Analyzing the many variables in a marine environment can be a challenge. Tidal patterns affect maritime traffic, coastal communities, and aquatic breeding periods; water temperatures affect algae growth which in turn affects marine animals and the fishing industry; the list goes on. To further complicate matters, an important aspect of marine data acquisition is time. Typically, data is gathered over time, as opposed to spatially, which results in large, multi-variate datasets that contain a wealth of information. However, the information is not very useful if it cannot be properly visualized.
Traditional 2D and 3D visualization techniques such as line/scatter or bar plots are adequate for basic analysis but fall short when one needs to analyze both large and small data patterns. Dr. Richard Koehler, founder of Visual Data Analytics, has spent his career demonstrating a different approach to visualizing data which better facilitates the understanding of complex ecosystems.
This 2D line graph is the daily passage of adult Chinook salmon. The X-axis represents the day of the year and the Y-axis represents the total number of salmon passing the gauge station. Each line plot represents a different year.
|The migration of Chinook salmon is typically displayed as a series of line graphs. Major patterns such as peak passage times are easy to discern, but smaller patterns or seasonal variability is hard to identify. Source: http://www.fpc.org/. Graph plotted in Golden Software’s 2D and 3D graphing software, Grapher.|
Major patterns, such as the peak passages, are easy to distinguish, but smaller patterns such as seasonal variability are difficult to identify. Additionally, the above graph represents only five years of data. What if five years isn’t enough? One can see how messy the graph would quickly become when adding another 10, 20, or 50 years of plots.
Dr. Koehler’s alternative visualization technique plots this time-series data as an image map, also known as a raster map. The date is turned into two temporal coordinates where the X-axis is a short term time step such as days or minutes, and the Y-axis is a longer time step such as months or years. The Z-value is represented in the cell location.
|An example of the grid format used to generate time maps (Koehler, 2004).
Using this time map technique, the Chinook passage data is plotted in Golden Software’s contouring, gridding, and 3D surface mapping software, Surfer. In this map, an additional 71 years of data is displayed. Despite the extra information, it is quite easy to distinguish temporal patterns, both large and small, that would otherwise be missed in the 2D line graph. The smaller spring and summer run patterns are now easier to see in this display. Also notable is the steady increase of salmon in recent years during the August through September run.
|The spring run can be seen in late April, and the summer run is in June with an obvious low count from years 1975 to 2000. The fall run is late August through September with notable increases in recent years. Data source: http://www.nwd-wc.usace.army.mil/. Map plotted in Surfer.|
Time maps are also useful when displaying ocean tide data. The Hawk Inlet on Admiralty Island in southeast Alaska is one of two ports in the Alaska Panhandle where ship arrival and departure is dependent on the tides. The timing of tides and daylight hours are critical factors for safe navigation; therefore, it is important for schedulers to know when the tides are favorable for vessel travel.
In these traditional views, high and low tides are distinguishable. But these displays are not sufficient for planning when ships can safely navigate in and out of the port that day, week or month. A significant amount of information is lost and obtaining a clear understanding of tidal patterns is practically impossible with these views.
|2D line chart plotting tide heights over time. As mentioned before, large patterns are distinguishable in traditional plots, but smaller patterns are difficult to decipher. Graph plotted in Grapher.|
However, by employing Dr. Koehler’s time map visualization technique, a whole new picture appears. Schedulers can quickly pinpoint days and times when tide levels are conducive for ship travel. This view also makes it easier to add extra information, such as the time of sunrise and sunset, which is represented by light grey lines. Despite having over half a million data points, this plot is clear and understandable. Not only do port schedulers benefit from tidal information, so do coastal managers preparing for nuisance flooding, recreational fishermen, surfers and boaters, and scientists working on habitat restoration projects, to name a few.
Multi-variate temporal analysis is another area in which time maps provide a clearer and more understandable visual representation of complex data. For example, several environmental factors must be considered when analyzing algae blooms. Such factors include sea surface temperature, sea surface salinity, air temperature, precipitation, stream flows, tidal height difference, upwelling, and wind speed. Individually, these factors do not have much of an impact on whether or not an algae bloom will occur. Instead, the factors must be analyzed as a whole which can be a troublesome task.
How can one recognize the windows of opportunity for these algae blooms? As noted before, traditional 2D and 3D views fall short for this type of task. Cue in Dr. Koehler’s time maps.
The first step is to display each environmental factor as a time map. In the example of the algae bloom, seven different environmental time maps will be created. Second, a binary filter is applied to the data that highlights days with favorable conditions for an algae bloom. A value of one equals days with favorable conditions and a value of zero equals days that are not favorable for an algae bloom. The binary filter is also displayed as a time map.
For example, one environmental factor is stream flow. Flows less than 350 cubic meters/second are favorable for an algae bloom. By viewing the binary image map, one can quickly pinpoint when stream flows are less than 350 cubic meters/second. These typically occur between August and October. One can also see in a drought occurred in 2001 resulting in less than normal stream flows for the months of January through April. All days represented in black are favorable stream flow conditions for an algae bloom.
|The time map on the left is the observed stream flow. The time map on the right is the binary filter highlighting days when stream flows are less than 350 cubic meters/second in black. Maps plotted in Surfer.|
Repeat these steps for all remaining environmental factors.
The final step is summing the binary data for the individual factors and displaying the resulting data as a time map. The summed data will contain values between zero and eight. Cells containing zero indicate on that particular day, none of the eight factors had conditions suitable for an algae bloom. Cells with a value of four indicate on that particular day, four out of the eight environmental factors had conditions suitable for an algae bloom. Cells with a value of eight indicate all eight environmental factors had conditions suitable for an algae bloom on that particular day. When displaying this combined data as a binary time map, values of eight means all criteria was met and values zero through seven mean at least one criterion was not met.
|This is the resulting time map depicting the days between 1993 and 2005 where all eight environmental factors were suitable for an algae bloom. Map plotted in Surfer.|
Of the 4,380 days between 1993 and 2005, there were 127 days where all eight environmental factors were favorable for an algae bloom.
The key to creating a time map is properly converting the typical date format, 5/1/2015, into two temporal coordinates, the day and year, within the input file. Using an Excel® spreadsheet, this chart shows the formulae using a calendar year. The X-axis is day and the Y-axis is year. The Z-value would be the time series such as salmon count, tidal height, or stream flow. Arrange your input data as column A = day (X), column B = year (Y), column C = value (Z).
|Convert your data to a format where you have three columns of data (X, Y and Z). The X column is the day of the calendar year, the Y column is the calendar year, and the Z column is the data variable.|
Once the data is in the correct format, the time map can be created in Surfer by gridding, also known as interpolating, the data and creating an image map from the gridded data.
Analyses can be greatly improved by incorporating data visualization. Depending on the need, traditional 2D and 3D views are satisfactory; however, by creating new ways to visualize both uni-variate and multi-variate data, unique and actionable insights will often be found. “The purpose of any graphic is to turn data into information,” states Dr. Koehler, “A well-crafted plot can be incredibly useful in understanding data and properly conveying your point. Having the right tool is critical to achieving this goal.” Dr. Koehler’s time maps do indeed achieve this goal. They convey information by transforming complex temporal data into clear, understandable visual models.