11-environmentalField.Rmd

# Applications for the Environmental Field {#Ch11}


\chapterauthor{Michael G. McManus, Marcus W. Beck, J{\"u}rgen Symanzik}


## Introduction {#Ch11-Introduction}
This chapter will give the purpose and description of linked micromaps as applied to environmental data using the R package **micromap**\index{R Packages!micromap}. We will ask what types of environmental data are most suited for summarizing and displaying as linked micromaps. Additionally, we will show how linked micromaps can be used in spatial data analysis by combining workflows with spatial sampling designs as a unique application to environmental data. Our linked micromap examples are drawn from federal and state environmental agencies in the United States.

The purpose of a linked micromap is to show spatial locations corresponding to statistical estimates [@COCPC1998]. Linked micromaps simultaneously summarize and display both statistical and geographic distributions by linking statistical summaries of polygons, or areal units, to a series of small maps [@PMWOK2015JSS]. Implicit with this purpose and description is that space, defined as spatial contiguity of polygons, provides some context to the statistical estimates, and that the statistical estimates are representative of the polygons. This spatial context to statistical estimates can be one of three patterns:  spatial similarity among neighbors, spatial dissimilarity among neighbors, or no spatial pattern among neighbors. @MPRG2016 illustrated that first pattern with data from the West Virginia Department of Environmental Protection (WVDEP), while the examples below by the Virginia Department of Environmental Quality (VDEQ) and Florida Department of Environmental Protection (FDEP) illustrate the second and third patterns, respectively [@SWSCSS2018;@VirginiaDEQ2020]. All three of those examples also addressed the issue of obtaining statistical estimates of environmental data that were representative of watersheds (polygons) used in the analysis. 

Public health statistics are often reported by administrative boundaries, such as counties, states, or countries, whereas environmental statistics are often reported by boundaries of natural features, such as watersheds, estuaries, or ecoregions. A census of environmental characteristics from such natural features is not possible. Consequently, environmental statisticians have recommended probability-based surveys or sampling to obtain representative estimates of natural resources [@PUW1999]. Probability-based surveys have three steps:  1) identify the total or target population of interest (e.g. all wadeable streams in Virginia), 2) select a random sample to ensure representativeness with the known probability of including any member of the target population, and 3) apply the sample weight, the inverse of the probability, to the environmental measurements so that inferences can be made to the target populations based on the samples [@PUW1999]. @SO2004 introduced spatially balanced probabilistic surveys as having sample sites distributed that correspond to the spatial distribution of the features in the target population of the natural resource and can be more efficient than a simple random sample of sites, in which clumping of sites may occur. The design and analysis of spatially balanced probabilistic surveys is done using the R package **spsurvey**\index{R Packages!spsurvey} [@DKOW2022;@OKP2012]. An example is provided to demonstrate how **spsurvey** is used for creating samples as a precursor to collecting data and summarizing them with linked micromaps. Statistical estimates of aquatic natural resources have been made from spatially balanced probabilistic surveys of watersheds within a state to the ecoregions conterminous  of the United States, as part US EPA’s National Aquatic Resource Surveys. The estimates from these surveys can be visualized with linked micromaps.

