Prediction Toolbox


AG predictor
AG predicton toolbox - overview

The purpose of this toolbox is to allow the application of extended statistical models generated by the CUBIST (TM) or JMP (TM) software for the moment to a set of auxiliary input parameters, thereby bypassing the need to i) manually calculate the equations using Raster calculator; ii) use a grid-to-point conversion process prior to applying the prediction model in the statistical software itself and then a point-to-grid conversion process.

The toolbox has been tested on a set of covariates rasters  at 100m resolution of the size of Nigeria by Odeh Inakwu of University of Sydney  during his stay at ISRIC World Soil Information. With the size of Nigeria of nearly 95 million hectares (approximately 1000 by 1000 km in extent), the prediction of the primary soil variables was done in under 1 hour, depending on the complexity of the model. Based on our experience, it is recommended to first predict using the coarse scale of 1km resolution covariates in order to select/adjust a model which passes the "good look test", before extending the model to 100-m resolution (i.e., the final resolution model).

The code itself can be improved in many aspects; the current version number is 1.0.2. We hope to improve/extend it in the next couple of month to support several statistical prediction model outputs. The speed of the code can be improved by at least a factor of 1000, if it is rewritten in a line processing manner and even in python/numpy itself instead of relying on AG functions .

 The zip file contains the toolbox for different ArcGIS (TM) versions as well as examples for model files for CUBIST (TM) or JMP (TM)model output. The needed input parameters:

Workspace: Directory which contains the input grid/GeoTiff Files

ModelFile: Directory+ file which contains the input mode file

OutGrid: Directory+ grid Name which contains the output file

Extension: If grids are to be used, leave <<#>> in there, if GeoTiff <<.tif>>, Erdas Imagine <<.img>>, or Envi <<.bsq>> are to be used, specify the necessary extension including the dot.

  Hannes and Odeh


Next developments: 1) Add a Neural Network parser, 2) include JMP procedure into the Toolbox

V1.01 02092011 HIR add catch for different JMP output files of the same versions on different continents.

V1.02 24012012 HIR added sample for worldgrids/covariates

AttachmentSizeHitsLast download
globalsoilmapDOTnetV1.zip37.74 KB3715 weeks 5 days ago
globalsoilmapDOTnetV1.01.zip37.55 KB2735 weeks 5 days ago
globalsoilmapDOTnetV1.02.zip45.94 KB2405 weeks 5 days ago
Your rating: None Average: 5 (1 vote)

In November 2008, an $18 million grant has been obtained from the Bill & Melinda Gates foundation and the Alliance for a Green Revolution in Africa (AGRA) to map most parts in Sub-Sahara Africa, and make all Sub-Saharan Africa data available. From this grant there are also funds for coordinating global efforts and for the establishment of a global consortium. Several institutions have assumed a leading role in this effort and have made substantial financial and in-kind contributions.

Our main donors: