The buttons and sliders below the Preset dropdown can be used to build your own selection or refine a preset selection. Clicking on an individual satellite in the 3d view will display a panel with detailed information.
Links to NASA's website are provided for additional information. In addition to displaying name and orbital details, the 3d view displays the satellite's future trajectory with respect to the Earth's surface. By default the trajectory is for one day but this can be changed to either one hour or one week.
Perhaps the most surprising fact for users of this application is the large proportion of orbital objects classified as junk.
It is important to note that satellite positions are derived from an ephemeris database downloaded in July 14, As such this app will not display satellites launched since then or reflect intentionally or unintentionally orbital adjustments. Similarly atmospheric friction and gravitational forces are likely to influence orbital position.
However the different between projected and actual position is unlikely to be perceptible at the scale used. If you encounter any problems or have suggestions for improvements please don't hesiste to let us know here. The number of electoral college votes won and the share of votes cast within a state or county are expressed in spatial terms, but ultimately it is the numbers of people within those regions that matter.
Second, the regions themselves are of wildly differing sizes, and they differ in a way that is not well-correlated with the magnitudes of the underlying votes. The map makers also face choices that would arise in many other representations of the data.
Do we want to just show who won each state in absolute terms this is all that matters for the actual result, in the end or do we want to indicate how close the race was? Do we want to display the results at some finer level of resolution than is relevant to the outcome, such as county rather than state counts? How can we convey that different data points can carry very different weights, because they represent vastly larger or smaller numbers of people? It is tricky enough to convey these choices honestly with different colors and shape sizes on a simple scatterplot.
Often, a map is like a weird grid that you are forced to conform to even though you know it systematically misrepresents what you want to show. This is not always the case, of course. Sometimes our data really is purely spatial, and we can observe it at a fine enough level of detail that we can represent spatial distributions honestly and in a very compelling way. But the spatial features of much social science are collected through entities such as precincts, neighborhoods, metro areas, census tracts, counties, states, and nations.
These may themselves be socially contingent. A great deal of cartographic work with social-scientific variables involves working both with and against that arbitrariness. The election dataset has various measures of the vote and vote shares by state. Here we pick some columns and sample a few rows at random. Would a two-color choropleth map be more informative than this, or less?
It extends to the county level with an additional four digits, so every county in the US has a unique six-digit identifier, where the first two digits represent the state. This dataset also contains the census region of each state.
Of course, spatial representations can be very useful, and sometimes absolutely necesssary. But we can start with a state-level dotplot, faceted by region. This plot brings together many aspects of plot construction that we have worked on so far, including subsetting data, reordering results by a second variable, and using a scale formatter. It also introduces some new options, like allowing free scales on an axis, and manually setting the color of an aesthetic.
We break up the construction process into several steps by creating intermediate objects p0 , p1 , p2 along the way. It makes the code a little more readable. Bear in mind also that, as always, you can try plotting each of these intermediate objects as well just type their name at the console and hit return to see what they look like. As always, the first task in drawing a map is to get a data frame with the right information in it, and in the right order.
This just a data frame. It has more than 15, rows because you need a lot of lines to draw a good-looking map. The map is plotted with latitude and longitude points, which are there as scale elements mapped to the x and y axes.
A map is, after all, just a set of lines drawn in the right order on a grid. We can map the fill aesthetic to region and change the color mapping to a light gray and thin the lines to make the state borders a little nicer. By default the map is plotted using the venerable Mercator projection.
Assuming we are not planning on sailing across the Atlantic, the practical virtues of this projection are not much use to us, either. If you glance again at the maps in Figure 7. This is because they are using an Albers projection. Look, for example, at the way that the US-Canadian border is a little curved along the 49th parallel from Washington state to Minnesota, rather than not a straight line. Normally it is left implicit. Maps are more complex.
Our locations and borders are defined on a more or less spherical object, which means must have a method for transforming or projecting our points and lines from a round to a flat surface. The many ways of doing this gives us a menu of cartographic options. The Albers projection requires two latitude parameters, lat0 and lat1. We give them their conventional values for a US map here.
Try messing around with their values and see what happens when you redraw the map. Now we need to get our own data on to the map.
Remember, underneath that map is just a big data frame specifying a large number of lines that need to be drawn. We have to merge our data with that data frame. Somewhat annoyingly, in the map data the state names in a variable named region are in lower case. We can create a variable in our own data frame to correspond to this, using the tolower function to convert the state names. This merge step is important!
