SmartScale: AI-Driven Cyclone Classification to Support Climate-Resilient and Multi-Scalar Spatial Planning Frameworks

TY - GEN

T1 - SmartScale

T2 - AI-Driven Cyclone Classification to Support Climate-Resilient and Multi-Scalar Spatial Planning Frameworks

AU - Forghaniallahabadi, Maryam

PY - 2025

Y1 - 2025

N2 - The increasing frequency and intensity of tropical cyclones under accelerating climate change present significant challenges for spatial planning systems across local, regional, and national scales. Effective planning requires timely, accurate, and scalable intelligence on cyclone behaviour; however, current planning frameworks rarely integrate data-driven hazard classification in ways that support multi-scalar decision-making. This paper introduces SmartScale, an AI-driven approach that classifies cyclone events using historical meteorological parameters and demonstrates how these classifications can directly inform climate-resilient spatial planning. Using the NOAA HURDAT2 dataset (1851–2014; n = 75,242 observations), I developed and evaluated two machine learning models i.e., Random Forest and a Deep Neural Network to classify 12 cyclone categories based on wind speeds, pressure, and quadrant-based wind radii. The Random Forest model achieved a validation accuracy of 0.906, outperforming the neural network (test accuracy = 0.881). Performance was strongest for high-frequency classes, with hurricanes (HU), tropical storms (TS), and tropical depressions (TD) classified with near-perfect precision and recall. Confusion matrices reveal that these well-performing classes show clear meteorological separation, while misclassification of rare classes reflects inherent dataset imbalance. The findings demonstrate that cyclone intensity and type can be predicted reliably using interpretable machine learning techniques. We argue that this capability can be embedded within a multi-scalar spatial planning framework, enabling national hazard zoning, regional emergency coordination, and local land-use and building regulation to draw from the same classification intelligence. SmartScale therefore provides a replicable pathway for integrating AI-enabled hazard assessment into spatial planning, strengthening climate resilience across spatial scales.

AB - The increasing frequency and intensity of tropical cyclones under accelerating climate change present significant challenges for spatial planning systems across local, regional, and national scales. Effective planning requires timely, accurate, and scalable intelligence on cyclone behaviour; however, current planning frameworks rarely integrate data-driven hazard classification in ways that support multi-scalar decision-making. This paper introduces SmartScale, an AI-driven approach that classifies cyclone events using historical meteorological parameters and demonstrates how these classifications can directly inform climate-resilient spatial planning. Using the NOAA HURDAT2 dataset (1851–2014; n = 75,242 observations), I developed and evaluated two machine learning models i.e., Random Forest and a Deep Neural Network to classify 12 cyclone categories based on wind speeds, pressure, and quadrant-based wind radii. The Random Forest model achieved a validation accuracy of 0.906, outperforming the neural network (test accuracy = 0.881). Performance was strongest for high-frequency classes, with hurricanes (HU), tropical storms (TS), and tropical depressions (TD) classified with near-perfect precision and recall. Confusion matrices reveal that these well-performing classes show clear meteorological separation, while misclassification of rare classes reflects inherent dataset imbalance. The findings demonstrate that cyclone intensity and type can be predicted reliably using interpretable machine learning techniques. We argue that this capability can be embedded within a multi-scalar spatial planning framework, enabling national hazard zoning, regional emergency coordination, and local land-use and building regulation to draw from the same classification intelligence. SmartScale therefore provides a replicable pathway for integrating AI-enabled hazard assessment into spatial planning, strengthening climate resilience across spatial scales.

KW - AI-driven hazard classification

KW - Machine learning

KW - Climate-resilient spatial planning

KW - SmartScale framework

KW - Multi-scalar planning framework

KW - Environmental modelling

KW - Artifical intelligence

UR - https://github.com/mforghani775-creator/SmartScale-AI-Driven-Cyclone-Classification-to-Support-Climate-Resilient-/blob/main/SmartScale_AI_Based_Cyclone_Classification.ipynb

U2 - 10.20933/100001469

DO - 10.20933/100001469

M3 - Other contribution

PB - University of Dundee

ER -