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

Accuracy of flood forecast is important to take appropriate preparedness measures for saving lives and livelihoods of people residing in the floodplains. Predictions from flood forecasting models are usually uncertain which can be improved by complementing the hydrological model with an error model that can capture the information which the operational hydrological model lacks. This paper presents the application of this approach for improving daily flow forecasts for flood warning in Karnali River Basin of Nepal. A conceptual rainfall-runoff model, TUWmodel has been developed to model the rainfall-runoff processes and to predict the runoff at the outlet of the basin at Chisapani. The model has been calibrated for the period 2008- 2011 with Nash-Sutcliffe Efficiency (NSE) 0.91 and percent bias (PBIAS) -0.7% and validated for the period 2012-2014 with NSE 0.88 and PBIAS -9.1% using observed temperature, precipitation and discharge data. A complementary ARIMA error model was developed from the error series for calibration set using automatic procedure and the predicted discharges were corrected using the error predictions from the error model. After error corrections, NSE and PBIAS were 0.95 and 0.1% respectively for calibration and 0.92 and 0.1% respectively for validation indicating significant improvements in the skill of forecasts.

This study aims to contribute to the conservation and sustainable management of the Logone-Birni floodplains in Cameroon. In total, 27 plots of 100 m x 100 m plots of were laid out in 3 land use types (protected area, grazing area and agricultural area). Within each plot, all woody individuals having ≥ 10 cm circumference at breast height (> 1.3 m) were identified, and all the stems with a circumference ≤ 10 cm were considered as seedlings. A total of 43 timber species belonging to 35 genera and 21 families were harvested. Shannon-Weaver diversity ranged from 1.77-2.39 and the Pielou equitability index from 0.84-0.88 with low values in agricultural areas and higher values in protected and grazing areas. The distribution of stems according to diameter and height showed that all stands exhibit classic exponential decay distribution ‘‘L’’ or ‘‘J’’ inverted, reflecting the predominance of individuals with small diameters. The regeneration rates of the stand vary from 0.81 to 51.56%, indicating a difficult regeneration due partly to anthropogenic activities. It also appears that given the low diversity and the difficult regeneration of the stand, it is important, even crucial, to manage this plain to ensure the conservation and sustainable management of the biological resources that it hosts.

Flood forecasting models are essential tools for mitigating the impacts of extreme hydrological events by providing early warnings and actionable insights. This study evaluates the performance of the Today's Earth (TE) model, JAXA's land surface simulation system developed under joint research with the University of Tokyo, by comparing its predictions against observed water level data from the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) measuring stations in Japan. The study also compares the results to Google’s SOTA hydrological model using the Google Runoff Reanalysis & Reforecast (GRRR) dataset. The GRRR model features a 7-day (168-hour) forecast window, while Today's Earth offers a shorter forecast window of 39 hours. The analysis focuses on major flood events in Japan (2020–2024), including typhoons and heavy precipitation events, to examine trends and accuracy in flood predictions over different lead times. This evaluation identifies strengths and areas for improvement in operational forecasting across diverse hydrological scenarios.The methodology integrates a comprehensive dataset of water level observations and forecast outputs, with an emphasis on lead time-dependent peak timing and magnitude error. Forecasts were evaluated based on their ability to capture observed flood peaks, with errors in both peak magnitude and timing quantified for varying lead times. Performance metrics such as Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), and Pearson correlation were calculated for each location and lead time.Results reveal that forecast accuracy varies significantly with lead time, playing a critical role in the effectiveness of early warning systems. While Google's hydrological model performs better under normal flow conditions, Today's Earth model performs considerably better in both peak magnitude and timing during flood events. The GRRR dataset consistently underestimated peak magnitudes, highlighting the difficulty of forecasting extremes. Another key finding is the marked improvement in forecasting accuracy for the Today’s Earth model between 2021 and 2022, which coincides with the incorporation of observed precipitation data. Initial results indicate a significant enhancement in peak flow timing predictions following this update. This study evaluates how this modification improved forecasting results, emphasising the potential to refine TE’s algorithms and integrate additional observational data.This research provides actionable insights into flood prediction reliability and demonstrates the value of leveraging Japan’s extensive network of water level gauges. Findings contribute to ongoing efforts to enhance flood forecasting systems globally and highlight the importance of targeted evaluations for improving model performance. The study's implications extend to disaster risk management, operational forecasting practices, and the broader pursuit of climate-resilient water management strategies.

