Global Stationarity Assumption
A single coefficient assumes the same relationship everywhere and at all times, potentially obscuring local health–environment mechanisms.
GLOBAL POSTERIOR · βBSTVC | Make Space and Time Explainable.
Turn spatiotemporal heterogeneityinto interpretable evidence.
Conventional global regression reduces complex spatiotemporal relationships to a single coefficient, while black-box GeoAI often explains predictions only after the fact. BSTVC (Bayesian Spatiotemporally Varying Coefficients) takes a different path—unifying local effects, global key-driver identification, dynamic prediction, and uncertainty quantification within a transparent Bayesian framework to reveal where, when, and why relationships change.
Local Effects.Global Insight.Dynamic Prediction.
The first open-source tool for spatiotemporal interpretability analysis.
3Response Types
Continuous · Binary · Count
Continuous · Binary · Count
50 / 95Bayesian Credible Intervals
Uncertainty-Informed Interpretation
Uncertainty-Informed Interpretation
12Empirical Studies
2018–Present · Health & Medical Geography
2018–Present · Health & Medical Geography
1 GBUpload Limit
Large Panels and Maps
Large Panels and Maps
01 / WHY BSTVC
Interpretability is more than
an importance ranking.
Where. When. Why.
Why does the same factor have different effects across places and periods?
Explore the GeoAI Paradigm ↗A global regression uses one coefficient for every place and period, creating a potential stationarity bias. BSTVC returns relationships to their geographic context while quantifying effect magnitude, direction, and uncertainty.
Spatiotemporally Nonstationary Relationships
Identify spatial and temporal heterogeneity simultaneously while preserving the geographic meaning of local relationships.
LOCAL COEFFICIENT FIELD · β(s, t)Evidence-Based Interpretation
Use credible intervals, model assessment, and STVPI show effect magnitude, evidential stability, and the most important drivers.
CREDIBLE EVIDENCE · 95%SPATIOTEMPORAL
INTERPRETABILITY
02 / THREE RESEARCH TASKS
From local mechanisms to global attribution,
then to dynamic prediction.
LOCAL INTERPRETABILITY · CORE FUNCTION
TCs · SCs · 50/95% CREDIBLE INTERVALSLocal Spatiotemporal Interpretability
Use STVC/STIVC to estimate the direction, magnitude, and uncertainty of covariate effects across places and periods—showing where, when, and how an effect operates.
GLOBAL INTERPRETABILITY · CORE FUNCTION
STVPI · RELATIVE CONTRIBUTION · ATTRIBUTIONGlobal Spatiotemporal Interpretability
Use STVPI to quantify each factor’s relative spatiotemporal contribution and identify key drivers and their spatial and temporal sources of variation.
DYNAMIC PREDICTION · CORE FUNCTION
PREDICTION · IMPUTATION · SMOOTHINGDynamic Spatiotemporal Prediction
Use STVI/STIVI and related models for missing-value imputation, spatiotemporal smoothing, and forecasting to construct continuous, reliable spatiotemporal data.
03 / GEOAI PARADIGM
Two geographic laws.
One white-box framework.
Go beyond asking why a model predicted an outcome: determine where the relationship occurs, when it changes, and how stable the evidence is.
Nearby entities tend to be more closely related. BSTVC represents spatial or spatiotemporal dependence across data, process, and parameter levels.
Phenomena and variable relationships differ across places. Local coefficients directly characterize this spatial and spatiotemporal nonstationarity.
INTERPRETABLE MODELSDATA → STRUCTURE → EVIDENCE
ANTE-HOC / INTRINSIC INTERPRETABILITY INTRINSIC INTERPRETABILITY
Definition:Explanatory variables, effect structures, and evidence outputs are specified before model fitting. Interpretation follows directly from model parameters and their posterior distributions, not from an external approximation added after training.
WHITE BOX / INTERPRETATION PATHObserved data Y → model structure Xβ(s, t) → posterior evidence
Interpretation comes directly from model structure and parameters, rather than an external approximation added after training.
The Model Structure Is the Explanation
Regression coefficients, credible intervals, and variance partitioning all arise directly from the Bayesian white-box model; mechanisms and predictions share the same statistical framework.
BSTVC CORE PARADIGMExplaining a Black Box with External Tools
The explanation is produced after training and usually approximates model behavior; it is not equivalent to a mechanism intrinsic to the statistical structure.
