Understanding how tree species resist drought is essential for resource conservation in the context of increasing climate stress. This study quantifies the spatiotemporal dynamics of drought impacts on two land classes within the Oum Er Rbia Basin in Morocco: irrigated orchards in the plain and evergreen forests in mountainous areas. The study focuses on the prolonged drought that began in September 2019, identified using the CHIRPS monthly Standardized Precipitation Index and the TerraClimate Palmer Drought Severity Index. To extract the tree mask, separate methodologies were applied. First, the Random Forest model, trained on Sentinel-2 data with spectral indices and phenological metrics, achieved an accuracy of 95.94% for the irrigated area. Second, for the forest extraction of evergreen forests, monthly NDVI time series from 01/2017 to 12/2018 were used, and pixels with values greater than 0.25 in all months were selected to distinguish forestland areas from cropland areas. This monthly NDVI threshold is ecologically justified by the evergreen nature of the dominant forest species in the Middle Atlas Mountains. The duration of vegetation resistance to drought (DVR) was quantified by calculating the number of months with NDVI values greater than 0.25 after drought onset in September 2019. This metric captures how long vegetation maintains physiological its function despite the prolonged drought conditions, reflecting the duration of vegetation resistance to drought. The results revealed that the DVR ranges from an average of 34 months in the irrigated trees to an average of 50.6 months in the mountain forest, indicating that irrigated orchards are significantly more vulnerable to prolonged drought than natural forest ecosystems. Within the irrigated perimeters, the Tassaout region showed the shortest DVR, directly linked to the complete interruption of the irrigation water supply from 2021 onwards. Within the mountain forest areas, south and southeast facing slopes showed the lowest DVR values, reflecting their greater exposure to hot and dry conditions, while north and northeast facing slopes demonstrated greater resistance to drought stress. Additionally, GRACE data, dam water flow for irrigation, and dam water area were analysed to determine the impact of irrigation on water resources. Field validation confirmed these trends. These integrated earth observation-field methods ultimately provide operational frameworks for adaptive drought management, supporting reforestation planning, irrigation scheduling, and resilience monitoring in semiarid regions globally.

In Morocco, snow constitutes a crucial freshwater resource, particularly in the Atlas Mountains, where seasonal snowpack significantly contributes to surface water availability, groundwater recharge, and down-stream water supply. However, snow monitoring in these regions remains challenging due to the scarcity and uneven distribution of ground-based snow depth measurements, especially at high altitudes. This lack of observations limits the accurate assessment of snowpack dynamics and hampers hydrological modeling and water resource management. In this study, we assessed the performance of an empirical approach to estimate snow depth from satellite-derived fractional snow cover (FSC) obtained from MODIS observations. Five empirical FSC snow depth models, including linear and nonlinear exponential formulations, are developed and applied across multiple regions of the Moroccan Atlas Mountains. Model coefficients are calibrated independently for each region using three complementary optimization techniques, nonlinear least squares regression, genetic algorithms, and simulated annealing. Model skill was evaluated during calibration and validation using the Kling-Gupta Efficiency (KGE), Pearson correlation coefficient (R), and absolute error metrics (RMSE and MAE). Results show substantial performance differences across formulations and regions. The most flexible exponential model achieved highest efficiency (KGE up to 0.87; R > 0.85) and 0.26 cm (MAE) under moderate snow conditions. Linear formulations exhibited limited robustness, whereas exponential models better captured snow depth dynamics, particularly in high-altitude areas with deep and persistent snowpacks. These results highlight the potential of FSC-based empirical modeling as a practical and operational solution for snow depth estimation in data-scarce mountainous regions of Morocco.

