Influence of data pre-processing on the behavior of spatial indicators

Abstract Spatial indicators are widely used to quantify the impact of climate and anthropogenic changes on species spatial distribution. These metrics are thus, determinant to decisions on the conservation measures to be implemented. In the current work, the effect of two common pre-processing methods: gridding and continuous interpolation, on the values given by five spatial indicators: index of aggregation, percentage of presence, center of gravity, inertia and isotropy was studied. Indicators were computed using empirical data of 32 species biomass distributions, obtained from time series of bottom trawl and of acoustic surveys, with different sampling designs. Spatial indicators computed using pre-processed data were compared with spatial indicators estimated without pre-processing the data using the difference between the two approaches. The pre-processing of the data consisted of a series of progressive increase of grid sizes, from 20 to 120 km, and a series of ten different interpolation methods: linear models, inverse distance weighting, bicubic spline, Generalised Additive Models, ordinary, universal kriging and geostatistical conditional simulations. Pre-processing the data, both by gridding or interpolation caused a change of several orders of magnitude on the indicator results, for the two surveys considered. Inertia showed opposite differences for trawl and acoustic datasets whereas the remaining indicators evidenced similar patterns of difference. An index of relative difference, was computed to verify whether the pre-processing effect on the indicator was higher or lower than the observed temporal variability. This index showed that for certain species, the variability of the indicators was over two-fold its respective inter-annual temporal variability, as it was the case of the percentage of presence and the index of aggregation, estimated using interpolated or gridded data. The most important factors explaining most of the difference between results with or without pre-processing the data were the indicator considered. For example, the percentage of presence was much more sensitive to pre-processing than inertia or isotropy. Additionally, the interpolation method (bi-cubic splines) and gridding size up to a certain level (