ارزیابی خطر سیلاب با استفاده از سیستم اطلاعات جغرافیایی رویکردی نوین در مدیریت بحران (مورد مطالعه: روستاهای شهرستان نیشابور)

Introduction

Hydro-meteorological hazards, particularly floods, constitute a profound threat to rural environments, disproportionately affecting agricultural livelihoods, infrastructure, and human life. In the context of disaster risk reduction (DRR) and rural spatial planning, accurate flood susceptibility mapping is an imperative prerequisite for mitigating vulnerabilities. Neyshabur County, characterized by a complex topographic configuration and intricate hydrological networks, frequently experiences catastrophic inundations. Despite the escalating frequency of extreme precipitation events driven by climate variability, micro-level hazard zoning for the county’s rural settlements remains critically underdeveloped. Consequently, existing crisis management frameworks lack the granular spatial intelligence required for proactive intervention. To bridge this epistemological and operational gap, this study introduces a robust methodological framework integrating Geographic Information Systems (GIS) with advanced ensemble machine learning—specifically the Gradient Boosting (GB) algorithm. By localizing multi-criteria evaluation metrics to the geomorphological realities of Neyshabur, this research aims to precisely delineate flood-prone rural territories, thereby providing empirical foundations for sustainable rural planning, spatial resilience, and strategic crisis management.
 

Methodology

This research adopts an applied, descriptive-analytical paradigm, synthesizing extensive documentary, geospatial, and empirical field data. The methodological architecture is predicated on evaluating spatial vulnerability across rural settlements using seven geo-environmental and hydrological indicators: elevation, slope, aspect, precipitation distribution, proximity to waterways, geological formations, and land-use typologies. High-resolution spatial datasets were procured from diverse authoritative sources, including Digital Elevation Models (DEM) derived from ALOS PALSAR, multispectral satellite imagery from Sentinel-2, empirical data from regional meteorological stations, and standard geological maps. To model flood susceptibility, the Gradient Boosting algorithm was selected due to its exceptional predictive accuracy in handling non-linear, multi-dimensional spatial datasets. The computational workflow commenced with rigorous data normalization, followed by partitioning the dataset into a 70% training subset and a 30% validation subset. Hyperparameter optimization was systematically executed utilizing a Grid Search cross-validation technique to maximize model generalization. The predictive efficacy and robust classification capabilities of the algorithm were empirically evaluated using standardized metrics, including Precision, Recall, Cohen’s Kappa coefficient, and Root Mean Square Error (RMSE). Furthermore, a Feature Importance analysis was embedded within the model to quantify the hierarchical influence of the localized independent variables on flood occurrence.
 

Findings

The spatial application of the Gradient Boosting model yielded highly precise flood hazard zonations across the study area. Predictive performance metrics validated the model’s exceptional reliability, registering a Kappa coefficient of 0.91and an RMSE of 0.19. The vulnerability assessment of rural settlements revealed alarming exposure levels: out of the analyzed settlements, 80 villages (comprising 20% of the total) are situated in “critical” hazard zones, while an additional 140 villages (35%) are located in “high-risk” zones. Consequently, a staggering 55% of the rural settlements in Neyshabur County are critically exposed to severe flood threats. Spatial correlations extracted from the model demonstrated that settlements situated within a <200 meter radius of primary waterways historically sustained the most devastating structural and economic damages due to rapid inundation. The Feature Importance analysis elucidated that topographical attributes (specifically elevation and slope) and precipitation intensity are the primary causative drivers of flood susceptibility. Anthropogenic land-use configurations and proximity to the hydrological network ranked as the most significant secondary determinants, highlighting the detrimental impact of unregulated rural encroachment into natural floodplains.
 

Discussion and Conclusion

The findings of this study unequivocally demonstrate that the convergence of topographical predisposition and unsustainable land-use practices has drastically amplified flood risks in the rural landscapes of Neyshabur County. The hybrid GIS and Gradient Boosting approach proved highly efficacious in identifying spatial vulnerabilities, offering a paradigm shift from reactive to proactive disaster management. The concentration of 55% of rural settlements in high to critical hazard zones necessitates immediate recalibration of regional rural planning policies. From a spatial planning perspective, it is evident that historical expansion patterns have severely encroached upon riparian buffer zones. To foster systemic resilience, it is strongly recommended that a statutory, non-negotiable safety buffer of at least 500 meters from major rivers and ephemeral waterways be institutionalized for all future rural development, infrastructure provisioning, and post-disaster reconstruction efforts. Furthermore, policymakers must prioritize the 80 critical-class villages for immediate structural interventions, including the construction of hydrological retention systems, ecological restoration of upstream watersheds, and the implementation of community-based early warning systems. Ultimately, integrating localized machine learning predictions into municipal master plans will empower rural planners to optimize land-use allocation, mitigate environmental hazards, and safeguard the socio-economic fabric of vulnerable rural communities.