A comparison of spatially explicit and classic regression modelling of live coral cover using hyperspectral remote-sensing data in the Al Wajh lagoon, Red Sea

Live coral is a key component of the Al Wajh marine reserve in the Red Sea. The management of this reserve is dependent on a sound understanding of the existing spatial distribution of live coral cover and the environmental factors influencing live coral at the landscape scale. This study uses remote-sensing techniques to develop ordinary least squares and spatially lagged autoregressive explanatory models of the distribution of live coral cover inside the Al Wajh lagoon, Saudi Arabia. Live coral was modelled as a response to environmental controls such as water depth, the concentration of suspended sediment in the water column and exposure to incident waves. Airborne hyperspectral data were used to derive information on live coral cover as a response (dependent) variable at the landscape scale using linear spectral unmixing. Environmental controls (explanatory variables) were derived from a physics-based inversion of the remote-sensing dataset and validated against field-collected data. For spatial regression, cases referred to geographical locations that were explicitly drawn on in the modelling process to make use of the spatially dependent nature of coral cover controls. The transition from the ordinary least squares model to the spatially lagged model was accompanied by a marked growth in explanatory power (R 2 = 0.26 to 0.76). The theoretical implication that follows is that neighbourhood context interactions play an important role in determining live coral cover. This provides a persuasive case for building geographical considerations into studies of coral distribution.