Applied Machine Learning for Flood Susceptibility Mapping in the Loukkos Basin (Northern Morocco): Validation Using the February 2026 Ksar El Kebir Flood Event

Mekkaoui, Oussama; Morarech, Moad; Bouramtane, Tarik; Akka, Hamza

doi:10.5194/egusphere-2026-1001

Preprints

https://doi.org/10.5194/egusphere-2026-1001

Preprints

07 Apr 2026

| 07 Apr 2026

Applied Machine Learning for Flood Susceptibility Mapping in the Loukkos Basin (Northern Morocco): Validation Using the February 2026 Ksar El Kebir Flood Event

Oussama Mekkaoui, Moad Morarech, Tarik Bouramtane, and Hamza Akka

Abstract. Flooding is a recurrent hazard in northern Morocco, where low-lying alluvial plains, strong river–floodplain connectivity, and winter storm sequences combine to produce damaging floods. This study develops an event-informed flood susceptibility assessment for the Loukkos Basin (Ksar El Kebir–Larache floodplain) by integrating satellite-derived flood evidence with geomorphometric and hydro-climatic predictors in a machine-learning framework. A binary flood inventory was derived from Sentinel-1 SAR change detection by contrasting pre-flood (October–November 2022) and flood-phase acquisitions (December 2022), and was validated using Sentinel-2 optical observations and field checks. Nine conditioning factors (elevation, slope, aspect, curvature, distance to rivers, drainage density, TWI, TPI, and CHIRPS-based rainfall) were compiled and standardized on a 10 m grid. Model training used a balanced sample of 220 points (110 flooded/110 non flooded) and was evaluated with a repeated hold-out strategy (10 iterations; 80 % training/20 % testing) using accuracy, precision, recall, and F1-score. Both Random Forest (RF) and Multilayer Perceptron (MLP) produced coherent susceptibility patterns, with the highest classes concentrated along the Loukkos river corridor and downstream floodplains. Mean test performance indicates strong generalization, with MLP outperforming RF (accuracy ≈ 0.909 vs. 0.864; F1 ≈ 0.909 vs. 0.870). Jackknife sensitivity analysis identifies elevation as the leading control for both models, while RF emphasizes terrain metrics (slope, drainage density) and MLP assigns a more balanced importance to topography and hydrological drivers (TPI, rainfall, distance to channel). Notably, flooded areas observed during the January–February 2026 flood episode (Sentinel imagery dated 14 February 2026) largely coincide with zones mapped as high susceptibility, providing an independent, qualitative post-study consistency check. The resulting maps offer a practical inspection tool to support land-use planning and prioritization of mitigation actions across the most flood-prone sectors of the Loukkos Basin.

Received: 21 Feb 2026 – Discussion started: 07 Apr 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Oussama Mekkaoui, Moad Morarech, Tarik Bouramtane, and Hamza Akka

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-1001', Anonymous Referee #1, 10 Aug 2026
Transformer, more than meets the eye: A deep learning approach to integrate rainfall time-series in multi-type landslide probability modeling

The paper models landslide type‑specific “pseudo‑probabilities” for the May 2023 Emilia‑Romagna events using slope units (SUs). Rainfall is interpolated to SU centroids and either aggregated over fixed windows or fed as daily/hourly time series into a Transformer; static terrain/geology features go through a DNN. Authors report AUCs >0.90 and state that daily time series generally perform best, with SHAP‑based “Expected Gradients” used for explanation. There are many issues such as followings,

The distribution map of landslide (by each type) should be included.

The authors used the limited input factors(predictors) In the topography, there are many factors from topographic and hydrological analysis (using SAGA GIS) and they are more related to landslide occurrence than the used factors. Also, soil factor is most importance to occur the landslide. But there is no soil related factor.

What are the criteria of selecting the factors? The relationship analysis (for example Frequency Ratio) between landslide and factors should be done and select the factors.