Seasonal snow in the Moroccan Atlas Mountains plays a key role in sustaining downstream water resources, yet its monitoring is hampered by sparse in situ observations and complex topography. The MODIS (Moderate Resolution Imaging Spectroradiometer) sensor provides daily Normalized Difference Snow Index (NDSI) data at 500 m resolution, offering valuable temporal coverage but no operational fractional snow cover (FSC) product and substantial sub-pixel heterogeneity, while Sentinel-2 delivers 10 m imagery suitable for detailed snow mapping but with lower temporal frequency. This study investigates whether artificial intelligence (AI) models can improve FSC estimation from MODIS NDSI over the Atlas relative to classical regression. We build a multi-year reference dataset from 69 Sentinel-2 scenes across six regions, aggregated to the MODIS grid and paired with cloud-gap-filled MODIS NDSI. Three AI regressors, Random Forest (RF), Support Vector Regression (SVR) and a Multilayer Perceptron (MLP) are trained and validated using image-wise data splitting and evaluated against a global linear regression baseline. Model skill is assessed using the correlation coefficient, RMSE, MAE and Kling-Gupta efficiency (KGE). All three AI models generalize well, with test R values of 0.82-0.83 and KGE between 0.77 and 0.83, and they clearly outperform the global regression baseline by reducing median RMSE and MAE by 10-15% and increasing mean KGE from 0.70 to around 0.80. SVR provides the best overall balance (test R=0.834, MAE = 0.141, KGE=0.832) and is retained as the reference model. When applied to MODIS time series, the SVR-based FSC product reproduces Sentinel-2-derived snow-covered area with very high accuracy (R=0.99, RMSE =8.13 km2, bias =-4.09 km2), indicating that AI-based FSC mapping from MODIS NDSI can reliably support long-term snow and water-resources monitoring in the Moroccan Atlas.

Study region: In the Oum Er Rbia watershed, Morocco, dam water resources play a crucial role in prolonged drought conditions, particularly in the case of the Al Massira Dam, which has been a strategic reservoir for drought resilience since its inauguration. Study focus: Optimized pipelines of explainable artificial intelligence (XAI) models were developed for monthly forecasts of water resource variations at the Al Massira dam, which has been affected by unprecedented drought since 2019. The architectures of the models developed incorporate Bayesian optimization via Optuna for identifying the best hyperparameters, advanced feature selection methods, and lagged regressors of teleconnection indices, drought indices, and hydroclimatic variables. The performance of the models was first evaluated in terms of their ability to forecast dam water volume up to 6 months ahead under near-normal hydroclimate conditions. Next, model performance was assessed under a scenario of unusual changes in time series. New hydrological insights for the region: The light gradient boosting machine (LightGBM) showed high uncertainty when forecasting water volumes under unusual drought conditions, with Skill= 75.1 % and NMAE= 11.2 %. The Bayesian probabilistic LSTM (ProbLSTM) reached the maximum predictive skill score (Skill=86.2 %, NMAE=3.6 %), followed by the generalized additive model (GAM) (Skill=85.3 % and NMAE=3.4 %). Overall, from an operational perspective, ProbLSTM and the GAM are preferable for seasonal forecasting because of their low performance variability under a scenario of unusual changes in time series and their high predictive performance.

In the era of climate change, drought is defined as one of the most severe natural catastrophes that affects the environment, crop growth, and water resources, leading to economic losses, migration, and risks to human life. Since 2019, Morocco has suffered one of the most severe droughts in its recording history, coinciding with a broader period of precipitation deficit across the Mediterranean basin, resulting in a significant reduction in reservoir storage levels and the suspension of irrigation provided by dams in some areas due to low or absent rainfall, making drought a serious challenge to natural resources in this country. This study focused on semi-arid regions, especially on the Tensift basin in Morocco, and was conducted between 2018 and 2024. The effects of drought on arboriculture were analyzed thanks to remote sensing by using the normalized difference vegetation index (NDVI), while NDVI, TCI (temperature condition index), VCI (vegetation condition index), VHI (vegetation health index), and SPI (standardized precipitation index) were employed to assess and evaluate the health of the vegetation area, especially the arboriculture area, and the effective water stress conditions in our agricultural study area. The analyses indicated that arboriculture cover decreased markedly, from 11.17% in 2020 to 7.40% in 2023. From the analyses, it was evident that the cover of this culture had significantly declined from 11.17% in 2020 to 7.40% in 2023. From 2018 to 2024, more than half of the area suffered from drought in the agriculture of varying intensity levels, from moderate to severe. The meteorology of the evaluations confirmed the above observations by showing that there was a considerable decline in precipitation levels from about 350 mm in 2019 to less than 50 mm in 2024, coupled with continuously negative SPI-6 indices. Moreover, it was established that there was a significant effect on tree crops such as olives and citrus. Degradation of land was at its worst during 2021 when it affected more than 3,000 hectares of olives and 2,250 hectares of citrus. Similarly, 80% of farmers engaged in the production of citrus registered a decline in yield from 37% to 41%. Consequently, this study provides new information concerning the need for comprehensive evaluation and monitoring of agricultural drought in Morocco, thereby emphasizing the importance of essential factors.

