Search

Our understanding of the Earth´s functional biodiversity and its imprint on ecosystem functioning, structure and resilience is still incomplete. Large-scale information on vegetation properties (‘plant traits’) is critical to assess functional diversity and its role in the Earth system. Such parameters are constantly changing due to variations in environmental conditions which makes extensive in-situ measurements in the field not logistically feasible. The advent of the upcoming space-borne hyperspectral missions will facilitate to map these properties. However, we are still lacking efficient and accurate methods to translate hyperspectral reflectance into large scale information on plant traits across biomes, land cover and sensor types. Yet, the absence of globally representative data sets on reflectance data and the corresponding in-situ measurements represents a bottleneck to develop empirical models for estimating plant traits from hyperspectral reflectance. Recent and ongoing initiatives (e.g. EcoSIS) provide a constantly growing source of hyperspectral data and plant trait observations from different vegetation types and sensors. In this study we integrated 29 data sets including four different ecosystem types spanning from Europe to north America. By combining these heterogeneous data sets, we propose multi-trait models based on Convolutional Neural Networks (CNNs) that simultaneously infer multiple plant traits from canopy spectra. We targeted a broad set of structural and chemical traits (n=20) related to light harvesting, growth, propagation and defense (e.g. leaf mass per area, leaf area index, pigments nitrogen, phosphorus). The performance of our multi-trait CNN models predicting these traits was compared to single-trait CNNs as well as single-trait partial least squares regression (PLSR) models. The results of the multi-trait models across a broad range of vegetation types (crops, forest, tundra, grassland, shrubland) and sensor types were promising and outcompeted state-of-the-art PLSR models. We found that the overall prediction performances significantly increased from single- to multi-trait CNN models and those of PLSR models. The key contribution of this study is to highlight the potential of weakly supervised approaches together with Deep Learning to overcome the scarcity of in-situ measurements and take a step forward in creating large-scale maps of Earth’s biophysical properties with the increase in availability of hyperspectral Earth observation data.

As global change accelerates, the urgency for a solid understanding of biosphere-environment interactions grows. However, we need more data on plant functional traits to test such relationships reliably across ecosystems. The TRY database contains an impressive collection of plant trait measurements for thousands of species already, and there have been some approaches to spatially extrapolate them using geospatial predictors and remote sensing data; however, the original data is spatially sparse so that extrapolations come with substantial uncertainties. At the same time, citizen scientists have collected increasingly dense observations of species occurrences around the globe. Here, we test if we can link species occurrences from the citizen science project iNaturalist with trait observations from TRY to produce global trait maps without the need for spatial extrapolation. We generated spatial grids for 18 traits, calculating a mean for each grid cell by averaging trait values associated with observations within that cell. We compared mean trait values from iNaturalist observations to community-weighted mean traits from sPlotOpen, a globally sampled dataset of vegetation plot data. Our results show correlations between the two datasets of up to r = 0.69, especially in biomes with higher iNaturalist observation density and those not dominated by trees. Also, we show that iNaturalist-derived maps have higher correlations to sPlotOpen-derived maps than previously published trait maps. This strong correlation between two fundamentally different datasets is astounding and unexpected. iNaturalist is noisy and heterogenous, sampled by citizen scientists who share the species they encounter and find interesting; sPlotOpen is a data collection of vegetation plots that were measured and recorded within the framework of specific research questions. The fact that these two datasets exhibit such a strong resemblance opens up a promising avenue for using the data treasure trove that is crowd-sourced data to help fill the gaps in plant trait data and demonstrates that crowd-sourced data, such as the iNaturalist observations, can be used to complement professional data collection efforts.

Vertical leaf angles and their temporal variation are directly related to multiple ecophysiological and environmental processes and properties. However, there is no efficient method for tracking leaf angles of plant canopies under field conditions.Here, we present AngleCam, a method to estimate leaf angle distributions from horizontal photographs acquired with timelapse cameras and deep learning. The AngleCam is a pattern recognition model based on convolutional neural networks and was trained with leaf angle distributions obtained from visual interpretation of more than 2500 plant photographs across different species and scene conditions.Leaf angle predictions were evaluated over a wide range of species, plant functional types and scene conditions using independent samples from visual interpretation (R2 = 0.84). Moreover, the method was evaluated using leaf angle estimates obtained from terrestrial laser scanning (R2 = 0.75). AngleCam was successfully tested under field-conditions for the long-term monitoring of leaf angles for two broadleaf tree species in a temperate forest. The plausibility of the predicted leaf angle time series was underlined by its close relationship with environmental variables related to transpiration. Moreover, showed that the variation in leaf angles resembles changes in several leaf-water related traits.The evaluations showed that AngleCam is a robust and efficient method to track leaf angles under field conditions. The output of AngleCam is compatible and relevant for with a range of applications, including functional-structural plant modelling, Earth system modelling or radiative transfer modelling of plant canopies. AngleCam may also be used to predict leaf angle distributions from existing data, such as curated in PhenoCam networks or citizen science projects.