Daily/hourly rainfall is interpolated to SU centroids using RBF with a linear kernel that forces exact station fit. There is no cross‑validated error analysis, no orographic/anisotropy treatment, no comparison to kriging or gauge‑radar fusion, and no check on station representativeness (188 stations over 11,670 km² is not “almost uniform” in complex terrain). Given the paper’s central claim rests on temporal detail, the spatial interpolation uncertainty should be quantified and propagated.

How to use the landslide as input data? Do you use the landslide location (highest 25^th percentile of elevation values as you mentioned) or SU (slope unit) itself? If you are use the SU as input data, it should be changed to the landslide location. This is not clear in the manuscript but the authors may used the SU as input data.

Also, why use the SU for the mapping. What is the size of SU, Do the SU can reflect the detail information of landslide location and factors? So, the most of study use the grid system.

The description of methodology including tools (S/W or code) for implementing Transformer Neural Network and evaluation should be introduced.

Why to divide 70%/30% for training and testing. Considering the number of data, 50%/50% is more reasonable. Also, how to divide the training and testing data?

Very Fast Kolmogorov-Arnold Network for Landslide Susceptibility Prediction: A Case Study in Gümüşhane, Türkiye

The paper compares six algorithms (XGBoost, LightGBM, NGBoost, CatBoost, MLP, FastKAN) to map landslide susceptibility in Gümüşhane Province using 20 conditioning factors. After a “multicollinearity analysis,” the authors train models on pixel samples (1,446,696 landslide pixels + the same number of non‑landslide pixels), split randomly 70/30, tune hyperparameters with grid search, evaluate using Accuracy/Precision/Recall/F1, interpret with SHAP, and “validate” by overlaying two 2025 landslides on the FastKAN map. They report 96.54% accuracy for FastKAN and declare it the first KAN application to landslide susceptibility. There are many issues such as followings,
This study utilized data from 65 landslide events that occurred in various locations within Gu¨mu¨¸shane Province to create an inventory map. A total of 1,446,696 landslide pixels and 1446,696 non-landslide pixels were generated for training and testing. For the application of ML, the number of landslides is a too few. Also, How 65 landslides become to 1,446,696 landslide pixels.

The only starting point and line of landslide should be used analysis. But the study looks to use the whole scar of landslide for the analysis.

The landslide locations in Fig. 2 are something wrong. There are many landslides in fluvial area and low slope areas. Also, the number of pixels is not 1,446,696.

The information of landslide such as type, reason, photo etc. should be included.

Training and testing site should be distribute equally in spatial

The fig 4 should be redrawn in color.

What are the criteria of selecting the factors? The relationship analysis (for example Frequency Ratio) between landslide and factors should be done and select the factors.

For the validation, the ROC-AUC method should be used.

For Method session, the description of how to use the algorithm, which code or SW were used, the process of the application of algorithms etc. should be include. There the only theory of the algorithms.

Estimating Temporal Probability of Rainfall-Induced Landslides Using Attention-Based Multiple Instance Learning and Daily Rainfall Data: Case Study Mt. Umyeon, Korea

There are several issues that should be addressed, as follows:

The study derives the annual exceedance probability (AEP) and the corresponding return period (RP = 1/AEP) using daily rainfall data from only one year (2011). However, the conventional frequency-based interpretation of return period generally requires multi-year observations and an assumption of (quasi-)stationarity. A single-year rainfall record represents only one realization of climate variability; therefore, it is not logically consistent to interpret the resulting AEP as a long-term mean annual exceedance probability or to present return-period classes such as 1–5 years or 5–10 years without additional multi-year evidence or explicit assumptions.

The reported area under the ROC curve (AUC) is 0.787 for the training data and 0.808 for the validation data. While small differences can occur due to sampling variability and regularization, it is not typical for validation performance to exceed training performance. The authors should clarify the evaluation protocol (e.g., the data-splitting strategy, whether spatially stratified cross-validation was used, and whether results were averaged over multiple runs) to rule out potential leakage or reporting inconsistencies.

