide (N2O) due to long-term flooding. This study selected three of the most common paddy soils in subtropical China, developing from quaternary red soil (R), lake sediment sand (S), and alluvial soil (C), to study their total N2O consumption and total nitrogen (N2) production using N2-free microcosm experiments. These paddy soils were treated with N2O addition (N2O treatment) or helium (He) addition (CK treatment) and incubated under flooding and anoxic conditions. The results showed that three alluvial soils (C1, C2, and C3) consumed over 99.93% of the N2O accumulated in the soil profile, significantly higher than R and S soils (p < 0.05). And the N2 production in three C soils was also significantly higher than other soils, accounting for 81.61% of the total N2O consumption. The main soil factors affecting N2O consumption in C, S, and R soils were soil clay content (p < 0.05), soil sand content (R2 = 0.95, p < 0.001), and soil available potassium (AK) (p < 0.01), respectively. These results indicate flooding paddy soils, no matter the parent materials developed, could consume extremely large amount of N2O produced in soil profiles.

Modeling of Flood Wave Propagation Due to Hemren Dam Failure

The occurrence of dam failures and the consequent floods pose substantial risks to submerged regions, and their unpredictability increases the hazard of their impact. This research simulates the flood event scenario due to the dam failure of the Hemrem dam in Diyala Governorate, Iraq. HEC-RAC model jas been used to modeling the Hemren dam failure. The generated flood waves are routed downstream until the confluence of the Diyala and Tigris rivers, covering a large study area of 5,378.5 km². The configuration of the model's geometry is formulated using the Digital Elevation Model (DEM) that encompasses the geographical extent of the Diyala River study area, from Hemren Dam to the confluence with the Tigris River south of Baghdad. The length of this reach is approximately 210 km. Based on the inundation maps, the region's geography is classified as having catastrophic constraints due to the water depth and flow rate. The results demonstrate how to predict the magnitude of floods and highlight the severity if Hemren Dam were to fail. This underscores the need for effective risk management. Furthermore, the maps created by this study can be utilized to develop long-term flood control plans.

Avoiding future surprises after acute shocks: long-term flood risk lessons catalysed by the 2021 summer flood in the Netherlands

In summer 2021, the Dutch Province of Limburg faced an acute high-water event caused by two days of heavy rainfall, resulting in €383 million worth of damage. Afterwards, various policy actors organised a number of evaluations. This study analyses these evaluations and the long-term lessons drawn from this flood event. Dealing with climate change requires insights into how lessons from acute shocks such as floods can stimulate the anticipation of climate change and solutions that can cope with future extremes. This study therefore applies policy learning theory—single, double and triple loop learning—to different layers of flood safety to analyse the extent to which governments prepare for future events and how.The results show that pre-2021 water and flood risk management policies—our point of comparison to derive what lessons emerge after the 2021 flood—focused mostly on flood prevention (layer 1) and from 1996 onwards with the start of the Room for the River programme also on spatial planning (layer 2), without paying much attention to crisis management (layer 3). Based on the 2021 flood, the official evaluation committee recommends adding two more layers to the multilayer flood safety approach: increasing water awareness (layer 0 as a base layer for the other layers) and focusing on climate-robust recovery after floods (layer 4) to improve long-term robustness. Based on our research, triple-loop learning as the deepest type of learning appeared in the form of applying the Room for the River approach to the tributary brooks systems. To prepare for the future, policy evaluations recommend improving stress tests to better deal with climate change. Futures scenarios and visions for the future are also recommended. Creating future visions can help to steer changes in existing land use and re-think areas to build or to use for water retention, for example. Governments are advised to strengthen their anticipatory capacities as part of their crisis management response and recovery systems. Furthermore, governments will need to apply the different layers for multilayer safety in an integrated way to ensure that regions are able to prepare for future floods.

Assessment of flood risk in a coastal city considering multiple socio-economic vulnerability scenarios

Abstract. The Intergovernmental Panel on Climate Change stresses the importance of vulnerability and exposure along with hazard in defining flood risk. Therefore, a novel approach for flood risk quantification is proposed that evaluates the impact of various realistic socio-economic scenarios on risk reduction. The flood hazard is derived from a hydrodynamic flood modelling framework over Mumbai, India's highly flood-prone coastal megacity. The socio-economic vulnerability is assessed by a multivariate approach based on principal component analysis and data envelopment analysis. Finally, the flood risk is quantified and mapped by aggregating hazard and vulnerability for different socio-economic scenarios, also key indicators contributing to a significant reduction in vulnerability and risk are identified. This non-structural long-term flood risk management approach will benefit densely populated urban areas, especially in developing, and underdeveloped countries.