Accurate information on the spatial distribution of plant species and communities is in high demand for various fields of application, such as nature conservation, forestry, and agriculture. A series of studies has shown that CNNs accurately predict plant species and communities in high-resolution remote sensing data, in particular with data at the centimeter scale acquired with Unoccupied aerial vehicles (UAV). However, such tasks require ample training data to generate transferable CNN models. Reference data are commonly generated via geocoded in-situ observations or labeling of remote sensing data through visual interpretation. Both approaches are commonly laborious and can present a critical bottleneck for CNN applications. An alternative source of training data is given by using knowledge on the appearance of plants in the form of plant photographs from citizen science projects such as the iNaturalist database. Such crowd-sourced plant photos are expected to be very heterogeneous, and often show a different perspective compared to the typical bird-perspective of remote sensing data. Still, crowd-sourced plant photos could be a valuable source to overcome the challenge of limited training data and reduce the efforts for field data collection and data labeling. Here, we explore the potential of transfer learning from such a crowd-sourced data treasure to the remote sensing context. Therefore, we investigate firstly, if we can use crowd-sourced plant photos for CNN training and subsequent mapping of plant species in high-resolution remote sensing imagery. Secondly, we test if the predictive performance can be increased by a priori selecting photos that share a more similar perspective to the remote sensing data. Therefore, we used three case studies to test our proposed approach using multiple RGB orthoimages acquired from UAV for the target plant species Fallopia japonica (F. japonica), Portulacaria Afra (P. afra), and 10 different tree species, respectively. For training the CNN models, we queried the iNaturalist database for photos of the target species and the surrounding species that are expected in the areas of each case study. We trained CNN models with an EfficientNet-B07 backbone. For applying these models based on the crowd-sourced data to the remote sensing imagery, we used a sliding window approach with a 10 percent overlap. The individual sliding-window-based predictions were spatially aggregated in order to create a high-resolution classification map. Our results demonstrate that CNN models trained with heterogeneous, crowd-sourced plant photos can indeed predict the target species in UAV orthoimages with surprising accuracy. Filtering the crowd-sourced photos used for training by acquisition properties increased the predictive performance. This study demonstrates that citizen science data can effectively anticipate a common bottleneck for vegetation assessments and provides an example on how we can effectively harness the ever-increasing availability of crowd-sourced and big data for remote sensing applications.

Solifluction is the slow downslope movement of soil mass due to freeze-thaw processes. It is widespread on hillslopes in Polar and Alpine regions and contributes substantially to sediment transport. As solifluction lobe movement is in the order of millimeters to centimeters per year, it is tricky to measure with a high spatial and temporal resolution and accuracy. We developed a semi-automated approach to monitor movement of three solifluction lobes with different degrees of vegetation cover along an elevational gradient between 2,170 and 2,567 m in Turtmann Valley, Swiss Alps. Subsequently, we compared movement rates and patterns with environmental factors.For solifluction movement monitoring, we applied a combination of the Phantom 4 Pro Plus and Phantom 4 RTK (Real Time Kinematic) drones, image co-alignment and COSI-CORR (Co-registration of Optically Sensed Images and Correlation) to track movement on orthophotos between 2017 and 2021. This drone data acquisition and co-alignment procedure enable a simple, time-saving field setup without Ground Control Points (GCPs). Our high co-registration accuracy enabled us to detect solifluction movement if it exceeds 5 mm with sparse vegetation cover. Dense vegetation cover limited feature tracking but detected movement rates and patterns still matched previous measurements using classical total station measurements at the lowest, mostly vegetated lobe. In contrast to traditional solifluction monitoring approaches using point measurements, our monitoring approach provides spatially continuous movement estimates across the complete extend of the lobe. Lobe movement rates were highest at the highest elevations between 2,560 and 2,567 m (up to 14.0 cm/yr for single years) and lowest at intermediate elevations between 2,417 and 2,427 m (up to 2.9 cm/yr for single years). We found intermediate movement rates at lowest elevations between 2,170 and 2,185 m (up to 4.9 cm/yr for single years). In general, movement had the highest rates at the solifluction lobes center and the lowest rates at the front of solifluction lobes. We linked observed movement patters to environmental factors possibly controlling solifluction movement, such as geomorphic properties, vegetation species and coverage, soil properties determined from electrical resistivity tomography (ERT), and soil temperature data. The least movement at the lobe front is characterized by coarse material and plant species stabilizing the risers or plant species growing here due to the stable risers. Most movement at the lobe center is characterized by fine material and no vegetation or plant species promoting movement. The soil temperature data further suggests that snow cover reduced freezing rates at solifluction lobes and potentially decreased solifluction movement at the lobe between 2,417 and 2,427 m. This study is the first to demonstrate the use of drone-based images and a semi-automated method to reach high spatiotemporal resolutions to detect subtle movements of solifluction lobes at timescales of years at sub-centimeter resolution. This provides new insights into solifluction movement and into drivers of and factors controlling solifluction movement and lobe development. Therefore, our semi-automated approach may have a great potential to uncover the fundamental processes to understand solifluction movement.