Limitations, notes, and caveats
Importing and displaying river network data
Basic distance calculation in a non-messy river network
Incorporating flow direction
Allowing different route-detection algorithms: a possible time-saver
Beyond individuals: summarizing or plotting at the dataset level
Editing a river network object, or fixing a messy one
Dealing with braided channels
The ‘riverdist’ package is intended as a free and readily-available resource for distance calculation along a river network. This package was written with fisheries research in mind, but could be applied to other fields. The ‘riverdist’ package builds upon the functionality of the ‘sp’ and ‘rgdal’ packages, which provide the utility of reading GIS shapefiles into the R environment. What ‘riverdist’ adds is the ability to treat a linear feature as a connected network, and to calculate travel routes and travel distances along that network.
Import a linear shapefile using
line2network(). The shapefile to import must be a linear feature - converting a polygon feature to a line feature will result in an outline, which will not be useable. The shapefile to import should be as simple as possible. If GIS software is available, both trimming the shapefile to the same region as the study area and performing a spatial dissolve are recommended.
Clean up the imported shapefile however necessary. Editing functions are provided, but the
cleanup() function interactively steps through all the editing functions in a good order, and is recommended in nearly all cases.
Convert point data to river locations using
ptshp2segvert(). This will snap each point to the nearest river network location.
Perform all desired analyses.
Care must be exercised in the presence of braided channels, in which multiple routes may exist between river locations. In these cases, the default route calculation provided by ‘riverdist’ will be the shortest travel route, but it is possible that the shortest travel route may not be the route of interest. Functions are provided to check for braiding, as well as select the route of interest. If no braiding is detected, route and distance calculation can switch to a more efficient algorithm.
Another important note is that only projected data can be used, both for river networks and for point data. The ‘riverdist’ environment in R does not share the ability of GIS software to project-on-the-fly, and treats all coordinates on a linear (rectangular) scale. Therefore, the projection of all data must also be the same.
Importing a river network using
In a typical workflow, a user will first import a projected polyline shapefile using
line2network(). This function reads the specified shapefile using the ‘sp’ and ‘rgdal’ packages, and adds the network connectivity, thus creating a river network object. If an unprojected shapefile is detected, an error will be generated. However, the
reproject= argument allows the
line2network() to re-project the shapefile before importing it as a river network.
While ‘riverdist’ does provide tools for editing a river network that do not rely on GIS software, it is strongly recommended to simplify the river shapefile as much as possible before importing into R. In particular, a spatial dissolve will likely be very helpful, if GIS software is available. This will create a few long line segments instead of many, many short segments.
library(riverdist) MyRivernetwork <- line2network(path=".", layer="MyShapefile") # Re-projecting in Alaska Albers Equal Area projection: AKalbers <- "+proj=aea +lat_1=55 +lat_2=65 +lat_0=50 +lon_0=-154 +x_0=0 +y_0=0 +datum=NAD83 +units=m +no_defs +ellps=GRS80 +towgs84=0,0,0" MyRivernetwork <- line2network(path=".", layer="MyShapefile", reproject=AKalbers)
Displaying a river network using
Basic plotting is provided using a method of the
plot() function. See
help(plot.rivernetwork) for additonal plotting arguments and graphical parameters. Shown below is the Gulkana River network, included as a dataset.
Once the river network has been imported, some cleanup may be necessary, depending on the structure of the shapefile. Ideally, there should be one segment between each endpoint or junction node, and the spatial extent of the shapefile should be limited to the region or network of interest. If the user has GIS software available, it may be simplest to format the shapefile as desired before importing into R. However, functions are included for river network formatting in R if necessary. The
cleanup() function interactively steps through the formatting functions in a good sequence, and may be a good way to get started. If there were spatial oddities with the parent shapefile (vertices out of order, or strange “jumps” in segments), the
cleanup_verts() function interactively steps through each individual segment, providing a means to edit the vertices of each segment if needed.
Checking connectedness using
For route and distance calculation to work, the topologies must be correct, with all the right segments being treated as connected. The
topologydots() function can check for this, and plots connected segment endpoints as green, and non-connected endpoints as red. This is shown below, and all appears good.
