Local library to support spatial predictions.

What are covariates and why are we interested in them?
Covariates are spatially extensive digital data sets that describe environmental conditions that are known, or assumed, to influence the distribution of some attribute or property whose spatial pattern is not known but which one wants to model and predict. To be useful for spatial prediction, covariate data sets must completely cover the entire extent of any area of interest. Covariates may describe any aspect of the environment but, in the case of modelling soil patterns, they typically describe one or more of the environmental factors known to influence the development and spatial distribution of soils in terms of the CLORPT model (Jenny, 1941): specifically Climate (C), Organisms (O), Relief (R), Parent Material (P) and Time (T). The Scorpan model (McBratney et al., 2003) is a recent update of the CLORPT model that additionally recognizes that known spatial patterns of soil (S) information can help to predict unknown patterns of soil properties or classes and that also includes explicit recognition of the influence of spatial structure and spatial factors (n) in the prediction equation.

We use information on spatial patterns that we have, in the form of covariates, to predict spatial patterns of classes or properties that we do not yet know. Covariates tend to be spatial data that we already possess or that are affordable and practical to create. These available, or easy to produce, data layers can then be used to predict the spatial distribution of more difficult to obtain attributes or classes by developing statistically valid relationships between observed values of soil properties or classes at known locations and the values of corresponding sets of environmental covariates at the same locations. .

Why do we want to produce and share covariate data sets?
Sharing maximizes the effectiveness and consistency of our modelling work. There is always some cost, in terms of time and effort, required to obtain, collate and organize data sets of environmental covariates so that they can be used as inputs to predictive models. It makes a lot of sense to make any prepared covariate data sets widely and freely available for others to use, once effort has been expended to prepare and assemble them. There is increasing recognition of the benefits of reusing and recycling previously prepared data sets of environmental covariates so as to obtain maximum benefit from the efforts that went into creating and assembling them. We want to reduce unnecessary efforts to recreate input layers that may have already been created and we want to recycle and reuse these assembled layers to obtain maximum use and value from them. Sharing previously prepared layers of covariate spatial data lets the next modeler build on the previously completed work of previous modelers, not repeat it, so that each successive effort can contribute new data to a shared repository rather than continually repeating the same work.

What are shared repositories of covariate data sets?
A shared repository of spatial covariate data sets is an accessible online platform, or location, where previously prepared covariate data sets can be archived, accessed, retrieved and made available for use by any interested party Ideally all previously produced covariate data sets can be freely accessed and freely used via a completely open platform. This is the desired outcome for all spatial data collected or produced for the AfSIS project. There may be some instances where producers of covariate data layers are not willing to share their data without restriction due to issues of confidentiality or proprietary concerns. However, we all benefit when we share and the default position ought to always be to share freely, fully and without reservation.

AfSIS has set up an open repository where all spatial data collected or produced by the AfSIS project can be located, accessed and downloaded (see
ftp://africagrids.net/) . ISRIC has set up a similar portal for all the global environmental covariates it has assembled within the World Grids repository
(see http://worldgrids.org/doku.php?id=project_summary). Anyone can access these portals and download and use the data. Equally, anyone can prepare new data layers of relevant environmental covariate data and contribute these new layers to the open repositories.