Real-time data can accelerate governments’ understanding of a disaster’s impact, as well as the urgent needs of a population much faster than traditional emergency data sources. The immediate days following a sudden onset emergency are typically focused on crucial decision making about the scope of impact, location of the most vulnerable and resource allocation. Following extensive discussions with relevant humanitarian agencies and stakeholders in Indonesia, Pulse Lab Jakarta and UNICEF identified a set of problem statements and have been working to outline solutions that integrated non-traditional real-time data sources.
Problem: Data on movement of people is expensive and time consuming to collect
Solution: Combine Facebook Geoinsights and Mobile Network Operator (MNO) data to produce real-time displacement maps
During disasters, Facebook provides anonymised, privacy-conscious aggregated real-time data on the movement of people at the town (Kecamatan) level. Under an existing agreement with Facebook, this data is available to UNICEF and some international NGO’s (IFRC), at no cost. Whilst rigorous benchmarking is required before deploying this data in decision making, in exploratory work done by PLJ and UNICEF it was revealed that the unweighted Facebook population movement data in the aftermath of the 2018 Palu earthquake was broadly representative of the later confirmed displacement data, and substantially more accurate than other early and informal data available in the 72 hours after the disaster.
Problem: Data preparedness is time consuming and under-resourced
Solution: Leverage a set of reproducible open source data pipelines that ingest data from disparate sources and presents that data via a common, documented, interoperable API.
Humanitarian Data Exchange (HDX) stood out from our survey of potential real-time data. HDX is an open source platform built by UNOCHA that allows UN agencies and NGOs to store, manage and access datasets for instant analysis, visualisation and comparison. The platform and tools are accessible to both technical and non-technical staff, and allow the design of visualisations and dashboards in a web browser. HDX also leads work on the Humanitarian Exchange Language (HXL) data tagging standard, which enables non-technical staff to easily annotate and validate data and create both summary and map based live data dashboards without the need for coding or use of additional data cleaning or validation resources. These dashboards are suitable for viewing on a range of devices and already function well in low-connectivity and offline environments.
Problem: 4W/5W Tools are cumbersome and take time away from disaster response
Solution: 5 Whats, Leverage existing technologies to create a standardised, massively streamlined 5Ws reporting workflow
The process of completing a 5W’s questionnaire can be complicated and is prone to data entry errors as it requires many data post-processing steps before becoming available (and soon after becomes unreliable or out of date). Using Natural Language Understanding NLU, 5Whats can understand responses and ask clarifying questions to ensure that activities are being recorded in the correct location and have sufficient detail, while reducing reliance on spreadsheets and manual, computer based data entry. UNICEF has invested extensively in, and open sourced the conversational reporting platform Rapidpro. Operating over SMS, Whatsapp and other instant messaging platforms, it allows for structured data collection in a conversational flow, without imposing additional workflow requirements or cognitive load on already time poor field and information management staff.
This research exploration has unearthed the need for the adoption of innovative tools that can surface relevant real-time data in a timely and interpretable fashion during times of crisis. Going forward, the goal is to make existing public data sets and ones available to UNICEF more accessible through reproducible data pipelines; test the prototype of our visualisation tool that demarcates affected areas; as well as gather user feedback and scope the feasibility of applying machine learning applications to existing data sets (such as predicting movement of people post-disaster using historical Facebook Geoinsights and Mobile phone CDRs).