Converting XY data to river locations using
Once an appropriate river network has been obtained, point data can be read. Function
xy2segvert() converts XY data into river locations, by “snapping” each point to the closest segment and vertex. Because of this, river locations can only be used in the context of the river network they belong to. In addition to the river locations, the snapping distance for each observation is also returned.
fakefish dataset includes coordinates from a sequence of telemetry flights.
## seg vert snapdist ## 1 1 595 329.34419 ## 2 1 399 40.27721 ## 3 1 352 402.52259 ## 4 1 116 525.06623 ## 5 1 806 355.32753 ## 6 1 505 11.34949
Spatial data can be read directly from a point shapefile using
pointshp2segvert(), which returns the resulting river locations added to the data table from the point shapefile.
Displaying point data in river locations using
riverpoints() function works essentially like
points() to overlay point data on an existing plot, but using river locations (segment and vertex). The
zoomtoseg() function produces a plot zoomed to the specified segment number, or vector of segment numbers.
In the plot below, the raw coordinates are displayed as red circles and the river locations are displayed as blue squares.
Computing network distance using
River network distance can be calculated directly with the
riverdistance() function. The
riverdistance() function calls
detectroute() internally, which the user will probably never need, but usage is shown below. The
riverdistance() function needs river locations (segment and vertex) for both starting and ending locations. Specifying
map=TRUE as shown below is not necessary, but can provide a check to verify that the function is working properly.
##  7 6 3 4 10 11 14
##  155435.2
Computing network distances between sequential observations of individuals using
River distance can be calculated manually between any two connected locations on a river network. However, a few common summary analyses were automated and are included in ‘riverdist’. First, the
riverdistanceseq() function returns a matrix of the distances between sequential observations for a set of individuals observed multiple times. In the example below, three fish were observed (or not) during 5 telemetry flights. The dataset is shown first, then the output from
riverdistanceseq(). Fish number 1 traveled 83.87 km between flights 3 and 4, and was not observed during flight 2.
## seg vert id flight ## 1 10 12 1 1 ## 2 2 17 1 3 ## 3 10 15 1 4 ## 4 1 641 1 5 ## 5 6 258 2 1 ## 6 11 393 2 2 ## 7 2 26 2 3 ## 8 9 354 2 4 ## 9 4 472 2 5 ## 10 2 25 3 1 ## 11 9 137 3 2 ## 12 9 181 3 4
## 1 to 2 2 to 3 3 to 4 4 to 5 ## 1 NA NA 83872.62 70277.57 ## 2 127558.14 109561.1 94177.71 26741.99 ## 3 86983.02 NA NA NA
Computing a matrix of network distances between all observations of an individual using
River distance can also be calculated between all observations of a single individual. In the output matrix shown below for individual 1, the element with row identifier
1 and column identifier
3 represents the river distance from the location observed in survey 1 to the location observed in survey 3, calculated as 83.6 km. An important note is that the distances reported are net distance between locations, and are not intended to be cumulative. Specifying
full=TRUE reports an output matrix that includes all observations, with values of
NA if the individual was not observed. This allows the output matrices for multiple individuals to have the same rows and columns, thus being directly comparable.
## 1 3 4 5 ## 1 0.0000 83588.99 283.6307 69993.94 ## 3 83588.9911 0.00 83872.6219 45678.35 ## 4 283.6307 83872.62 0.0000 70277.57 ## 5 69993.9398 45678.35 70277.5705 0.00
Computing minimum observed home range for individuals using
The minimum observed (linear) home range for each individual can be calculated using
## Minumum home ranges associated with each individual ## ## ID range ## 1 1 99914.27 ## 2 2 141136.24 ## 3 3 88669.52
Maps can be produced by calling
plot() on an object returned from
cumulative=TRUE will create plots with line thickness varying by the number of times an individual would have traveled a given section of river. For a cumulative plot, either a vector of survey identifiers must be used, or else the data locations must be in chronological order for each individual.
The overlap between the respective home ranges of all individuals can be described using
homerangeoverlap() This returns three matrices: first, the linear distance represented in the union of the home ranges of each pair of individuals (
$either), then the linear distance represented in the intersection (
$both), then the proportional overlap (
$prop_both), defined as intersection/union. Approximately 78% of the linear home range occupied by either individual 1 or 3 is shared by both of them.
