Problems of geometry, sampling, and scale in gridded biodiversity data, and proposed solutions

Petr Keil , Friederike Wölke, Gabriel Ortega-Solis , Florencia Grattarola, François Leroy , Florian Jansen, Carmen Soria

Grids, and gridded biodiversity data such as regional or country-level atlases, play a prominent role in ecology, particularly in the study of spatial patterns of species occupancy, geographic ranges, biodiversity, and their drivers and temporal dynamics. However, managing, exploring, and analyzing data in grids comes with problems. Here, we review the problems with gridded data, and the existing solutions. We focus on grid-specific problems of sampling (e.g. varying sampling method and effort in space and time, imperfect detection), geometry (e.g. varying grid cell area and shape, positional errors), and scale (e.g. spatial grain and temporal extent). A first group of solutions can be implemented prior to gridding of the data. This includes the selection of an appropriate geographic projection, grid grain, and grid cell shape. The second type of solution involves the manipulation and processing of the gridded data. Examples include aggregating cells to coarser grains or removing cells that fail to meet certain quality criteria. The third type of solution is implemented during the analysis of the data. The most important is the quantification of the problem for use in statistical models or machine learning algorithms as a covariate. We hope to provide guidance particularly to early-career ecologists who may otherwise struggle to make sense of the various solutions scattered through the literature.