EXTERNAL EXPLANATION TOOLLOCAL + GLOBAL + DYNAMIC
From Local Mechanisms and Global Attribution to Dynamic Prediction
BSTVC organizes three research tasks into a continuous evidence chain: determine where, when, and how factors operate; compare the relative contributions of key drivers; then perform imputation, smoothing, and prediction within the same Bayesian framework.
Locate effect direction, magnitude, and uncertainty for every place and period.
STVC · STIVC → coefficient surfacesCompare candidate-factor contributions and identify key drivers and their temporal–spatial sources.
STVPI → variance partitioningPerform missing-value imputation, smoothing, and forecasting in one posterior framework.
STVI · STIVI → posterior predictionGLOBAL + LOCAL
Integrate Global Importance and Local Interpretation in the Study Design
Interpretability includes both population-level key-driver identification and local mechanisms for individuals, places, and time slices. BSTVC extends local interpretation to β(s, t) and connects it to global attribution and dynamic prediction.
GLOBAL INTERPRETABILITYWhich factors matter most overall?VARIABLE RANKING / CONTRIBUTION
βGLOBAL → LOCAL
LOCAL INTERPRETABILITYWhere, when, and how do effects vary?
β(s, t) / LOCAL CREDIBLE INTERVALS
DATA + PROCESS + PARAMETER
Interpretation Is Intrinsic from the Data Level to the Parameter Level
BSTVC organizes observed data, spatiotemporal processes, local parameters, and posterior uncertainty in a unified Bayesian hierarchical framework; predictions, coefficients, and contributions arise from the same statistical structure.
FULL MAP
ALL LOCAL UNITS FIT IN ONE MODEL
Observations, covariates, and spatial relationships
Temporal, spatial, and space–time interaction processes
Local parameters and credible intervals
Interpretation, attribution, and prediction
04 / MODEL SYSTEM
One System for interpretation and prediction.
From intercepts to coefficients, independent variation to space–time interactions, and spatial cross-sections to spatiotemporal panels—the research question determines the model scale.
HOW TO CHOOSEPredict only the spatiotemporal dynamics of Y?STVI / STIVI →Interpret local effects of X?STVC / STIVC / BSVC →Compare relative factor contributions?STVPI
Dynamic Predictionŷ
STVI
Spatiotemporally Varying Intercepts
Characterizes the response variable’s spatiotemporal evolution for missing-value imputation, smoothing, and dynamic prediction.
- DATA
- Spatiotemporal panel Y(s, t)
- BOUNDARY
- Focuses on response dynamics; does not interpret local X effects
Interaction-Aware Predictiont×s
STIVI
Spatiotemporally Interacting Varying Intercepts
Adds space–time interaction to a varying-intercept structure to represent coupled spatial and temporal dynamics.
- DATA
- Panel Y(s, t) with space–time interaction
- BOUNDARY
- For complex dynamic prediction; not a substitute for varying-coefficient attribution
Core Modelβst
RECOMMENDED FIRST MODEL · STVCSTVC
Spatiotemporally Varying Coefficients
Fits local spatiotemporal coefficients for explanatory variables to identify temporal and spatial nonstationarity in their relationships.
- DATA
- Y(s, t) + multiple X(s, t)
- BOUNDARY
- Spatial and temporal effects vary independently
Model Extensionβ×
STIVC
Spatiotemporally Interacting Varying Coefficients
Uses a space–time interaction assumption to reveal more complex local effect mechanisms and evolutionary patterns.
- DATA
- Hierarchical or interaction-based spatiotemporal panel
- BOUNDARY
- Requires an explicit SSH/TSH interaction assumption
Spatial Cross-Sectionβs
BSVC
Bayesian Spatially Varying Coefficients
Identifies spatially heterogeneous relationships and local driving patterns in single-time-point or spatial cross-sectional data.
- DATA
- Single-period spatial cross-section Y(s), X(s)
- BOUNDARY
- Explains spatial heterogeneity without estimating temporal change
Global AttributionΣ%
STVPI
Spatiotemporal Variance Partitioning Index
Transforms model results into relative contribution percentages to identify key factors and their temporal and spatial sources.
- DATA
- BSTVC posterior variance components
- BOUNDARY
- For relative contribution; not a substitute for interpreting local coefficient direction
Simultaneously quantifies spatially and temporally heterogeneous associations between Y and multiple X variables while incorporating spatial and temporal autocorrelation.
