When estimating biophysical plant traits the observational scale can be essential. As a consequence, the spatial and temporal resolution that we use for our remote sensing observations can be of key importance – but may vary for each trait.
Luckily, the development in robotics, computer vision, and sensor technology has provided new remote sensing tools to capture unprecedented ultra-high spatial and high temporal and spectral resolution with unmanned aerial vehicles (UAVs). This development has led to a revolution in geospatial data collection and opened new ways to gather vegetation traits beyond what has been possible with classical remote sensing. (Aasen et al., 2018). Within the PhenoFly team (www.phenofly.net) we investigate how reliable information about vegetation can be gathered from these sensing systems. To investigate scaling effects, we combine UAV observations with measurements from our field-phenotyping platform (FIP) - cable-suspended multi-sensor system that can freely be moved across 1 ha of land (Kirchgessner et al., 2017) – and other observations such as phenocams. This gives us the opportunity to investigate the spatial scaling effects starting from sub millimetre resolution and temporal scaling effects from daily to weekly observations.
In this contribution we will show examples from estimation phenology and LAI, two plant traits within the (proposed) essential biodiversity variables species traits phenology and morphology (Kissling et al., 2018). In addition, we highlight the challenges that can occur when comparing results from different sensing approaches.

Bigger plants weigh more. This simple maxim provides the foundation for allometry, whereby dimensional attributes can be used to infer plant biomass. Around the world, there is a need for high quality measurements of aboveground biomass and associated carbon stocks, in order to quantify forage availability for livestock and wildlife, fluxes of carbon between different pools, and ecosystem responses to disturbances such as fire, for example. The diversity of vegetation structures has led to the development of a wide range of survey approaches; these can yield reasonable predictions of plant mass but are often very labour intensive to implement, resulting in limited spatiotemporal resolution of observations. Consequently, biomass stocks can suffer from undersampling, particularly in heterogeneous and/or temporally dynamic low stature ecosystems such as grasslands and shrublands.

Fine-grained proximal remote sensing has widely been heralded as a potential solution to these undersampling issues, with the advent of ‘low-cost’ lightweight remotely piloted aircraft, ‘low-cost’ imaging sensors and accessible structure-from-motion (SfM) photogrammetry resulting in drone-based landscape surveys becoming increasingly popular within both academia and industry. This explosion in do-it-yourself proximal remote sensing has seen the development and adoption of workflows that are often appropriate for specific applications, but which yield datasets and findings that can be difficult to compare, hindering cross-site syntheses and transferable understanding about the ecosystem properties retrievable from remote sensing observations.

Here we present an ongoing global experiment to investigate the transferability of allometric relationships through space and time and taxa. Working with an ever-growing global network of drone-equipped scientists, we have used SfM photogrammetry to measure vegetation structure calibrated with biomass observations from destructive harvests plots. High standards of both quality and comparability are ensured through a standardised and comprehensive sampling protocol (available at https://tinyurl.com/yavrdy9x), followed by centralised processing and analysis. To date (November 2018), 36 groups worldwide are planning data contributions (Figure 1). Eight teams have already collected 16 different datasets, consisting of 303 observations (harvest plots) representing 18 different species/communities across three continents. Our preliminary findings indicating strong coefficients of determination (R2 ranging from 0.64 to 0.95) for various taxa, and reveal new insights into the (dis)similarity of standardised allometric functions. We will share our aspirations of how this new and expanding global network will enable further coordinated experiments to continue advancing observations of plant trait relationships around our planet.

Figure 1. Pending and completed sampling sites in the drone allometry experiment

Drylands, one of the planet’s largest terrestrial biomes, are a combination of sparse vegetation, biological soil crusts (biocrusts), i.e. soil surface communities of cyanobacteria, mosses and/or lichens, that can cover up to 70% of the surface, and bare soils. Monitoring the spatial distribution and the phenological cycle of dryland cover can provide highly valuable information, as they control key ecosystem processes. In this study, we examine the potential of ESA Sentinel-2 (S2) images to monitor the temporal succession of typical dryland ecosystem components (i.e. biocrusts, annual and perennial plants) in the Negev desert. The optical properties of biocrust samples typical of the study area were characterized in laboratory simulating a precipitation event by wetting the samples with about 5 mm of distilled water. Optical properties of dry and wet samples were measured by means of a hyperspectral imaging spectrometer (Hyperspec VNIR, Headwall Photonics, USA) covering the visible and near infrared spectral regions (400-1000 nm). The high spatial resolution of the hyperspectral images collected on the samples allowed to investigate the spectral behavior of biocrusts and soil patches during their rehydration (i.e. greening). A robust index sensitive to biocrust greening and not affected by soil moisture was defined based on the depth of the chlorophyll a absorption feature in the red region. The index was then applied to a time series of twenty S2 images acquired between August 2015 until August 2017. The S2 time series was discussed in relation to ground precipitation data and field data on the presence of biocrusts, annuals and perennials plants (i.e. dominant species and fractional cover) from eight study sites along a precipitation gradient in the Negev area. A change detection analysis was then performed in order to emphasize the biocrust greening at the beginning of the rainy season and the maximum vegetation development of the desert cover, a combination of biocrust and vascular plants. Results showed that Sentinel 2 data were successful in identifying biocrust greening in the northern Negev (the semi-arid region), while the identification of biocrust greening was more difficult in the more dry southern area. Promising results were also obtained in the analysis of the vegetation phenology: the vegetation development along the Negev latitude gradient showed clear differences between 2016 and 2017 rainy season, in relation to the different total precipitation registered. These results show Sentinel 2 potentiality for monitoring dryland ecosystems and impacts of climate change, in particular, on biocrusts that are known to be especially vulnerable to changing climatic conditions.

