The rapid growth of spatial data across various fields has made data-driven machine learning (ML) methods increasingly essential for spatial analysis. The performance of ML methods heavily depends on hyperparameter tuning (HPT), which becomes particularly challenging when dealing with spatial stratified heterogeneity (SSH) that causes significant variability in statistical characteristics across sub-regions and demands separate local models. Traditional HPT methods either apply uniform hyperparameters or treat each model independently, ignoring spatial associations between adjacent models. To address this gap, we propose a geographic evolutionary (GeoEvo) framework with multi-task optimization to account for spatial associations in collaborative HPT of local ML models under SSH. GeoEvo formulates the problem of a single objective optimization with multi-task constraints to jointly optimize hyperparameters across multiple local models. The framework introduces an evolution operator for geographic proximity differential (Geo-DE) to enable collaboration among spatially adjacent models and a geographic selection operator (Geo-SL) to promote hyperparameter sharing, improve resource utilization and accelerate convergence. Extensive experiments based on soil organic carbon stocks and PM2.5 concentration datasets demonstrated that GeoEvo enhanced accuracy and stability while maintaining computational efficiency. Subsequent visual analytics revealed that considering the diversity and spatial correlations within the hyperparameter space enhanced the optimization process and improved prediction accuracy.

Although numerous models have been proposed to predict the intensity of human activities in urban areas, two major issues hamper the performance of existing models: (1) fail to incorporate appropriate prior knowledge instrumental for improving accuracy and interpretability; (2) fail to integrate probabilistic and deterministic predictions to achieve complementary strengths, namely uncertainty quantification and high predictive accuracy. To address these challenges, we proposed a prior-enhanced dual-mode spatiotemporal graph neural network (PED-STGNN) to support both probabilistic and deterministic predictions. Specifically, we introduced a hypergraph node-to-vector (hypernode2vec) method to capture the multivariate functional similarity prior derived from complex and multivariate relations between urban regions. This functional similarity characterizes urban systems more precisely than existing methods relying on first-order pairwise relations. It improves accuracy and interpretability while enabling spatial modeling of higher-order multivariate relations beyond first-order pairwise relations. We also designed a plug-and-play probabilistic prediction module that enables switches between probabilistic and deterministic modes. Experiments based on the human activity intensity in Fuzhou, China, demonstrated the advantages in accuracy, interpretability and multi-scenario applicability.

Monitoring real-time air quality is essential for safeguarding public health and fostering social progress. However, the widespread deployment of air quality monitoring stations is constrained by their significant costs. To address this limitation, we introduce AirRadar, a deep neural network designed to accurately infer real-time air quality in locations lacking monitoring stations by utilizing data from existing ones. By leveraging learnable mask tokens, AirRadar reconstructs air quality features in unmonitored regions. Specifically, it operates in two stages: first capturing spatial correlations and then adjusting for distribution shifts. We validate AirRadar’s efficacy using a year-long dataset from 1,085 monitoring stations across China, demonstrating its superiority over multiple baselines, even with varying degrees of unobserved data.

Spatial interpolation is essential for handling sparsity and missing spatial data. Current machine learning-based spatial interpolation methods are subject to the statistical constraints of spatial stratified heterogeneity (SSH), normally involving separate modeling of each stratum and simple weighted averaging to integrate intra-stratum and inter-strata features. However, these models overlook the different contributions of inter-strata features to different locations within a stratum (heterogeneous inter-strata associations, HIA) and the explanation of spatial effects on the interpolation process, leading to suboptimal and unreliable interpolation outcomes. This article proposes a novel explainable spatial interpolation method considering SSH (X-SSHM). Spatial and environmental features are utilized to describe intra-stratum and inter-strata information, which are fed into random forest-based learners to achieve high-level semantic feature mapping. Geographically weighted regression is employed to integrate intra-stratum and inter-strata features to achieve a unified expression of SSH and HIA, obtaining the final interpolation result. Geographically weighted Shapley (GSHAP) is proposed to decompose the marginal contributions of intra-stratum and inter-strata features. Model performance is evaluated on simulated and soil organic matter datasets. X-SSHM outperformed five baselines regarding interpolation accuracy. Moreover, statistical methods validated X-SSHM’s ability to elucidate the mechanisms by which SSH, spatial autocorrelation and HIA affect the model interpolation process.