Defines interaction random effects through spatial stratified heterogeneity (SSH) or temporal stratified heterogeneity (TSH), supporting two-level coupled data such as county–province or day–month structures while controlling model complexity.
Uses variance partitioning to express heterogeneous effects as interpretable percentages, screening key factors while distinguishing temporal and spatial sources of contribution.
05 / METHOD ADVANTAGES
BSTVC Advantages:
Comparable, Quantifiable, Reproducible.
Its advantages arise not only from local regression, but from the combination of unified full-map modeling, Bayesian uncertainty, and a complete evidence chain.
Unified Full-Map Modeling
Fits all local parameters in one complete Bayesian hierarchical framework rather than separating map units into unrelated small models.
Directly Comparable Local Coefficients
Unified data, process, and parameter levels make regression coefficients statistically comparable across places and periods.
Explicit Uncertainty Quantification
Local predictions, TCs, and SCs include 50% inner and 95% outer credible intervals.
Integrated Local–Global Interpretation
Identify nonstationary spatiotemporal effects from local coefficients, then compare factor contributions using STVPI.
Three Response Types
Supports continuous (log-Gaussian), binary (logistic), and count (Poisson) responses within one system.
No-Code Desktop Workflow
Connects data conversion, field checks, spatial-order alignment, model fitting, result export, and visualization guidance.
TCs · SCs · CI
STVPI · attribution
impute · smooth · forecast
ONE METHOD SERIESAnalyze Factors → Identify Key Drivers → Predict Dynamically
From local mechanism interpretation and global key-driver identification to dynamic prediction, BSTVC connects research questions, statistical evidence, and decision-ready communication in one coherent methodology.
06 / EMPIRICAL WORKFLOW
A Real-Case-Driven
desktop modeling workflow.
Each response type is demonstrated with a real software interface. Import and validate data first, then check spatial order, configure parameters, run the model, and export results.
MODEL SPECIFICATION PRINCIPLEChoose the likelihood and link function from the response range, distributional properties, and data-generating mechanism—not researcher preference.
- 01Data InputTable + Complete Map
- 02Validate AlignmentFields + Spatial Order
- 03Configure ParametersY / X / Time / Space
- 04Run ModelsBSTVC / BSVC
- 05Export ResultsSix Excel Output Types
ACTUAL DESKTOP SCREEN
log-Gaussian RESPONSE
Global Health-Adjusted Life Expectancy (HALE)
2000–2020 · 177 COUNTRIES AND TERRITORIES
SUITABLE FORExplaining why continuous indicators—such as life expectancy, resource quantities, concentrations, and rates—vary across places and periods.Gaussian likelihood on log-transformed response · identity link
- INPUT
- TestData_Continuous. csv + complete TestMap_Continuous Shapefile set
- RESPONSE
- Y = HALE
- EXPLANATORY X
- High BMI, health expenditure, natural environment, health behavior, PM2.5, and related covariates
- OUTPUT
- TCs · SCs · STVPI · local predictions and credible intervals
ACTUAL DESKTOP SCREEN
logistic RESPONSE
COVID-19 Occurrence / Risk Status
MONTHLY SPATIOTEMPORAL PANEL · 0/1 RESPONSE
SUITABLE FORStudying spatiotemporal determinants of 0/1 events such as disease occurrence, target attainment, or risk status.Bernoulli / Binomial likelihood · logit link
- INPUT
- Covid_Binary_Panel(1). xlsx + complete province-level map
- RESPONSE
- Y = covid
- EXPLANATORY X
- PM2.5, temperature, and other environmental health factors
- OUTPUT
- Probability prediction · TCs · SCs · ROC/AUC
ACTUAL DESKTOP SCREEN
Poisson RESPONSE
Acute Hepatitis B Case Counts
2012–2023 · U. S. SPATIAL UNITS
SUITABLE FORInvestigating local spatiotemporal mechanisms for non-negative count outcomes such as cases, events, or facilities.Poisson likelihood · log link
- INPUT
- BSTVC_panel_converted. xlsx + complete us_a5 map
- RESPONSE
- Y = COUNT
- EXPLANATORY X
- Population density, uninsured rate, poverty rate, precipitation, and related factors
- OUTPUT
- Count prediction · TCs · SCs · Poisson deviance
RESPONSE CHECK / ContinuousAVOID MIS-SPECIFICATION
Download Three Example Datasets ↓Do not treat strongly skewed or overdispersed counts, or 0/1 outcomes, as continuous variables. Transform data and inspect distributions when appropriate.
