Methods and tools to improve understanding and application of biodiversity data based on satellite Earth observation: Examples from Benin, Ghana and Brazil

Abstract:

People from a variety of backgrounds need to understand and apply biodiversity data as part of community leadership roles, such as tribal elders, entrepreneurs, researchers, city managers, business people and labor organizers. New technology trends are changing options for using biodiversity information due to the increasing use of low-cost aerial vehicles and in-situ sensors for scientific observation, the combination of data from commercial and government satellites, and increased use of artificial intelligence to interpret large datasets. The Space Enabled Research Group (MIT) and our collaborators design Decision Support Systems that use biodiversity data to inform policy and estimate the impacts on humans, plants and animals. The presentation highlights three projects: 1) collecting data using active and passive satellite, airborne and water sensors to monitor invasive plants in Benin; 2) using government and commercial satellite data to estimate and validate findings on deforestation due to mining in Ghana; and 3) combining field data and satellite observation to map biodiversity and socioeconomic change in mangrove forests in Brazil. In each of these examples, the analysis asks how human actions such as deforestation, fishing, and fires influence biodiversity. Experience from these projects reveals patterns that may be common. First, the projects seek to build information systems that meet local needs; this requires large investments of time by teams to exchange knowledge about local context. Second, the projects seek to build online tools with customized presentations of biodiversity data for local users. Simultaneously, the projects strive to make it easy to re-use the analytical framework for other geographic regions and application areas. Third, the projects apply tools from artificial intelligence and cloud computing to harness long-term data sets. The team also invests significant time to develop methods to quantify the errors when machine learning is used for feature classification. Fourth, these projects generate new data streams with low-cost sensors; this process can be complex to start in an unfamiliar context. The presentation will highlight preliminary results from the ongoing research projects that illustrate how these patterns impact the biodiversity community.