TESTING THE TEMPORAL AND SPATIAL VALIDITY OF SITE-SPECIFIC MODELS DERIVED FROM AIRBORNE REMOTE-SENSING OF PHYTOPLANKTON

Validations of predictive models are necessary for the accurate applic ation of remotely sensed imagery within ecological research and fisher ies management. Multiple regression models derived from airborne image ry on 16 May 1990 accurately depicted phytoplankton biomass and turbid ity within aquaculture impoundments. To examine their temporal validit y, the exact models, as well as identical model forms, were fit to sim ilar imagery and in situ data collected on 20 June 1990. None of the e xact models for 16 May accurately predicted in situ data on 20 June; h owever, model forms were robust for describing in situ variables. To e xamine their spatial validity, identical model forms were fit to in si tu data partitioned among phytoplankton composition and biomass. The f it of the models and the contribution of imagery variables to the mode ls varied among in situ variables. Although imagery variables explaine d all of the observed variability for turbidity, regression tree model ing indicated that a significant proportion of the variability in chlo rophyll distribution both among and within impoundments was explained through both imagery variables and phytoplankton biomass. Consequently , universal models for the airborne remote sensing of water-quality va riables in systems having distinct optical signatures is unlikely. Rat her, robust site-specific models will need to be developed.