## Recent Geovisualizations with Environmental Data {#Ch11-Geovisualizations}

Virginia Department of Environmental Quality and Florida Department of Environmental Protection both implemented spatially balanced probabilistic surveys of aquatic resources and visualized the results of the surveys using linked micromaps.  Under the Clean Water Act in the United States, states must report on the condition or status of their lakes, rivers, streams, and estuaries.  To meet this requirement, VDEQ has conducted such surveys of perennial, non-tidal, wadeable streams and rivers from 2001-2018 and has collected aquatic macroinvertebrates at stream sites for biological monitoring [@VirginiaDEQ2020]. The aquatic macroinvertebrates indicate the ecological conditions of a stream and are expressed as a Stream Condition Index (SCI) on a numeric scale. From the stream surveys, estimates of the SCI score can be made among the 11 major river basins in Virginia and compared to a statewide threshold value [@VirginiaDEQ2020]. Higher SCI scores suggest better ecological condition of the streams.

```{r Ch11-micromapVDEQ, fig.width = 7, fig.height = 5, fig.cap = 'A linked micromap of stream condition for 11 basins from spatially balanced probabilistic surveys done by Virginia Department of Environmental Quality. Higher scores for Virginia Stream Condition Index (VSCI) and Virginia Coastal Plains Macroinvertebrate Index (VCPMI) indicate better ecological condition of the streams. Sample size, n, is indicated for each basin and median estimates and their interquartile ranges are shown in the statistical graphics coloumn. The dashed vertical line at 60 is the threshold for biological impairment.', eval = TRUE, echo = FALSE}
library(micromap)

VA_stats <- readRDS("data/stats.RDS")
VA_map_table <- readRDS("data/map.table.RDS")

mmplot(
  stat.data = VA_stats,
  map.data = VA_map_table,
  map.link = c("Basin", "ID"),
  panel.types = c("dot_legend", "labels", "labels", "dot_cl", "map"),
  panel.data = list(NA, "Basin", "n", list("x50", "x25", "x75"), NA),
  ord.by = "x50",
  grouping = c(4, 3, 4),
  vertical.align = "center",
  colors = brewer.pal(3, "Spectral"),
  rev.ord = TRUE,
  panel.att = list(
    list(
      1,
      point.type = 20,
      point.border = TRUE,
      point.size = 2
    ),
    list(
      2,
      header = "Basin",
      panel.width = .6,
      align = "left",
      text.size = .9
    ),
    list(
      3,
      header = "n",
      panel.width = .2,
      align = "left",
      text.size = .9
    ),
    list(
      4,
      header = "Estimated Median VSCI/VCPMI Score\nand Associated Interquartile Range",
      graph.bgcolor = "lightgray",
      point.size = 1.5,
      xaxis.ticks = list(40, 50, 60, 70, 80),
      xaxis.labels = list(40, 50, 60, 70, 80),
      add.line = 60,
      add.line.col = "black",
      add.line.typ = "dashed",
      xaxis.title = "VSCI Score",
      panel.width = 1.2
    ),
    list(
      5,
      header = "Light Gray Means\nPreviously Displayed",
      map.all = TRUE,
      fill.regions = "aggregate",
      active.border.color = "black",
      active.border.size = 1.0,
      inactive.border.color = gray(.7),
      inactive.border.size = 1,
      panel.width = 1.0
    )
  )
)
```

Unlike a chloropleth map that simply maps a summary statistic to color aesthetics of polygons, the SCI linked micromap produced by VDEQ displays *both* a measure of central tendency and a measure of variation for each of the 11 major river basins (Figure \@ref(fig:Ch11-micromapVDEQ)). Specifically, the statistical graphics column of the linked micromap shows the estimated median SCI score and the interquartile range (IQR), as the estimated 25th and 75th percentiles, with the basins ranked by their medians. Those statistics are compared to a statewide threshold SCI of 60 as shown by the vertical, black, dashed line. Close inspection of the linked micromap reveals a spatial dissimilarity for the southwestern basins of Holston, Clinch-Powel, and Big Sandy, all of which are part of the Tennessee River drainage. Holston has excellent condition based on its SCI score, Clinch-Powell has good condition, and Big Sandy has severe stress. Also, worth noting, is that the 75th percentile of the poorest scoring basin, Rappahannock, extends past the statewide threshold indicating that some sites might benefit from conservation efforts as opposed to poorer scoring sites, which might need remediation to improve their condition. The VDEQ Report included additional linked micromaps having two statistical graphics column, with one column showing the basin estimates of a water quality stressor,paired with another column of the SCI [@VirginiaDEQ2020]. Pairing a potential stressor with the SCI response provided a visualization of stressors that might have a larger impact on aquatic macroinvertebrates relative to other basins [@VirginiaDEQ2020]. The VDEQ linked micromap provided a means to show measures of variation, along with comparison to a reference value. 