Landslides present a prominent hazard in the Alpine region of Italy, and mapping and monitoring them is essential for risk assessment and planning. Presently, widespread landslide mapping is often approached by visual interpretation of aerial photogrammetric surveys and various land surface movement measurement techniques. The novelty of this study is in the use of timeseries analysis of spectral indices of vegetation dynamics from the ESA Sentinel 2 satellites in conjunction with UAV surveys for landslide detection and monitoring. This approach should prove particularly useful in densely vegetated areas where land surface movement detection is more challenging. Coupling ecosystem and landslide research, this analysis can unveil new relationships between vegetation behavior and slower landslide processes.
The primary data source is the timeseries of vegetation indices derived from Sentinel 2 satellite imagery for the period 2015-2018 and a regional landslide database, augmented by UAV surveys of selected landslides. The UAV surveys have been performed at different stages of vegetative cycle and are used to enhance the resolution of the available data and provide textural and structural assessment of the vegetation canopy with orthophoto and structure from motion techniques.
We present the various approaches for establishing the effects of landslide activity on vegetation behavior, and the results of landslide detection based on this analysis. The analysis is performed by establishing a range of baseline vegetation behavior statistics and comparing with the vegetation dynamics on the landslide bodies using different baseline criteria, normalizing factors and scale of analysis. To single out the effects of landslide activity from other factors, different methods are tested for establishing vegetation baseline and normalizing the analysis for topographic variables, vegetation types. Furthermore, UAV data acquisition on selected landslides is explored to augment the analysis with tree species classification, canopy texture and structure characterization.
The results show the link between landslide activity and vegetation behavior in contrast to the vegetation baseline, the constraints of the methodology for large scale landslide mapping and the effect to which the approach may be useful.

Biodiversity includes multiscalar and multitemporal structures and processes, with different levels of functional organization, from genetic to ecosystemic levels. One of the mostly used methods to infer biodiversity is based on taxonomic approaches and community ecology theories. However, gathering extensive data in the field is difficult due to logistic problems, especially when aiming at modelling biodiversity changes in space and time, which assumes statistically sound sampling schemes. In this context, airborne or satellite remote sensing allows information to be gathered over wide areas in a reasonable time. Most of the biodiversity maps obtained from remote sensing have been based on the inference of species richness by regression analysis. On the contrary, estimating compositional turnover (beta-diversity) might add crucial information related to relative abundance of different species instead of just richness. Presently, few studies have addressed the measurement of species compositional turnover from space. Extending on previous work, in this talk, we will propose novel techniques to
estimate beta-diversity from airborne or satellite remote sensing, mainly based on: i) multivariate statistical analysis, ii) the spectral species concept, iv) multidimensional distance matrices, and iv) the Rao's Q diversity, v) the Rényi generalised entropy.
Each of these measures addresses one or several issues related to turnover measurement. This talk will encompass most of the available methods for estimating beta-diversity from remotely sensed imagery and potentially relating them to species diversity in the field.

The Cerrado (Brazilian savannah), is the most diverse of all of the world’s savannahs. While holding a high diversity and endemism of species, this biome is mostly unprotected and understudied. Species functional traits, as well as trait or functional diversity bring an extra dimension to biodiversity, which remains even less studied, although it plays a major role in e.g. the resilient and integrity of ecological systems. Monitoring species traits is therefore of great importance for biodiversity conservation. In this study, we propose to map woody plant traits using spaceborne hyperspectral (Hyperion) data combined with field-based plant inventory data and plant trait data. We demonstrate our concept in one study site in a natural vegetation area in the Cerrado. The proposed methodology consists in a two-stage approach. Initially, the community (taxonomic) turnover patterns were mapped via Sparse Generalized Dissimilarity Modelling (SGDM). This method combines the use of Generalized Dissimilarity Modelling for community turnover mapping, with the capability of Sparse Canonical Correspondence Analysis for dealing with hyper-dimensional data, such as hyperspectral imagery. This results in community dissimilarities between pairs of samples (sites), which can be ordinated in a Nonmetric Multidimensional Scaling, to derive the main community variation (turnover) axes for the sampled sites, which, via a subsequent knn-imputation can be extrapolated to the whole study area, thus resulting in a community turnover map. The extracted community axes were used in matrix calculations to assign the sampled species to the derived turnover axes, and their specific traits. Here we demonstrate this concept by mapping several woody plant traits throughout our study area. These maps can form the bases of e.g. maps of functional trait diversity, which in turn can support informed and effective conservation actions to protect the biodiversity of this important biome. The approach is also suitable to be applied in other systems and, with the advent of forthcoming spaceborne hyperspectral missions, could prove extremely useful for monitoring and conservation of natural systems over large extents.