Accurate and reliable traffic flow data are essential for intelligent transportation systems; however, limitations arising from hardware and communication costs often lead to missing data. Tensor decomposition is widely used to address these issues. However, existing imputation methods employ a fixed geographic feature similarity matrix to constrain the tensor decomposition process, which fails to accurately capture the spatial heterogeneity of traffic flows, thus limiting the imputation accuracy and robustness. This study proposes a tensor decomposition method embedded with geographic meta-knowledge (Meta-TD) to accurately determine the spatial heterogeneity of traffic flows. The key innovation is establishing a dynamic relationship between the geographic meta-knowledge and spatial heterogeneity of traffic flows, and then using the spatial heterogeneity of the traffic flows to constrain the tensor decomposition process. Experimental results based on real urban traffic flows demonstrated the superiority of Meta-TD over fifteen baseline models under random, block, and long time-series missing patterns, achieving reductions in MAE, RMSE, and MAPE of 6.97–97.05%, 3.33–94.68%, and 0.72–90.89%, respectively. Notably, Meta-TD maintained high accuracy for sudden changes in traffic flow states, evidencing its robustness to varying missing data rates and distribution patterns. This adaptability makes it highly suitable for complex and dynamic urban traffic environments.

Spatiotemporal learning plays a crucial role in mobile computing techniques to empower smart cites. While existing research has made great efforts to achieve accurate predictions on the overall dataset, they still neglect the significant performance heterogeneity across samples. In this work, we designate the performance heterogeneity as the reason for unfair spatiotemporal learning, which not only degrades the practical functions of models, but also brings serious potential risks to real-world urban applications. To fix this gap, we propose a model-independent Fairness-aware framework for SpatioTemporal Graph learning (FairSTG), which inherits the idea of exploiting advantages of well-learned samples to challenging ones with collaborative mix-up. Specifically, FairSTG consists of a spatiotemporal feature extractor for model initialization, a collaborative representation enhancement for knowledge transfer between well-learned samples and challenging ones, and fairness objectives for immediately suppressing sample-level performance heterogeneity. Experiments on four spatiotemporal datasets demonstrate that our FairSTG significantly improves the fairness quality while maintaining comparable forecasting accuracy. Case studies show FairSTG can counter both spatial and temporal performance heterogeneity by our sample-level retrieval and compensation, and our work can potentially alleviate the risks on spatiotemporal resource allocation for underrepresented urban regions.

Ensemble learning synthesizes the advantages of different models and has been widely applied in the field of spatial prediction. However, the nonlinear constraints of spatial heterogeneity on the model ensemble process make it difficult to adaptively determine the ensemble weights, greatly limiting the predictive ability of the ensemble learning model. This paper therefore proposes a novel geographical spatial heterogeneous ensemble learning method (GSH-EL). Firstly, the geographically weighted regression model, geographically optimal similarity model, and random forest model are used as three base learners to express local spatial heterogeneity, global feature correlation, and nonlinear relationship of geographic elements, respectively. Then, a spatially weighted ensemble neural network module (SWENN) of GSH-EL is proposed to express spatial heterogeneity by exploring the complex nonlinear relationship between the spatial proximity and ensemble weights. Finally, the outputs of the three base learners are combined with the spatial heterogeneous ensemble weights from SWENN to obtain the spatial prediction results. The proposed method is validated on the PM2.5 air quality and landslide dataset in China, both of which obtain more accurate prediction results than the existing ensemble learning strategies. The results confirm the need to accurately express spatial heterogeneity in the model ensemble process.

Human mobility data play a crucial role in many fields such as infectious diseases, transportation, and public safety. Although the development of Information and Communication Technologies (ICTs) has made it easy to collect individual-level positioning records, raw individual trajectory data are still limited in availability and usability due to privacy issues. Developing models to generate synthetic trajectories that are statistically close to the real data is a promising solution. This study proposed a novel trajectory generation method called Act2Loc (Activity to Location), which combined machine learning and mechanistic models. First, an activity-sequence generation model was constructed based on machine learning models (i.e. K-medoids and Transformer) to generate individual activity sequences aligning with human activity patterns. Then, a spatial-location selection model was proposed based on mechanistic models (e.g. Universal Opportunity model) to explicitly determine the specific locations of the activities in each generated sequence. Experimental results showed that compared to baselines based on purely machine learning or mechanistic models, Act2Loc can better reproduce the spatio-temporal characteristics of the real data, with additional advantage of low data requirements for training, proving its potential for generating synthetic trajectories in practice. This research offers new insights on knowledge-guided GeoAI models for human mobility.

Explainable spatio-temporal prediction gains attraction in the development of geospatial artificial intelligence. The neural ordinal differential equation (NODE) emerges as a new solution for explainable spatio-temporal prediction. However, challenges still need to be solved in most existing NODE-based prediction models, such as difficulty modeling spatial data and mining long-term temporal dependencies in data. In this study, we propose a spatio-temporal attentional NODE (STA-ODE) to address the two challenges above. First, we define a spatio-temporal ordinary differential equation to predict a value at each time iteratively by a novel spatio-temporal derivative network. Second, we develop an attention mechanism to fuse multiple prediction values for capturing long-term temporal dependencies in data. To train the STA-ODE model, we design a loss function that aligns the prediction results in spatial dimension with prediction results in temporal dimension to calibrate the parameters of the model. The proposed model was validated with three real-world spatio-temporal datasets (traffic flow dataset, PM2.5 monitoring dataset, and temperature monitoring dataset). Experimental results showed that STA-ODE outperformed seven existing baselines regarding prediction accuracy. In addition, we used visualization to demonstrate the sound interpretability and prediction accuracy of the STA-ODE model.