07 / READ THE OUTPUT
Read
the spatiotemporal evidence chain.
Do not interpret a coefficient in isolation. Combine TCs, SCs, STVPI, local predictions, credible intervals, and model assessment to build an interpretable conclusion.
01 / RESULT
Map Local Effect Direction and Magnitude
02 / HOW TO READSpatial regression coefficients are exported by map unit for visualization in R, ArcGIS, ArcGIS Pro, or QGIS. Use consistent classes, color scales, and spatial extents across covariate maps.
RECOMMENDED: ArcGIS Pro / QGIS / R sf03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.
01 / RESULT
Read How Relationships Change over Time
02 / HOW TO READThe mean curve represents the temporal regression coefficient; the darker band is the 50% credible interval and the lighter band is the 95% credible interval. Whether an interval crosses zero helps assess effect direction and evidential stability.
RECOMMENDED: R ggplot2 / Excel03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.
01 / RESULT
Move from Absolute Ranking to Relative Contributions
02 / HOW TO READSTVPI compares overall contributions and separates their temporal and spatial sources while retaining Bayesian uncertainty for key-driver identification.
RECOMMENDED: R ggplot2 / Excel03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.
01 / RESULT
Deliver Predictions, Imputations, and Uncertainty Together
02 / HOW TO READlocal. prediction contains observed y, imputed fill_y, model-based predict_y, and 50%/95% credible intervals, enabling R², RMSE, and response-specific performance metrics.
RECOMMENDED: R / ArcGIS Pro / Excel03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.
01 / RESULT
Evaluate Discrimination for Binary Outcomes
02 / HOW TO READThe ROC curve shows the trade-off between sensitivity and false-positive rate. A larger AUC generally indicates stronger discrimination and should be interpreted with local probability maps and uncertainty.
RECOMMENDED: R pROC / Python scikit-learn03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.
01 / RESULT
Assess Fit, Complexity, and Predictive Accuracy Together
02 / HOW TO READmodel. evaluation summarizes DIC, WAIC, eff, pd2, and LS. For candidate models, smaller DIC and WAIC usually indicate better fit; eff and pd2 reflect effective parameters and complexity; LS values closer to zero generally indicate better predictive accuracy.
RECOMMENDED: Excel / R03 / INTERPRETATION
Assess direction and magnitude first, then the inner and outer credible intervals; finally integrate model assessment, mapped patterns, and substantive context.
Bayesian credible intervals express posterior parameter uncertainty and should not be mechanically equated with frequentist significance tests or reduced to significant/non-significant labels.Lower values generally indicate better fit.
↓ LOWER IS GENERALLY BETTERQuantifies effective parameters and model complexity.
↔ INTERPRET WITH MODEL FITEvaluates accuracy and discrimination by response type.
LS → 0 · AUC ↑Reads temporal and spatial nonstationary effects separately.
DIRECTION + MAGNITUDE + CrI08 / SELECTED PUBLICATIONS
A methods-to-applications
evidence chain.
From methodological development to validation in health, social, and environmental settings, representative studies continue to test and extend the interpretive capabilities and application boundaries of the BSTVC model family.
Scientific Reports
01Estimating Missing Values in China’s Official Socioeconomic Statistics Using Progressive Spatiotemporal Bayesian Hierarchical Modeling.
DOI ↗Science of the Total Environment
02Exploring Spatiotemporal Nonstationary Effects of Climate Factors on Hand, Foot, and Mouth Disease Using Bayesian STVC.
DOI ↗Annals of GIS
03Spatiotemporally Varying Coefficients Model: A Bayesian Local Regression to Detect Spatial and Temporal Nonstationarity.
DOI ↗International Journal of Disaster Risk Reduction
04Spatiotemporal Disparities in Regional Public Risk Perception of COVID-19 Using Bayesian STVC Series Models.
DOI ↗Journal of Cleaner Production
05Spatiotemporal Heterogeneity in Associations of National Population Ageing with Socioeconomic and Environmental Factors.
DOI ↗BMC Public Health
06Revealing Spatiotemporal Inequalities, Hotspots, and Determinants in Healthcare Resource Distribution.
DOI ↗09 / RESEARCH APPLICATIONS
Twelve Research Cases
Connected to Real-World Questions.
STVC
Hand, Foot, and Mouth Disease and Climate
Spatialized epidemiological odds ratios across 36 months in Sichuan to identify local seasonal trends and hotspots in disease–climate relationships.
- Research Question
- Hand, Foot, and Mouth Disease and Climate
- Model
- STVC
- Core Outputs
- TCs + SCs + credible intervals
STVC
Healthcare Resources and Socioeconomic Conditions in Northeast China
Validated the general STVC formulation with ten years of county data and compared local spatiotemporal relationships between healthcare resources and socioeconomic conditions.
- Research Question
- Healthcare Resources and Socioeconomic Conditions in Northeast China
- Model
- STVC
- Core Outputs
- TCs + SCs + credible intervals
STVC
County-Level Hospital Beds in Southwest China
Revealed local environmental and socioeconomic effects on healthcare-resource inequality and produced a continuous county-level map series of hospital beds.
- Research Question
- County-Level Hospital Beds in Southwest China
- Model
- STVC
- Core Outputs
- TCs + SCs + credible intervals
STVC
Multisource Drivers of Urban Tourism
Compared globally stationary and locally nonstationary relationships using 30 candidate factors across 343 cities from 2008 to 2017.
- Research Question
- Multisource Drivers of Urban Tourism
- Model
- STVC
- Core Outputs
- TCs + SCs + credible intervals
STVC / STIVC
COVID-19 Risk Perception
Estimated the regional public risk perception index (PRPI) across 366 Chinese cities and performed cluster and outlier analysis.
- Research Question
- COVID-19 Risk Perception
- Model
- STVC / STIVC
- Core Outputs
- TCs + SCs + credible intervals
STVC + STVPI
Attribution of Global Population Ageing
Used STVC to identify local associations and STVPI to screen key drivers across 189 countries and territories over two decades.
- Research Question
- Attribution of Global Population Ageing
- Model
- STVC + STVPI
- Core Outputs
- Local coefficients + relative contributions
BSTVC + STVPI
County-Level Healthcare Resources in China
Combined resource inequality, hotspot detection, and spatiotemporal determinants using a panel of hospital beds across 2,308 counties.
- Research Question
- County-Level Healthcare Resources in China
- Model
- BSTVC + STVPI
- Core Outputs
- Local coefficients + relative contributions
MSTVI / BSTVC
County-Level Maternal Mortality
Characterized long-term MMR patterns at a small-area scale and examined heterogeneous effects of multilevel health determinants.
- Research Question
- County-Level Maternal Mortality
- Model
- MSTVI / BSTVC
- Core Outputs
- Spatiotemporal trends + dynamic prediction
BSTVC
Multiple Urban Healthcare Resources
Compared spatiotemporal differences in resource patterns and socioeconomic effects across multiple urban healthcare-resource indicators in China.
- Research Question
- Multiple Urban Healthcare Resources
- Model
- BSTVC
- Core Outputs
- TCs + SCs + credible intervals
BSTVC / STVPI
Healthcare Resource Allocation Efficiency
Evaluated small-area spatiotemporal patterns and determinants of healthcare-resource allocation efficiency using a Sichuan county panel.
- Research Question
- Healthcare Resource Allocation Efficiency
- Model
- BSTVC / STVPI
- Core Outputs
- Local coefficients + relative contributions
BSTVC SERIES
Reanalysis of Hand, Foot, and Mouth Disease
Extended research on local HFMD risk and driving mechanisms using a new spatiotemporally interpretable model.
- Research Question
- Reanalysis of Hand, Foot, and Mouth Disease
- Model
- BSTVC SERIES
- Core Outputs
- TCs + SCs + credible intervals
STVI / BSTVC
Maternal and Under-Five Mortality
Jointly characterized spatiotemporal changes in MMR and U5MR and connected dynamic prediction with interpretation of maternal and child health indicators.
- Research Question
- Maternal and Under-Five Mortality
- Model
- STVI / BSTVC
- Core Outputs
- Spatiotemporal trends + dynamic prediction
01 / 12SWIPE HORIZONTALLY TO BROWSE CASES
PRIMARY SOURCE
For complete methods, original case figures, and updates, visit the BSTVC Model Homepage ↗
10 / FREE DOWNLOAD
Download BSTVC Desktop.
