It is useful to start from the definition of latent heat flux (LHF); on the NASA website mentioned in your question, we can read:

“Latent heat flux is the amount of energy moving from the surface to the air due to evaporation (positive values) or from the air to the land due to condensation (negative values)”

-End quote

LHF values ​​in the case of hurricanes should be positive indicating the source of the energy that leads to the development of the convective system.

The most relevant quote in the context of the negative values ​​of your data is:

“The highest positive latent heat flux occurs during hurricanes and typhoons, as these events are powered by the movement of heat energy from the warm ocean to the atmosphere, seen here in Hurricane Frances and Typhoon Songda. Significant negative latent heat flux is somewhat rare and occurs over the ocean only during certain configurations of air and surface conditions.”

-End quote

In addition, according to Trenberth et al.

https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2006JD008303

latent heat flux values ​​for a hurricane are positive (Figure 5b)

Based on the above, I am convinced that the negative values ​​of LHF are not correct.

I recommend visiting the site

https://reanalyses.org/atmosphere/overview-current-atmospheric-reanalyses

You can ask questions about any reanalysis dataset (including NARR).

I hope this site will solve the problem of unexpected negative LHF values ​​in NARR.

Thanks to everyone for the responses, I saw them and then I started work on your recomendations.

According to Mesinger and Ebisuzaki, the latent heat flux from NARR data has the wrong sign. So if we have negative values they should be interpreted as positive, we have an evaporation process, and the same occurs for positive values(they should be negative).

They didn't change the sign because the users from the Eta model are old as well as new.

Hope this can help others that use this kind of data from NARR.

Thanks, again, for your help.