While the VDEQ R code builds off Chapter \@ref(Ch2), some distinct arguments and layouts are worth mentioning. The VDEQ linked micromap features asymmetrical perceptual grouping as specified by `grouping = c(4, 3, 4)`, followed by `vertical.align = "center"` that aligns the rows within a perceptual group, rather than the default alignment of "top". The center alignment is helpful when perceptual groups contain different number of rows. This is a five panel linked micromap showing a unique aspect of two label panels by using this syntax: `panel.types = c("dot_legend", "labels", "labels", "dot_cl", "map")`. The first label specification corresponds to the basin in the `panel.data`, while the second label specification corresponds to the sample size, n.  From both of those calls, the linked micromap displays the basin names and their sample size. The VDEQ ecologists applied a novel use of `dot_cl` by reporting a three-number summary of estimates of the median, first quartile, and third quartile in their linked micromap.  The `dot_cl` argument is typically used to report a central tendency estimate, such as a mean or median, and then the lower and upper confidence limits, as @MPRG2016 did. Finally, in that same statistical graphics column, the overall statewide median estimate of SCI was displayed by using these arguments:
`add.line = 60, add.line.col = "black", add.line.typ = "dashed"`.

Finally, the R code used to create Figure \@ref(fig:Ch11-micromapVDEQ) is shown below, which uses the _stats_\index{Datasets!stats} and _VA_map_table_\index{Datasets!VA\_map\_table} datasets.

```{r, Ch11-micromapVDEQ, eval = FALSE, echo = TRUE}
```

