Using Earth Observation & Machine Learning to optimise Road Condition Monitoring in Tanzania
The exponential growth of urban population in developing nations — driven by a desire for a better quality of life — has resulted in the mass migration of people from rural communities into urban areas, seeking employment opportunities. This growth is exerting immense pressure on infrastructure and resources with many countries struggling to cope. The provision of efficient and effective transport infrastructure is vital for economic growth, the continued delivery of vital services and for reducing inequality. As such, there is a responsibility to make cities more accessible to all socio-economic groups. To do so involves the creation of new road infrastructure and as well as the management of existing road networks. This however is both time and cost intensive and thus problematic for national and regional governments who are often time poor and resource stretched.
New approaches are required to support governments with monitoring, maintaining and targeting improvements for existing and new road infrastructure networks that are both less labour intensive and capable of providing rapid and automated condition assessments. Advancements in Earth observation (EO) technologies capable of imaging these environments at resolutions down to 0.25m from satellites and centimetres from drones are providing governments in developing countries with the opportunity for rapid visual inspection of remote regions over large areas. EO data comes in various forms; however is essentially any remote observation of the Earth using sensors — satellite and aerial photos, UAV imagery and Synthetic Aperture Radar (SAR). The value of this data though is truly unleashed through coupling this data with advancements in automated image analytics such as Machine Learning (ML). Exploitation of ML analytical techniques creates the opportunity to develop technology and data driven solutions that can process and analyse significant volumes of EO data over large areas in a timely manner, at a comparatively low cost, and provide outputs appropriate for informing asset managers and encouraging the uptake of data driven policy.
Funded by DFID, The Satellite Applications Catapult has been supporting Frontier Technology Livestreaming (FTL) to scope and design a project investigating how EO and ML technologies can be exploited for monitoring road surface conditions. This has been developed with a focus on Tanzania, specifically the island of Zanzibar. The island suffers from the typical problems such as rapid urban expansion, low-quality infrastructure, and heavy seasonal tropical rainfall responsible for frequently damaging existing roads.
Focusing primarily on the monitoring of road condition, the immediate beneficiary of this study will be 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. However, the condition of low volume 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.
What also makes Zanzibar a suitable test case is the potential to collaborate with the Zanzibar Mapping Initiative which is collecting ultra-high resolution imagery acquired by UAVs (Unmanned Aerial Vehicles).
As part of our activities, we visited Tanzania and Zanzibar to engage with DFID and the World Bank, and the Ramani Huria and Zanzibar Mapping Initiative (ZMI) teams responsible for mapping Dar es Salaam and Unguja and Pemba Islands. During the visits, we experienced first-hand the problems of poor-road quality and accessibility experienced by local people, including associated challenges with informal settlements. Poor sanitation, poor quality roads, frequent flooding and limited mobility within the city due to road infrastructure that struggles to serve the basic requirements of the local population.
During our visit, we held a workshop in Dar es Salaam during the Urban Resilience Tanzania Conference (URTZ), which brought together several key stakeholders in the field of urban resilience and marked the launch of the Tanzania Urban Resilience Program (TURP). Using our expertise in the exploitation of EO and advanced image analytics, we engaged with these stakeholders to ensure that we incorporated and identified requirements in the project scope that were both realistic and achievable, whilst still being innovative. The project identified that:
- Any solution developed needs to be low cost, timely and accessible to all.
- The solution needs to be transferable to other locations, and ultimately facilitate improved ground operations for road survey teams.
- It is vital for the solution to be compatible with existing Road Asset Management Systems (RAMS) already in use in Tanzania.
These requirements have been used to design the scope of the project based on the resources and information we had available. We agreed that the project should explore a key component of high-tech automated monitoring solutions for low volume roads in Tanzania, through qualifying the capability of ML and deep learning in assessing road condition using EO imagery, primarily acquired from UAVs however also supplemented where necessary, by satellite imagery.
We concluded that in-field ground calibration activities should form a key component of the project, and that outputs from the road quality survey when coupled with imagery data could form a good quality training dataset. However, this posed a problem, we would need to engage with suppliers who have the expertise in both using machine learning with EO data and expertise in road condition surveying in the field. In line with best practices — we endorse the development of open source solutions, which are available to all for further development, thus making the technology accessible to users in developing nations and international donor agencies, where the need is greatest.
Increased availability of EO data and low-cost UAV data means the opportunity exists for expansion of this project into other areas of Tanzania, as well as other developing nations driving continued long-term impact.
Techniques applied as part of this project can be adapted and transferred to various datasets covering a variety of environmental conditions. In addition, advancements in computing power coupled with cloud based services means that future application of machine learning using EO data can be undertaken rapidly and at lower cost, with frequent updates. We actively encourage open sharing of any algorithms or tools generated as part of this project, with a view to inspire others to develop applications suitable elsewhere.
We look forward to working alongside FTL, DFID and the successful implementing agent in the coming stages of the this one-year project by providing support throughout the duration, and are excited to see how it develops.