The landslide inventory contains only 163 initiation points, which is a relatively small dataset for splitting into separate training (70%) and validation (30%) subsets, especially for a deep-learning framework. A more robust validation strategy (e.g., repeated spatial cross-validation or bootstrapping) is recommended to improve the reliability of the performance assessment.

The relationships between landslide occurrence and the conditioning factors should be examined more rigorously. The analysis should focus on predictors that are demonstrably relevant, or the manuscript should provide stronger justification for retaining all variables. Additional feature-selection and/or sensitivity analyses would strengthen the physical interpretability and reduce the risk of including irrelevant factors.

Most figures are difficult to read because the panels and fonts are too small, and the overall graphical quality is poor. The figures should be revised to improve resolution, font sizes, and general readability.

The manuscript is lack in scientific aspect. There is no survey/collection, data process and analysis and just show the existing results. The manuscript looks like report. So, the manuscript is not suitable to published in “Geosciences Journal”.

There is big problem in input data. The salinity, temperature, rainfall and tide data were used for the study. But the station of the data is very few considering the study area. Moreover, the data used after interpolation and this is not accepted because the station is too few. Also, the selection of input data should be justified. To apply the machine learning, the data should be enough. If the data is not enough, the machine learning can’t be used. So, the manuscript is not suitable to published in “Geosciences Journal”.

I can’t find the new findings in data and methodology and the manuscript is not novel. So, the manuscript can’t be published in “Geosciences Journal”

Major Comments
The dataset is too small to support such a broad model comparison. The study uses only 31 annual observations to train and tune 15 models, which creates a serious risk of overfitting, especially for more flexible models such as the 100-unit MLP. The authors should either expand the dataset or narrow the analysis to a smaller set of parsimonious models. In addition, key forecasting baselines are missing. Naïve, drift, linear-trend, exponential-smoothing, ARIMA, and lagged-CO₂ models should be included, because year-only and lagged-CO₂ baselines are essential for determining whether the proposed machine-learning models actually improve upon the historical trend.
The validation and forecasting procedures also require substantial clarification. The current validation strategy is unclear and appears unsuitable for time-series data, since random splitting or shuffled cross-validation may introduce temporal leakage. Model tuning and evaluation should instead rely on rolling-origin or expanding-window validation, preferably within a nested framework. The 2021–2030 forecasts are also not reproducible. Because the models require future values of domestic credit, GDP, population, renewable energy consumption, and trade openness, the manuscript must clearly explain how these future predictor values were generated, forecasted, or assumed under specific scenarios.
Several methodological details and interpretation issues should be addressed before the results can be considered reliable. Preprocessing and hyperparameter tuning are insufficiently described; the manuscript should report scaling procedures, search spaces, validation criteria, random seeds, software versions, and complete model settings. Preprocessing must be fitted only on the training data, and all models should receive comparable tuning effort. The model-selection approach is also inconsistent: although Bagging performs best in Table 5, other models are later selected as the “most plausible” mainly because their forecast curves visually resemble the historical trend. Model selection should follow predefined out-of-sample performance criteria rather than visual judgment. Finally, the XAI analysis is unclear and appears overinterpreted. The manuscript should specify which model was explained using SHAP, LIME, and permutation importance, and given the small sample size and strong correlations among predictors, feature rankings should be treated as unstable predictive associations rather than causal effects.

Integrating Multi-Temporal Synthetic Aperture Radar Observations and Boosting Ensemble Learning for High-Resolution Flood Susceptibility Mapping in Napa County, California

The manuscript has lack in novelty and completeness in many parts. Please check the followings,

For the conditioning factor, how to choose the factors. There should be reason to selecting the factors. The hillshade map can’t be one of factor.