Identifying road freight cargo types is crucial for regional economic interaction and transportation optimization. Existing methods primarily rely on manual labeling and the rule, neither of which can achieve automated semantic enhancement of large-scale road freight trajectories. Consequently, this study proposes a semi-supervised trajectory semantic enhancement method for identifying cargo types based on trajectory feature extraction and point-of-interest (POI) association. The raw trajectories are segmented and enriched with the closest POIs. The sample labeling method with POI semantic enhancement is then proposed using company registration information. Finally, the spatiotemporal and sequential features of labeled freight trips are extracted to build a self-training semi-supervised model for identifying the cargo type of road freight. Experimental studies on real trajectory data demonstrate superior accuracy and robustness compared to existing methods, with accuracy and F1 values reaching 81.4 and 0.77%, respectively. The proposed sample labeling method improves representativeness and universality, increasing accuracy by 7.8–14.4% and F1 value by 8.5–34.5% compared to the rule-based method. The semi-supervised model improves accuracy by 8.9% and F1 value by 29.1% compared to the supervised model when only 10.0% of samples were labeled. This method enables automatic and full-sample cargo type identification in real-world large-scale transportation systems.

Short-term traffic forecasting is important for the development of an intelligent traffic management system. Critical to the performance of the traffic prediction model utilized in such a system is accurate representation of the spatiotemporal traffic characteristics. This can be achieved by integrating spatiotemporal traffic information or the dynamic traffic characteristics in the modeling process. The currently employed spatiotemporal k-nearest neighbor (STKNN) model is based on the spatial heterogeneity and adaptive spatiotemporal parameters of the traffic to improve the prediction accuracy. However, the non-stationary characteristics of the traffic cannot be fully represented by simply modeling the entire time range or all the time partitions based on experience. We therefore developed a dynamic STKNN model (D-STKNN) for short-term traffic forecasting based on the non-stationary spatiotemporal pattern of the road traffic. The different traffic patterns along the road are first automatically determined using an affinity propagation clustering algorithm. The Warped K-Means algorithm is then used to automatically partition the time periods for each traffic pattern. Finally, the D-STKNN model is developed based on the three-dimensional spatiotemporal tensor data models for the different road segments with different traffic patterns during different time periods. The D-STKNN model was verified through extensive experiments performed using actual vehicular speed datasets collected from city roads in Beijing, China, and expressways in California, U.S.A. The proposed model outperforms existing seven baselines in different time periods under different traffic patterns. The results confirmed the imperative of considering the non-stationary spatiotemporal traffic pattern in developing a model for short-term traffic prediction.

Missing data is a common problem in the analysis of geospatial information. Existing methods introduce spatiotemporal dependencies to reduce imputing errors yet ignore ease of use in practice. Classical interpolation models are easy to build and apply; however, their imputation accuracy is limited due to their inability to capture spatiotemporal characteristics of geospatial data. Consequently, a lightweight ensemble model was constructed by modelling the spatiotemporal dependencies in a classical interpolation model. Temporally, the average correlation coefficients were introduced into a simple exponential smoothing model to automatically select the time window which ensured that the sample data had the strongest correlation to missing data. Spatially, the Gaussian equivalent and correlation distances were introduced in an inverse distance-weighting model, to assign weights to each spatial neighbor and sufficiently reflect changes in the spatiotemporal pattern. Finally, estimations of the missing values from temporal and spatial were aggregated into the final results with an extreme learning machine. Compared to existing models, the proposed model achieves higher imputation accuracy by lowering the mean absolute error by 10.93 to 52.48% in the road network dataset and by 23.35 to 72.18% in the air quality station dataset and exhibits robust performance in spatiotemporal mutations.

Fine-grained prediction of urban population is of great practical significance in many domains that require temporally and spatially detailed population information. However, fine-grained population modeling has been challenging because the urban population is highly dynamic and its mobility pattern is complex in space and time. In this study, we propose a method to predict the population at a large spatiotemporal scale in a city. This method models the temporal dependency of population by estimating the future inflow population with the current inflow pattern and models the spatial correlation of population using an artificial neural network. With a large dataset of mobile phone locations, the model’s prediction error is low and only increases gradually as the temporal prediction granularity increases, and this model is adaptive to sudden changes in population caused by special events.