Long-Term Flooding Mitigates the Linkage between Paddy Soil Organic Carbon Sink and Climate Factors

Long-Term Flooding Mitigates the Linkage between Paddy Soil Organic Carbon Sink and Climate Factors

Dissecting lncRNA-mRNA networks underlying physiological alterations in Taxodium hybrid ‘Zhongshanshan’ leaves in acclimation to prolonged partial submergence

Long non-coding RNAs (lncRNAs) play essential roles in plant adaptation to various abiotic stresses. However, lncRNAs-mRNA networks underpinning physiological alterations in trees responding to long-term flooding remain unknown. Here, cuttings of Taxodium hybrid ‘Zhongshanshan 703’ (T. mucronatum × T. distichum; T. ‘Zhongshanshan’), which is highly tolerant to flooding, were irrigated to keep the field capacity (C) or partial-submerged (S) for two months. The leaves above (AL) and below (BL) flooding-water surface were separately harvested. In total, 294 and 975 lncRNAs (DELs) as well as 1503 and 6170 mRNAs (DEMs) were significantly differentially expressed in AL and BL of T. ‘Zhongshanshan’ treated with S than those exposed to C, respectively. 493 DELs were predicted to target 517 DEMs through co-location way, and 906 DEMs were identified as putative target genes of 314 DELs through co-expression way. Furthermore, co-expression networks of DELs and DEMs that are closely correlated with the changes in concentrations of organic acids, sugars, amino acids and phytohormones and activities of anaerobic respiratory enzymes and enzymatic antioxidants were established. For instance, the transcript levels of an lncRNA (TCONS_00030503) and its target gene ADH1 which is involved in alcoholic fermentation were elevated in the BL of T. ‘Zhongshanshan’ exposed to S. This is in line with increased ADH activity in BL due to flooding. Moreover, eight hub DELs and five hub DEMs that are related to photosynthesis, starch metabolism, glycolysis, amino acid metabolism and hormone metabolism were identified. Overall, these results suggest that lncRNA-mRNA networks play significant roles in regulating TCA cycle, fermentation, glycolysis, reactive oxygen species detoxification, and metabolism of carbohydrate, alanine and phytohormone in T. ‘Zhongshanshan’ leaves, and these processes play pivotal roles in acclimation to long-term submergence.

Using Copula functions to predict climatic change impacts on floods in river source regions

Flood frequency in river source regions is significantly affected by rainfall and snowmelt as part of climatic changes. A traditional univariate flood frequency analysis cannot reflect the complexity of floods, and when used in isolation, it can only underestimate flood risk. For effective flood prevention and mitigation, it is essential to consider the combined effects of precipitation and snowmelt. Copula functions can effectively quantify the joint distribution relationship between floods and their associated variables without restrictions on their distribution characteristics. This study uses copula functions to consider a multivariate probability distribution model of flood peak flow (Q) with cumulative snowmelt (CSm) and cumulative precipitation (CPr) for the Hutubi River basin located in northern Xinjiang, China. The joint frequencies of rainfall and snowmelt floods are predicted using copula models based on the Coupled Model Intercomparison Project Phase 6 data. The results show that Q has a significant positive correlation with 24-d CSm (r = 0.559, p = 0.002) and 23-d CPr (r = 0.965, p < 0.05). Flood frequency will increase in the future, and mid- (2050–2074) and long-term (2075–2099) floods will be more severe than those in the near-term (2025–2049). The probability of flood occurrence is higher under the SSP2-4.5 and SSP1-2.6 scenarios than under SSP5-8.5. Precipitation during the historical period (1990–2014) led to extreme floods, and increasing future precipitation trends are found to be insignificant. Snowmelt increases with rising temperatures and occurs earlier than estimated, leading to an earlier flood period in the basin and more frequent snowmelt floods. The Q under the joint return period is larger than that during the same univariate return period. This difference indicates that neglecting the interaction between precipitation and snowmelt for floods leads to an underestimation of the flood risk (with underestimations ranging from 0.3% to 22%). The underestimations decrease with an increase in the return period. The joint risks of rainfall or snowmelt according to various flood periods should be considered for rivers with multi-source runoff recharge in flood control design. This study reveals the joint impact of precipitation and snowmelt on extreme floods under climate change in river source regions. This study also provides a scientific basis for regional flood prevention and mitigation strategies, as well as for the rational allocation of water resources.

