Ordinary kriging vs inverse distance weighting: spatial interpolation of the sessile community of Madagascar reef, Gulf of Mexico

Abstract

Information about the distribution and abundance of the habitat-forming sessile organisms in marine ecosystems is of great importance for conservation and natural resource managers. Spatial interpolation methodologies can be useful to generate this information from in situ sampling points, especially in circumstances where remote sensing methodologies cannot be applied due to small-scale spatial variability of the natural communities and low light penetration in the water column. Interpolation methods are widely used in environmental sciences; however, published studies using these methodologies in coral reef science are scarce. We compared the accuracy of the two most commonly used interpolation methods in all disciplines, inverse distance weighting (IDW) and ordinary kriging (OK), to predict the distribution and abundance of hard corals, octocorals, macroalgae, sponges and zoantharians and identify hotspots of these habitat-forming organisms using data sampled at three different spatial scales (5, 10 and 20 m) in Madagascar reef, Gulf of Mexico. The deeper sandy environments of the leeward and windward regions of Madagascar reef were dominated by macroalgae and seconded by octocorals. However, the shallow rocky environments of the reef crest had the highest richness of habitat-forming groups of organisms; here, we registered high abundances of octocorals and macroalgae, with sponges, Millepora alcicornis and zoantharians dominating in some patches, creating high levels of habitat heterogeneity. IDW and OK generated similar maps of distribution for all the taxa; however, cross-validation tests showed that IDW outperformed OK in the prediction of their abundances. When the sampling distance was at 20 m, both interpolation techniques performed poorly, but as the sampling was done at shorter distances prediction accuracies increased, especially for IDW. OK had higher mean prediction errors and failed to correctly interpolate the highest abundance values measured in situ, except for macroalgae, whereas IDW had lower mean prediction errors and high correlations between predicted and measured values in all cases when sampling was every 5 m. The accurate spatial interpolations created using IDW allowed us to see the spatial variability of each taxa at a biological and spatial resolution that remote sensing would not have been able to produce. Our study sets the basis for further research projects and conservation management in Madagascar reef and encourages similar studies in the region and other parts of the world where remote sensing technologies are not suitable for use.

Spatial benthic community analysis of shallow coral reefs to support coastal management in Culebra Island, Puerto Rico

“…Exploration of the spatial structure of sample points via the average nearest neighbor function revealed that the observed values (11.9 units) were smaller than those expected by a random dispersion (53.1 units), which generated a nearest neighbor ratio of 0.225 and indicated a clustering tendency. The global autocorrelation quantified through Moran's I covariance index equaled 0.923 (p = 0.000005 at a distance of 24.5 m), which indicated a very high positive spatial dependency and validated the appropriateness of performing IDW on tightly spaced samples over similar depths (Li & Heap, 2011;Zarco-Perello & Simões, 2017). Cross-validation analysis optimized the power value (a) for IDW interpolations by both maximining the regression function between expected and observed values and minimizing its root mean squared error (Table S6).…”

Peer Review #1 of "Spatial benthic community analysis of shallow coral reefs to support coastal management in Culebra Island, Puerto Rico (v0.1)"

“…Exploration of the spatial structure of sample points via the average nearest neighbor function revealed that the observed values (11.9 units) were smaller than those expected by a random dispersion (53.1 units), which generated a nearest neighbor ratio of 0.225 and indicated a clustering tendency. The global autocorrelation quantified through Moran's I covariance index equaled 0.923 (p=0.000005 at a distance of 24.5m), which indicated a very high positive spatial dependency and validated the appropriateness of performing IDW on tightly spaced samples over similar depths (Li & Heap, 2011;Zarco-Perello & Simões, 2017). Cross-validation analysis optimized the power value (α) for IDW interpolations by both maximining the regression function between expected and observed values and minimizing its root mean squared error (Table S6).…”

A data-driven geospatial workflow to improve mapping species distributions and assessing extinction risk under the IUCN Red List

“…spatial autocorrelation), and the local influence of points (weighted average) diminishes with distance. IDW has been used for mapping invasive plants (Roberts et al 2004) and to understand the distribution of coral reef sessile organisms, with a higher accuracy than more advanced geostatistical interpolation methods (Zarco-Perello & Simões 2017). Recently, Gomes et al (2018) found that Maxent SDMs models largely overlapped with abundance maps derived from using IDW based inventory plots for 227 hyper-dominant Amazonian tree species.…”