recharge, influences groundwater reserves in Morocco's main basins, focusing on assessing the relative importance of snowmelt and rainfall. Remote sensing data from MODIS for snow cover area (SCA), the Global Gravitybased Groundwater Product (G3P) for groundwater storage anomalies (GWSA), and the Integrated Multi-satellitE Retrievals for GPM (IMERG) for rainfall were used for this purpose from September 2002 to August 2023. Trend analysis, correlation metrics, and a regression-based framework were used to assess temporal variability and quantify recharge sources. Results showed that G3P-derived GWSA accurately captured groundwater level variations, achieving high correlation (up to 0.93) and low error (as low as 0.72), confirming its reliability. During the study period, most basins showed significant GWSA declines, closely related to decreasing SCA. A 6 to 8-month lag was observed between SCA and GWSA peaks. Furthermore, SCA depletion and GWSA change rates revealed distinct hydrological responses across basins. In Oum Er-Rbia and Draa, their strong correlation highlighted the significant role of snowmelt in groundwater recharge. In contrast, moderate correlations in other basins indicate additional influences. On average, snowmelt contributed 5 %-50 % to groundwater recharge, highlighting its important, yet secondary, role in seasonal snow regions. These results highlight the role of snowmelt in groundwater recharge and the vulnerability of groundwater to variations in snow cover and precipitation. Our results further underscore the need for adaptive water management, particularly in semi-arid and arid regions.

SnowMapPy is an open-source Python-based package developed to streamline the collection, processing, and analysis of MODIS snow cover data from Terra and Aqua satellites. By automating essential steps, including data clipping, reprojection, filtering, and time series generation, SnowMapPy significantly enhances the efficiency and precision of snow hydrology research. The package allows users to collect and process data directly from Google Earth Engine, enabling efficient data acquisition and processing tailored to the needs of snow hydrology, water resource management, and climate change studies. Designed for accessibility and flexibility, SnowMapPy supports large-scale, high-resolution snow cover analysis with minimal configuration. Through its modular structure, the package facilitates customized workflows, making it a valuable tool for researchers aiming to understand snow dynamics and its impact on seasonal water resources. The first application of SnowMapPy was in the Moroccan Atlas Mountains, where snowmelt is essential for water resources. The tool effectively captures spatial and temporal snow cover variations, offering valuable insights for water management and climate impact assessments. This initial implementation highlights its potential for broader application in other snow-covered regions worldwide.

In North African countries where water scarcity and limited data prevail, employing predictive hydrological modeling is crucial to gain accurate insights into current and future water reserves. Hence, these models parameters exhibit instability in this context due to the climate variability observed through basins. Therefore, our efforts focus on using a combination of measured data, remote sensing information, and reanalysis data for calibration and validation, to check the improvement in the result accuracy. Through this study, we simultaneously investigate the spatiotemporal stability of the HBV model in several sub-catchments of Oum Er-Rbia Basin, by improving the performance of a bucket-type conceptual model. We created a Nested Cross-Validation (NCV) framework to assess spatiotemporal stability. The framework uses optimal parameters from a donor catchment of the Hydrologiska Byråns Vattenbalansavdelning (HBV) model as inputs for target catchment parameter ranges. In particular, we evaluated HBV's capacity for prediction over time and space, as well as its impact on model parametrization throughout the regionalization process in the setting of sparse data catchments. As results, the HBV model is spatially transferable from one basin to another, with NSE ranging from 0.5 to 0.8 and KGE values between 0.1 to 0.9, meaning a moderate to high performance. The HBV optimum parameter sets exhibit unpredictable behavior over space. On the contrary, their inter-annual behavior is nearly identical. It also detected a decrease in the model's predictive skills over time, which can be explained by the research area's tendency to dry out year after year. Furthermore, employing KGE for calibration rather than NSE improves model predictive performance significantly. The model calibration process with the KGE outperformed those with the NSE metric, especially when simulating high flows. Furthermore, the findings demonstrate a significant relationship between high model performance and high values of several optimal parameter sets throughout the calibration and validation periods.Keywords: HBV model, poorly gauged basin, arid and semi-arid region, KGE, NSE.