The amount of overlap in home ranges can also be plotted using
## $either ## 1 2 3 ## 1 99914.27 157177.9 106065.85 ## 2 157177.89 141136.2 141136.24 ## 3 106065.85 141136.2 88669.52 ## ## $both ## 1 2 3 ## 1 99914.27 83872.62 82517.95 ## 2 83872.62 141136.24 88669.52 ## 3 82517.95 88669.52 88669.52 ## ## $prop_both ## 1 2 3 ## 1 1.0000000 0.5336159 0.7779879 ## 2 0.5336159 1.0000000 0.6282548 ## 3 0.7779879 0.6282548 1.0000000
Computing a matrix of network distances between all observations using
A matrix of the river network distance between every observation and every other observation can also be calculated using
riverdistancemat(). A use for this function might be if the user wishes to calculate all distances at once, and subset later.
## 1 2 3 4 5 6 7 ## 1 0 83589 284 69994 101953 25605 83956 ## 2 83589 0 83873 45678 22200 109194 367 ## 3 284 83873 0 70278 102236 25322 84239 ## 4 69994 45678 70278 0 64042 95599 46045 ## 5 101953 22200 102236 64042 0 127558 22567 ## 6 25605 109194 25322 95599 127558 0 109561 ## 7 83956 367 84239 46045 22567 109561 0
logical argument can be used for subsetting, if the full network distance matrix is not needed. This capability is shown below, calculating the distance matrix only for observations occurring on segment number 2. The
ID argument can be used with a vector of observation labels, to display row and column labels that may be easier to interpret than observation indices.
## id1-flight3 id2-flight3 id3-flight1 ## id1-flight3 0.0000 366.62343 326.28884 ## id2-flight3 366.6234 0.00000 40.33459 ## id3-flight1 326.2888 40.33459 0.00000
Computing a matrix of network distances between all observations of two different datasets using
A matrix of river network distance between two location datasets can be calculated using
riverdistancetofrom(), which can be similarly subsetted using the
logical2 arguments. Row and column names can be added using the
An example of the use of this function might be computation of distance between observations of instrumented fish and a set of fixed river locations, such as contaminant sites. In this case, a similar function
upstreamtofrom() (discussed later) might be used, which would give the means to calculate up- or downstream distance from contaminant sites, with the option of examining flow-connectedness.
streamlocs.seg <- c(2,2,2) streamlocs.vert <- c(10,20,30) streamlocs.ID <- c("loc A","loc B","loc C") logi2 <- (smallset$seg==2) obsID <- paste0("id",smallset$id,"-flight",smallset$flight) riverdistancetofrom(seg1=streamlocs.seg, vert1=streamlocs.vert, seg2=smallset$seg, vert2=smallset$vert, ID1=streamlocs.ID, ID2=obsID, logical2=logi2, rivers=Gulk)
## id1-flight3 id2-flight3 id3-flight1 ## loc A 129.95819 496.5816 456.2470 ## loc B 99.42056 267.2029 226.8683 ## loc C 479.65380 113.0304 153.3650
Defining flow direction of a river network using
Flow direction and directional (upstream) distance can also be calculated. For this to be accomplished, the segment and vertex of the river “mouth”, or lowest point must first be identified. The segment containing the river mouth can be visually identified from a plot of the river network. In the Gulkana River example, the lowest segment happens to be segment 1. Identifying the lowest vertex of segment 1 can be done using the plot produced by
showends(), shown below. In this case, the mouth vertex happens to be vertex 1. This will not necessarily be the case. After importing a shapefile into R, the segment vertices will be stored in sequential order, but not necessarily by flow direction. Specifying the segment and vertex coordinates of the river network mouth can be done using
setmouth() as shown below, though it can also be set manually by direct assignment.
