geo-spattime >> pkgs, minor edits

Author: ercbk

qmd/diagnostics-probabilistic.qmd

@@ -20,7 +20,7 @@
 
 -   [{]{style="color: #990000"}[topmodels](https://topmodels.r-forge.r-project.org/){style="color: #990000"}[}]{style="color: #990000"} currently supported models:
 
-    -   `lm`, `glm`, `glm.nb`, `hurdle`, `zeroinfl`, `zerotrunc`, `crch`, `disttree`, and models from [{disttree}]{style="color: #990000"}, [{crch}]{style="color: #990000"}
+    -   `lm`, `glm`, `glm.nb`, `gamlss`, `bamlss`, `hurdle`, `zeroinfl`, `zerotrunc`, `crch`, `betareg`, and models from [{disttree}]{style="color: #990000"}, [{crch}]{style="color: #990000"}
     -   Also see video, [Probability Distribution Forecasts: Learning from Random Forests and Graphical Assessment](https://www.youtube.com/watch?v=iMBgmfdKs8g)
 
 -   `autoplot` produces a ggplot object that can be used for further customization

qmd/geospatial-modeling.qmd

@@ -189,7 +189,7 @@
 -   Packages
     -   [{]{style="color: #990000"}[BMEmapping](https://cran.r-project.org/web/packages/BMEmapping/index.html){style="color: #990000"}[}]{style="color: #990000"} - Spatial Interpolation using **Bayesian Maximum Entropy (BME)**
         -   Ssupports optimal estimation using both precise (hard) and uncertain (soft) data, such as intervals or probability distributions
-    -   [{]{style="color: #990000"}[gstat](https://cran.r-project.org/web/packages/gstat/index.html){style="color: #990000"}[}]{style="color: #990000"} - **OG**; Has various interpolation methods.
+    -   [{]{style="color: #990000"}[gstat](https://cran.r-project.org/web/packages/gstat/index.html){style="color: #990000"}[}]{style="color: #990000"} - **OG**; Has idw and various kriging methods.
     -   [{]{style="color: #990000"}[intamap](https://cran.r-project.org/web/packages/intamap/index.html){style="color: #990000"}[}]{style="color: #990000"} ([Paper](https://www.sciencedirect.com/science/article/abs/pii/S0098300410002190?via%3Dihub)) - Procedures for **Automated Interpolation** (Edzer Pebesma et al)
         -   One needs to choose a model of spatial variability before one can interpolate, and experts disagree on which models are most useful.
     -   [{]{style="color: #990000"}[interpolateR](https://cran.r-project.org/web/packages/InterpolateR/index.html){style="color: #990000"}[}]{style="color: #990000"} - Includes a range of methods, from traditional techniques to advanced **machine learning** approaches

