dpabon/YAXArraysToolbox.jl: Zenodo publication

The YAXArraysToolbox.jl package allows users to efficiently process and visualize large spatio-temporal gridded data. The backbone of the package is the YAXArrays.jl package that allows to easily and efficiently apply functions in large gridded datasets that cannot be loaded into RAM. The YAXArrays.jl package also allows users to easily upscale analysis from laptops to clusters without the necessity to re-write code. However, we consider that functions and analyses that are generic enough can also be part of the Julia Data cubes ecosystem in a new package. The YAXArraysToolbox.jl functions on which users interact do not follow the YAXArrays.jl style, but instead focus on ready to use functions similar to those found in the Raster package of the R ecosystem. For visualization, the YAXArraysToolbox package relies on the Makie ecosystem, specifically, for plotting maps, relies on the GeoMakie package. The YAXArraysToolbox.jl package is structured in three main modules where one of the modules connects to the YAXArrays.jl package. The users will interact with the modules “Basic operations” and “Spatio-temporal analysis/functions”. In the “Basic operations” module, the user will be able to apply functions to easily manipulate, and visualize large gridded data. In the spatio-temporal analysis module, scientists and students can easily perform different analysis to evaluate the relationship between different variables present in large gridded data. So far in the Spatio-temporal module we make available two techniques. The first one is the Space for time technique  (space4time) introduced in the geoscience field by (Duveiller et al., 2018). The space for time technique estimates the average change of local climate if the land cover changes from one class to another. For example, savannas are usually hotter than neighboring forest, then contrasting the local climate conditions we can evaluate the potential effect of the transition from forest to savannas. The second technique is the Spatio-temporal folds generation presented by (Meyer et al., 2018). Spatial and temporal autocorrelation can produce bias when evaluating the performance of Machine learning models, for this reason, it is necessary to implement cross validation strategies that consider the spatial and temporal context when defining folds for training and testing models. The spatio-temporal cross validation scheme implemented in the YAXArraysToolbox package allows the user to define three strategies: Leave-Location-Out (LLO), Leave-Time-Out (LTO) and Leave-Location-and-TimeOut (LLTO) as proposed by (Meyer et al., 2018).