The validation strategy is not sufficiently rigorous. The 1,126 flood and non-flood points were randomly divided into training and testing subsets despite likely clustering within flood polygons and events. This creates spatial and event-level leakage and can substantially inflate the reported 97.70% accuracy. The models should be reassessed using spatial-block cross-validation, event-based holdout testing, and an independent flood inventory. Moreover, there are no flood inventory (training/testing) map.

The reported performance statistics are internally inconsistent. Table 3 labels the results as validation performance, whereas the text and Figure 5 describe training performance. The ROC values in Table 3 do not match the AUC values in Figure 5. The ROC plots also appear to use only a few operating points. All metrics should be recalculated from held-out prediction probabilities, with confidence intervals, confusion matrices, PR-AUC, calibration measures, and defined decision thresholds.

The flood inventory is inadequately documented. The manuscript does not provide event dates, Sentinel-1 acquisition details, orbit direction, polarization, preprocessing steps, change-detection thresholds, permanent-water masking, or data-merging rules. Because the inventory covers 2000–2018, the treatment of pre-2014 events must be clarified, since Sentinel-1 could not have observed them.

The claimed 10 m resolution is misleading. Several predictors have much coarser native resolutions, including the FAO soil map and station-interpolated rainfall, and the graphical abstract identifies SRTM 3-arc-second DEM data, whereas the methods claim a Sentinel-1-derived DEM. Resampling coarse inputs to 10 m does not create 10 m information. Native resolutions, resampling methods, coordinate systems, and scale-sensitivity analyses are required.

The model configuration raises concerns about overfitting and implementation validity. LightGBM uses 2,903 trees, 342 leaves, and a maximum depth of 180 for only 900 training samples. The scale_pos_weight of 182.332 is unexplained despite an apparently balanced dataset. In addition, RF is implemented with RandomForestRegressor for a binary classification problem. The objectives, class weights, tuning ranges, nested validation procedure, random seeds, and probability calibration must be reported.

The manuscript repeatedly conflates flood susceptibility with flood risk. Exposure, vulnerability, damage, population, infrastructure, and event probability are not modeled; therefore, the outputs cannot support risk-based claims or recommendations for levees, detention dams, diversion structures, artificial lakes, or hydropower generation. These recommendations require hydraulic modeling, economic assessment, and environmental evaluation.

Several equations and interpretations require correction. The SPI equation cannot normally produce negative values as described, the TRI expression appears incorrectly formulated, and the kappa terminology is imprecise. The discussion also oversimplifies flood mechanisms by combining slow-rise river flooding and flash flooding within one static model.

For the result (Fig.4). what is the criteria to dived the HR, R, MR, LR, Vl? The distribution is something wrong. How to calculate the optimized parameter as table 2. Do you use metaheuristic algorithm?

Feature-importance results are insufficiently defined and overinterpreted. The claim that four variables explain 70.65% of flood occurrence is unjustified without specifying whether importance is based on gain, split frequency, permutation, or SHAP values. Correlated DEM-derived predictors also require grouped importance, ablation tests, and uncertainty analysis.

Reproducibility is inadequate. The study should provide code, event-level inventory metadata, train-test identifiers, hyperparameter search records, raster sources, acquisition dates, software versions, and final prediction layers. Reference-list inconsistencies, missing variables in Figure 2, ambiguous map dates, and inconsistent terminology should also be corrected.

Engineering Applications of Artificial Intelligence
An environmental similarity-guided attention neural network for landslide susceptibility mapping in underrepresented environments

After checking the manuscript, the manuscript can’t be published in present form because there are some major issues such as followings,
The authors developed the new algorithms, EGS-NN. To validate the accuracy, reliability and transparency of the EGS-NN, the authors should more analysis especially validation such as followings, SHAP attribution and cross-model consensus, statistical comparison of random-test AUC, companion patch-based ablation and fold-prediction dispersion and cross-model agreement.

For the conditioning factor, how to choose the factors? There should be reason to selecting the factors.

The information of landslide such as type, reason, photo etc. should be included.

