Predictive Mapping of Reef Fish Species Richness, Diversity and Biomass in Zanzibar using IKONOS Imagery and Machine-learning Techniques

Remote Sensing of Environment, Volume 114, Issue 6, 15 June 2010, Pages 1230-1241

Anders Knudby, Ellsworth LeDrew, and Alexander Brenning

“During the last three decades, the large spatial coverage of remote sensing data has been used in coral reef research to map dominant substrate types, geomorphologic zones, and bathymetry. During the same period, field studies have documented statistical relationships between variables quantifying aspects of the reef habitat and its fish community. Although the results of these studies are ambiguous, some habitat variables have frequently been found to correlate with one or more aspects of the fish community. Several of these habitat variables, including depth, the structural complexity of the substrate, and live coral cover, are possible to estimate with remote sensing data. In this study, we combine a set of statistical and machine-learning models with habitat variables derived from IKONOS data to produce spatially explicit predictions of the species richness, biomass, and diversity of the fish community around two reefs in Zanzibar. In the process, we assess the ability of IKONOS imagery to estimate live coral cover, structural complexity and habitat diversity, and we explore the importance of habitat variables, at a range of spatial scales, in the predictive models using a permutation-based technique. Our findings indicate that structural complexity at a fine spatial scale (not,  vert, similar 5 to 10 m) is the most important habitat variable in predictive models of fish species richness and diversity, whereas other variables such as depth, habitat diversity, and structural complexity at coarser spatial scales contribute to predictions of biomass. In addition, our results demonstrate that complex model types such as tree-based ensemble techniques provide superior predictive performance compared to the more frequently used linear models, achieving a reduction of the cross-validated root-mean-squared prediction error of 3–11%. Although aerial photographs and airborne lidar instruments have recently been used to produce spatially explicit predictions of reef fish community variables, our study illustrates the possibility of doing so with satellite data. The ability to use satellite data may bring the cost of creating such maps within the reach of both spatial ecology researchers and the wide range of organizations involved in marine spatial planning.”

One thought on “Predictive Mapping of Reef Fish Species Richness, Diversity and Biomass in Zanzibar using IKONOS Imagery and Machine-learning Techniques

Comments are closed.