Hice experimentos y creo que la única solución es congelar el valor de m para el EON. Simplemente, no hay suficiente calidad de datos para ajustar el modelo completo de 5 parámetros. Si ponemos m = 1.0 que es aproximadamente el mediano de las demás regiones, entonces funciona muy bien. Aproveché también para congelar el seeing en cero (aunque con cualquier velo pequeño sería igual)
Copy of mail message to Javier: