How to use DSSM
how-to-use-DSSM.RmdWelcome to Pandora & IsoMemo Data Search and Spatiotemporal Modeling!
The “Data” tab serves for selection of data from various isotopic databases.
With the “Interactive map” tab, the selected data can be viewed in an interactive interface on the world map.
The “Modeling” tab serves for the computation of models using the selected data (or user data provided in a data-file) for estimating a spatial average model, a spatio-temporal average model or a spread model.
For a video introduction please have a look at:
https://www.youtube.com/watch?time_continue=1&v=B-isiVu09zQ
Data
On the left panel the user can select the databases and load them into the app by using the “Load data” button. Under “Select categories” the user may select categories or groups of fields. Each category has multiple fields or variables which can be selected individually by clicking on the drop down menu.
A certain selection of fields can be saved and re-loaded by the “Save data selection” and “Load data selection” buttons.
Once the data is loaded, a spreadsheet appears on the screen. It can be sorted using the values by certain fields by clicking on the field name and filtered by clicking on the entry below the field name.
The selected and filtered data set in the spreadsheet can be exported
in .xlsx, .csv, or .json format
using the “Export” button at the bottom of the left
panel.
Radiocarbon Calibration
On the left panel the calibration options for radiocarbon dates can be found. By selecting a method all uncalibrated radiocarbon dates in the displayed spreadsheet are calibrated by the selected method. The coverage of the credible intervals in percent can be selected in the “Calibration range” field.
The calibration is done with the help of the Bchron R package.
Interactive Map
Once data is loaded into the dashboard view in the “Data” tab, all locations of the selected data are
displayed on the world map. Keep in mind that the longitude
and latitude variables have to be available and
selected.
By clicking on a location the field values selected under “Select categories” in the “Data” tab can be seen.
On the right panel one or two numeric fields of the selected data can be chosen. One- and two-dimensional summary statistics and plots (with different options) are computed and displayed. The statistics can be copied into the clipboard for further use. The plots can be exported in various formats.
Modeling
Local Average Model
Here, a spatial or local average of a user-defined independent variable is modeled. A GAM (Generalized Additive Model) or GAMM (Generalized Additive Mixed Model) is fitted here.
Left Panel - Data and Model Options
-
Data selection: Under “Data” source the user can
select if they want to use the data from the “Data” tab or upload a
.xlsxor.csvfile. For uploads, the first row must contain variable names, and the coordinate format must be chosen from: “decimal degrees”, “degrees decimal minutes”, or “degrees minutes seconds”. If a.csvis selected, column and decimal separators must also be specified. - Variable selection: Specify the independent variable, its uncertainty in standard deviations (optional), the longitude and latitude variable, and optionally a site/ID variable. If selected, a random intercept is added to the model.
-
Model options:
- The “Number of basis functions” governs the number of basis functions used for the approximation of the surface, so called approximate thin-plate-splines (see e.g. doi:10.1007/3-540-47977-5_2 or doi:10.1111/1467-9868.00374 for further information). A higher number delivers more exact results and enables the estimation of more complex 2-D surfaces, but uses much more computational power. In general, there is little gain from very high numbers and there is no additional gain from selecting a higher number than there are unique combinations of longitude and latitude.
- A Bayesian option is available, but increases computation time. Select how much MCMC (Markov Chain Monte Carlo) iterations should be computed. Higher numbers deliver more exact results at the expense of computation time (usually 1000 iterations already give a pretty usable results for testing). The smoothing degree can be adjusted, the model already selects an optimal amount of smoothing itself, but this provides the option to estimate a more or less smoother surface.
- Click “Run” to estimate the model.
Central Panel - Display, Export, and Map Selection
The estimated spatial average is displayed in the center. An export button is provided below, together with sliders for map section selection.
Right Panel - Graphic Options
- Options to control the displayed plot include range restriction settings (0–1 or 0–100), surface type (terrestrial, aquatic, or complete), markers, grid, scale, north arrow.
- The “Apply convex hull” checkbox marks, whether estimates of locations within the convex hull of the data points should be displayed. In general, this is indicated as extrapolation beyond areas with available data is highly speculative.
- The slider “Display up to max standard error” governs the area where we have a standard error below the selected mark to prevent areas with low certainty being shown.
- Right under the colour options, a slider “Plot resolution” is available. A higher number delivers a higher quality picture of the coloured area with less pixaleted borders but needs much more drawing time. A lower value is recommended for setting the right picture options and map section and a high value for publication.
Location/Batch Estimate
The options on the bottom of the rightern panel enable the user to get an estimate of a specified location (“Center point latitude/longitude”) within a certain radius. The estimate with standard error will appear below the x-axis in the displayed plot.
For getting estimates for a batch of points the “Batch Estimates” button opens a pop-up window, where a user-specified data set can be uploaded. A local estimate including standard error and 95 percent uncertainty intervals is added and can be exported.
Spatio-Temporal Average Model
Here, a spatio-temporal average of a user-defined independent variable is modeled. A GAM (Generalized Additive Model) or GAMM (Generalized Additive Mixed Model) is fitted.
Left Panel - Data and Model Options
Data and variable selection: Same setup as above, but with additional date variables and a “Date type” selection (Interval, Mean + 1 SD, or Single point). Depending on type, appropriate fields must be filled.
Model options: Modeling options are identical: basis functions, Bayesian model toggle, MCMC iterations, and smoothing level.
Central Panel - Display, Export, and Map Selection
The spatial average is displayed, exportable via button below. Map section adjustable with sliders.
Spread Model
Here, a time spread of a user defined date variable is modeled. An extremal quantile regression model (see e.g. doi:10.1214/009053604000001165 is fitted here to get an estimate very close to the local minimum or maximum. This is done using a Bayesian quantile approach (see e.g. doi:10.1016/S0167-7152(01)00124-9) with a quantile set near 0/1.
Left Panel - Data and Model Options
Data and variable selection: The user selects the data source and specifies the date variables, longitude, latitude, and optionally a site/ID variable, see setup as above. The “Date type” must be specified (Interval, Mean ± SD, Single point).
Model options: Model direction (Min or Max) must be chosen. Then, basis functions, Bayesian toggle, MCMC count, and smoothing can be set, see setup as mentioned above
Central Panel - Display, Export, and Map Selection
The estimated spatial average is shown in the center. Export options are available. A “Time selection” slider allows selection of the estimate’s time point. Sliders adjust the map section.