Make spatiotemporal relationships interpretable.
SPATIOTEMPORAL
INTERPRETABILITY
INTERPRETABILITY
English desktop release for Windows 10/11, with direct GitHub access to the installer, source repository, reproducible case interface, and supporting tools.
Download BSTVC Desktop
English installer · Windows x64 · Version 2026.8.1 · Released 1 August 2026
BSTVC_desktop_EN_win_x64_26.08.01_setup. exe SHA-2564f98423277254676d88eb64072e01920a597eaeddfaed1ee889bdcc0585c977c
Download from GitHub Release ↗
BSTVC-Desktop Repository
Releases, source, issue tracking, and version history
View GitHub Project ↗
English User Guide
The module and integration point are reserved for the forthcoming English guide.
Coming soon · Link reserved
Reproducible Case Dataset
BSTVC Process Monitor
Reserved interface for example data, maps, and expected outputs
Future GitHub Release · Interface reservedCompare CPU, memory, and thread settings locally
View GitHub Project ↗https://pan.baidu.com/s/5h7zSPjXEvBINShxLVLjirA
- 01 Windows x64 installer
- 02 English user interface
- 03 Continuous-response example
- 04 Binary-response example
- 05 Count-response example
- 06 Data, maps, and visualization guidance
11 / USER GUIDE KNOWLEDGE HUB
From Desktop to R:
build a complete research workbench.
Use the tabs to navigate data preparation, nine parameters, six output types, current limitations, R-package installation, documentation, and support.
Organize Tables and Maps into a Verifiable Spatiotemporal Panel
Files
Supports CSV, XLSX, and XLS. Import the complete Shapefile set rather than a standalone. shp file.
Spatial Order
The spatial-unit order in every time slice must match the map. The desktop app can compare, reorder, and export validated data automatically.
Fields
The table and map must use spatial identifiers with matching names and types; represent missing values consistently as NA.
Auxiliary Tool
Built-in data conversion, field conversion, and custom spatial matrices support wide-to-long reshaping and numeric field conversion.
Align Every Modeling Parameter with the Study Design
Data Source
Choose Original Data if spatial order is unchanged; choose Validated Data after automatic reordering.
Variables and Indices
Specify response Y, one or more explanatory variables X, the Time field, and the Space field.
Response Type
Choose continuous, binary, or count according to the data; each response type uses a different likelihood.
Computation Settings
Standardize explanatory variables where appropriate. The default is six threads; spatial weights can use QUEEN-B or a custom matrix.
Generate a Complete Interpretive Evidence Package in One Run
model. evaluation
Model fit, complexity, and predictive assessment using DIC, WAIC, eff, pd2, and LS.
local. prediction
Local predictions, missing-value imputations, and 50%/95% Bayesian credible intervals.
random effects
Spatial and temporal random effects for explanatory variables, with corresponding credible intervals.
TCs / SCs / STVPI
Temporal coefficients, spatial coefficients, and relative spatiotemporal contribution percentages, downloadable as a multi-sheet Excel workbook.
Understand the Current Version Boundaries before Formal Analysis
Data Scale
The upload limit is 1 GB. Computational demand increases with spatial units, time points, explanatory variables, response type, and thread settings.
Spatial Matrix
For point maps, use distance-based or k-nearest-neighbor matrices. If a kNN matrix triggers an INLA error, adjust its construction or use another matrix type.
STVPI
The current binary and count models do not include a residual term, so an overall explained-percentage metric is not yet available.
Nonstationarity Assumption
BSTVC currently assumes spatiotemporal nonstationarity for all explanatory variables; more flexible variable-level settings are planned.
Use the Desktop App, R Package, Documentation, and Model Homepage Together
BSTVC R package
bayesianstvc/BSTVC-R supports continuous, binary, and count responses within a full-map framework, producing local mechanisms, key-driver evidence, and dynamic predictions.
Installation and Dependencies
The current README uses devtools:: install_github("songbi123/BSTVC") or remotes:: install_github("songbi123/BSTVC"); install INLA first according to its official instructions.
Versions and License
DESCRIPTION specifies R ≥ 4.4.0 and INLA ≥ 24.06.27 under GPL-3; model assessment covers DIC, WAIC, pd, and LS.
Documentation and Collaboration
The repository provides GetStart. Rmd, English and Chinese PDFs, preprocessing scripts, and GitHub Issues. Desktop supports no-code reproduction; the R package supports batch research.