Study region: The Oum Er Rbia watershed, Morocco, is a region facing severe water stress conditions associated with the simultaneous occurrence of meteorological and hydrological droughts. Study focus: This study proposes a new hydrometeorological drought composite index (HDCI) by adapting an explainable artificial intelligence (XAI) approach for synergistic integration of multisource drought-related indicators. The streamflow anomalies, hydroclimatic coefficients and water variations in dams were comparatively explored as response variables for the selection and weighting of the HDCI components using Shapley additive explanations theory (SHAP). New hydrological insights for the region: The severity of hydrometeorological drought in Oum Er Rbia watershed is regulated by the interaction of several factors, among which the contribution of terrestrial water storage to hydrometeorological drought related to streamflow anomalies tends to become more pronounced as the number of influencing factors decreases. The new composite index is highly correlated with the reference hydrometric variables. However, heteroscedasticity between hydrometric stations influences the performance of the HDCI. Therefore, the integration of factors based on spatial dependencies represents a potential avenue for reducing the influence of spatial heterogeneities. Overall, by integrating exclusively geospatial and reanalysis data, HDCI has advantages for the assessment of hydrometeorological drought conditions at the pixel scale compared with conventional methods, which use direct measurements of hydrometric variables but are often discontinuous and unavailable in real time.

Accurate prediction of daily streamflow is critical for effective water resource management. In this paper, we present a novel strategy for deep learning generalization and adaptation to improve daily streamflow prediction in a mountainous region. The process is realized using Bayesian optimization; a Cross Basin Transfer approach to select the stable model; finally, improving predictions using knowledge transfer. The Gated Recurrent Unit model was found to be stable in a complex and diverse sub-basin, while, the Long Short-Term Memory model is robust in a more homogeneous environment. In contrast, the One-Dimensional Convolutional Neural Network had poor accuracy and was unable to capture the flow dynamics. Moreover, the transferred model showed improvement, with the efficiency increasing from 0.20 to 0.51. Our strategy of adapting deep learning models based on the generalization of different areas may represent a valuable approach in the application related to water resources management.

Snow water equivalent (SWE) is a critical variable for understanding water availability and snowmelt-driven streamflow in mountainous regions. Yet, its spatial and temporal estimation is constrained by scarce in situ measurements and the inherent challenges of deriving SWE directly from satellite observations. Thus, accurate SWE assessment is essential for predicting the spatial distribution of snowpack and its temporal contributions to downstream outflow, particularly in semi-arid snow-fed basins like Morocco's High Atlas regions. In this study, we simulate the local and spatial distribution of SWE and outflow at 500 m using Snow17 model, ERA5-Land and satellite-derived fractional Snow Cover Area (fSCA) from Moderate Resolution Imaging Spectroradiometer (MODIS) for the period 2000-2022. The reanalysis data was downscaled and bias corrected using machine learning models (e.g. random forest). To validate results, we compared simulated snow cover area (fSCA) (transformed from SWE simulation) with fSCA issued from MODIS. The methodology was tested in the Rheraya sub-basin (Tensift basin) and applied in Ait Ouchene and Tillouguite sub-basins (Oum Er Rbia basin) in Morocco's High Atlas Mountains. Statistical analysis shows strong model performance, with Nash-Sutcliffe Efficiency (NSE) values exceeding 0.84 for snow depth (SD) simulations. Moreover, spatio-temporal analysis revealed that SWE and snow depth are significantly higher above 2,500 m elevation, with SWE exceeding 300 mm and SD surpassing 60 cm in Tillouguite and Rheraya sub-basins. Findings also demonstrated that snowmelt contributions to outflow varied significantly with elevation, accounting for 40-46% of annual outflow above 2,500 m and playing a dominant role during spring (55-57% of seasonal outflow). Our research provides a framework for enhancing SWE/outflow estimation and understanding snowpack dynamics in semi-arid mountainous regions, highlighting the vital role of high-altitude snowpacks in water resource sustainability and management under climate change.