In the Far-North Region of Cameroon, the high rate of weathering and erosion of rocks do not allow to map the hydrological and geological structures in the field. This paper aims to test the combined use of field topographic data and Shuttler Radar Topography Mission (SRTM) data to define watershed morphometry, highlight the different erosion factors, and characterize the structural lineaments of the Mayo-Kani Division, within the Far-North Region of Cameroon. All these are done using; watersheds, hydrographic network, slopes, manual lineaments and automatic lineaments maps. These treatment techniques used (filtering, enhancement and slope analyses) allows: to morphologically characterize the ten identified Mayo-Kani watersheds, to measure the erodibility rate of the watersheds using the Universal Soil Loss Equation (USLE) and to compute the fracture maps. This last operation was validated using ground truth field observations and the literature review. Fracture networks were analyzed using statistical analysis techniques by studying fracture length distribution laws.

Accurate rainfall-runoff modeling is crucial for disaster prevention, mitigation, and water resource management. This study aims to enhance precision and reliability in predicting runoff patterns by integrating physical-based models like HEC-HMS with data-driven models, such as LSTM. We present a novel hybrid model, Ia-LSTM, which combines the strengths of HEC-HMS and LSTM to improve hydrological modeling. By optimizing the “initial loss” (Ia) with HEC-HMS and utilizing LSTM to capture the effective rainfall-runoff relationship, the model achieves a substantial improvement in precision. Tested in the Yufuhe basin in Jinan City, Shandong province, the Ia-LSTM consistently outperforms individual HEC-HMS and LSTM models, achieving notable average Nash-Sutcliffe Efficiency (NSE) values of 0.873 and 0.829, and average R2 values of 0.916 and 0.870 for calibration and validation, respectively. The study shows the potential of integrating physical mechanisms to enhance the efficiency of data-driven rainfall-runoff modeling. The Ia-LSTM model holds promise for more accurate runoff estimation, with wide applications in flood forecasting, water resource management, and infrastructure planning.

A real-time probabilistic channel flood-forecasting model based on the Bayesian particle filter approach

An approach for improving the capability of a coupled meteorological and hydrological model for rainfall and flood forecasts

The accuracy and sufficiency of precipitation data play a key role in environmental research and hydrological models. They have a significant effect on the simulation results of hydrological models; therefore, reliable hydrological simulation in data-scarce areas is a challenging task. Advanced techniques can be utilized to improve the accuracy of satellite-derived rainfall data, which can be used to overcome the problem of data scarcity. Our study aims to (1) assess the accuracy of different satellite precipitation products such as Tropical Rainfall Measuring Mission (TRMM 3B42 V7), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), PERSIANN-Climate Data Record (PERSIANN-CDR), and China Meteorological Assimilation Driving Datasets for the SWAT Model (CMADS) by comparing them with gauged rainfall data; and (2) apply them for runoff simulations for the Han River Basin in South Korea using the SWAT model. Based on the statistical measures, that is, the proportion correct (PC), the probability of detection (POD), the frequency bias index (FBI), the index of agreement (IOA), the root-mean-square-error (RMSE), the mean absolute error (MAE), the coefficient of determination (R2), and the bias, the rainfall data of the TRMM and CMADS show a better accuracy than those of PERSIANN and PERSIANN-CDR when compared to rain gauge measurements. The TRMM and CMADS data capture the spatial rainfall patterns in mountainous areas as well. The streamflow simulated by the SWAT model using ground-based rainfall data agrees well with the observed streamflow with an average Nash-Sutcliffe efficiency (NSE) of 0.68. The four satellite rainfall products were used as inputs in the SWAT model for streamflow simulation and the results were compared. The average R2, NSE, and percent bias (PBIAS) show that hydrological models using TRMM (R2 = 0.54, NSE = 0.49, PBIAS = [−52.70–28.30%]) and CMADS (R2 = 0.44, NSE = 0.42, PBIAS = [−29.30–41.80%]) data perform better than those utilizing PERSIANN (R2 = 0.29, NSE = 0.13, PBIAS = [38.10–83.20%]) and PERSIANN-CDR (R2 = 0.25, NSE = 0.16, PBIAS = [12.70–71.20%]) data. Overall, the results of this study are satisfactory, given that rainfall data obtained from TRMM and CMADS can be used to simulate the streamflow of the Han River Basin with acceptable accuracy. Based on these results, TRMM and CMADS rainfall data play important roles in hydrological simulations and water resource management in the Han River Basin and in other regions with similar climate and topographical characteristics.