The headline gaps (e.g., 0.919 vs. 0.912; 0.872 vs. 0.857) are smaller than the reported standard deviations, so the authors should apply paired tests (Wilcoxon signed-rank, DeLong for AUC) with confidence intervals to establish that the improvement is statistically meaningful.

The random split likely inflates performance through spatial autocorrelation—especially since ESG-NN relies on near neighbors—so the models should be re-evaluated under spatial (or environmental) block cross-validation.

The paper motivates generalization to unfamiliar environments but only tests a single-area random split; the authors should either add a cross-region experiment or scale back the transfer narrative to match the experiment.

The Discussion cites closely related methods (geographical GAT, GGPR, CatGWR) but compares only against ANN/RF/LR; a quantitative comparison with at least one of these is expected.

The weights use fixed similarity with softmax (not learned attention), and the method aggregates features rather than susceptibility values, so both the "attention" terminology and the Third Law framing should be stated more precisely.

All map is too small and poor. There should be redrawn.

Geoscience Frontiers
From spatial dependence to temporal memory: A unified graph neural network for spatiotemporal landslide susceptibility prediction

Before the check the model and validation, there are serious missing of basic information such as followings,
The information of landslide such as distribution map, number, type, reason and photo etc. should be included.

For the conditioning factor, there is no information such as map, scale and source etc. How to choose the factors. There should be reason to selecting the factors.

As the validation, Spatial-block CV is recommend and calculation of similarity and importance of factors should be considered.

So, the manuscript can’t be reviewed and published to the journal.

- The manuscript addresses a relevant topic and the authors have made a clear effort to respond to the previous reviewers. However, I still have a major concern regarding the fundamental design of the study.
- The main issue is that the sediment yield used as the target variable is not based on direct observations. Instead, it was estimated from a discharge–sediment rating relationship. The authors now acknowledge this limitation and the resulting structural dependence between the target and several hydrological predictors.
- This makes the very high machine-learning performance difficult to interpret. In particular, runoff volume, peak discharge, and their derived variables are closely related to the discharge information already used to generate the proxy sediment yield. Therefore, an RF R2 of about 0.94 may partly reflect this built-in relationship rather than true predictive capability.
- The addition of 5-fold cross-validation improves the analysis, but it does not resolve this underlying circularity.
- I am also concerned about the small number of independent events (69) relative to the large number of tested model and feature combinations. Some configurations include up to 196 predictors, which further raises concerns about overfitting and the robustness of the reported results.
- In my view, these limitations affect the central conclusions of the paper and cannot be adequately addressed by further textual revision alone. Independent sediment observations would be necessary to properly validate the proposed framework.

The used models are not new and novel and all data except of rainfall have made using DEM.

The manuscript addresses an important flood-management problem and combines Sentinel-1 flood mapping, GIS-derived conditioning factors, and two machine-learning models for the Loukkos Basin. However, several methodological weaknesses currently limit the reliability of the conclusions.

The title claims validation using the February 2026 Ksar El Kebir flood, but the 2026 event is used only as a qualitative visual comparison. No independent flood boundary, confusion matrix, ROC-AUC, intersection-over-union, or class-capture statistics are presented. The authors should either conduct a quantitative event-transfer validation or replace “Validation Using” with “Qualitative Comparison with.”

The 2022 Sentinel-1 flood inventory is insufficiently documented. The threshold value, orbit information, acquisition dates, preprocessing workflow, speckle filtering, terrain correction, permanent-water masking, and quantitative validation accuracy are not reported. These details are essential because the inventory defines the dependent variable.

Only 220 balanced samples were used, and the models were evaluated through repeated random 80/20 hold-out partitions. Because flooded samples are spatially clustered, nearby points may occur in both training and testing sets, inflating model performance. Spatial block cross-validation is required.

Both RF and MLP achieved perfect training performance, indicating possible overfitting. Reported test metrics should include the mean, standard deviation, and confidence interval across all ten repetitions rather than single values.