Calculating flow direction using
riverdirection() and directional network distance using
If the flow direction has been established by specifying the river network mouth, river direction can be calculated using
riverdirection(), and upstream distance can be calculated using
upstream(). If the input river locations are flow-connected,
riverdirection() returns “up” if the second location is upstream of the first and “down” if downstream. In the flow-connected case,
upstream() returns the network distance as positive if the second location is upstream of the first, and negative if downstream. If the input locations are not flow-connected,
riverdirection() returns “up” if the total upstream distance is greater than the total downstream distance, and “down” otherwise. In the non flow-connected case, upstream distance in
upstream() can return one of two things, depending on the user’s research intent. Specifying
net=TRUE will return the “net” distance (upstream distance - downstream distance between the two locations). Specifying
net=FALSE (the default) will return the total distance between the two locations, with the sign depending on whether the upstream distance exceeds the downstream distance.
For example, the route between two points goes downstream along a river for 100m, then up 20m on a tributary. Specifying
net=TRUE will return a distance of -80m. Specifying
net=FALSE will return a distance of -120m.
flowconnected=TRUE in both
upstream() will only return distances or directions if the input locations are flow-connected, and will return
##  "down"
##  -66235.32
##  -28636.75
##  NA
Automations of flow direction and directional network distance
River direction and upstream distance are also applied in functions
upstreamseq() which work like
upstreammatbysurvey() which work like
upstreammat() which work like
upstreamtofrom which work like
upstreammat() use the additional
net= argument, and all river direction and upstream distance functions use the additional
Currently three algorithms are implemented to detect routes between river locations, with the functions automatically selecting the most appropriate unless another is specified. Dijkstra’s algorithm is used by default (
algorithm="Dijkstra"), which returns the shortest route in the presence of braiding. The sequential algorithm (
algorithm="sequential") may be used, which returns the first complete route detected. The sequential algorithm is much slower and is not recommended in nearly all cases, but is retained as an option for certain checks.
If many distance calculations or a more sophisticated analysis is to be conducted, it is highly recommended to run
algorithm="segroutes" to be used. This adds route and distance information to the river network object, greatly simplifying distance calculation, and reducing processing time for each distance calculation. The
buildsegroutes() function also includes an option to calculate a distance lookup table as well. This may take a few seconds to run, but will reduce computation times even further - cutting the already-fast segment route algorithm time by 50-80%. Lookup tables can be calculated directly using
buildlookup(), but may be extremely slow to calculate without running
In the example below, distance is calculated between the two points in a complex river network, using all three route detection algorithms, and calculating the time requirement for a single calculation.
##  68937.76
## Time difference of 0.1534309 secs
##  68937.76
## Time difference of 0 secs
##  68937.76
## Time difference of 0.01698399 secs
In this case, building segment routes and distance lookup tables takes a little more than a second, even with a fairly complex river network. Calculating a single distance afterwards takes about 80 microseconds using segment routes and about 25 microseconds using the lookup tables. This dramatically saves time if multiple distances are to be computed, such as in a large matrix of distances, or multiple analyses. Kernel density and K-function analysis is probably prohibitively slow otherwise.
Calculating and plotting kernel density using
A method has been provided to display kernel density calculated from point data, using river network distance. The
makeriverdensity() function calculates scaled kernel density at approximately regularly-spaced river network locations, with the linear resolution specified by the optional
resolution= argument. A gaussian (normal) kernel is used by default, but a rectangular (simple density) kernel can be used as well. The
survey= argument may also be used with a vector of survey identifiers corresponding to the point data. If the
survey= argument is used,
makeriverdensity() will calculate separate densities for each unique survey, and a method of
plot() will produce a separate plot for each unique survey.
plot() function method can display densities using line thickness, color, or both. For additional plotting arguments including possible color ramps, see
Densities for nine of the ten Fakefish surveys are shown below.
data(Gulk, fakefish) fakefish_density <- makeriverdensity(seg=fakefish$seg, vert=fakefish$vert, rivers=Gulk, survey=fakefish$flight.date, resolution=2000, bw=10000) par(mfrow=c(3,3)) plot(x=fakefish_density, ramp="blue", dark=0.85, maxlwd=15, whichplots=c(1:8,10)) # showing only nine plots for clarity
An alternate use of
homerange() to give the spatial data spread for each flight event
homerange() function could also be used to give a measure of the spatial spread of observations for each flight event, by using the flight identifier instead of individual identifier in the
In this case, the observations are the most closely spaced in the first flight event, and the most widely spaced in the August 11 flight event. Interestingly, the density plot of the August 11 flight shows many of the the individuals to be located near one another, but spread to multiple tributaries, with little presence in between. Considering this event to be the largest spread may still be appropriate, depending on the study, since all individuals were observed as concentrated in the lower mainstem on April 1.