qmd/geospatial-spat-temp.qmd

@@ -3,10 +3,18 @@
 ## Misc {#sec-geo-sptemp-misc .unnumbered}
 
 -   Packages
+    -   [{]{style="color: #990000"}[cubble](https://huizezhang-sherry.github.io/cubble/){style="color: #990000"}[}]{style="color: #990000"} ([Vignette](https://www.jstatsoft.org/article/view/v110i07)) - **Organizing** and **wrangling** space-time data. Addresses data collected at unique fixed locations with irregularity in the temporal dimension
+        -   Handles **sparse grids** by nesting the time series features in a tibble or tsibble.
     -   [{]{style="color: #990000"}[magclass](https://cran.r-project.org/web/packages/magclass/index.html){style="color: #990000"}[}]{style="color: #990000"} - Data Class and Tools for Handling Spatial-Temporal Data
         -   Has **datatable** under the hood, so it should be pretty fast for larger data
         -   **Conversions, basic calculations and basic data manipulation**
-    -   [{]{style="color: #990000"}[mlr3spatiotempcv](https://mlr3spatiotempcv.mlr-org.com/){style="color: #990000"}[}]{style="color: #990000"} - Extends the **mlr3** machine learning framework with **spatio-temporal resampling methods** to account for the presence of spatiotemporal autocorrelation (STAC) in predictor variables
+    -   [{]{style="color: #990000"}[mlr3spatiotempcv](https://mlr3spatiotempcv.mlr-org.com/){style="color: #990000"}[}]{style="color: #990000"} - Extends the **mlr3** machine learning framework with spatio-temporal **resampling methods** to account for the presence of spatiotemporal autocorrelation (STAC) in predictor variables
+    -   [{]{style="color: #990000"}[rasterVis](https://oscarperpinan.github.io/rastervis/){style="color: #990000"}[}]{style="color: #990000"} - Provides three methods to **visualize** spatiotemporal rasters:
+        -   `hovmoller` produces Hovmöller diagrams
+        -   `horizonplot` creates horizon graphs, with many time series displayed in parallel
+        -   `xyplot` displays conventional time series plots extracted from a multilayer raster.
+    -   [{]{style="color: #990000"}[sdmTMB](https://sdmtmb.github.io/sdmTMB/){style="color: #990000"}[}]{style="color: #990000"} - Implements spatial and spatiotemporal **GLMMs** (Generalized Linear Mixed Effect Models)
+    -   [{]{style="color: #990000"}[sftime](https://r-spatial.github.io/sftime/){style="color: #990000"}[}]{style="color: #990000"} - A complement to [{stars}]{style="color: #990000"}; provides a generic data format which can also handle **irregular (grid)** spatiotemporal data
     -   [{]{style="color: #990000"}[spStack](https://span-18.github.io/spStack-dev/){style="color: #990000"}[}]{style="color: #990000"} - Bayesian inference for point-referenced spatial data by assimilating posterior inference over a collection of candidate models using stacking of predictive densities.
         -   Currently, it supports point-referenced Gaussian, Poisson, binomial and binary outcomes.
         -   Highly parallelizable and hence, much faster than traditional Markov chain Monte Carlo algorithms while delivering competitive predictive performance
@@ -15,11 +23,13 @@
         -   Models: Bayesian Gaussian Process (GP) Models, Bayesian Auto-Regressive (AR) Models, and Bayesian Gaussian Predictive Processes (GPP) based AR Models for spatio-temporal big-n problems
         -   Depends on [{spacetime}]{style="color: #990000"} and [{sp}]{style="color: #990000"}
     -   [{]{style="color: #990000"}[stars](https://r-spatial.github.io/stars/){style="color: #990000"}[}]{style="color: #990000"} - **Reading, manipulating, writing and plotting** spatiotemporal arrays (raster and vector data cubes) in 'R', using 'GDAL' bindings provided by 'sf', and 'NetCDF' bindings by 'ncmeta' and 'RNetCDF'
+        -   Only handles **full lattice/grid** data
         -   It supercedes the [{]{style="color: #990000"}[spacetime](https://cran.r-project.org/web/packages/spacetime/index.html){style="color: #990000"}[}]{style="color: #990000"}, which extended the shared classes defined in [{sp}]{style="color: #990000"} for spatio-temporal data. [{stars}]{style="color: #990000"} uses PROJ and GDAL through [{sf}]{style="color: #990000"}.
 -   Resources
     -   [SDS, Ch.13](https://r-spatial.org/book/13-Geostatistics.html)
     -   [Geodata and Spatial Regression, Ch. 14](https://ruettenauer.github.io/Geodata_Spatial_Regression/09_spatiotemporal.html)
     -   [Spatial Modelling for Data Scientists, Ch. 10](https://gdsl-ul.github.io/san/10-st_analysis.html)
+    -   Spatio-Temporal Statistics in R (See R \>\> Documents \>\> Geospatial)
 -   Papers
     -   [INLA-RF: A Hybrid Modeling Strategy for Spatio-Temporal Environmental Data](https://arxiv.org/abs/2507.18488)
         -   Integrates a statistical spatio-temporal model with RF in an iterative two-stage framework.
@@ -89,7 +99,7 @@
         -   Points: Data having points support should use the SpatialPoints class for the spatial feature
         -   Polygons: Values reflect aggregates (e.g., sums, or averages) over the polygon (SpatialPolygonsDataFrame, SpatialPolygons)
         -   Grids: Values can be point data or aggregates over the cell.
-    -   `stConstruct` creates "an [STFDF]{.arg-text} object if all space and time combinations occur only once, or else an object of class [STIDF]{.arg-text}, which might be coerced into other representations."
+    -   `stConstruct` creates "a [STFDF]{.arg-text} (full lattice/full grid) object if all space and time combinations occur only once, or else an object of class [STIDF]{.arg-text} (irregular grid), which might be coerced into other representations."
 -   [Long]{.underline} - The full spatio-temporal information (i.e. response value) is held in a single column, and location and time are also single columns.
     -   [Example]{.ribbon-highlight}: Private capital stock (?)
 