Integrated small RNA, transcriptome and physiological approaches provide insight into Taxodium hybrid 'Zhongshanshan' roots in acclimation to prolonged flooding.

Although Taxodium hybrid 'Zhongshanshan' 406 (Taxodium mucronatum Tenore×Taxodium distichum; Taxodium 406) is an extremely flooding-tolerant woody plant, the physiological and molecular mechanisms underlying acclimation of its roots to long-term flooding remain largely unknown. Thus, we exposed saplings of Taxodium 406 to either non-flooding (control) or flooding for 2months. Flooding resulted in reduced root biomass, which is in line with lower concentrations of citrate, α-ketoglutaric acid, fumaric acid, malic acid and adenosine triphosphate (ATP) in Taxodium 406 roots. Flooding led to elevated activities of pyruvate decarboxylase, alcohol dehydrogenase and lactate dehydrogenase, which is consistent with higher lactate concentration in the roots of Taxodium 406. Flooding brought about stimulated activities of superoxide dismutase and catalase and elevated reduced glutathione (GSH) concentration and GSH/oxidized glutathione, which is in agreement with reduced concentrations of O2- and H2O2 in Taxodium 406 roots. The levels of starch, soluble protein, indole-3-acetic acid, gibberellin A4 and jasmonate were decreased, whereas the concentrations of glucose, total non-structural carbohydrates, most amino acids and 1-aminocyclopropane-1-carboxylate (ACC) were improved in the roots of flooding-treated Taxodium 406. Underlying these changes in growth and physiological characteristics, 12,420 mRNAs and 42 miRNAs were significantly differentially expressed, and 886 miRNA-mRNA pairs were identified in the roots of flooding-exposed Taxodium 406. For instance, 1-aminocyclopropane-1-carboxylate synthase 8 (ACS8) was a target of Th-miR162-3p and 1-aminocyclopropane-1-carboxylate oxidase 4 (ACO4) was a target of Th-miR166i, and the downregulation of Th-miR162-3p and Th-miR166i results in the upregulation of ACS8 and ACO4, probably bringing about higher ACC content in flooding-treated roots. Overall, these results indicate that differentially expressed mRNA and miRNAs are involved in regulating tricarboxylic acid cycle, ATP production, fermentation, and metabolism of carbohydrates, amino acids and phytohormones, as well as reactive oxygen species detoxification of Taxodium 406 roots. These processes play pivotal roles in acclimation to flooding stress. These results will improve our understanding of the molecular and physiological bases underlying woody plant flooding acclimation and provide valuable insights into breeding-flooding tolerant trees.

Reducing the Susceptibility of Flood Supplier Areas in Gondang Micro-Catchment, Bojonegoro, East Java Province: a Flood Disaster Mitigation Effort

Flood disaster mitigation begins with reducing the susceptibility of flood supplier areas. This task can be accomplished by mapping the flood-supplier susceptibility and then deciding on long-term flood mitigation actions following the susceptibility level of the land and the legal status of its area. Pacal reservoir has Gondang as one of its catchment areas. During the rainy season, the spillway releases water immediately because the inflow exceeds capacity. To maximize water storage, reduce Pacal reservoir input, and prevent downstream flooding, the land cover must be improved. This paper will investigate the Gondang micro-flood-prone catchment’s locations and propose mitigation strategies. The legal status of the land is used, as well as Climate Hazards Group Infrared Precipitation with Station data (CHIRPS), Digital Elevation Model (DEM), and Google Imagery from 2014 and 2020. Paimin’s Equation was used to assess the flood risk. The findings indicate that the Gondang micro-catchment is susceptible to flood suppliers on a low to a high level. Changes in land cover during 2014-2020 resulted in alterations to the susceptibility level of the area. The legal status of the area has had an impact on land cover management. The simulation results of reforestation and maintenance of dense forests in forest areas, as well as regreening activities in agricultural areas, can improve land with high susceptibility to moderate levels, which accounts for 20.9% of micro-catchment areas. All of these proposed activities are expected to control the Pacal Reservoir’s inflow during the rainy season by increasing soil infiltration and reducing runoff.

Flooding Length Mediates Fencing and Grazing Effects on Soil Respiration in Meadow Steppe.