You need to take care that the values of the key variables you are matching on really do exactly correspond to one another. If they do not, missing values NA codes will be introduced into your merge, and the lines on your map will not join up.
If the keys have different names in each data set you can specify that, if needed. To reiterate, it is important to know your data and variables well enough to check that they have merged properly. Do not do it blindly. Maps that look broken when you draw them are usually caused by merge errors. But errors can also be subtle. For example, perhaps one of your state names inadvertently has a leading or, worse, a trailing space as a result of the data originally being imported from elsewhere and not fully cleaned.
So, be careful. We have now merged the data. Now that everything is in one big data frame, we can plot it in a map. To complete the map, we will use our party colors for the fill, move the legend to the bottom, and add a title.
You can also see the code for the map theme in the Appendix. With the map data frame in place, we can map other variables if we like. The default color used in the p1 object is blue. In addition, the gradient runs in the wrong direction. In our case, the standard interpretation is that a higher vote share makes for a darker color. We fix both of these problems in the p2 object by specifying the scale directly.
For election results, we might prefer a gradient that diverges from a midpoint. Alternatively, we can re-specify the mid-level color along with the high and low colors. We will make purple our midpoint, and use the muted function from the scales library to tone down the color a little.
This is because Washington DC is included in the data, and hence the scale. Even though it is barely visible on the map, DC has by far the highest points margin in favor of the Democrats of any unit of observation in the data. This brings out the familiar choropleth problem of having geographical areas that only partially represent the variable we are mapping. In the U. The problem is evident at the state level, as we have seen, also arises even more at the county level.
County-level US maps can be aesthetically pleasing, because of the added detail they bring to a national map. But they also make it easy to present a geographical distribution to insinuate an explanation.
The results can be tricky to work with. When producing county maps, it is important to remember that the states of New Hampshire, Rhode Island, Massachussetts, and Connecticut are all smaller in area than any of the ten largest Western counties. Many of those counties have fewer than a hundred thousand people living in them.
Some have fewer than ten thousand inhabitants. The result is that most choropleth maps of the U. The other big variable, in the U. The procedure is essentially the same as it was for the state-level map. We need two data frames, one containing the map data, and the other containing the fill variables we want plotted.
Because there are more than three thousand counties in the United States, these two data frames will be rather larger than they were for the state-level maps.
The datasets are included in the socviz library. The county map data frame has already been processed a little in order to transform it to an Albers projection, and also to relocate and rescale Alaska and Hawaii so that they fit into an area in the bottom left of the figure. This is better than throwing away two states from the data.
The steps for this transformation and relocation are not shown here. It looks the same as our State map data frame, but it is much larger, running to almost , rows. The id field is the FIPS code for the county. Next, we have a data frame with county-level demographic, geographic, and election data:. This data frame includes information for entities besides counties, though not for all variables. If you try out the p1 object you will see that ggplot produces a legible map, but by default it chooses an unordered categorical layout.
We could recode it so that R is aware of the ordering. We will learn more about this scale function in the next chapter. We also tweak how the legend is drawn using the guides function to make sure each element of the key appears on the same row. Again, we will see this use of guides in more detail in the next chapter.
We can now do exactly the same thing for our map of percent Black population by county. Figures 7. Between the two of them, population density and percent Black will do a lot to obliterate many a suggestively-patterned map of the United States. The first is an effort to replicate a poorly-sourced but widely-circulated county map of firearm-related suicide rates in the United States.
The rates are binned into six categories. We will match the color palettes between the maps, but for the population map we will flip our color scale around so that less populated areas are shown in a darker shade. We do this by using a function from the RColorBrewer library to manually create two palettes. The rev function used here reverses the order of a vector.
The brewer. The colors are specified in hexadecimal format. Again, we will learn more about color specifications and how to manipulate palettes for mapped variables in Chapter 8. Having drawn the gun plot, we use almost exactly the same code to draw the reverse-coded population density map. However, the visual impact of the first has a lot in common with the second. The dark bands in the West except for California stand out, and fade as we move toward the center of the country.
There are some strong similarities elsewhere on the map too, such as in the Northeast. The gun-related suicide measure is already expressed as a rate. However, this sort of standardization has its limits. In particular, when the event of interest is not very common, and there is very wide variation in the base size of the units, then the denominator e.
Before tweeting this picture, please read the text for discussion of what is wrong with it. Third, and more subtly, the data is subject to reporting constraints connected to population size. If there are fewer than ten events per year for a cause of death, the Centers for Disease Control CDC will not report them at the county level because it might be possible to identify particular deceased individuals.
0コメント