The next example is from FDEP and uses several statistical graphics columns to summarize water quality and land cover data among watersheds. Studies of freshwater aquatic ecosystems often need to summarize water quality results from watersheds, as VDEQ did, along with land cover in those watersheds, which is how FDEP used linked micromaps.  Spatially balanced probabilistic surveys were done by FDEP to study emerging contaminants from different types of waterbodies, canals, rivers, streams, small and large lakes, and unconfined aquifers, in 29 drainage basins [@SWSCSS2018]. The results for one of the contaminants, the widely used pesticide imidacloprid, were pooled across waterbody types and summarized with a 3-statistical graphics column linked micromap of the 29 drainage basins  displaying:  1) percentage of surveyed sites having detections of imidacloprid, 2) percentage of urban and agricultural land cover in a basin, and 3) a five-number summary, minimum, first quartile, median, third quartile, and maximum, of imidacloprid concentrations shown as a box plot [@SWSCSS2018]. As an exploratory spatial data analysis tool, @SWSCSS2018 noted from the linked micromap that neighboring drainage basins showed no geographic patterning in percentages of detections, which was confirmed by a nonsignificant spatial autocorrelation using Moran’s Index.  This application of linked micromaps was novel in two regards: first, by summarizing data from different sources (i.e., land use, water quality) for the polygons of interest, and second, by formally testing if there was geographic patterning in the results.

```{r Ch11-micromapFDEP, fig.width = 7, fig.height = 6, fig.cap = 'A linked micromap summarizing results for the pesticide Imidacloprid for 29 Florida basins from spatially balanced probabilistic surveys of freshwater bodies by Florida Department of Environmental Protection.  Data are sorted by percentage of samples in a basin having detectable concentrations of the pesticide. The second statistical graphics column shows percent of agricultural and urban land cover in a basin. The third statistical graphics column displays boxplots of concentrations of the pesticide, with single vertical bars indicating samples having concentrations below instrument detection limits.', eval = TRUE, echo = FALSE}
library(micromap)

# data frames
FL_pesticide <- readRDS("data/FL_pesticide.RDS")
FL_basins <- readRDS("data/Florida_Basins.rds")

# code to handle NA values in rankings
FL_pesticide$Ranked_Imida <- rank(
  FL_pesticide$Imidacloprid,
  na.last = FALSE,
  ties.method = "first"
)

mmplot(
  stat.data = FL_pesticide,
  map.data = FL_basins,
  map.link = c("GROUP_NAME", "ID"),
  panel.types = c(
    "dot_legend", "labels", "labels", "dot",
    "dot", "box_summary", "map"
  ),
  panel.data = list(
    NA, "GROUP_NAME", "Sum_PNT_COUNT", "Imidacloprid", "Percent_Ag_Ur_Tr",
    list("Imida_Min", "Imida_25", "Imida_Med", "Imida_75", "Imida_Max"),
    NA
  ),
  ord.by = "Ranked_Imida",
  rev.ord = TRUE,
  grouping = c(6, 6, 5, 6, 6),
  vertical.align = "center",
  plot.panel.spacing = 2,
  colors = c("red", "orange", "yellow", "green", "blue", "purple"),
  panel.att = list(
    list(
      1,
      point.type = 20,
      point.border = TRUE,
      panel.width = 1,
      point.size = 2
    ),
    list(
      2,
      header = "Basin\nName",
      panel.width = 1.1,
      align = "left",
      left.margin = -1.6,
      text.size = .65
    ),
    list(
      3,
      header = "Sample\nSize",
      panel.width = .1,
      align = "right",
      left.margin = -1.6,
      text.size = .65
    ),
    list(
      4,
      header = "Imidacloprid\nDetected",
      graph.bgcolor = "lightgray",
      point.size = 1,
      xaxis.ticks = list(0, 25, 50, 75),
      xaxis.labels = list(0, 25, 50, 75),
      xaxis.title = "Percent",
      left.margin = -1.0,
      panel.width = .4
    ),
    list(
      5,
      header = "Ag + Urban\nLand Use",
      graph.bgcolor = "lightgray",
      point.size = 1,
      xaxis.ticks = list(0, 25, 50, 75),
      xaxis.labels = list(0, 25, 50, 75),
      xaxis.title = "Percent",
      left.margin = -1.4,
      panel.width = .4
    ),
    list(
      6,
      header = "Imidacloprid Boxplot\n(min-25th-Med-75th-max)",
      graph.bgcolor = "lightgray",
      xaxis.ticks = list(-3, -2.5, -2, -1.5, -1, -0.5, 0),
      xaxis.labels = list(-3, -2.5, -2, -1.5, -1, -0.5, 0),
      xaxis.title = "Log micrograms per liter",
      graph.bar.size = 0.4,
      left.margin = -1.4,
      panel.width = .85
    ),
    list(
      7,
      header = "Gray Means\nPreviously Displayed",
      map.all = TRUE,
      fill.regions = "aggregate",
      active.border.color = "black",
      active.border.size = 1,
      left.margin = -0.8,
      panel.width = .7
    )
  )
)
```

In Figure \@ref(fig:Ch11-micromapFDEP), the syntax: `panel.types = c("dot_legend", "labels", "labels", "dot", "dot", "box_summary", "map")` is used in the creation of boxplots.  The `box_summary` requires that _FL\_pesticide_\index{Datasets!FL\_pesticide} dataset have columns containing the following summary statistics for imidalcloprid concentrations for each drainage basin:  minimum, first quartile, median, third quartile, and maximum.  The boxplots are displayed in the third statistical panel, with the width of the boxplot specified with the argument: `graph.bar.size = 0.4`. The R code used to create Figure \@ref(fig:Ch11-micromapFDEP) is below. The _FL\_pesticide_\index{Datasets!FL\_pesticide} dataset had an NA entry for the Florida Keys regarding the percentage of samples having detectable imidacloprid concentrations as none of the freshwater bodies sampled by FDEP occurred there. Users may have to explicitly code how to handle NA values.  In this case, the syntax
`FL_pesticide$ranked_imida <- rank(FL_pesticide$Imidacloprid, na.last = FALSE, ties.method = "first")` is used to create a new variable that ranked NA values last. The new variable was then used with the `ord.by` argument. A challenge mentioned in Chapter \@ref(Ch2) and illustrated with the Florida linked micromap is that of small subareas, which is the Perdido basin, the westernmost basin in Florida.  

```{r, Ch11-micromapFDEP, eval = FALSE, echo = TRUE}
```

