Public and private organizations are equally excited about future prospects of machine learning (ML) and artificial intelligence (AI). Over last few years, we have been seeing so many success stories of machine learning applications in marketing, image classification, natural language processing (NLP). However, remote sensing is a field has not come into machine learning focus yet. It has got a wealth of untapped data in form of highly available satellite imagery which can be integrated with socio-economic data in order to build environmental and social threat prediction models. In addition to web and social media data, remote sensing applications can hugely benefit by building their building their own user communities and encouraging them to contribute by verifying assessments.
Using these such models, one can then build demand-driven applications to help public and private organizations understand, prepare and respond to economic, social, and humanitarian losses in timely manner. These applications can be made near-real-time remote sensing applications by adding additional layer of response tools, like alerts and advisories etc. Ideally these applications should have the capability to be scaled from local to global level for disaster and risk management.
Recently a case study remote sensing application has been built for floodplain prediction in Senegal. It has been noticed that floods cause more economic, social, and humanitarian losses than any other threat. It leveraged satellite imagery data using Google's cloud based Earth Engine and turned turn data into insight for decision-makers on the ground. Following are fewer of its objectives:
- disaster risk reduction
- event prediction for timely response and recovery
- fill information gaps about exposure and vulnerability to flooding
- allow stakeholder to better target investments and protect assets.
Official census data were one of key component of training set, as reported its variation is effectively explain by five key features: 1) a lack of basic informational resources, (2) age of population, (3) disabilities, (4) distance to main population centres and (5) population increase due to migration. These factors reportedly explain 69% of the variation in the census. These features combined with remotely sensed satellite data were used to train machine learning model for floodplain prediction. More about this study can be found under https://link.springer.com/chapter/10.1007/978-3-319-65633-5_16