Essential RF and MLP implementation details are missing, including tree number, depth, hidden layers, neurons, activation function, optimizer, learning rate, epochs, regularization, early stopping, and random seeds.

The CHIRPS rainfall layer represents cumulative rainfall for 2020–2024 rather than event-specific rainfall for either the 2022 training flood or the 2026 comparison event. Resampling it to 10 m does not increase its intrinsic spatial information.

Model outputs should be described as relative susceptibility scores, not true flood probabilities, because the models were trained using an artificially balanced case–control sample.

Overall, quantitative 2026 validation, spatially independent model evaluation, and substantially improved methodological reproducibility are necessary before the manuscript can be considered for publication.

The used models are not new and novel and all data except of rainfall have made used DEM.
Citation: https://doi.org/10.5194/egusphere-2026-1001-RC1
- CC1: 'Reply on RC1', Hamza Akka, 10 Aug 2026
  
  Dear Referee 1,
  Thank you for the review and for the comments provided on our manuscript.
  Before preparing our detailed point-by-point response, we would like to clarify one issue regarding the referee report. The submitted report appears to contain several sections referring to studies that are unrelated to our manuscript. In particular, some comments discuss sediment-yield prediction, landslide susceptibility, CO₂ forecasting, and other datasets and methodologies that are not part of our study.
  Our manuscript concerns flood susceptibility mapping in the Loukkos Basin using Sentinel-1-derived flood information, nine conditioning factors, and Random Forest and Multilayer Perceptron models. We clearly recognize the section of the report that specifically addresses these elements, including the comments concerning the 220 training/testing samples, the 80/20 repeated hold-out strategy, spatial validation, the 2022 flood inventory, CHIRPS rainfall, and the February 2026 flood event.
  We would therefore be grateful if the Editor could confirm whether only this Loukkos-specific section should be considered as the referee comments applicable to our manuscript, or whether a corrected version of the referee report will be provided.
  Once this is clarified, we will be pleased to address all relevant comments carefully and provide a detailed response and revised manuscript.
  Kind regards
  
  Citation: https://doi.org/10.5194/egusphere-2026-1001-CC1
RC2:
'Comment on egusphere-2026-1001', Anonymous Referee #2, 24 Aug 2026
Dear authors,
Thanks for your effort on such nice work. Please find my comments below:
General comments
In general, the article is well-structured and clear, which helps the reader follow the article flow easily. It also makes a novel contribution to flood management and preparedness by proposing a method for flood risk assessment that uses advanced techniques, such as machine learning, to produce flood maps with identified flood risk factors in Morocco, a challenge in this country. There is also a practical need to promote scientific research in risk management. Therefore, the risk maps generated by the model can be used to enhance stakeholders' and affected communities' understanding of flood risk. The research provides meaningful recommendations for flood management in the Loukkos River basin.
Methods:
The method is well presented, with clear flow, steps, and explanation. This is one of the cases of downscaling research on flood management at the basin scale.
In my view, for flood risk assessment, even with post-event validation that supports the general result, a single event may reflect only a specific case at a given time and not capture the overall flood situation in a place like the Loukkos River basin. Other references can be used to enrich the data for assessment; for example, Loudyi et al. (2022) reported that event severity is calculated using rainfall station data from each of the nine catchments, including Loukkos.
Additionally, relying on a single event might underestimate the pressures of climate change, an emerging challenge that calls for climate-related time series that capture changes over time and variation across events to manage natural hazards effectively. The authors used post-event data from 2026 for validation, but I think the inundation areas could overlap even though the flood conditions were different, as it was rather a compounding flood event due to the impacts of the tidal scheme; therefore, the drivers of flooding were different as well. This point is crucial for flood management.
Factors influencing flooding
The research will be more convincing if the authors treat vegetation cover (not included in the Exposition), land use, and infrastructure (such as dams/reservoirs) as driving factors in their flood risk analysis. The research may need to address/discuss interactions among fluvial discharge, coastal waters, and nearshore conditions that may be associated with compound flooding susceptibility in downstream areas, as it is also mentioned in the Study Area part (line 164, page 4) and some other relevant articles.
Regarding rainfall, could you please explain why you treat it as (lines 267 – 268, page 8) “…Daily precipitation data were aggregated into monthly totals and then summed to obtain cumulative rainfall over the five-year period…” while, in my understanding, the Loukkos River basin mainly faces flash floods due to daily intensive rainfall.
Result:
Even though the flood map results from both models classify the study area into four susceptibility categories: low (0–0.25), medium (0.25–0.5), high (0.5–0.75), and very high (0.75–1), I still expect a more convincing discussion/argument on the differences in the ranking of roles of the flood drivers from the two models. I think it is important for decision-makers and affected communities to understand the risk factors to prioritize and allocate resources for risk management, or to avoid flood risks through safer master planning or integrated planning for the whole basin.
Dicussion
In my view, aside from discussing your results, some parts of this section provide additional explanation of what the methods do not address/include and what the flood susceptibility maps have not shown, which are really useful and should be on the maps/results. They include:
Vegetation coverage: as mentioned earlier, I think it should be treated as a separate influencing factor in the model.