Analysis of the temporal relationship between meteorological drought and hydrological drought is crucial in monitoring water resource availability. This study examined the linear and lagged relationships of the spread of meteorological drought to hydrological drought and their joint effects on low-flow drought variability in the Oum Er-Rbia (OER) watershed. To this end, random forest (RF) model and statistical methods were used to study the characteristics of the temporal relationships between meteorological and hydrological drought indices at monthly, seasonal, and annual scales. The various analyses revealed that the relationship between hydrological and meteorological drought is mainly a function of the time scale considered, the choice of indices to describe each type of drought and the season considered. The hydrological drought of surface water and snow cover is synchronized with the meteorological drought at the monthly, seasonal, and annual scales. In contrast, the transition from meteorological drought to groundwater drought has a lag time of 1 month and is statistically significant up to t − 5 and t + 5, i.e., 6 months. The linear correlation between the annual rainfall deficit and the monthly groundwater storage index was the lowest (0.15) in December and the highest (0.83) in March. This suggests a seasonal response of groundwater drought to the cumulative effects of precipitation deficits. The RF analysis highlighted the importance of the cumulative characteristics of meteorological drought regarding the severity of low-flow drought. The meteorological drought indices at longer time scales have a greater impact on the severity of low-flow drought, with a contribution of approximately 10% per index. However, the relative contributions of meteorological factors and hydrological indices rarely exceed 5%. Thus, by exploring for the first time the complex interactions among the severity of low-flow regimes, meteorological and hydrological drought indices and meteorological factors, this study provides a new perspective for understanding the characteristics of propagation from meteorological to severe hydrological drought.

Flood rank among the most destructive natural disasters, and their impact is being exacerbated by climate change and rapid global urbanization, which have led to a growing number of people residing in disaster-prone areas. Early assessment and mitigation of flood risks are crucial for reducing vulnerability. In response, there has been a marked increase in flood studies leveraging remote sensing and Geographic Information System (GIS) technologies. This review article aims to explore risk, flood susceptibility, and the knowledge gaps in flood hazard management, with a focus on applying machine learning techniques. It offers a comprehensive overview of published studies that utilize machine learning algorithms for spatial flood susceptibility mapping. Through an extensive global analysis, we compare various machine learning methods, highlighting their accuracy, strengths, and limitations based on specific evaluation criteria. Among these, tree-based ensemble algorithms consistently demonstrate superior performance compared to other machine learning approaches.

This study examines the temporal relationships between meteorological and agricultural drought indices using lagged and linear correlations, the Mann–Kendall trend test, and machine learning (random forest – RF and deep neural network – DNN). On a seasonal and annual scale, the results revealed that the resonance of agricultural drought is strongly synchronized with the temporal variability of meteorological drought. At the monthly scale, the resonance of agricultural drought reflected by the vegetation condition index and the soil moisture condition index (SMCI) has an obvious latency time of at least one month and is statistically significant up to three months. For both agricultural drought indices, their statistical relationships with meteorological drought indices are highly variable, depending on the month of the agricultural season, the time scale and the type of meteorological drought index. The correlations between the SMCI and Palmer drought severity index were the most stable. They ranged from 0.7 to 0.86, whereas the linear correlations between the SMCI and the precipitation conditions index varied from 0.5 to 0.16 in the first and last months of the agricultural season, respectively. Despite this high correlation variability, analysis of historical trends on an annual scale demonstrated the existence of obvious similarities of very negative trends in the spatiotemporal changes in agricultural and meteorological drought indices. Similarly, machine learning models highlighted the importance of the positive relative contribution of their joint occurrence to the annual variability in agricultural yields. Overall, the RF model achieved optimal performance with a relatively small number of predictors, whereas the DNN model was more dependent on the number of features used.

