Bibliometric Analysis of Research Trends and Knowledge Structure of Machine Learning Applications in Drought Prediction

Abstract

Drought, one of the most destructive natural disasters, profoundly impacts ecosystems, agricultural production, water resource security, and socio-economic development. In recent years, machine learning has emerged as a central approach to drought prediction, owing to its capacity to integrate multi-source data and improve accuracy. Using the Web of Science Core Collection and VOSviewer, this study conducts bibliometric analyses to trace research development, clarify knowledge structures, and identify emerging directions of machine learning in drought prediction from 1999 to 2025. Results show a steady upward trend in publications and citations, with rapid growth since 2019, reflecting sustained global scholarly attention. Analysis of the productivity of countries, institutions, journals, and authors reveals that China and the USA lead in research output, supported by several highly productive institutions and widely recognized journals. Author keyword co-occurrence analysis identifies three thematic clusters: (1) traditional meteorological and hydrological drought studies focused on indices such as the SPI and the SPEI; (2) integrated frameworks combining machine learning with remote sensing; and (3) advanced deep learning approaches, including LSTM and Transformer models, incorporating variables such as soil moisture, precipitation, and runoff, along with climate drivers such as the ENSO and extreme events. Overall, these results demonstrate a transition from single-index models to integrated prediction systems combining diverse datasets with advanced algorithms, alongside increasing focus on climate-drought interactions and extremes. These findings guide enhancing model interpretability, advancing real-time early warning, and supporting drought management under climate change.