This post is part of our Dutch series featuring articles from the NRC Climate Blog written by our own ENP researchers.
You can find the original post (in Dutch), published on the NRC Climate Blog on July 2020 here.
Read the translated text in English directly below.
Big data can easily be misinterpreted
Big data is increasingly being used to map the effects of climate change. Huge amounts of data, beautiful graphs and impressive images are showing intriguing patterns, such as migration flows after or during a cyclone. This is then enthusiastically shared and widely picked up on (social) media and in science, because hey, big data is a lot of data, I guess it must be right!
Unfortunately, this is not always the case. If incorrect assumptions underlie a big data analysis or if the big data is not interpreted correctly, it creates a false impression.
This is how I fell for the beautiful, impressive pictures of big data in a scientific article about climate adaptation in Bangladesh. Using anonymous mobile phone data (call detail records), this article examined people's behaviour during Cyclone Mahasen, which hit the south of Bangladesh in May 2013.
The data consists of the times a person's phone contacted another mobile phone tower, which could indicate that the person was moving. That data showed something strange: many people in the affected areas were on the move during the cyclone in the middle of the night. The authors of the article thought it was a last-minute evacuation. The government had been very late in informing people in the area about the approaching cyclone, so unexpected evacuation seemed likely.
Excited by their results, I reached out to one of the article’s lead authors: David Wrathall from Oregon State University in the United States. We agreed that I would spend two months in the affected area collecting more details about what the data actually showed.
We still thought it was an evacuation, although we noticed that many of the detected mobile phone movements moved towards the coast instead of inland. That seems illogical when you're looking for safe shelter.
Once I arrived at the destination, interviews and conversations revealed that something completely different had happened (at least, in that location). Very few people had sought safe shelter because of the cyclone. The dikes were strong enough to hold back the water. In addition, there were enough shelters within reach of the existing phone tower, so even the movements of those who had evacuated would not have been visible.
But the data did show movements of hundreds of people in the middle of the night, during the cyclone. They turned out to be fishers. There are many large wooden fishing boats in the local harbours. Depending on their size, about 10-20 people work on such a boat. During the storm, hundreds of fishers went to the coast to protect their boats. They tied the boats tight to the coast and stayed onboard.
They also called fellow fishers to come help or rescue their own boats. This often happened in shifts, so they could also go home if they needed to help their family. That created a lot of telephone traffic and many fishermen were travelling back and forth in the affected area during the cyclone.
As a next step, together with Ruben Dahm from the Deltares research institute, I investigated an area where few abnormal movements were detected during Cyclone Mahasen. That area suffers from severe coastal erosion, causing the dikes to be broken and weak. Coupled with interviews, an analysis of satellite data showed how this area already had lost several kilometres of land to the River Meghna in the last 20 years. Cyclone Mahasen appeared to have had an enormous impact here. This also involved forced evacuation. The water poured into the village and people described how they climbed on their roofs or fled to nearby schools or shelters. Those locations were often within the range of one mobile phone tower, so no movements were detected in the big data analysis.
Big data can indeed provide us with clues that lead to better understanding of how climate change and natural disasters affect people and how they respond to them. But data alone is of little use; it must be interpreted with on-site research. If that is missing, data can easily be interpreted the wrong way. People don't flee every cyclone, and a lack of mobility is not necessarily a good sign either.
For this research, a Dutch Research Council (NWO) Veni project, with use of satellite data co-financed by Deltares, Ingrid Boas (Environmental Policy Group, Wageningen University) worked with Ruben Dahm (Deltares) and David Wrathall (Oregon State University). For the full article, click here: https://doi.org/10.1111/gere.12355