In semi-arid regions of the Mediterranean, snowmelt and precipitation are vital water sources for downstream communities. Here, snow-covered mountain peaks serve as natural water reservoirs, playing a crucial role in regulating river flow and replenishing groundwater. This research leverages remote sensing to compensate for the lack of ground-based hydroclimatic data, focusing on the latest version of the MODIS snow cover product (version 6, V6). The study aims to refine the Normalized Difference Snow Index (NDSI) threshold and develop localized models for fractional snow cover (FSC) estimation tailored to the Moroccan Atlas Mountains. For this purpose, 448 Sentinel-2 scenes across six different regions in the Atlas Mountains were used to adjust the NDSI threshold and develop FSC models. Moreover, 8419 MOD10A1 and 7561 MYD10A1 images covering the period from March 2000 to June 2023 were processed to improve cloud filtering and generate a high-precision daily snow cover product for the region. Significant improvements were achieved in reducing cloud-covered pixels from 25.7% to 0.4%. Two NDSI MODIS threshold selection schemes were tested: the standard global threshold of 0.4 and a locally optimized threshold of 0.2. The local threshold demonstrated superior accuracy, significantly reducing snow cover estimation errors compared with the global threshold (0.4) for both Terra and Aqua MODIS images. The newly developed FSC models demonstrate high accuracy, displaying high correlation coefficients (average of 0.84) and low error measures when comparing MODIS-derived FSCs with high-resolution Sentinel-2 data. The improved daily snow cover product was compared with high-resolution snow maps obtained from Sentinel-2 satellite imagery in different regions of the Moroccan Atlas. On average, the product showed a mean correlation coefficient of 0.96, a mean absolute error of 0.22%, and a mean reasonable negative bias of -0.17%. This research concludes that the improved daily snow cover product offers a robust understanding of the spatio-temporal dynamics of snow extent. These advancements offer considerable potential improvements to modelling snowmelt contribution to the water balance, supporting efficient water resource management in the southern Mediterranean region.

Floods are the most common natural hazard, causing major economic losses and severely affecting people’s lives. Therefore, accurately identifying vulnerable areas is crucial for saving lives and resources, particularly in regions with restricted access and insufficient data. The aim of this study was to automate the identification of flood-prone areas within a data-scarce, mountainous watershed using remote sensing (RS) and machine learning (ML) models. In this study, we integrate the Normalized Difference Flood Index (NDFI), using Google Earth Engine to generate flood inventory, which is considered a crucial step in flood susceptibility mapping. Seventeen determining factors, namely, elevation, slope, aspect, curvature, the Stream Power Index (SPI), the Topographic Wetness Index (TWI), the Topographic Ruggedness Index (TRI), the Topographic Position Index (TPI), distance from roads, distance from rivers, stream density, rainfall, lithology, the Normalized Difference Vegetation Index (NDVI), land use, length slope (LS) factor, and the Convergence Index were used to map the flood vulnerability. This study aimed to assess the predictive performance of gradient boosting, AdaBoost, and random forest. The model performance was evaluated using the area under the curve (AUC). The performance assessment results showed that random forest (RF) achieved the highest accuracy (1), followed by random forest and gradient boosting ensemble (RF-GB) (0.96), gradient boosting (GB) (0.95), and AdaBoost (AdaB) (0.83). Additionally, in this research study, we employed the Shapely Additive Explanations (SHAP) method, to explain machine learning model predictions and determine the most contributing factor in each model. This study introduces a novel approach to generate flood inventory, providing significant insights into flood susceptibility mapping, and offering potential pathways for future research and practical applications. Overall, the research emphasizes the need to integrate urban planning with emergency preparedness to build safer and more resilient communities.

SnowMapPy is a Python-based package developed to streamline the collection, preparation, and analysis of MODIS NDSI data, specifically from the Terra and Aqua satellite products. By automating essential steps (data clipping, reprojection, filtering, and time series generation), SnowMapPy improves the efficiency and precision of snow hydrology research. The protocol allows users to work with both local and Google Earth Engine cloud-based datasets, enabling flexible data acquisition and processing tailored to the needs of snow hydrology, water resource management, and climate change studies. Designed for accessibility and flexibility, SnowMapPy supports large-scale, high-resolution snow cover analysis with minimal configuration. The package facilitates customized workflows through its modular structure, making it a valuable tool for researchers aiming to understand snow dynamics and their impact on seasonal water resources.

