inst/app/help/3_outlier_speed.md

In xhdong-umd/ctmmweb: Web App For ctmm Package

• Every point is assigned an estimated speed value. Points with extreme speeds could be outliers due to error.
• Standardized Device Error specifies the error in meters for data with HDOP=1 (regular data).
  • The usual values are 10 meters for GPS, 100 meters for VHF, 1000 meters for ARGOS.
  • Note the input have to be numeric. A regular numericInput is not used here because it will respond to mouse wheel, so scrolling after input could trigger a series of unintended updates.
  • The calculated speed could become all 0s if the device error value was too large.
• The UI is very similar to the distance to median center tab, except there are two checkboxs for selected points (selected rows in table of selected range):
  • Draw Path Around Selected Rows will add paths with arrows for points before and after the selected points. You can zoom in plot for more details.
  • Label Row Number in Path further add row number of points, which is sorted by timestamp.
• The speed value can have more sparse distribution, with majority of points at low speed, which could result:
  • Most points could be almost transparent in color. You can select a range in histogram to highlight them if needed.
  • The speed unit is determined by the median value in all speed values to ensure all units to be same in dataset, so the high value outliers can be seen with lower unit.
• Removed Outliers table includes the points removed both in distance and speed tabs.
• The speed estimation algorithm is quite complex to deal with some special cases:
  • Some points have duplicated timestamp which cause zero time and infinite speed value
  • If A is far from the points before and after it in time, both the speed value of point coming to A and A coming to next point are huge. The algorithm need to assign the extreme speed value to appropriate point.
  • It's possible that some data set have special cases that were not handled perfectly by the algorithm.
  • It's often better to use the distance tab to filter some obvious outlier first, then the case left for speed can be simplified.
  • See discussion about the algorithm, and some algorithms used by Movebank.
  • If the sophisticated method failed for any reason, it will fall back to a simpler method with less accurate estimation but more robust for data.

xhdong-umd/ctmmweb documentation built on May 29, 2019, 11:59 a.m.