At usual conversion rates of cover and errors inherent in change detection with sequential, mid-resolution satellite imagery, a decade would be the minimum length for cover changes larger than the errror: see attached

Article What Drives Conversion of Tropical Forest in Carrasco Provin...

Article Forecasting the pattern and pace of Fagus forest expansion i...

Article A comparison of fine resolution census and image-based natio...

This strongly depend on area and your objectives. Learn carefully a historical reviews of the place you study. Maybe some significant events took place in certain years and might considerably change land cover features. If you won't find any drastic events during 1986-2014, you will have to define all criteria based on detailed study of various kinds of changes with its statistical analysis.  

In many cases the lengths of distinct periods defined by influence of antropogenic factors for instance onset of urbanization processes or other types of land use, or any natural catastrophic events.

Another question of concern, what if the changes that occurred within this period are irregular in terms of time. Could I use epochs that area not constant. E.g. one epoch has ten years another 7 years another years or they must be constant?

They mustn't be constant.   And definition of their length s may be one of research objectives

Interesting question and responses. Besides the ecological and anthropogenic considerations raised, depending on where you are working, the data density in the Landsat archive should also be consulted. Peruse the USGS Landsat pages for some info on that. Can also look at this excellent publication:

Article The global availability of Landsat 5 TM and Landsat 7 ETM+ l...

Interesting discussion. Yes, I depends on the dynamic of land use in your study site and the objective of your study. 

In addition to the other answers, your time interval will fully depend on your objective. Say, if you want to see a dramatic change of forest area after fire so your time interval can be determined based on prior and after the fire event. But, if your aim is to see the gradual change of land cover during 1986-2014 without any specific type of change, it's better to have time interval between 5 - 10 years. You should also consider about the spatial resolution. You can expect to detect more change if you use a higher resolution imagery with the same time interval. 

I have the similar concerns with you, and I think it depends on your objectives. Even detecting the land use and land cover change at 5 or 10 years interval, it may also change during these 5 or 10 years . As my objective is to detect the gradual changes of land use and land cover, so I try to detect the land use and land cover at every year.

The objectives of the study is a very crucial consideration. However, depending on the study area, availability of data in terms of cloud-free Landsat imageries could also determine what is useful for the study. Therefore, it may be impossible in many instances, to have a regular interval.

Cheers

As noted before, defining your objectives is the starting point to answer your question. Yet, in many projects it is reasonable to assume that the first objective might be indeed to understand what are the intra/inter-annual LUCC most significant in your area of study and field of research (and from these determine temporal and spatial resolution, that might be different for different processes as you noted).

At the risk of stating the obvious: you might also want to consider the actual ammount of funding available for image acquisition, the time to finish your work, and storage/processing resources at hand. All knowledge that you have about your area of study, including relevant (for your research) land uses, their histories, field-collected data, etc., can be of relevance. At times, understanding what objectives can be fullfilled with available resources and what are the tradeoffs and uncertainties of your work can be of great importance.

it is depends upon application i.e. in which class you are interested. For example if you need change in build-up land, you can choose five years and you can model it. Next thing is availability of data. i.e cloud free, day time observation etc.