Search

Abstract Earth observation data is key for monitoring vegetation dynamics across temporal and spatial scales. The most widely used method to estimate vegetation properties from Earth observation data is vegetation indices. However, temporal dynamics in vertical leaf angles can strongly alter reflectance signals and, hence, vegetation indices. Here, we derive leaf angles from plant photographs to simulate the effect on vegetation indices with radiative transfer models. We show that leaf angle dynamics systematically confound widely applied vegetation indices. Moreover, we demonstrate that these effects are not random but tightly linked to abiotic environmental conditions. These systematic effects of vertical leaf angles have implications for monitoring plant properties, biodiversity, and ecosystem functional properties. We discuss the related challenges and opportunities to assess spatio-temporal vegetation dynamics with remote sensing and vegetation indices.

Plant traits play a pivotal role in steering ecosystem dynamics. As plant canopies have evolved to interact with light, spectral data convey information on a variety of plant traits. Machine learning techniques have been used successfully to retrieve diverse traits from hyperspectral data. Nonetheless, the efficacy of machine learning is restricted by limited access to high-quality reference data for training. Previous studies showed that aggregating data across domains, sensors, or growth forms provided by collaborative efforts of the scientific community enables the creation of transferable models. However, even such curated databases are still sparse for several traits. To address these challenges, we investigated the potential of filling such data gaps with simulated hyperspectral data generated through the most widely-used radiative transfer model (RTM) PROSAIL. We coupled trait information from the TRY plant trait database with information on plant communities from the sPlot database, to build a realistic input trait dataset for the RTM-based simulation of canopy spectra. Our findings indicate that simulated data can alleviate the effects of data scarcity for highly underrepresented traits. In most other cases, however, the effects of including simulated data from RTMs are negligible or even negative. While more complex RTM models promise further improvements, their parameterization remains challenging. This highlights two key observations: firstly, RTM models, such as PROSAIL, exhibit limitations in producing realistic spectra across diverse ecosystems; secondly, real-world data repurposed from various sources exhibit superior retrieval success compared to simulated data. As a result, we advocate to emphasize the importance of active data sharing over secrecy and overreliance on modeling to address data limitations.

Abstract Phenological shifts across plant species is a powerful indicator to quantify the effects of climate change. Today, mobile applications with automated species identification open new possibilities for phenological monitoring across space and time. Here, we introduce an innovative spatio‐temporal machine learning methodology that harnesses such crowd‐sourced data to quantify phenological dynamics across taxa, space and time. Our algorithm links individual phenological responses across thousands of species and geographical locations, using a similarity measure. The analysis draws on nearly ten million plant observations collected through the AI‐based plant identification app Flora Incognita in Germany from 2018 to 2021. Our method quantifies changes in synchronisation across the annual cycle. During the growing season, synchronised behaviour can be encoded by a few characteristic macrophenological patterns. Nonlinear spatio‐temporal changes of these patterns can be efficiently quantified using a data compressibility measure. Outside the growing season, the phenological synchronisation diminishes introducing noise into the patterns. Despite biases and uncertainties associated with crowd‐sourced data, for example due to human data collection behaviour, our study demonstrates the feasibility of deriving meaningful indicators for monitoring plant macrophenology from individual plant observations. As crowd‐sourced databases continue to expand, our approach holds promise to study climate‐induced phenological shifts and feedback loops.

Increasing tree mortality due to climate change has been observed globally. Remote sensing is a suitable means for detecting tree mortality and has been proven effective for the assessment of abrupt and large-scale stand-replacing disturbances, such as those caused by windthrow, clear-cut harvesting, or wildfire. Non-stand replacing tree mortality events (e.g., due to drought) are more difficult to detect with satellite data – especially across regions and forest types. A common limitation for this is the availability of spatially explicit reference data. To address this issue, we propose an automated generation of reference data using uncrewed aerial vehicles (UAV) and deep learning-based pattern recognition. In this study, we used convolutional neural networks (CNN) to semantically segment crowns of standing dead trees from 176 UAV-based very high-resolution (

Unoccupied aerial vehicles (UAV) with RGB-cameras are affordable and versatile devices for the generation of a series of remote sensing products that can be used for forest inventory tasks, such as creating high-resolution orthomosaics and canopy height models. The latter may serve purposes including tree species identification, forest damage assessments, canopy height or timber stock assessments. Besides flight and image acquisition parameters such as image overlap, flight height, and weather conditions, the focal length, which determines the opening angle of the camera lens, is a parameter that influences the reconstruction quality. Despite its importance, the effect of focal length on the quality of 3D reconstructions of forests has received little attention in the literature. Shorter focal lengths result in more accurate distance estimates in the nadir direction since small angular errors lead to large positional errors in narrow opening angles. In this study, 3D reconstructions of four UAV-acquisitions with different focal lengths (21, 35, 50, and 85 mm) on a 1 ha mature mixed forest plot were compared to reference point clouds derived from high quality Terrestrial Laser Scans. Shorter focal lengths (21 and 35 mm) led to a higher agreement with the TLS scans and thus better reconstruction quality, while at 50 mm, quality losses were observed, and at 85 mm, the quality was considerably worse. F1-scores calculated from a voxel representation of the point clouds amounted to 0.254 with 35 mm and 0.201 with 85 mm. The precision with 21 mm focal length was 0.466 and 0.302 with 85 mm. We thus recommend a focal length no longer than 35 mm during UAV Structure from Motion (SfM) data acquisition for forest management practices.