Grassland management affects soil respiration (Rs, consists of heterotrophic respiration and autotrophic respiration) through soil micro-ecological processes, such as hydrothermal, plant root, organic carbon decomposition and microbial activity. Flooding, an irregular phenomenon in grasslands, may strongly regulate the response of soil respiration and its components to grassland management, but the regulatory mechanism remains unclear. We conducted a 3-year experiment by grassland management (fencing and grazing) and flooding conditions (no flooding (NF), short-term flooding (STF) and long-term flooding (LTF)) to study their effects on Rs and its components in a meadow steppe in the Hui River basin of Hulunbuir. We found differences in the patterns of Rs and its components under grassland management and flooding conditions. In 2021-2023, the temporal trends of Rs, heterotrophic respiration (Rh) and autotrophic respiration (Ra) were generally consistent, with peaks occurring on days 190-220, and the peaks of grazing were higher than that of fencing. In NF, Rs of grazed grassland was significantly higher than that of fenced grassland in 2021-2022 (p < 0.05). In STF and LTF, there was no significant difference in Rs between fenced and grazed grassland (p > 0.05). The dependence of Rs on soil temperature (ST) decreased with increasing flooding duration, and the dependence of Rs on ST of grazed grassland was higher than fenced grassland under NF and STF, but there was no difference between fenced grassland and grazed grassland under LTF. In addition, Rh was more sensitive to ST than Ra. This may be due to the different pathways of ST effects on Rs under grazing in different flooding conditions. Our study indicates that the effect of flooding on Rs is the key to the rational use of grassland under future climate change. To reduce regional carbon emissions, we recommend grazing on flooding grassland and fencing on no-flooding grassland.

An Internet of Things and AI-Powered Framework for Long-Term Flood Risk Evaluation

An Internet of Things and AI-Powered Framework for Long-Term Flood Risk Evaluation

Modelling of Riverine Flooding Using HEC-RAS-A Case Study of Badri Khawar River in KPK, Pakistan

Floods are one of the most occurring disasters in the world. Pakistan gets affected by flooding due to its diverse geography and has to lose several lives of people every year. The main objective of this study was to carry out modeling of riverine flooding using HEC-RAS program to estimate the safe channel capacity of the project reach and identify the vulnerable areas along the project reach for different flooding scenarios. A 9.21 Km long reach of Badri Khawar river, a right bank tributary of Indus River in KPK province has been taken as a case study. Based on the available data, one dimensional steady flow analysis in HEC-RAS has been carried out for different flooding scenarios. By analyzing the simulated results, the safe channel capacity of the study reach was found to be in the range of 125m3/s to 100m3/s from upstream to downstream. Based on the extent of flooding in different scenarios, vulnerable areas have been delineated on a flood map devised by the help of Google Earth. There is more inundation at the left overbank as compared to the right overbank. For the worst-case scenario, the maximum estimated flood extension on left bank side was about 425m while on the right bank side it was 204m. Based on the results of this research, it is suggested to enhance the existing flood protection measures along the project reach and improve the flood warning system and emergency evacuation mechanism in the vulnerable areas for long-term flood safety.

A comparative study of Machine Learning and Deep Learning methods for flood forecasting in the Far-North region, Cameroon

Flood crises are the consequence of climate change and global warming, which lead to an increase in the frequency and intensity of heavy rainfall. Floods are, and remain, natural disasters that result in huge loss of lives and material damage. Flood risks threaten all countries of the globe in general. The Far-North region of Cameroon has suffered of flood crises on several occasions, resulting in significant loss of human lives, infrastructural and socio-economic damage, with the destruction of homes, crops and grazing areas, and the halting of economic activities. The models used for flood forecasting in this region are generally physical-based, and produce unsatisfactory results. The use of artificial intelligence based methods for flood forecasting in order to limit its consequences is a way to be explored in the Far-North region of Cameroon. The aims of the present research work is to design and compare the performance of Machine Learning and Deep Learning based models such as one dimensional Convolutional Neural Network, Long and Short Term Memory and Multi Layer Perceptron for short-term and long-term flood forecasting in the Far-North region of Cameroon. The models designed take as input the temperature and rainfall time series recorded in this region. Performance criteria used for evaluating models are Nash–Sutcliffe Efficiency, Percent Bias, Coefficient of Determination and Root Mean Squared Error. As the results of the design and performance comparison of the models, the best model for short-term flood forecasting is the LSTM model , and the best model for long-term flood forecasting is still the LSTM model. The best models obtained from the comparisons have satisfactory performance and good generalization capabilities, as reflected by the performance criteria. The results of our research work can be used for implementation of floods warning systems and for definition of an effective and efficient flood risk management policies in order to make the Far-North region of Cameroon more resilient to flood crises.