Rainfall ‘Line 458 – 463’: indicating the strong role of rainfall in triggering floods, and it may help represent event intensity. This confuses me as to why, in your method, you still treat the rainfall as it is in lines 267 – 268, page 8, as ‘Daily precipitation data were aggregated into monthly totals and then summed to obtain cumulative rainfall over the five-year period’. I am concerned about this because, in my understanding, flash flood susceptibility in the Loukkos River is emerging.

From line 504, page 16 – line 529, page 17 and lines 505 – 518, page 18: these parts discuss/explain compound susceptibility to flooding in coastal areas. This is also what I expected to see on your maps after reading some references about flood management in the Lokkous River basin, but unfortunately it is not. Therefore, this part is quite disconnected because the issue is not mentioned in the introduction, influencing factors, or method. It would be more concise if you briefly introduced it and explained in your methods or factors sections why you do not cover or address it in your research, along with the limitations.

Thank you and best regards,
Citation: https://doi.org/10.5194/egusphere-2026-1001-RC2

Oussama Mekkaoui, Moad Morarech, Tarik Bouramtane, and Hamza Akka

Viewed

Total article views: 684 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
420	230	34	684	36	39

HTML: 420
PDF: 230
XML: 34
Total: 684
BibTeX: 36
EndNote: 39

Views and downloads (calculated since 07 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	164	63	16	243
May 2026	116	85	4	205
Jun 2026	18	10	3	31
Jul 2026	55	35	2	92
Aug 2026	40	25	2	67
Sep 2026	27	12	7	46

Cumulative views and downloads (calculated since 07 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	164	63	16	243
May 2026	116	85	4	205
Jun 2026	18	10	3	31
Jul 2026	55	35	2	92
Aug 2026	40	25	2	67
Sep 2026	27	12	7	46

Viewed (geographical distribution)

Total article views: 649 (including HTML, PDF, and XML) Thereof 649 with geography defined and 0 with unknown origin.

Country	#	Views	%

Cited

Latest update: 15 Sep 2026

Short summary

This study presents a machine learning–based flood susceptibility assessment for the Loukkos Basin (northern Morocco) using Sentinel-1 SAR–derived flood inventory and geo-environmental factors. Random Forest and Multilayer Perceptron models were applied with balanced samples and validation, with MLP showing better performance. Results emphasize topographic and hydrological controls, and maps are validated using the February 2026 flood event.


Total:	0
HTML:	0
PDF:	0
XML:	0

Applied Machine Learning for Flood Susceptibility Mapping in the Loukkos Basin (Northern Morocco): Validation Using the February 2026 Ksar El Kebir Flood Event

Viewed

Viewed (geographical distribution)

Cited

1 citations as recorded by crossref.