@@ -105,7 +115,7 @@
         -   Each row is a single time unit and space unit combination.
         -   This is likely not the row order you want though. I think these spatio-temporal object creation functions want ordered by time, then by space. (See Example)
 
-    -   [Example]{.ribbon-highlight}: Create full grid spacetime object from a long table (7.2 Panel Data in [{spacetime}]{style="#990000"} vignette)
+    -   [Example]{.ribbon-highlight}: Create full grid spacetime object from a long table (7.2 Panel Data in [{spacetime}]{style="color: #990000"} vignette)
 
         ::: panel-tabset
         ## Input Data
@@ -151,9 +161,9 @@
 
         ``` r
         st_data <- 
-          STFDF(geom_states, 
-                vec_time, 
-                df_data)
+          STFDF(sp = geom_states, 
+                time = vec_time, 
+                data = df_data)
 
         length(st_data)
         #> [1] 816
@@ -183,6 +193,10 @@
         ```
 
         -   So combining of these objects into a spacetime object doesn't seemed be based any names (e.g. a joining variable) of the elements of these separate objects.
+        -   STFDF arguments:
+            -   [sp]{.arg-text}: An object of class [Spatial]{.arg-text}, having n elements
+            -   [time]{.arg-text}: An object holding time information, of length m;
+            -   [data]{.arg-text}: A data frame with n\*m rows corresponding to the observations
         -   Converting back to a dataframe adds 6 new columns to [df_data]{.var-text}:
             -   [V1]{.var-text}, [V2]{.var-text}: Latitude, Longitude
             -   [sp.ID]{.var-text}: IDs within the spatial geomtry object (geom_states)
@@ -237,7 +251,7 @@
 
         -   See Long \>\> Example \>\> Create full grid spacetime object for a more detailed breakdown of creating these objects
         -   **The order of station geometries in [geom_stations]{.var-text} should match the order of the columns in [mat_wind_velos]{.var-text}**
-            -   In the vignette, the data used to create the geometry object is ordered according to the matrix using `match`. See [R, Base R \>\> Functions](r-base-r.qmd#sec-r-baser-funs){style="color: green"} \>\> `match`
+            -   In the vignette, the data used to create the geometry object is ordered according to the matrix using `match`. See [R, Base R \>\> Functions](r-base-r.qmd#sec-r-baser-funs){style="color: green"} \>\> `match` for the code.
 
         ## STFDF Object
 
@@ -279,7 +293,7 @@
         #> 78888  605634.5   6136859    12 1978-12-31 12:00:00 1979-01-01 12:00:00      6574 1.0835638
         ```
 
-        -   [space]{.arg-text} has a confusing description. Here it's just a numerical index for the number of columns of [mat_wind_velos]{.var-text} or the geometries in [geom_stations]{.var-text}
+        -   [space]{.arg-text} has a confusing description. Here it's just a numerical index for the number of columns of [mat_wind_velos]{.var-text} which would match an index/order for the geometries in [geom_stations]{.var-text}.
         -   [interval = TRUE]{.arg-text}, since the values are daily mean wind speed (aggregation)
             -   See Time and Movement section
         -   [df_st_wind]{.var-text} is in long format

scrapsheet.qmd

@@ -52,6 +52,24 @@ title: "Scrapsheet"
         6.  Calculate CI variant on the distribution of averaged scores
     -   Questions
         -   How many bootstrap iterations?
+-   Empirical Orthogonal Functions (EOFs)
+    -   A form of PCA applied to spatiotemporal data that is useful for understanding patterns in fields like climate science, oceanography, and meteorology.
+
+    -   Components
+
+        -   Spatial Patterns (EOFs) - Shows where *variability* is concentrated
+
+        <!-- -->
+
+        -   Time series (principal components) - Shows when each pattern is *active* and with what *amplitude*
+
+    -   Process
+
+        -   Locations are variables and observations are time series
+        -   Preprocess (scale) and perform PCA
+        -   Each EOF is an eigenvector ($V$) representing a spatial pattern, and its eigenvalue tells you how much variance that pattern explains. The PC is a time series ($\text{PC1}_i = V_{(,1)} \cdot A_{(i,)}$)
+
+    -   Example: In sea surface temperature data, EOF1 might reveal the El Niño/La Niña pattern - showing warming/cooling in the tropical Pacific. The associated PC would be a time series showing El Niño events as positive peaks and La Niña as negative troughs.
 
 ## Multivariable Geostatistics