The IDEAMAPS Network is delighted to announce the release of our Modeller Platform with pre-processed spatial training and covariate datasets for modellers to experiment with and share "slum" modelling methods. As of March 2021, we have prepared dozens of training and covariate datasets for the greater Lagos (Nigeria) area, and are working to finalise similar data in Accra (Ghana) and Nairobi (Kenya). All IDEAMAPS data are aligned with the IDEAMAPS Domains of Deprivation Framework, which we hope might inspire and inform your modelling decisions.
We are making the IDEAMAPS Modeller Platform available to anyone who is interested in contributing to a collaborative, open model development process. The platform is hosted by University of Twente's ITC Big Geodata Service Portal, and provides a robust computing environment. We are seeking modellers from any background (e.g. statistics, geography, data science) to experiment with a variety of different approaches to produce a 100x100m surface of deprivation, scaled from most-deprived to least-deprived across a city. We will work with local stakeholders in the three pilot cities and modellers to compare varied modelled outputs, and develop a framework for a "good" IDEAMAPS models which is not only accurate, but also that scales well across cities.
Replicated storage, shared workspaces, GPU computing capabilities, computing cluster, ready-to-use scientific and geospatial packages, as well as supportive services such as code and data repositories, and database (e.g. PostgreSQL) and map servers (e.g. Geoserver).
Training data (100x100m raster): 300 diverse field-referenced locations across the city classified as deprived, mixed, not-deprived or non-residential with 30+ additional characteristcs (e.g. presence of open drains, school, streetlights, etc). We also have a ground photo taken at each of the 300 locations.
Covariate data (100x100m raster): More than 50 fine-scale publicly available datasets resampled to 100x100m raster format with full coverage of the city extent.
Contextual features - Covariates (10x10m raster): These features have full coverage of the city extent and characterise building, road, built area, and other physical characteristics at fine resolution from free imagery. See Engstrom et al. (2019) for details, and GitHub for example code.
