Radiant MLHub Spotlight Q&A: Renate Thiede
Combining mathematical statistics, geospatial data, and artificial intelligence in support of global development.
Happy New Year! Meet our Community Voice for the first quarter of 2022, Renate Thiede. Renate is a doctoral candidate in spatial statistics at the University of Pretoria, where she teaches part-time.
Spatial statistics is a niche area, combining geography and statistical analysis. Renate’s career trajectory as a Spatial Statistician started with her undergraduate studies in mathematical statistics, where she incorporated geospatial modules. In 2017, she won the STATOMET Prize for being the best statistics student and the Vice Chancellor and Principal’s medal for being the best undergraduate in the faculty. Renate started incorporating remote sensing technologies in her postgraduate studies. She created an algorithm to detect unpaved footpaths and roads in informal settlements between improvised buildings known as “shacks” (Thiede, R.N. et al., page 41). An indicator of poverty, informal roads are notoriously difficult to identify from satellite images, limiting policy-based decisions derived from geospatial analysis.
Renate is an alumnus of Radiant Earth’s first virtual ML4EO training of trainers bootcamp that focused on using machine learning with satellite data. The bootcamp ran from May 3–14, 2021, thanks to a grant from the GIZ FAIR Forward- Artificial Intelligence for all program, which the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) implements on behalf of the German Federal Ministry for Economic Cooperation and Development (BMZ).
In this Q&A, we sat down with Renate to discuss her journey combining statistics, geospatial data, and AI. She also shares her experience participating in the ML4EO Bootcamp.
“Radiant Earth is paving the way for a brighter future!”
A previous biographical introduction mentioned you knew early on that you wanted to pursue mathematical statistics and geoinformatics. What sparked the interest?
Statistics has always been an interest of mine. I’ve always enjoyed it — even as a child. I was fascinated by questions like: “How many blue Smarties are there in an average box? How many occurrences of a given letter are there on an average page?” The ability to derive complex insights from limited data isn’t much different from magic!
My brother, a GIS professional, converted to geoinformatics (GIS)! But converting me was easy, as GIS provides a way to visualise statistics in real-world space. The visual aspect was what got me interested at first. Soon enough, I realised that it also allowed me to solve real-world spatial problems. This is when I fell irreversibly in love with the combination of statistics and GIS.
You incorporate AI in spatial statistics. What type of applications have you built by combining spatial data, statistics, and AI?
Currently, I’m part of a research team working on a convolutional neural network to identify informal roads. This work is an outflow from my masters research. The trouble is that training datasets are not available for this type of road. So we’re in the process of building one by the painstaking process of manual digitisation. Preliminary results are very promising! I’m also supervising a masters student doing statistical research related to land cover classification in the Western Cape, South Africa.
Let’s talk about the ML4EO Training of Trainers Bootcamp, which you completed in October 2020. What made you want to participate? And how has the course helped you advance your skills in ML4EO?
The course has opened my eyes to a range of applications. The fact that it was fully online was a huge bonus. The focus on developing countries also made me eager to participate. I’ve found that researchers from developed countries cannot always appreciate the challenges and circumstances of the developing world, even in the field of EO. This includes challenges regarding EO, such as different environmental and infrastructural circumstances, and challenges regarding ML, such as limited access to computing power. The lecturers took us through complete working examples, which helped me ground my knowledge.
The ML4EO Bootcamp included the requirement to train at least five other people. The majority of your trainees came from the South African world of statistics and had no background in Earth observations. From your experience as a researcher and teacher, what value does AI add to spatial analysis in an African context?
The greatest value that comes to mind is speed and efficiency. The African continent is huge! Applications such as land cover classification or road detection cannot be done manually. Classification algorithms that are fine-tuned to specific environments are not versatile enough to cover the heterogeneous environments within our ecologically diverse countries. I believe AI models trained on African data can offer great insights into the state of conditions on the ground in a cost- and time-efficient way.
What was the biggest surprise you had teaching the ML4EO course materials to others?
The biggest surprise was the interest and participation from people with no or little prior knowledge of remote sensing and GIS. On the one hand, I get it: maps and spatial images grab people’s interest because they’re visually engaging. But I also think it’s a testament to the bootcamp design and the usefulness of interactive discussions. Our bootcamp sessions were safe spaces where people were welcome to ask anything, and trainees tapped into their diverse backgrounds to come up with answers.
You’re an advocate for open-source software and public data and plan to make your research data and codes available publicly. Why is it important to contribute to an open data ecosystem?
For me, there are two main reasons. First, it’s about equity. If data, software, and research are stuck behind paywalls, only the already privileged have access. This, in turn, widens the gap between the developed and developing world in terms of research, depriving the whole world of the valuable expertise and insights the developing world has to offer. Second, it has the potential for increasing software and data quality.
The great reproducibility crisis in academia clearly demonstrates the negative effects of a culture of no sharing. When research, software, and data are shared as freely as possible, this places an onus on the initial creators to make their product as accessible and reusable as possible. It also allows users of the products to identify and fix mistakes. This is hugely empowering!
As you know, Radiant Earth has various open training datasets available on Radiant MLHub. What does a data infrastructure such as Radiant MLHub mean to you as a researcher in a developing country?
It’s a game-changer! These freely available, easily accessible datasets allow me to educate students on both ML and EO and to further my research. The fact that most of the data is collected for Africa is amazing, as this is a continent that is generally underrepresented. This kind of data infrastructure is empowering for researchers in developing countries. It equalises the playing field and allows us to participate in international research and derive solutions applicable to our own countries.
From your perspective, what innovations in spatial analysis and AI can we expect in the next decade, and how do you think it will improve lives in Africa?
I think AI models will become increasingly commonplace. This, in turn, will inspire the creation of high-quality training datasets and increase the number of providers of computational resources. I believe education will become more accessible through the internet, empowering Africans to become informed, responsible AI practitioners. The rise of African-based AI practitioners will improve education on the continent and ensure that local problems are addressed with all the expertise of those who understand conditions on the ground. This will lead to practical solutions in various sectors, including agriculture, environmental protection and rehabilitation, urban planning and service provision, disaster management, and much more. However, this will only be possible through collaboration and sharing of models and data. In this regard, Radiant Earth is paving the way for a brighter future!