Study regions: The study area encompasses two distinct sub-basins within the High Atlas Mountains: Oukaimeden in the Rheraya and Tichki in the Mgoun Valley. Study focus: The research integrates remote sensing data, particularly the Normalized-Difference Snow Index (NDSI) from the MODIS Sensor, with machine learning (ML) and deep learning (DL) models to predict daily snow depth (DSD) at a local scale. The models evaluated include two ML approaches: Support Vector Regression (SVR) and eXtreme Gradient Boosting (XGBoost) and four DL models: 1-Dimensional Convolutional Neural Network (1D-CNN), Long Short-Term Memory Networks (LSTM), Gated Recurrent Unit (GRU), and Bi-directional Long Short-Term Memory Network (Bi-LSTM). The dataset was processed and normalized for optimal performance, and hyperparameters were fine-tuned using a randomized search method. New hydrological insights for the region: The Results highlight the efficacy of AI-based approaches for snow depth prediction, with SVR achieving the best performance (Root Mean Square Error of 2-5 cm and an average coefficient of determination of 0.97). This study reveals that incorporating lag times of snow depth data significantly enhances predictive accuracy. These findings underscore the potential of integrating remote sensing with AI techniques to improve hydrological modeling and water resource planning in data-scarce regions like the Atlas Mountains.

The lack of knowledge about the temporal variability, or snow cover phenology and its spatial variation poses enormous challenges to water resource managers who mostly rely on a few weather stations with limited spatial coverage which prevents them from having a complete understanding of snow changes as a whole. Meanwhile, the free availability, wide-coverage, frequent updating, and long-term time horizon make data from programs such as Landsat and Sentinel-2 a valuable data source for reliable snow data information at an unprecedented spatial scale.In this context, this research aims to derive the snow phenology parameters (first day of snowfall, last day of snow melt; and snow duration) over Morocco’s Atlas Mountains by combining over 10,000 images from Landsat-8 and Sentinel-2 satellites for four hydrological years (2016-2021) to create a harmonized product with a time interval of about 3 days using Google Earth Engine platform. The time series produced allowed us to create detailed maps of snow cover and extract a homogeneous normalized difference snow index (NDSI) profile over the four years whereby we were able to determine the optimal threshold to separate the presence of snow from its absence. The results showed that derived seasonality snow metrics provide considerable variation in both time and space, where an increase in snowpack measurement values at higher elevations can be observed. The experimental results demonstrate that the proposed workflow can accurately derive snow seasonality timing with almost a day and a half delay than the in-situ observed dates and with an overall accuracy equal to 0.96. We expect these results to benefit various applications such as hydrological modeling, natural hazards, and regional climate change studies.

Data-driven methods, such as machine learning (ML) and deep learning (DL), play a pivotal role in advancing the field of snow hydrology. These techniques harness the power of algorithms to analyze and interpret vast datasets, allowing researchers to uncover intricate patterns and relationships within the complex processes of snow dynamics. In snow hydrology, where traditional models may struggle to capture the nonlinear and dynamic nature of snow-related phenomena, data-driven methods provide a valuable alternative. Using data-driven methods (ML and DL) requires advanced skills in various fields, such as programming and hydrological modeling. In response to these challenges, we have developed an open-source Python package named MorSnowAI that streamlines the process of building, training and testing artificial intelligence models based on machine learning and deep learning methods. MorSnowAI not only automates the building, training, and testing of artificial intelligence models but also significantly simplifies the collection of data from various sources and formats, such as reanalyzing datasets (ERA5-Land) from Copernicus Climate Data and remote sensing data from Modis, Landsat, and Sentinel datasets to calculate Normalized Difference Snow Index (NDSI). It can also utilize local datasets as inputs for the model. Among other features available in the MorSnowAI package, it provides pre-processing and post-processing methods that users can choose, along with visualization and analysis of the available time series. The scripts developed in the MorSnowAI package have already undergone evaluation and testing in various snow hydrology applications. For instance, these applications include predicting snow depth, streamflow, snow cover, snow water equivalent, and groundwater levels in mountainous areas of Morocco. The automated processes within MorSnowAI contribute to advancing the field, enabling researchers to focus on refining model inputs, interpreting results, and improving the overall understanding of complex hydrological systems. By bridging the gap between hydrology and advanced data-driven techniques, MorSnowAI fosters advancements in research, offering valuable insights for resource management in regions heavily influenced by snow dynamics.