Deep learning models for drought susceptibility mapping in Southeast Queensland, Australia

• Published in: Stochastic environmental research and risk assessment Vol. 39; no. 10; pp. 4849 - 4865
• Main Authors: Rezaie, Fatemeh, Panahi, Mahdi, Jun, Changhyun, Dayal, Kavina, Kim, Dongkyun, Darabi, Hamid, Kalantari, Zahra, Seifollahi-Aghmiuni, Samaneh, Deo, Ravinesh C., Bateni, Sayed M.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2025; Springer Nature B.V
• Subjects: Agricultural drought; Agricultural production; Algorithms; Annual precipitation; Aquatic Pollution; Artificial neural networks; Availability; Chemistry and Earth Sciences; Clay; Climate change; Computational Intelligence; Computer Science; Convolutional neural network; Deep learning; Drought; Earth and Environmental Science; Earth Sciences; Environment; Environmental impact; Food security; Long short-term memory; Machine learning; Math. Appl. in Environmental Science; Neural networks; Original Paper; Physics; Precipitation; Probability Theory and Stochastic Processes; Rain; Root-mean-square errors; Runoff; Sensitivity analysis; Socioeconomics; Soil depth; Soil moisture; Statistics for Engineering; Stream flow; Topography; Variables; Vegetation; Waste Water Technology; Water availability; Water Management; Water Pollution Control; Water shortages; Weather hazards; Queensland Australia; Australia; Deep learning; Drought; Soil moisture; Food security; Convolutional neural network
• ISSN: 1436-3240; 1436-3259; 1436-3259
• DOI: 10.1007/s00477-024-02879-w

Drought is a global phenomenon with significant negative impacts on water availability, agricultural production, livelihoods, and socioeconomic conditions. Despite its destructive effects, spatially predicting drought hazards remains a challenging task. This study developed an innovative framework by leveraging two state-of-the-art deep learning models: convolutional neural networks (CNNs) and the long short-term memory (LSTM) model. Key predictive factors, including the topographic wetness index, soil depth, mean annual precipitation, elevation, slope, sand content, clay content, and plant-available water-holding capacity (PAWC), were carefully selected for analysis. An agricultural drought inventory map was generated based on the relative departure of soil moisture. The performance of the CNN and LSTM models was evaluated using root mean square error (RMSE), standard deviation (StD), and the area under the receiver operating characteristic curve (AUC). The results indicated that certain parts of the research area were highly susceptible to drought. Both models performed well, achieving AUC values of 81.9% (CNN) and 81.7% (LSTM). The RMSE and StD further confirmed the predictive capabilities of these models. Sensitivity analyses highlighted the importance of PAWC, mean annual precipitation, and clay fraction in detecting drought-prone areas. The drought susceptibility map provides valuable insights into the vulnerability and likelihood of an area experiencing drought conditions, offering essential information for decision-makers to effectively prioritize resources and mitigate drought impacts.