Wildlife detection, counting and survey using satellite imagery: are we there yet?

Abstract

ABSTRACTWildlife surveys are key to assessing the health of global biodiversity. Traditional field and aerial methods however have significant limitations, including high costs, substantial time investment, and potentially biased estimates. The increasing availability of high-throughput monitoring sensors in recent years has opened new perspectives for wildlife studies. Very-high-resolution (VHR) satellite sensors promise large spatial and temporal coverage while seemingly being less costly than traditional methods. Deep learning (DL) has shown increasingly impressive capabilities for processing remote sensing imagery, suggesting good prospects for imagery-based wildlife surveys. We reviewed all taxa and geographic area studies that use satellite imagery for wildlife detection, counting and surveys. Through an analysis of 49 peer-reviewed papers, this study examined the sensors and resolutions employed along with the methods used to detect, count and survey wildlife in various biomes. Results have revealed an increasing trend of publications. Mammals and birds are the focus of most of the papers, mainly in polar/alpine and pelagic ocean waters biomes. Visual interpretation is the most common method used for wildlife detection and counting while total count is mostly used for surveying. Most of the papers present a proof of concept to detect, count and survey wildlife. Technological advances are expected to enhance the spatial and temporal resolutions of satellite imagery, as well as image processing capabilities. Three main bottlenecks preventing the development of on-demand operational approaches for wildlife surveys were identified: 1) the business model of VHR satellite imagery providers is not conducive to wildlife studies; 2) satellite imagery is rarely shared; and 3) the training of multidisciplinary highly qualified personnel is underdeveloped. In response, this review presents key research priorities for advancing remote sensing for wildlife monitoring. They include wildlife-dedicated satellite constellations at enhanced spatial and temporal resolutions, increased data accessibility and sharing, adapted survey strategy, development of foundational DL model and multidisciplinary integration. We believe that progress in these directions will foster new survey strategies that are certain to revolutionize wildlife monitoring in the decades to come.

Keywords

• Wildlife
• remote sensing
• satellite imagery
• survey
• deep learning