Time Series and Spatial Modelling

Time series and spatial modelling encompass statistical and computational frameworks for analysing data that vary over time or across geographic space. Time series approaches characterise temporal dependencies, trends and seasonality in sequential observations, supporting tasks such as forecasting, anomaly detection and causal inference. Key methods range from classical autoregressive and state-space models to modern machine-learning architectures that handle irregular sampling, multivariate interactions and long-range dependencies. Spatial modelling addresses the representation, aggregation and interpretation of phenomena distributed over continuous or discrete geographic units. Challenges include spatial autocorrelation, non-stationarity and the modifiable areal unit problem (MAUP), in which results depend on the choice of spatial aggregation. Contemporary spatial methods extend regression to account for spatial lag and error, employ hierarchical and Bayesian frameworks to capture multiscale variation, and leverage graph-based and clustering techniques to define adaptive spatial units. When combined as spatio-temporal models, these approaches enable the joint analysis of how processes evolve in both dimensions, underpinning applications in environmental monitoring, public health, urban planning and beyond.

Research from Nature Portfolio

An automated tool for flexible aggregation of adjacent spatial units has been introduced to satisfy minimum area or attribute requirements while mitigating boundary-induced bias. This algorithm dissolves neighbouring polygons according to user-defined criteria, updating attributes to preserve data fidelity and seamlessly integrating heterogeneous resolutions. Applications include fine-scale pollution exposure estimation by merging global grids with local administrative boundaries, and delineation of metropolitan extents through population-driven municipality aggregation.

A novel variant of the Self-Organizing Map, driven by Dynamic Time Warping, has been proposed for unsupervised pattern discovery in highly time-resolved environmental exposure data. By optimally aligning temporal features, this method uncovers detailed diurnal patterns in indoor and outdoor temperature and particulate matter time series. It outperforms traditional map algorithms by reducing quantisation error and revealing exposure profiles that may correlate with health outcomes.

Research from all publishers

A simulated spatial testbed of high-resolution areal units has been developed to interrogate MAUP effects systematically. By generating multiple levels of spatial resolution and numerous zonal configurations, this sandbox enables controlled studies of how scale and boundary design influence aggregation, disaggregation, model training and prediction accuracy. It also supports non-parametric estimation of confidence intervals via resampling.

A multi-scale population analysis framework constructs adaptive analysis units driven by scene features and temporal patterns in population activity. Through temporal decomposition and feature-based clustering, this method yields homogeneous regions that stabilise phase patterns in population time series and enhance interpretability in both routine and emergent urban scenarios.

In the time-series domain, a graph spatio-temporal process based on neural controlled differential equations has been devised for multivariate anomaly detection with missing values. Incorporating a distribution-based scoring mechanism, this approach models both spatial and temporal dependencies in irregularly sampled series, demonstrating robust outlier detection in power-grid and industrial control applications.

Time Series and Spatial Modelling publication trend

The graph below shows the total number of articles in time series and spatial modelling across all publications each year (not limited to Nature Index journals).

Technical terms

Time series: A sequence of observations recorded at successive time points, often equally spaced, used to model temporal dynamics.

Spatial autocorrelation: The tendency for values at nearby locations to be similar, violating independence assumptions and requiring specialised models.

Modifiable Areal Unit Problem (MAUP): Sensitivity of spatial analysis results to the scale and configuration of aggregation units, which can bias inference.

Dynamic Time Warping (DTW): A distance measure that aligns time series by non-linearly warping the time axis to minimise dissimilarity.

Self-Organizing Map (SOM): An unsupervised neural network that projects high-dimensional data onto a low-dimensional grid while preserving topological features.

References

1. A smart and flexible approach for aggregation of adjacent polygons to meet a minimum target area or attribute value. Scientific Reports (2023).
2. Using dynamic time warping self-organizing maps to characterize diurnal patterns in environmental exposures. Scientific Reports (2021).
3. A simulated ‘sandbox’ for exploring the modifiable areal unit problem in aggregation and disaggregation. Scientific Data (2024).
4. Multi-scale population analysis unit construction method considering scene feature variability and long/short-term patterns in spatiotemporal population activities. International Journal of Digital Earth (2024).
5. Graph spatiotemporal process for multivariate time series anomaly detection with missing values. Information Fusion (2024).