Table 1. Description of and limitations about the stated assumptions in parameters of long-term and short-term dynamic occupancy models.

Model (parameter)		Assumption		Justification	Limitations
Dynamic occupancy models
Probability of Occupancy (ψ)	Changes across 15 year periodsa and annuallyb		Probability of occupancy derived conditional on detection probability, estimation for short-term data better than long-term			Occupancy estimates are conservative because of negative bias in detectability, but are preferable to overestimates
Detection probability (p)	a constant within 5-year periods, b assumed to change annually and as per method used in sighting or feeding trail detection		Model explicitly estimates detection probability, i.e. the probability of having false negatives in the data; estimation far more robust for current short-term data than for long-term data			False negatives expected to dominate the long-term datasets, so estimates of detection probability are conservative (typically with slight negative biased)
Probability of persistence/Colonization-Extinction (φ, γ)	Changes after 15 years, constant over 5-year secondary replicates; assumed to change annuallyb		Assumption based on our own field observations of 3 identified individual dugongs, and from home ranges reported by De Iongh et al. (1998)			Sheppard et al. (2007) suggest that dugong movements may be more individualistic and longer ranges may be covered, however given our observations, this seemed relatively unlikely.
Ecological covariates (β)	a Fixed site-specific covariates (e.g.) exposure, depth that would not change at ecologically significant scales over time; b site-specific covariate data on seagrass meadows and anthropogenic threats based on annual monitoring		Covariates assumed to be static and not changing over time for long-term models; Mean and standard deviations of covariate values used over three years			Unable to use other covariates related to human disturbance, etc. for long-term models, due to gaps and missing data
Survey coverage	a Data from about 60% of known extant seagrass meadows, in the absence of data on the condition of past meadows b Nearly 80-85% of the total Lakshadweep archipelago surveyed for seagrasses		Bias in parameter estimates possibly differs between sites			Model cannot account explicitly for these differences, so only locations with minimum of three data points included. Patchy sampling coverage might also negatively bias detectability, but considering the scale of the study, it is a logistical constraint

Key: ^a Long-term dynamic occupancy models; ^b Short-term dynamic occupancy models.