## Minumum home ranges associated with each individual ## ## ID range ## 1 2015-04-01 66026.33 ## 2 2015-04-19 145231.00 ## 3 2015-06-01 84856.73 ## 4 2015-06-21 136195.56 ## 5 2015-07-07 176480.46 ## 6 2015-08-11 235446.27 ## 7 2015-09-05 171769.53 ## 8 2015-09-20 130524.91 ## 9 2015-10-29 140431.19 ## 10 2015-11-25 124827.73
Plotting clustering or dispersal using K-functions, with
K-functions, defined here as the average proportion of additional individuals located within a given distance of each individual, can be a useful tool for investigating evidence of clustering or dispersal. Calling
kfunc() and specifying the flight indentifier will result in a sequence of plots, each displaying the K-function associated with each survey.
Unless otherwise specified, each plot will be overlayed with a confidence envelope, calculated by resampling all within-survey distances, thus creating a null distribution under the assumption that clustering is independent of survey. Thus, a K-function above the envelope at a small distance range can be seen as evidence that locations were more clustered than expected for that survey; conversely, a K-function below the envelope at a small distance range can be seen as evidence that locations were more dispersed than expected for that survey.
The plots below show relatively high amounts of clustering for the 2015-04-01 and 2015-09-20 surveys. The 2015-07-07 and especially 2015-08-11 surveys show possible clustering at small distances and very strong dispersal at medium distances, which can be corroborated with the density maps shown above..
Summarizing up-river position, defined as distance from mouth, using
mouthdist() for one observation and
mouthdistbysurvey() for a dataset
In some cases, a meaningful summary measure may be the up-river position of each observation. Distance between an individual observation and the river mouth can be calculated using
mouthdist(). In the case of multiple observations of a set of individuals, a summary matrix can be calculated using
mouthdistbysurvey(), which returns a matrix of distances from the river mouth, with each row corresponding to a unique individual and each column corresponding to a unique survey.
##  56016.35
## 1 2 3 4 5 6 7 8 9 10 ## 1 26535 121619 NA 131182 162899 NA NA 105966 101516 65373 ## 2 10333 NA NA 105166 NA NA NA 99972 NA NA ## 3 NA 7581 106455 NA 150986 164013 125509 106237 102427 14046 ## 4 NA 108482 117334 NA 145338 131182 146021 NA NA 4878 ## 6 NA 125610 NA NA NA 165423 105344 NA 108398 NA ## 7 10153 75961 88672 155589 131218 NA NA NA NA NA ## 8 56016 NA 106901 NA NA NA 166606 104745 95707 64041 ## 9 46998 96425 130164 106331 NA NA NA 105344 NA 122739 ## 10 36393 NA NA 99958 132170 NA NA 82537 75916 NA ## 11 76180 NA NA NA NA 139313 NA 104850 11434 107265 ## 12 NA 63984 NA 106901 NA NA 130796 NA NA 97886 ## 13 NA NA NA 120679 145370 174795 142849 NA NA NA ## 14 31502 NA 80465 105835 144800 140778 106050 NA 126070 56293 ## 15 NA 92050 131218 NA 147101 130009 NA NA 52751 NA ## 16 45619 NA NA NA NA NA 106656 104879 117856 8714 ## 17 38397 NA 105165 NA 133124 160212 NA NA NA NA ## 18 NA 83037 86457 153561 NA NA 90940 NA NA 129706 ## 19 NA 103833 NA 107267 NA 196411 NA 130493 89983 NA ## 20 NA NA 106260 NA 136400 141994 104879 29574 NA NA
Plotting the summary matrix from
mouthdistbysurvey() or other distance sequence using
The resulting matrix of up-river position returned from
mouthdistbysurvey(), or another distance sequence returned from
upstreamseq(), can be plotted using
plotseq(). A few types of plots are available, with two shown below. Additional
type= arguments can be seen by calling
The user is cautioned to use any plots returned as descriptive tools only, as ANOVA-type inference would likely be inappropriate without accounting for repeated-measures and/or serial autocorrelation.