Remote sensing precipitation or precipitation from numerical weather prediction (NWP) is considered to be the best substitute for in situ ground observations for flood simulations in transboundary, data-scarce catchments. This research was aimed to evaluate the possibility of using a combination of a satellite precipitation product and NWP precipitation for better flood forecasting in the transboundary Chenab River Basin (CRB) in Pakistan. The gauge-calibrated satellite precipitation product, i.e., Global Satellite Mapping of Precipitation (GSMaP_Gauge), was selected to calibrate the Integrated Flood Analysis System (IFAS) model for the 2016 flood event in the Chenab River at the Marala Barrage gauging site in Pakistan. Precipitation from the Global Forecast System (GFS) NWP, with nine different lead times up to 4 days, was used in the calibrated IFAS model to predict the flood hydrograph in the Chenab River. The hydrologic simulations, with global GFS forecasts, were unable to predict the flood peak for all lead times. Then, the Weather Research and Forecasting (WRF) model was used to downscale the precipitation forecasts with one-way and two-way nesting approaches. In the WRF model, the CRB was centered in two domains of 25 km and 5 km resolutions. The downscaled precipitation forecasts were subsequently supplied to the IFAS model, and the predicted simulations were compared to obtain the optimal flood peak simulation in the Chenab River. It was found in this study that the simulated hydrographs, at different lead times, from the precipitation of two-way WRF nesting exhibited superior performance with the highest R2 and Nash–Sutcliffe efficiency (NSE) and the lowest percent bias (PBIAS) compared with one-way nesting. Moreover, it was concluded that the combination of GFS forecast and two-way WRF nesting can provide high-quality precipitation prediction to simulate flood hydrographs with a remarkable lead time of 96 h when applying coupled hydrometeorological flow simulation.

The objectives of this paper are to present a procedure for parameter updating that can be combined with any conceptual rainfall–runoff model for flood-forecasting purposes and to study the improvements in flood forecasting induced by the assimilation of soil moisture information into rainfall–runoff models. The main feature of this methodology is that it carries out updating by reference to not only recent streamflow observations, as in the case of classic procedures, but also soil moisture measurements, which can be retrieved either from time domain reflectometry (TDR) probes or from satellite remote sensing using synthetic aperture radar (SAR) systems. The aim of the research was to assess the usefulness of this additional soil moisture information. To this end, an approach has been suggested that gradually introduces the additional information and can detect a threshold above which this information benefits the whole flood-forecasting procedure. This methodology was put forward for use in the European AIMWATER project on four catchments within the Seine River basin upstream from Paris and on the Arade catchment in southern Portugal.

Floods are one of the most common natural disasters that occur frequently causing massive damage to property, agriculture, economy and life. Flood prediction offers a huge challenge for researchers struggling to predict floods since long time. In this article, flood forecasting model using federated learning technique has been proposed. Federated Learning is the most advanced technique of machine learning (ML) that guarantees data privacy, ensures data availability, promises data security, and handles network latency trials inherent in prediction of floods by prohibiting data to be transferred over the network for model training. Federated Learning technique urges for onsite training of local data models, and focuses on transmission of these local models on the network instead of sending huge data set towards central server for local model aggregation and training of global data model at the central server. In this article, the proposed model integrates locally trained models of eighteen clients, investigates at which station flooding is about to happen and generates flood alert towards a specific client with five days lead time. A local feed forward neural network (FFNN) model is trained at the client station where the flood has been expected. Flood forecasting module of local FFNN model predicts the expected water level by taking multiple regional parameters as input. The dataset of five different rivers and barrages has been collected from 2015 to 2021 considering four aspects including snow melting, rainfall-runoff, flow routing and hydrodynamics. The proposed flood forecasting model has successfully predicted previous floods happened in the selected zone during 2010 to 2015 with 84 % accuracy.

