Dear Ari,

Thanks for your interest in MGET.

All of MGET’s parameters are described in the help that is available inside the tools. If you do not see that documentation, click the Show Help button:

Now you will get this:

So, to answer your question. The Cayula-Cornillon algorithm operates by passing a moving window over the image and computing histograms of the cells in the window. It looks for a bimodal distribution of temperatures—a necessary condition for there being two distinct masses of water in the window. If it finds a bimodal distribution, it computes a temperature that best separates the two populations, and then computes the mean temperature of each population. The threshold is the minimum difference in temperature for the tool to assume that a front could be present and proceed to the next part of the algorithm, which determines whether the two temperature populations are spatially coherent (i.e. separated such that the hot cells are all in one place and the cold cells are in another, rather than them being all mixed up).

If you use a small threshold, many fronts will be detected, although the algorithm might also be more susceptible to noise. If you use a large threshold, fewer fronts will be detected: only the ones with large temperature differences.

The threshold means something different between the tools you mentioned. In the Data Products tools, the threshold is in degrees C. In the Oceanographic Analysis tools, it is more complicated. Those tools are intended for use with arbitrary SST data. Because the Cayula-Cornillon algorithm can only operate on integer values, not floating-point values, it is necessary to convert floating point rasters to integers before running the algorithm. You can do this yourself before running the MGET tool, or you can give the MGET tool a Map Algebra _expression_ that allows it to perform the conversion itself. The Map Algebra _expression_ parameter is under the Input Raster Options, and the documentation gives examples of how to use it. Anyway, the algorithm operates on integers and the MGET tool does know how much SST each integer represents. For example, each integer might be 0.075 C, as is traditional with NOAA 4km AVHRR Pathfinder SST, or it might be 0.000717185 C for NASA MODIS L3 SST. Because of this, the Oceanographic Analysis tools require you to provide the threshold on the scale of the integers rather than on the scale of degrees C. So, if each integer represented 0.075 C, and you wanted the mean temperature difference to be 0.5 degrees C, you would provide the threshold 6.666667, which is what 0.5 degrees C is in the integer scale.

Regarding 3-day resolution. The tool is limited to whatever temporal resolutions that NASA produces, which currently are daily, 8-day, monthly, and annual. If you are trying to address the problem of clouds, I recommend you find fronts in daily images, then combine the fronts images together. That approach works better than using a single composite SST image such as 8-day or monthly. There are two problems with the composite images:

1.    When there are few clouds during the time period, then many temperatures are averaged together. If fronts have been moving around, then they will be smoothed (blurred) by the averaging. The algorithm will have a harder time detecting the front because the histograms will show less distinct populations: the smoothing will make some of the cold temperatures a bit warmer, and the warmer ones a bit colder.

2.    But when there are many clouds, a different problem can happen. The composite images are made by averaging together all the non-cloudy pixels. If one area is cloudy for all but the end of the time period, then it will receive the average temperature from the end of the time period. If an adjacent area is not cloudy in the beginning but then cloudy for the rest of the time, it will receive the average temperature from the beginning of the time period. The result is a false front—a warm area next to a cold area that only happened due to the pattern of clouds. This can be detected by the algorithm.

So, instead of that, find fronts in each image, then average the fronts images together to obtain a frontal frequency map. The fronts images contain 1 where there was a front, 0 where there was no front, and No Data where there was land or too many clouds to execute the algorithm. By averaging these together for daily images, you obtain a frequency.

Finally, one last hint. The default parameters of the algorithm are configured as they were in Cayula and Cornillon’s original paper. But that paper operated on ~1.5 km resolution SST data—it was using AVHRR level 2 swaths for a specific region, not global grids like most of the products available through MGET’s Data Products tools, which usually have ~4 km resolution or coarser. In my experience, the original parameter values do not work the best on this lower resolution data, and not enough fronts are detected. If this seems true for your situation, I can suggest some alternative parameter values that might work better.

Best regards,

Jason

Greetings,

I would like to thank you for MGET availability and ability.
I use MGET especially Cayula-Cornillon tool to predict fronts area. Fronts area detected used as a parameter of potential fishing ground (pfg) prediction.

There are two tools Find Fronts, first in Oceanographic Analysis and the second one in Data Products. Currently i use the first one with MODIS SST Level 3 data, but then I found the second one.

First I want to ask, what is threshold mean in both tools? Is it Δtemperature, or is it function of Δtemperature to something?

And the second, is there any Find Cayula-Cornillon Fronts in PO.DAAC MODIS L3 SST for 3day temporal resolution?

Thank you for you response.

--
Ari Murdimanto
RS-GIS Ocean Remote Sensing
Balai Penelitian dan Observasi Laut
Perancak, Jembrana
BALI - INDONESIA
http://www.bpol.litbang.kkp.go.id/