In the example below, the plot shows evidence of the instrumented fish beginning the sampling period near the river mouth, then migrating upriver, and finally ending the sampling period near the mouth.
Depending on the study,
plotseq() may be useful in plotting the upstream distance between observations of all individual, as calculated in
upstreamseq(). It is worth noting that the outputs from both
riverdistanceseq() may have many empty cells, if individuals were missed on any flights. For the purpose of clarity, the default plot type is used below, which produces boxplots. However, in this case, a jittered dotplot may be more appropriate due to the small sample sizes, and can be produced by specifying
In this case, the plots show evidence that movement was generally up-river between flights 1 and 2 and between flights 2 and 3, that there was the greatest variability in upstream movement between flights 5 and 6, and that movement was generally down-river in the latter part of the study.
## 1 to 2 2 to 3 3 to 4 4 to 5 5 to 6 6 to 7 7 to 8 8 to 9 9 to 10 ## 1 95084 NA NA 31717 NA NA NA -54220 -36143 ## 3 NA 98874 NA NA 161737 -136261 -78484 -3810 -88381 ## 4 NA 8852 NA NA -14156 14839 NA NA NA ## 6 NA NA NA NA NA -62068 NA NA NA ## 7 65807 12711 66918 -24372 NA NA NA NA NA ## 8 NA NA NA NA NA NA -62652 -9038 -31666 ## 9 49427 33739 -83233 NA NA NA NA NA NA ## 10 NA NA NA 32212 NA NA NA -6621 NA ## 11 NA NA NA NA NA NA NA -93416 95831 ## 13 NA NA NA 112787 166903 -52141 NA NA NA ## 14 NA NA 33038 41936 -132316 -93566 NA NA -69777 ## 15 NA 70005 NA NA -123847 NA NA NA NA ## 16 NA NA NA NA NA NA -1777 69473 -109142 ## 17 NA NA NA NA 140074 NA NA NA NA ## 18 NA 3419 67104 NA NA NA NA NA NA ## 19 NA NA NA NA NA NA NA -67214 NA ## 20 NA NA NA NA 125132 -38174 -75305 NA NA
Plotting all upstream movements using
In a more generalized application, the
matbysurveylist() function calculates the distances or upstream distances, between each pair of observation events, for all individuals. Functionally, this is a summary of the outputs of
upstreammatbysurvey(), but for all individuals in a dataset. The output from
matbysurveylist() can then be plotted using
plotmatbysurveylist(), providing a one-plot summary of all movements within a dataset. It is recommended to use the default
matbysurveylist(), which calculates directional (upstream) distances, which will result in a more informative plot.
The output plot is the upper triangle of a matrix of plots, in which plot
[i,j] represents the upstream distances traveled between observation
i and observation
j, for all individuals observed in those observations. Each plot is overlayed with a horizontal line at an upstream distance of zero, for an illustration of up-river or down-river movement trend for that pairwise movement. It is worth noting that all sequential pairings of events (first to second, second to third, etc.) fall on the lower edge of the triangle, and this sequence of plots is the same as those given by
plotseq() in the previous example.
Three types of plots can be produced using the
type= argument. The default boxplot type (
type="boxplot") is shown below. Alternately, if the
type= argument is set to
"confint", each plot gives a line showing the extent of an approximate 95% confidence interval for the mean upstream distance traveled. Inference should be made with caution, as sample sizes are likely to be small, and no attempt has been made at using a family-wise confidence level. However, plotting confidence intervals may be further illustrative of trend. A jittered dotplot may be produced by specifying
type="dotplot", which will be the most appropriate in the case of small sample sizes. The default boxplot is shown below for the purpose of clarity, but a jittered dotplot would likely be a better choice in this instance.
Ideally, the shapefile used to define a river network object should be refined in GIS before importing into R. That being said, there are sure to be instances in which it is advantageous or necessary to make changes to the river network object within R.
All-purpose river network cleanup using
In many cases, the
cleanup() function will be the recommended first step in fixing a messy river network object after importing it. The
cleanup() function should be called within the console, and interactively calls the editing functions in sequence. It then returns a new, edited river network object which can be edited further, or used as is. It may even be the most straightforward to call
cleanup() multiple times. If there were spatial oddities with the parent shapefile (vertices out of order, or strange “jumps” in segments), the
cleanup_verts() function interactively steps through each individual segment, providing a means to edit the vertices of each segment if needed.