R PACKAGE / GITHUBbayesianstvc/BSTVC-R ↗
Source, releases, installation, and Issues.
CHINESE GUIDE / PDFChinese User Guide ↗Data, modeling, outputs, and visualization.
ENGLISH GUIDE / PDFGetStart-English ↗R-package examples and reproducible workflows.
DEPENDENCY / INLAINLA Installation Guide ↗Dependencies and version notes for large Bayesian computations.
MODEL / HOMEPAGEBayesian STVC Model Homepage ↗Theory, methodological evolution, and research cases.
TOOL / MONITORBSTVC Process Monitor ↗Monitor CPU, memory, and thread settings.
12 / AUXILIARY TOOL
BSTVCMONITOR
BSTVC Process Monitor:
Find the Right Thread Setting for Model Runs.
A lightweight resource monitor for BSTVC Desktop that tracks CPU, memory, and thread states during model computation to support faster, more stable settings.
WINDOWS / LOCAL ONLYv0.5.7
More Threads
Are Not Always Faster.
Different data scales, model structures, and hardware configurations have different resource sweet spots. Start with the BSTVC default of six threads, then compare eight or 12 using real trajectories from similar models.
BSTVC RECOMMENDED START6 THREADS
Start with the default six threads, then adjust using monitor results from comparable models.
BSTVC PROCESS MONITORLIVE / LOCAL
RESOURCE TRAJECTORYpast 60 min
TOTAL CPU MEMORY PRESSURE
PROCESS / PIDRUNTIMESTATUS
BSTVC / INLA PID 428000:18:42RUNNING
BSTVC / INLA PID 391600:11:07STOPPED
LIVE RESOURCE SIGNALS
See How BSTVC Modeling Uses Your Computer
Monitor one or more inla. exe processes and continuously record system-wide CPU, memory, threads, PID, runtime, and status. The main CPU series matches the system-wide metric shown by Windows Task Manager.
- CPU mean / P95 / proportion above threshold
- Current and peak thread counts and memory use
- Distinguish multiple processes by PID, start time, and runtime
THREAD SETTING EVIDENCE
Compare Thread Settings with Similar BSTVC Models
More threads are not always faster. Balance sustained CPU activity, memory pressure, variability, and solve time for your computer or workstation; the default remains six threads.
- Past-hour or full-recording time windows
- Composite resource score for relative comparison
- Completed processes remain visible and clearly labeled
LOCAL-ONLY / PRIVACY
Keep run records local for review and reproducibility
The monitor runs locally through PowerShell and a browser, requires no cloud service, and uploads no process data. Each run can save CSV, JSON, and local HTML reports.
- Default directory: %LOCALAPPDATA%\INLA-Resource-Monitor\runs
- Customizable output directory, sampling interval, and thresholds
- No administrator account or permanent PowerShell policy changes required
THREAD STARTING POINTS06 08 12 More threads do not guarantee faster runs; test with comparable models.
Recommended StartTypical Desktop Modeling
Advanced ComparisonMany-Core Workstation
Test with CautionWatch Memory and Throttling
THREAD DECISION
Start with the default six threads. Try eight threads only when memory is sufficient and the task is large. Reserve 12 threads for many-core, high-memory systems and confirm stable headroom with the monitor.
OPEN-SOURCE / MIT
Published by bayesianstvc. Runs on Windows 10/11 with PowerShell 5.1 or 7 and a modern browser. View v0.5.7 Release ↗
13 / CONTACT & FEEDBACK
Contact Us
to Improve the Methods and Software.
Project lead, statistical-methodology advisor, and copyright holder.
chaosong.gis@gmail.com ↗Software developer responsible for system design and user support.
tangxt.me@gmail.com ↗ANONYMOUS VISITOR MAPCONNECTING / ANONYMOUS TOTAL—
GITHUB ORGANIZATIONbayesianstvc
BSTVC ORGANIZATION
Advancing Spatiotemporal
Interpretability.
Bayesian Spatiotemporally Varying Coefficients
PRODUCT ECOSYSTEM
One framework. Multiple ways to explore.
BSTVC-RModeling package
BSTVC-DesktopNo-code analysis
BSTVC-ShinyInteractive application
BSTVC-MonitorResource diagnostics
Explore the BSTVC Organization ↗