The case for remote sensing of individual plants

Abstract

Remote sensing has greatly advanced our understanding of the land surface and the role of biology within it (Tucker and Sellers, 1986). But our ability to generate observations from remote sensing data at scales clearly aligned with biological processes has been limited. The problem is that biological processes like natural selection, metabolism, and resource allocation vary within and among individuals and change on scales of space and time that are finer than the granularity of traditional remote sensing measurements. Advances in technology are poised to overcome this problem by generating data from tower mounted, airborne and satellite sensors at scales of space and time aligned with biological understanding. Miniaturized sensor designs focus on lightweight instruments that can be carried by drones or operated on field platforms. And constellations of small sensors called cube-sats are now working together in space to image the entire land surface of our planet every day at resolutions fine enough to resolve individual plants. The quantitative step forward represented by these new technologies is significant, but the most important advance is conceptual. Measurements from remote sensing at ultra-high spatial and temporal resolution open the door to characterizing phenomena that have been beyond our grasp, including population dynamics (Kellner and Hubbell, 2017, 2018), high-spatial-resolution phenology (Wu et al., 2016), and physical quantities that can be related to organismal condition, like foliar chemistry, canopy temperature and solar-induced fluorescence (Daumard et al., 2010; Porcar-Castell et al., 2014). These new measurements cross thresholds of scale in space, time, and biological organization that are clearly aligned with decades of understanding in plant biology (Gamon et al., 1992; Demmig-Adams and Adams, 2006). Drones operate at low altitude and can produce measurements at densities orders of magnitude greater than traditional airborne remote sensing (Kellner et al., 2019). These new measurements fundamentally alter our scope of inference by allowing us to work not with area-based summaries, but with individual plants. The cost of operating a drone is about an order of magnitude less than a traditional airborne program, which makes it possible to acquire measurements frequently, on demand. Lidar sensors record the return-time of emitted laser pulses to produce a physically accurate three-dimensional point cloud (Disney, 2019). Measurement densities from drone lidar are easily in the thousands of points per square meter (Fig. 1). This important distinction has allowed high-density point clouds from drones to resolve branch and stem structure within individual trees, and to associate individual plants with remote sensing data (Brede et al., 2017; Trochta et al., 2017). Optical sensors on drones can produce pixels at the leaf level, resulting in many thousands of measurements within single trees from imaging spectrometers, multispectral cameras, and traditional digital cameras (Fig. 1). Increasing the density of measurements vastly increases the information content in remote sensing data about plant structure and condition and reduces or eliminates the mixed-pixel problem, which occurs when large pixels contain aggregations of multiple objects, diminishing our ability to interpret remote sensing data in biological terms. Combining information from multiple sensors at these new scales of observation is a priority for research and discovery. One of the most exciting applications of remote sensing of individual plants is the measurement of chlorophyll a fluorescence in sunlight. This emission, called solar-induced fluorescence (SIF), is quantifiable using optical remote sensing within narrow spectral bands. Measurement of SIF is important for two reasons. Unlike remote sensing metrics based on reflected light, SIF is a biologically regulated emission. Because SIF originates from the light reactions of photosynthesis, measurement of SIF can help to constrain instantaneous estimates of photochemistry. The interpretation of SIF is challenging because there is one observed quantity (SIF) and two unknowns (photochemistry and heat dissipation, Porcar-Castell et al., 2014). Studies using satellite observations have demonstrated that ecosystem gross primary production and crop yields correlate positively with SIF (Guanter et al., 2014; Li et al., 2018) and that SIF declines in the presence of water stress (Daumard et al., 2010). More work is needed to determine when SIF correlates positively with photochemical conversion, as opposed to heat dissipation or canopy structure, and whether the relationship is consistent across leaf, organism, and ecosystem scales. Understanding the biological basis of this relationship will strengthen our understanding of correlations between SIF and ecosystem gross primary production (Magney et al., 2019). Drone remote sensing can help to overcome these uncertainties by generating diurnal time series observations of SIF at leaf, organism, and ecosystem scales that are coupled with other measurements. Answers to these questions are at the core of our understanding of plant biology because they characterize the efficiency with which absorbed sunlight converts into chemical energy. These questions also have direct implications for national space agency programs advancing the measurement of SIF as a proxy for photosynthesis and for understanding the role of plant biology in the Earth system (Drusch et al., 2017; Stavros et al., 2017; Sun et al., 2017). Dense image time series from cube-sats can provide the combination of high temporal frequency and fine spatial resolution to map individual trees and quantify population dynamics throughout whole continents. Species with conspicuous phenology are model organisms for questions in large-area analysis of tree populations because they exhibit characteristics that allow individuals to be detected and tracked over time using high-resolution satellite data (Fig. 2). Repeated attempts to observe individual trees in a remote sensing time series result in a sequence that indicates whether a given individual was detected or not on a given date. Detection using conspicuous phenology, such as flowering or fruiting, unambiguously affirms that an individual is living. An undetected tree may be alive but not detected, for any number of reasons, or undetected because it is dead. Consequently, the analysis of these data requires a statistical framework that can accommodate variable probabilities of detection and missing observations, similar to capture–recapture analysis (Kellner and Hubbell, 2017). Studies using repeated detections of individuals from remote sensing time series have quantified rates of mortality, recruitment, and population growth (Thomas et al., 2013; Kellner and Hubbell, 2017, 2018). Quantifying how these rates change across spatial gradients could characterize the demographic determinants of species range boundaries. Extracting the most important information from cube-sat time series requires the ability to identify and monitor objects, like individual plants, in contrast to pixel-based summaries. This new capability for object-based remote sensing at planetary scale is entirely novel and transformative. It is also a decisive break from traditional image processing, which is almost exclusively pixel-based. Developing the capacity for object-based remote sensing at this scale will require new machine-learning pipelines and substantial computing resources to align millions of satellite images, identify objects within image subsets, and track objects through time (Gorelick et al., 2017). The promise of these high-density time series is the ability to quantify how forests are changing over time with the granular resolution of individual plants, to test biological hypotheses across regions and continents, and to interpret high-resolution image time series from a biological point of view. New technology is strengthening the connection between remote sensing and the biological sciences. The importance of this connection rests on the idea that measuring the land surface at scales of space and time that align with biological knowledge and individual organisms will lead to new insights. Developing this connection is important because biological systems are fundamental components of the Earth system, governing fluxes of heat, water, and carbon between the land surface of our planet and the atmosphere. Feedbacks between biological processes and the Earth system are a source of uncertainty in global biogeochemical cycles and climate forecasting (Schimel et al., 2015). Remote sensing observations of individual plants can advance fundamental plant biology and Earth-system science. This work was supported by Brown University, the National Science Foundation (DEB 1852710), and funds provided to Brown University by Peggy and Henry D. Sharpe Jr. and Peter S. Voss. We thank Markus Birrer, Pamela K. Diggle, Christoph Eck, Cristoph Falleger, Benedikt Imbach, Kamil Král, Martin Krůček, Henry Johnson, Joe Mascaro, Carlos Silva, Jan Trochta, Orlando Vargas-Ramirez, Tomáš Vrška, Carlo Zgraggen, Stephen Porder, Dov Sax and one anonymous reviewer.