Usage will look like
Issue: the river network contains unneeded or unconnected segments - fixes using
This is a very likely issue, particularly if the network was imported without changes in a GIS environment. River network segments can be manually removed using
trimriver(), as shown below. Using the argument
trim= removes the specified segments, and using the argument
trimto= removes all but the specified segments. The example below is fairly simplistic, but illustrates the usage of the
It is also possible to trim a river network to include only segments that are within a spatial tolerance of a set of X-Y points. The
trimtopoints() function offers three methods of doing this. Specifying
method="snap" (the default) returns a river network made up only of the closest segments to the input points. This is the simplest method, but may result in spatial gaps, as shown in the example below. Specifying
method="snaproute" returns a network of the closest segments to the input points, but also includes any segments necessary to maintain a connected network. Specifying
method="buffer" returns a river network made up of segments with endpoints or midpoints located within a specified “buffer” distance of the input points. This may be advantageous if the user wants to include segments that are near, but not directly proximal, to the input points.
data(Kenai3) x <- c(174185, 172304, 173803, 176013) y <- c(1173471, 1173345, 1163638, 1164801) Kenai3.buf1 <- trimtopoints(x=x, y=y, rivers=Kenai3, method="snap") Kenai3.buf2 <- trimtopoints(x=x, y=y, rivers=Kenai3, method="snaproute") Kenai3.buf3 <- trimtopoints(x=x, y=y, rivers=Kenai3, method="buffer", dist=5000) plot(x=Kenai3, main="original") points(x, y, pch=15, col=4) legend(par("usr"), par("usr"), legend="points to buffer around", pch=15, col=4, cex=.6)
Automatically removing all segments not connected to the river network mouth can be done using
removeunconnected(). This may be useful if extraneous lines are retained from an import from GIS. This function may take some time to process, and simplifying the network using
dissolve() is recommended.
data(Koyukuk2) Koy_subset <- trimriver(trimto=c(30,28,29,3,19,27,4),rivers=Koyukuk2) Koy_subset <- setmouth(seg=1,vert=427,rivers=Koy_subset) Koy_subset_trim <- removeunconnected(Koy_subset) par(mfrow=c(1,2)) plot(x=Koy_subset, main="original") plot(x=Koy_subset_trim, main="unconnected segments removed")
Issue: the river network contains sequential “runs” of segments that do not otherwise branch - a fix using
Using an unnecessarily complex river network can greatly increase processing time. Runs of segments can be combined using
dissolve(), which works much like a spatial dissolve within GIS.
Issue: the river network segments do not break where they should - a fix using
This issue is problematic, as it directly affects how connectivity is detected within the river network. Without appropriate connectivity, routes and distances cannot be calculated. To address this issue,
splitsegments() automatically breaks segments where another segment endpoint is detected.
In the example below, segments 7, 8, 13, and 16 need to be split in multiple places. Since connectedness is not detected for the associated tributaries,
topologydots() shows the endpoints as red, or unconnected. In this case, calling
splitsegments() breaks the segments in the appropriate places, allowing for the network to be connected as it should be. It is worth noting that the user can specify which segments to split, with respect to which, which may aid in processing time and specificity.
Issue: segments that should connect do not - a fix using
Segments (or vectors of segments) can be manually “attached” at their endpoints or closest points using
data(Koyukuk0) Koyukuk0.1 <- connectsegs(connect=21, connectto=20, rivers=Koyukuk0) par(mfrow=c(1,2)) plot(Koyukuk0, ylim=c(1930500,1931500), xlim=c(194900,195100), main="original") topologydots(Koyukuk0, add=TRUE) plot(Koyukuk0.1,ylim=c(1930500,1931500), xlim=c(194900,195100), main="connected") topologydots(Koyukuk0.1, add=TRUE)
Issue: the river network contains segments that are smaller than the connectivity tolerance - a fix using
This is an issue that may be difficult to recognize, and may cause mysterious problems with network topology, sometimes preventing route calculation. These “microsegments” can be removed using
Issue: A long straight-line section of the river network does not contain vertices between endpoints - a fix using
In some cases, such as when a river network contains a lake, the shapefile will contain long straight-line sections, with vertices retained only for the beginning and end. If point data exist in these regions, when they are converted to river locations, they will be snapped to the nearest vertex - in this case, one of the endpoints of the straight stretch. Since it is likely that a greater degree of precision is desired in distance calculations,
addverts() provides a method of inserting additional vertices, wherever the river network contains vertices that are spaced at a greater distance than a specified threshold.
