Despite recent developments in spatiotemporal deep learning, most of them have been on time series and spatiotemporal forecasting, far less attention has been paid to spatial/spatiotemporal interpolation.
In this repository, a list of papers (with available codes) related to spatial interpolation is given, which will be updated ASAP once the papers are announced in the corresponding AI conferences/journals. Hope this list can help those interested in AI for Spatial/Spatiotemporal Interpolation.
Welcome to contribute the related papers. Please open an issue or email me.
📧: jlidw[AT]connect.ust.hk
Spatial Interpolation; Spatial Extrapolation; Spatiotemporal Interpolation; Spatiotemporal Extrapolation;
Spatial Interpolation#1 is the traditional method for spatial estimation, which is a process of using values at observed locations to estimate values at unobserved ones in geographic space.
#1Some works may use another term extrapolation. Strictly speaking, interpolation and extrapolation are similar things but different ranges. Interpolation means predicting the values within the spatial range of the known locations, while extrapolation will predict the values outside the spatial range of the known locations. We can optionally use one of these two items when the range of predicted locations is not considered a constraint.
Here, we follow the geoscience1 to use the term interpolation, as all spatial interpolation methods can generate an extrapolation.
Spatiotemporal Interpolation#2 is an extension of spatial interpolation, which adds a time dimension to spatial data and estimates values at unobserved locations given the values from observed locations during a period.
#2Some AI works use Spatiotemporal Kriging to represent the spatiotemporal interpolation task, which is conceptually ambiguous. Kriging2 is a general term denoting a number of geostatistical techniques for interpolation. Essentially, kriging is a method rather than a task. Particularly, the term Spatiotemporal Kriging itself is a method of spatiotemporal interpolation based on spatiotemporal covariogram3, which is an extension of Kriging.
-
Spatial prediction is a larger term, including any method that incorporates spatial dependence to estimate variables at unknown locations.
-
Spatial interpolation is more specific and mainly uses point-based observed variables and sample coordinates to estimate variables at unknown locations.
Spatial interpolation is to “predict” data for any locations with no historical observations according to sparse station observations. This problem is fundamentally different and more challenging than multivariate time-series imputation, which assumes data at certain locations is partially missing across time.
Will add a table later.
- SSIN: Self-Supervised Learning for Rainfall Spatial Interpolation, in SIGMOD 2023. [Paper], [Code]
- Rainfall Spatial Interpolation with Graph Neural Networks, in DASFAA 2023. [Paper], [Code]
- Deep geometric neural network for spatial interpolation, in SIGSPATIAL 2022. [Paper], [Code]
- Kriging Convolutional Networks, in AAAI 2020. [Paper], [Code]
- Inductive and Adaptive Graph Convolution Networks Equipped with Constraint Task for Spatial–temporal Traffic Data Kriging, in KBS 2024. [Paper], [Code]
- Graph Neural Processes for Spatio-Temporal Extrapolation, in KDD 2023. [Paper], [Code]
- INCREASE: Inductive Graph Representation Learning for Spatio-Temporal Kriging, in WWW 2023. [Paper], [Code]
- Decoupling Long-and Short-Term Patterns in Spatiotemporal Inference, in TNNLS 2023. [Paper], [Code]
- Inductive Graph Neural Networks for Spatiotemporal Kriging, in AAAI 2021. [Paper], [Code]