Hub

  • In order to provide you with a better experience, this website uses cookies — by using our website, you agree to our use of cookies OK

ITT: Machine Learning and Remote Sensing in Tanzania

Technology for efficient exploitation of satellite imagery has advanced rapidly in recent years. The availability of very high-resolution satellite and aerial imagery, coupled with enormous computing power to process these images is opening up new opportunities to survey roads. This proof of concept research project will explore whether machine learning and high resolution Earth Observation (EO) imagery can reduce the cost and increase the speed of rural road surveying in Tanzania. It will build upon a growing body of research moving toward a method for the automatic identification of road layout and road condition survey from EO data.

Context/Objectives

This ITT is part of a three-year programme (Frontier Technology Livestreaming (FTL)) designed to help DFID apply frontier technologies to the biggest challenges in development. Focusing primarily on the monitoring of unpaved road condition, the immediate beneficiary of this study will be the national authorities who would have the means to access reliable data on the condition of the road network at increased speed and efficiency. Better data can improve the prioritisation of scarce maintenance resources, allowing targeted intervention, and would allow authorities to more confidently advocate for increased resources and investment. The poor condition of rural roads is one of the biggest blockages to inclusive growth in Tanzania. The majority of roads in Tanzania are unpaved rural roads, whose condition can currently only be ascertained by slow and expensive survey teams covering approximate 50km/day. The rural road network is over 100,000km long so even with 10 teams a survey from scratch would take more than a year. However, the condition of unpaved roads in Tanzania is subject to rapid change, therefore during the course of a year the road condition may change significantly due to heavy rain or traffic volume, thus rendering the survey data obsolete and out of date before it is completed.

The Department for International Development (DFID) is already supporting the AfCAP programme, which is investigating the use of satellite imagery coupled with human operators who visually identify road condition - without the application of automated techniques. This study seeks to take a significant step further to demonstrate a method of automatically identifying road condition - focussing primarily on unpaved roads. State of the art computer vision techniques are being pioneered by specialist research organisations to build models that predict features, such as road quality, from aerial images using a body of accurate data for ground truth to learn from. These models can then be used to predict features in new areas for which no ground truth data were available.

Through the DFID-supported Ramani Huria (open map) programme in Tanzania, Dar es Salaam and the surrounding area has one of the largest free resource of very-high resolution aerial EO imagery acquired by UAV of any developing city so is an excellent test ground of what is possible. A similar UAV survey programme has recently been completed for Zanzibar’s Unguja Island. In both Dar es Salaam and Zanzibar many roads remain unpaved and can be used to develop new survey methods, which could in future be extended to rural low volume rural roads.

This study will seek to demonstrate that the use of Automated Classification Techniques for Low Volume Road Condition is viable from high resolution EO imagery and to investigate methods by which they can be incorporated into future road condition surveys. There will always remain a role for manual verification to ground truth the automated survey, spot-checking and fill in detail for critical areas. A key benefit will be to better target manual surveys through understanding uncertainties across traditional and potential automated approaches. In the long term, a major attraction is the possibility of continuous monitoring. The frequency of new imaging is rapidly increasing, allowing broader scope with more observations then traditional road surveys. This shall further improve the understanding of how frequently road maintenance is required.

If a reliable method can be demonstrated, it could be used to augment conventional road surveys resulting in considerable savings in cost and time. While the immediate beneficiaries of this programme will be in Tanzania, FTL will also seek opportunities to apply learnings from the research project to other DFID programmes within both Tanzania and elsewhere.

Estimated timetable of the key stages:

  • Deadline for Receipt of Questions: Wednesday 5 July 2017 (submitted to ftlenquiries@imcworldwide.com)

  • Answers to Questions Shared with all Tenderers: Monday 10 July 2017

  • Submission Deadline: Thursday 20 July 2017

  • Provisional Selection of a Preferred Tender: Friday 28 July 2017

  • Contract Issued: Friday 25 August 2017

  • Mobilisation date: Monday 11 September 2017