The example below first shows how the Skilak Lake section of the Kenai River network was originally read from a shapefile, with the vertices of segment 74 overlayed. The second plot shows the same section with the vertices of segment 74 overlayed, after adding vertices every 200 meters to the full river network.
Most of the utility of the ‘riverdist’ package was designed assuming a truly dendritic river network, in which there is no braiding of channels and only one path exists between one river location and another. If this is not the case and a braided network is used, the user is strongly cautioned that the distances reported may be inaccurate. In the event of braiding, the shortest route between two locations is returned, but the possibility exists that this may not be the route desired.
Checking for braiding using
Braiding can be checked for in a river network as a whole using
checkbraided(), which can take a while to run on a large or complex network. In the example below, no braiding exists in the Gulkana River network, and severe braiding exists in the Killey River West channel network.
## ## No braiding detected in river network.
## ## Braiding detected in river network. Distance measurements may be inaccurate.
Braiding can also be checked for specific routes. In the example below, braiding does not exist between segments 1 and 7, but does exist between segments 1 and 5.
## Note: any point data already using the input river network must be re-transformed to river coordinates using xy2segvert() or ptshp2segvert().
## No braiding detected between segments.
## Braiding detected between segments. Distance measurements may be inaccurate.
Investigating multiple routes using
If the user wishes to explore the possibility of multiple routes beween two locations,
routelist() detects a list of routes from one segment to another, which is applied by function
riverdistancelist() to calculate the distance along the routes that were detected. This was by means of randomization in a previous version, but now uses an algorithm that returns a complete list of possible routes. The
riverdistancelist() function, shown below, returns a list of all routes detected by ascending distance, and the corresponding distances.
## $routes ## $routes[] ##  1 2 4 15 16 ## ## $routes[] ##  1 2 4 5 6 8 9 11 12 14 16 ## ## $routes[] ##  1 3 4 15 16 ## ## $routes[] ##  1 3 4 5 6 8 9 11 12 14 16 ## ## $routes[] ##  1 2 4 5 7 8 9 11 12 14 16 ## ## $routes[] ##  1 2 4 5 6 8 10 11 12 14 16 ## ## $routes[] ##  1 3 4 5 7 8 9 11 12 14 16 ## ## $routes[] ##  1 3 4 5 6 8 10 11 12 14 16 ## ## $routes[] ##  1 2 4 5 6 8 9 11 13 14 16 ## ## $routes[] ##  1 2 4 5 7 8 10 11 12 14 16 ## ## $routes[] ##  1 3 4 5 6 8 9 11 13 14 16 ## ## $routes[] ##  1 3 4 5 7 8 10 11 12 14 16 ## ## $routes[] ##  1 2 4 5 7 8 9 11 13 14 16 ## ## $routes[] ##  1 2 4 5 6 8 10 11 13 14 16 ## ## $routes[] ##  1 3 4 5 7 8 9 11 13 14 16 ## ## $routes[] ##  1 3 4 5 6 8 10 11 13 14 16 ## ## $routes[] ##  1 2 4 5 7 8 10 11 13 14 16 ## ## $routes[] ##  1 3 4 5 7 8 10 11 13 14 16 ## ## ## $distances ##  6044.006 6068.785 6147.616 6172.396 6230.349 6272.589 6333.960 6376.199 ##  6386.100 6434.153 6489.710 6537.763 6547.664 6589.903 6651.274 6693.514 ##  6751.468 6855.078
The shortest and longest routes detected are mapped below.
##  6044.006
##  6855.078
It is worth noting that the default functions for route and distance calculation do return the shortest route, in the presence of multiple possible routes. The default-calculated route is shown below, and is the same as the shortest route determined by
##  1 2 4 15 16
##  1 2 4 15 16