## Spatial Surveys and Linked Micromaps {#Ch11-Spatial-Surveys}

Spatially balanced probabilistic surveys are often used to draw samples for creating summary statistics for areal units, such as watersheds or ecoregions, and we will illustrate the drawing of such a survey sample. For more details on the design and analysis of those surveys, the readers should consult @SO2004, @OKP2012, and @DKOW2022.  A spatially balanced probabilistic survey is connected to a linked micromap in that the survey provides the representative samples of the natural feature (lakes, rivers, streams, estuaries, ecoregions) from which estimates are made for the areal units. This is a unique application to environmental data by demonstrating how the **spsurvey** package can be used to draw samples that can be used to collect and summarize data with micromaps.

Three steps are required to draw a spatially balanced probabilistic survey using the R package **spsurvey**\index{R Packages!spsurvey}: 1) reading in the sample frame, which is a GIS file in the format of shapefile or an R simple feature, or sf, object, 2) specifying the design, and 3) selecting the sample [@OKP2012].  The example sample frame we will use contains points for 3,190 lakes in the state of Oregon in the United States and is the same sample frame used by @OKP2012, where readers can get more details on the Generalized Random Tessellation Stratified (GRTS) algorithm. Unlike other spatially balanced sampling algorithms, GRTS can draw samples from sample frames of points (such as lakes), linear networks (such as rivers and streams), or polygons (such as estuaries) [@DKOW2022].  For the Oregon lakes, we will draw an equal probability sample of 50 lakes where each site has an equal inclusion probability. This equal probability sample can more accurately represent the lake target population by accounting for the natural spatial aggregation of the lakes on the landscape.

We begin by plotting the Oregon border and lakes together in Figure \@ref(fig:Ch11-micromap-oregon-lakes). Note that both spatial layers are simple features objects and the same coordinate reference system or CRS is required to plot them together. We ensure the CRS of each sf object using the syntax `st_crs(OR) == st_crs(OR_lakes)`.  

```{r Ch11-micromap-oregon-lakes, eval = TRUE, echo = FALSE, fig.cap = 'Map of Oregon lakes, with lakes represented as their centroids. The simple features object of the 3190 lakes is the sample frame from which the Generalized Random Tessellation Stratified function draws an n = 50 lakes for a spatially balanced probabilistic survey. '}
library(spsurvey)
library(tidyverse)

load("data/OR_spsurvey.RData")

ggplot() +
  geom_sf(
    data = OR_lakes,
    color = "blue"
  ) +
  geom_sf(
    data = OR,
    color = "black",
    fill = NA
  ) +
  labs(title = "Oregon Lakes:  Albers Equal Area Projection") +
  theme_bw()
```

The _OR_lakes_\index{Datasets!OR\_lakes} dataset is the sample frame that will be used with the `grts` function from the **spsurvey**\index{R Packages!spsurvey} R package to draw a spatially balanced probabilistic sample. That function explicitly incorporates spatial locations into the survey design.  Consequently, the sample frame needs to be in a projected coordinate reference system such as an Albers or UTM projection so a spatially balanced sample can be drawn [@OKP2012]. An equal-probability draw of 50 lakes is done by specifying `eqprob <- grts(OR_lakes, n_base = 50)`. That sample of 50 lakes is shown in Figure \@ref(fig:Ch11-micromap-spsurvey-lakes).

```{r Ch11-micromap-spsurvey-lakes, eval = TRUE, echo = FALSE, fig.cap = 'Oregon lakes Sample Frame & Equal Probability Sites in Projected Coordinates. Red points are the GRTS sample and blue points are the sample frame.'}
ggplot() +
  geom_sf(
    data = OR_lakes,
    color = "blue"
  ) +
  geom_sf(
    data = eqprob_base,
    color = "red"
  ) +
  geom_sf(
    data = OR,
    color = "black",
    fill = NA
  ) +
  labs(title = "Oregon Lakes Sample Frame & GRTS Equal Probability Sites") +
  theme_bw()
```

We demonstrated a practical example of how to import a target population dataset and used the spsurvey to create a spatially balanced sample.  This sample can then be used to collect and summarize data that can be further displayed with a linked micromap.  Next, we demonstrate the value of using spsurvey for increasing confidence in summary statistics with an example from West Virginia Department of Environment Protection (WVDEP) [@MPRG2016].

The survey designs by VDEQ and FDEP that led to the linked micromaps (Figures \@ref(fig:Ch11-micromapVDEQ) and \@ref(fig:Ch11-micromapFDEP)) were more sophisticated than the example of an equal probability draw.  Using a spatially balanced probabilistic survey drawn from **spsurvey**\index{R Packages!spsurvey} has two advantages.  First, from a design perspective, one obtains a good spatial coverage of the features to be sampled.  As @OKP2012 noted, users designing a spatially balanced survey often underestimate the effort required to build a single GIS layer having all the attributes needed for such a design.  Second, from an analytical perspective, if a spatially patterned response is observed among the sample sites then the local neighborhood variance estimator in **spsurvey** tends to outperform other alternatives by providing smaller variances, and consequently, confidence limits around the estimated means, totals, percentiles, and cumulative distribution functions from the survey [@SO2004;@OKP2012]. This advantage of the local neighborhood variance estimator over an independent random sample variance estimator has been shown in river and stream survey estimates of a fish community index of biotic integrity and a qualitative habitat evaluation index, and in reservoir survey estimates of chlorophyll-a (a measure of productivity), temperature, secchi depth (a measure of water clarity), and methane emission rates (a greenhouse gas) [@SO2003;@SO2004;@BMN2016].  Consequently, in a linked micromap of spatially balanced probabilistic survey results, one could have smaller confidence limits displayed around the means or percentiles because of the improved performance of the local neighborhood variance estimate.  This result is seen below for 25 basins in West Virginia with median estimates and 95% confidence limits for specific conductance, a measure of the ionic content of the sampled rivers and streams (Figure \@ref(fig:Ch11-micromapWVDEP)).  The left statistical panel has confidence limits using the local neighborhood variance estimator, and the right panel used the simple random sample variance estimator.  Some of the percentage reductions in confidence limits were quite modest. The Lower New basin had only ~ a 3% reduction; whereas others reductions were quite large, such as the Upper Ohio North/Upper Ohio South basin having a ~56% reduction.  The average percentage reduction among the 25 basins was ~28%.  The independent random sample variance estimator does not take into account any spatial patterning in the sampled survey responses [@SO2003].  If there is spatial patterning, such that responses at two sites close to each other are more similar than responses at two sites further apart, then that patterning is incorporated into the local neighborhood variance estimator [@SO2003].

```{r Ch11-micromapWVDEP, fig.width = 7, fig.height = 6, fig.cap = 'A linked micromap of specific conductance, a measure of the ionic content of the sampled rivers and streams, for 25 basins from spatially balanced probabilistic stream surveys done by West Virginia Department of Environmental Protection.  The first statistical panel displays the median and 95% confidence limits based on the default variance local neighborhood estimator.  The second statistical panel displays the median and 95% confidence limits based on specifying the simple random sample (SRS) variance estimator.', eval = TRUE, echo = FALSE}
library(sf)
library(micromap)

# sf data frame with geospatial and statistical data
WV_basins <- readRDS("data/wvbasins2.rds")

mmplot(
  map.data = WV_basins,
  panel.types = c("dot_legend", "labels", "dot_cl", "dot_cl", "map"),
  panel.data = list(
    NA,
    "alt.name",
    list("Cond_med", "Cond_LCB95Pct", "Cond_UCB95Pct"),
    list("Cond_med_srs", "Cond_LCB95Pct_srs", "Cond_UCB95Pct_srs"),
    NA
  ),
  ord.by = "Cond_med",
  rev.ord = TRUE,
  grouping = 5,
  colors = brewer.pal(6, "Blues")[-1],
  panel.att = list(
    list(
      1,
      panel.width = 1.3,
      point.type = 20,
      point.size = 2,
      point.border = FALSE,
      xaxis.title.size = 1
    ),
    list(
      2,
      header = "WVDEP\nSubbasins",
      panel.width = 1.25,
      align = "left",
      text.size = .8
    ),
    list(
      3,
      header = "Specific\nConductance Local",
      graph.bgcolor = "white",
      xaxis.ticks = list(0, 250, 500, 750),
      xaxis.labels = list(0, 250, 500, 750),
      xaxis.labels.size = 1,
      xaxis.title = "[\u03BCS/cm]",
      xaxis.title.size = 1,
      panel.width = 1.5,
      point.border = FALSE,
      line.width = 1.25,
      point.size = 1.5
    ),
    list(
      4,
      header = "Specific\nConductance SRS",
      graph.bgcolor = "white",
      xaxis.ticks = c(0, 250, 500, 750),
      xaxis.labels = c(0, 250, 500, 750),
      xaxis.labels.size = 1,
      xaxis.title = "[\u03BCS/cm]",
      xaxis.title.size = 1,
      panel.width = 1.5,
      point.border = FALSE,
      line.width = 1.25,
      point.size = 1.5
    ),
    list(
      5,
      header = "Light Gray Means\n Highlighted Above",
      inactive.border.color = gray(.7),
      inactive.border.size = 2,
      panel.width = 1.5
    )
  )
)
```

