FastMapping v2.0: a tool to automate depuration and mapping of spatial data

PACCIORETTI, P.; CÓRDOBA, M.; BRUNO, C.; AGUATE, F.; BALZARINI, M.

Jornada; VII Jornadas Integradas de Investigación, Extensión y Enseñanza de la Facultad de Ciencias Agropecuarias; 2017

In the last decade, technological advances in precision agriculture have made yield monitoring one of the most widely used precision farming technologies. The output of these monitors are often datasets with georeferenced records at several sites within a field. The spatial interpolation for the prediction of non-sampled sites is a common practice in the processing of yield monitor data. However, these data contain measurement errors that are usually associated with the mapping process itself. Artifacts may occur as extraneous data points that lie outside the general range of the dataset (global outliers) or as data values that differ significantly from neighboring data values, but lie within the general range of the data (spatial outliers). Consequently, datasets need to be pre-proceeded or depurated before the analysis of spatial variability. The optimal exploitation of georeferenced data depends on the capacity for efficiently pre-process and mapping spatial variability. The FastMapping v2.0 software was developed to automate, on a user-friendly platform, the data pre-processing and the geostatistical analysis focused on spatial variability in continuous domains. The application was programmed in R language with a graphical user interface that is freely accessible through a browser using internet connection. The pre-process first eliminates border data points with a selected distance to remove the frequently noisy borders and end-of-field yield monitor errors. Outliers are removed in two-steps: 1) the dataset is constrained to sensible threshold limits 2) the mean and standard deviation are calculated to identify probable erroneous data. Finally, the application uses the local Moran index of spatial autocorrelation to identify and remove spatial outliers. FastMapping allows the fitting of several spatial correlation models and the automatic selection of the one that best performs in terms of spatial prediction. The prediction accuracy is quantified through k-fold cross-validation at each of the spatial correlation models. Once the best predicting model is selected, several maps can be visualized: the adjusted experimental and theoretical semi-variograms, the spatial variability map obtained by kriging interpolation, and the map of prediction variance. With FastMapping, the user can easily set several adjustments and models; with or without mean trend, ordinary (without trend) or universal (with trend) kriging interpolation. Predictions can be point or block type, in a global or local neighborhood. FastMapping can open text files (.txt) with many formats, and results can be exported to tables with comma separated values (.csv files). The results include the coordinates of the predicted points, the predicted values, and the prediction variance. The spatial variability maps can be exported as georeferenced data (GeoTIFF files). An illustration of the software is presented here; it is used to obtain a spatial variability map from an intensively sampled soybean yield dataset in a crop field. Free R-based applications like FastMapping, with user-friendly interfaces, expand the adoption of improved methodological and computational spatial data analysis.