spatPomp, an R package for statistical inference for spatiotemporal partially observed Markov processes

spatPomp concerns inference for nonlinear partially-observed Markov process models having a spatial unit structure in addition to the temporal Markovian property of the latent process. Such a model is called a SpatPOMP. For example, ecological metapopulation models consist of spatial units corresponding to distinct, interacting sub-populations. Each unit may itself consist of multiple interacting species. In spatPomp, models can be represented by specifying the latent dynamic process and how it is measured. The spatPomp package builds on pomp, and its base class, spatPomp extends the pomp class pomp. Therefore, all algorithms and methods provided by pomp are accessible in spatPomp. However, practical analysis of high-dimensional systems can take advantage of the additional unit structure that POMP models do not necessarily possess.

Algorithms currently provided by spatPomp include the following:

• Guided intermediate resampling filter, girf
• Block particle filter, bpfilter
• Adapted bagged filter, abf
• Ensemble Kalman filter, enkf
• Iterated GIRF for parameter estimation, igirf
• Iterated bpfilter for parameter estimation, ibpf
• Iterated ensemble Kalman filter for parameter estimation, ienkf

In addition, the particle filter provided by pomp as pfilter and the corresponding iterated filter, mif2, are useful for validating spatPomp models and workflows on small subsets the spatial units.

spatPomp is currently in development. All are welcome to contribute methods or models, or any other feedback. Please let the developers know if you find spatPomp useful and if you publish results obtained using it!

The latest development version of spatPomp is available on GitHub and versions are occasionally uploaded to CRAN. Other relevant resources are:

• A tutorial on spatiotemporal partially observed Markov process models via the R package spatPomp
• A tutorial on the iterated block particle filter

Papers using spatPomp include the following:

• Ionides, E. L., Asfaw, K., Park, J., and King, A. A. (2023). Bagged filters for partially observed interacting systems. Journal of the American Statistical Association, 118, 1078-1089. doi. arxiv. code.
• Ionides, E. L., Ning, N. and Wheeler, J. (2022). An iterated block particle filter for inference on coupled dynamic systems with shared and unit-specific parameters. Statistica Sinica, prepublished online. doi. arxiv. code.
• Li, J., E. L. Ionides, A. A., King, M. Pascual and N. Ning (2024). Inference on spatiotemporal dynamics for networks of biological populations. arxiv. code.
• Ning, N. and Ionides, E. L. (2023). Iterated block particle filter for high-dimensional parameter learning: Beating the curse of dimensionality. Journal of Machine Learning Research 24(82), 1-76. pdf. arxiv.
• Wheeler, J., Rosengart, A. L., Jiang, Z., Tan, K., Treutle, N. and Ionides, E. L. (2024). Informing policy via dynamic models: Cholera in Haiti. To appear in PLOS Computational Biology. arxiv. code. software.
• Zhang, B., Huang, W., Pei, S., Zeng, J., Shen, W., Wang, D., Wang, G., Chen, T., Yang, L., Cheng, P., Wang, D., Shu, Y., & Du, X. (2022). Mechanisms for the circulation of influenza A (H3N2) in China: A spatiotemporal modelling study. PLoS Pathogens, 18(12), e1011046. doi.