Introduction:

The mapping of areas with higher forest fire density can be developed through kernel density estimation, which requires the selection of a function and bandwidth (h). The h value, when defined by subjective (visual) processes, will depend on the knowledge and experience of the person making the selection.

Objective:

To propose a statistical alternative, based on forest fires information (2005-2013) from Jalisco, Mexico, for the selection of h as support for kernel density estimation.

Materials and methods:

A total of 13 h values were defined using seven techniques. The h value was selected using the following statistics: root mean square error, root mean integrated squared error, coefficient of variation and comparative percentage.

Results and discussion:

The h values obtained with the techniques analyzed were between 2 550 and 41 906 m. There was great variation in the results; the range between the maximum and the minimum value was 39 356.34 m with an average of 10 936.74 ± 9 955.04 m. The above implies that there is no single and universal process for all cases. According to the validation criteria, the statistically most adequate h value is between 5 300 and 5 900 m; the closest result was obtained with the mean random distance technique (5 395 m).

Conclusion:

It is possible to select h under a practical statistical perspective, avoiding the use of subjective criteria.