Wildness estimation of large rivers

Wild rivers are an invaluable resource that play a vital role in maintaining healthy ecosystems and providing ecosystem services. These rivers provide habitat for a wide variety of plant and animal species. However, the increasing pressure of human activities has been causing a rapid decline of biodiversity and ecological function. But there is currently no map available that identifies the river segments that remain under good conditions, which would be worth protecting and conservation. The quality of the river in terms of wildness is multidimensional and difficult to measure with simple remote sensing products such as land cover and human modification products. However, by using remote sensing images with machine learning methods, we were able to better improve our abilities to provide a detailed map of river wildness with high spatial resolution. Citizen science, which involves engaging non-experts in scientific research, has become an increasingly popular tool for collecting data and expanding the scope of scientific inquiry. In this study, we combine citizen science, machine learning and remote sensing images to map the wildness of large rivers globally. We built a reference database of annotated images thanks to the contribution of citizen scientists through a web application. The application asks each participant to rank two images based on their wilderness for multiple rounds. Then, the rankings were then used to assign a wildness score to each image using the true skill algorithm. Finally, we used this dataset to train an artificial neural network to identify the wildness of river sections. The performance of the model was evaluated using a preliminary test on a geographically separated subset of the data. This river wildness map will provide invaluable insights for conservation initiatives and enable targeted actions to safeguard and restore the remaining wild rivers.