Two final points about the WVDEP linked micromap should be noted. First, we used a simple features object, **sf**\index{R Packages!sf}, that contained both the geospatial data and the statistical summaries of the basins. Consequently, we only had to specify the `map.data` argument, in contrast to the Virginia and Florida examples, where we used both the `map.data` and `stat.data` syntax.  Second, the basins in the linked micromap were smoothed; that is their polygon boundaries generalized, so they would draw faster as mentioned in Chapter \@ref(Ch2).


## Summary and Further Reading {#Ch11-FurtherReading}


Building off of the foundation in Chapter \@ref(Ch2), this chapter covered environmental applications of linked micromaps. Such applications have two unique aspects. One is the use of geographic boundaries from natural features, such as ecoregions or watersheds, instead of administrative or political boundaries that are more commonly used in other applications. See also Chapter \@ref(Ch12) for medical applications that deal with non-standard geographic boundaries, such as health service areas and census tracts. The other is the strong link between spatial survey design to obtain a representative sample from the natural features so that survey estimates and sample statistics can be displayed in linked micromaps. Data collected from spatially balanced probabilistic surveys by state environmental agencies were used in the linked micromap examples from this chapter. The FDEP example used a linked micromap for initial exploratory spatial data analysis and subsequent testing for spatial autocorrelation in pesticide detection among basins [@SWSCSS2018]. @MPRG2016 used linked micromaps for a multivariate spatial analysis of WVDEP water quality data.  The examples in this chapter were of static linked micromaps, but Chapter \@ref(Ch7) covers interactive exploration of spatial data, including one by VDEQ.  


## Acknowledgements {-}


 The manuscript has been subjected to the U.S. Environmental Protection Agency’s (USEPA) peer and administrative review and approved for publication. However, the views expressed are those of the authors and do not represent the views or policies of the USEPA. The mention of trade names or commercial products does not constitute endorsement or recommendation for use. We appreciate reviews provided by Michael Dumelle and Marc Weber at USEPA. We also want to acknowledge the data and input provided by Jason Hill and Emma Jones, both at VDEQ, James Silvanima at FDEP, and Michael Whitman at WVDEP.


\printbibliography[segment=\therefsegment,heading=subbibliography]