Determine the appropriate threshold value by calculating the niche overlap and the distance of two niche centroids.

Note:

The data used in this case study can be downloaded via this link.

From Fig.2 in Case 1, we can see that ‘Equal training sensitivity and specificity’ (blue ellipsoid) is closer to the original N (white) intuitively. We can also calculate the similarities quantitatively in NicheA.

In the ‘Toolbox’ menu, ‘Quantifying niche similarity’ submenu, we can find the function to quantify niche similarity by calculating the overlap of two ellipsoids (Fig. 1). Because we want to calculate the overlap of the original N with all the other results, we select ‘virtual_species’ in the first drop-down list, and check ‘Loop Comparison’ on the right of the next drop-down list. Because we want to calculate the overlap of ellipsoids, the ‘Calculation Method’ is set to ‘MVE’.

Figure 1. Quantifying niche similarities of the original N and Maxent’s result with different threshold values.

Then we can get the overlap results (Table 1).

Table 1. The niche similarities between the original N and the Maxent result with different threshold values.

VS1 VS1 MVE Volume VS2 (Threshold method) VS2 MVE Volume Overlap Overlap/VS1 Overlap/VS2
origin 39.290 Balance training omission, predicted area and threshold value 144.882 39.290 1 0.271
origin 39.290 Maximum training sensitivity plus specificity 90.619 39.290 1 0.434
origin 39.290 10 percentile of training presence 74.425 39.176 0.997 0.5264
origin 39.290 Equal training sensitivity and specificity 57.021 37.697 0.959 0.6614

So, if we require the thresholded result to contain at least 95% information of the original N with a highest threshold value, the 'Equal training sensitivity and specificity' method is likely to be the ‘best’ method.

Note:

The sentence above is only an example, but not a standard to measure the 'best' threshold method. You must pick your own method from your specific scientific question.

We can also choose the ‘best’ threshold value by calculating the distance to the niche centroid. Click ‘Toolbox’–‘Show N attributes’, where the coordinates of the ellipsoid will be available (Fig. 2).

Figure 2. Show N attributes. The content in the red circle is the coordinates of the ellipsoid’s center (MVE).

After reading the coordinates and calculating the Euclidean distances from the two centers, we can get Table 2 and 3. In this manual, we won’t interpret the results in detail, because each case is unique, and should be interpreted comprehensively.

Table 2. The coordinates of the ellipsoid’s centroid.

Ellipsoid label (Threshold method) x y z
Original 2.471252 -1.15762 0.37744
Balance training omission, predicted area and threshold value 2.522364 -1.08732 0.304633
Maximum training sensitivity plus specificity 2.547513 -1.20877 0.467751
10 percentile of training presence 2.548166 -1.30262 0.492427
Equal training sensitivity and specificity 2.64899 -1.40261 0.645102

Table 3. The center distances between the original N and Maxent’s result with different threshold values.

Balance training omission, predicted area and threshold valueMaximum training sensitivity plus specificity 10 percentile of training presence Equal training sensitivity and specificity
Original 0.113383 0.128792 0.200403342 0.404044427