In a flood-prone region, quick and accurate flood forecasting is imperative. It can extend the lead time for issuing disaster warnings and allow sufficient time for habitants in hazardous areas to take appropriate action, such as evacuation. In this paper, two hybrid models based on recent artificial intelligence technology, namely, the genetic algorithm-based artificial neural network (ANN-GA) and the adaptive-network-based fuzzy inference system (ANFIS), are employed for flood forecasting in a channel reach of the Yangtze River in China. An empirical linear regression model is used as the benchmark for comparison of their performances. Water levels at a downstream station, Han-Kou, are forecasted by using known water levels at the upstream station, Luo-Shan. When cautious treatment is made to avoid overfitting, both hybrid algorithms produce better accuracy in performance than the linear regression model. The ANFIS model is found to be optimal, but it entails a large number of parameters. The performanc...

While concept of boosting ensemble data mining techniques has been recently attracted a lot of attention for flood forecasting, mainly on non-urbanised river basins or reservoirs [1,2], time-series boosting, i.e., contribution of last timestep prediction to the next forecasting model is a new era, especially for real-time operation of flood forecasting models in the shape of early warning systems.This study aims to provide time-series boosting for ensemble flood forecasting model through adding forecasted water level of one timestep before as an input of training base models to previous proposed rainfall feature, especially rainfall duration, intensity, evidence of past rainfall and season occurrence [3]. Several weak learner data mining techniques are developed for various forecast lead times and recorded in data cube structure that can be used for developing time-series boosted ensemble model. This novel model was tested for real case study of Hanwell urban drainage systems located in the west London, UK for a period of 20 years data with 15min intervals. Confusion matrix is employed for performance assessment and the model is compared by conventional benchmark gradient boosted models.Results shows the added feature can significantly increase the accuracy of overflow detection of all developed base models, especially for longer timesteps. More specifically, adding the new feature to the model can increase the accuracy rate from 84% for the best developed base model to 93% in 3hrs-ahead predictions. More importantly, the model can decrease underestimation miss rate from 45% to only 21% for the same forecast lead time. Furthermore, new time-series boosted ensemble model can noticeably increase overflow detection rate, where hit rate increase from 78% to 88% in 3hrs-ahead predictions. Overall, the concept of time-series boosted ensemble modelling can overcome the problem of missing and false alarm of real-time operation by adding the previous situation of catchment to the forecasting procedure.

Abstract. A simple two-dimensional rainfall model, based on advection and conservation of mass in a vertical cloud column, is investigated for use in short-term rainfall and flood forecasting at the catchment scale under UK conditions. The model is capable of assimilating weather radar, satellite infra-red and surface weather observations, together with forecasts from a mesoscale numerical weather prediction model, to obtain frequently updated forecasts of rainfall fields. Such data assimilation helps compensate for the simplified model dynamics and, taken together, provides a practical real-time forecasting scheme for catchment scale applications. Various ways are explored for using information from a numerical weather prediction model (16.8 km grid) within the higher resolution model (5 km grid). A number of model variants is considered, ranging from simple persistence and advection methods used as a baseline, to different forms of the dynamic rainfall model. Model performance is assessed using data from the Wardon Hill radar in Dorset for two convective events, on 10 June 1993 and 16 July 1995, when thunderstorms occurred over southern Britain. The results show that (i) a simple advection-type forecast may be improved upon by using multiscan radar data in place of data from the lowest scan, and (ii) advected, steady-state predictions from the dynamic model, using "inferred updraughts", provides the best performance overall. Updraught velocity is inferred at the forecast origin from the last two radar fields, using the mass-balance equation and associated data and is held constant over the forecast period. This inference model proves superior to the buoyancy parameterisation of updraught employed in the original formulation. A selection of the different rainfall forecasts is used as input to a catchment flow forecasting model, the IH PDM (Probability Distributed Moisture) model, to assess their effect on flow forecast accuracy for the 135 km2 Brue catchment in Somerset. Keywords: rainfall forecasting, flood forecasting, weather radar, satellite, storm model

<scp>RMetS M</scp>eteorological <scp>M</scp>asterclass <scp>S</scp>eries: Anticipating floods, droughts and heatwaves