Ensemble learning for spatiotemporal data: methods, applications, challenges and opportunities

2026-06-03

PRE

Wenhui Zhang

Lizeng Wang

Shifen Cheng *

Feng Lu *

DOI：10.1080/13658816.2026.2681076

Original

Original request for help

Save

Abstract

En 中文

Ensemble learning (EL), one of the most popular modern machine learning technologies, has been increasingly adopted in spatiotemporal data analysis. Spatiotemporal dependence, heterogeneity and scale effects have motivated the development of spatiotemporal ensemble learning (ST-EL). However, current literature lacks a systematic review, as existing surveys remain largely application-specific. This paper addresses these gaps by framing the review around how spatiotemporal properties reshape EL and provides a methodological roadmap toward trustworthy and scalable ST-EL systems. We traced the evolution of ST-EL along two complementary dimensions: (i) base learners, from homogeneous to heterogeneous designs and (ii) ensemble strategies, from global and static aggregation to spatiotemporally adaptive schemes. Existing approaches were then discussed around three analytical levels: spatiotemporal-enhanced input representation, spatiotemporal-aware base learner design and training, and spatiotemporal-informed ensemble strategies. Key tasks and applications were also reviewed with associated code resources. Finally, open issues were discussed on bias evaluation, base learner selection, uncertainty quantification and model interpretability. Future directions were outlined from multi-scale modeling, multi-source data fusion, model recalibration, to the integration with spatiotemporal foundation models.

Keywords:

Ensemble learning

spatiotemporal data

spatiotemporal properties

geospatial intelligence

machine learning

Ensemble learning for spatiotemporal data: methods, applications, challenges and opportunities

Abstract

Journal

Organization

Cited Papers

Citing Papers