Ecosystem restoration and reforestation often operate at large scales, whereas monitoring practices are usually limited to spatially restricted field measurements that are (i) time- and labour-intensive, and (ii) unable to accurately quantify restoration success over hundreds to thousands of hectares. Recent advances in remote sensing technologies paired with deep learning algorithms provide an unprecedented opportunity for monitoring changes in vegetation cover at spatial and temporal scales. Such data can feed directly into adaptive management practices and provide insights into restoration and regeneration dynamics. Here, we demonstrate that convolutional neural network (CNN) segmentation algorithms can accurately classify the canopy cover of Portulacaria afra Jacq. in imagery acquired using different models of unoccupied aerial vehicles (UAVs) and under variable light intensities. Portulacaria afra is the target species for the restoration of Albany Subtropical Thicket vegetation, endemic to South Africa, where canopy cover is challenging to measure due to the dense, tangled structure of this vegetation. The automated classification strategy presented here is widely transferable to restoration monitoring as its application does not require any knowledge of the CNN model or specialist training, and can be applied to imagery generated by a range of UAV models. This will reduce the sampling effort required to track restoration trajectories in space and time, contributing to more effective management of restoration sites, and promoting collaboration between scientists, practitioners and landowners.

Abstract Plant functional traits (‘traits’) are essential for assessing biodiversity and ecosystem processes, but cumbersome to measure. To facilitate trait measurements, we test if traits can be predicted through visible morphological features by coupling heterogeneous photographs from citizen science (iNaturalist) with trait observations (TRY database) through Convolutional Neural Networks (CNN). Our results show that image features suffice to predict several traits representing the main axes of plant functioning. The accuracy is enhanced when using CNN ensembles and incorporating prior knowledge on trait plasticity and climate. Our results suggest that these models generalise across growth forms, taxa and biomes around the globe. We highlight the applicability of this approach by producing global trait maps that reflect known macroecological patterns. These findings demonstrate the potential of Big Data derived from professional and citizen science in concert with CNN as powerful tools for an efficient and automated assessment of Earth’s plant functional diversity.

Deep learning and particularly Convolutional Neural Networks (CNN) in concert with remote sensing are becoming standard analytical tools in the geosciences. A series of studies has presented the seemingly outstanding performance of CNN for predictive modelling. However, the predictive performance of such models is commonly estimated using random cross-validation, which does not account for spatial autocorrelation between training and validation data. Independent of the analytical method, such spatial dependence will inevitably inflate the estimated model performance. This problem is ignored in most CNN-related studies and suggests a flaw in their validation procedure. Here, we demonstrate how neglecting spatial autocorrelation during cross-validation leads to an optimistic model performance assessment, using the example of a tree species segmentation problem in multiple, spatially distributed drone image acquisitions. We evaluated CNN-based predictions with test data sampled from 1) randomly sampled hold-outs and 2) spatially blocked hold-outs. Assuming that a block cross-validation provides a realistic model performance, a validation with randomly sampled holdouts overestimated the model performance by up to 28%. Smaller training sample size increased this optimism. Spatial autocorrelation among observations was significantly higher within than between different remote sensing acquisitions. Thus, model performance should be tested with spatial cross-validation strategies and multiple independent remote sensing acquisitions. Otherwise, the estimated performance of any geospatial deep learning method is likely to be overestimated.

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 Convolutional Neural Networks (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 often require ample training data, which is commonly generated in the field via geocoded in-situ observations or labeling remote sensing data through visual interpretation. Both approaches are 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 photographs typically exhibit very different perspectives and great heterogeneity in various aspects, yet the sheer volume of data could reveal great potential for application to bird’s eye views from remote sensing platforms. 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 photographs 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 photographs that share a more similar perspective to the remote sensing data. We used two case studies to test our proposed approach with multiple RGB orthoimages acquired from UAV with the target plant species Fallopia japonica and Portulacaria afra respectively. Our results demonstrate that CNN models trained with heterogeneous, crowd-sourced plant photographs can indeed predict the target species in UAV orthoimages with surprising accuracy. Filtering the crowd-sourced photographs 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.

Leaf mass per area (LMA), leaf dry matter content (LDMC) and leaf water content/ equivalent water thickness (EWT) are commonly used functional plant traits in ecology. Whereas spectroscopy has recently proven to be a powerful tool to collect such functional trait information across large scales, it remains unclear whether these reflectance-based trait predictions are accurate enough to reliably model trait variation at the intraspecific level (i.e. across individuals of one species). We explored the potential of hyperspectral leaf reflectance-based methods to predict LMA, LDMC and EWT at the intraspecific level for two herbs (Hieracium umbellatum and Jacobaea vulgaris) and two shrubs (Rosa rugosa and Rubus caesius), based on 2400 leaf samples. More specifically we tested i) inversion of the PROSPECT-D radiative transfer model, ii) a generic PLSR approach using the multibiome LMA PLSR model and iii) a data-specific PLSR approach at the species level. For the latter approach we furthermore assessed both model transferability across species and the trade-off between sample size and model accuracy. Although the PROSPECT-D model inversion and the multibiome LMA PLSR model were relatively accurate for intraspecific LMA predictions of shrubs (R2 > 71 and 76%, respectively, however NRMSE = 33–47%), their performance was lower for herbs (R2 70%, NRMSE

Abstract Plant invasions can result in serious threats for biodiversity and ecosystem functioning. Reliable maps at very‐high spatial resolution are needed to assess invasions dynamics. Field sampling approaches could be replaced by unmanned aerial vehicles (UAVs) to derive such maps. However, pixel‐based species classification at high spatial resolution is highly affected by within‐canopy variation caused by shadows. Here, we studied the effect of shadows on mapping the occurrence of invasive species using UAV‐based data. MaxEnt one‐class classifications were applied to map Acacia dealbata, Ulex europaeus and Pinus radiata in central‐south Chile using combinations of UAV‐based spectral (RGB and hyperspectral), 2D textural and 3D structural variables including and excluding shaded canopy pixels during model calibration. The model accuracies in terms of area under the curve (AUC), Cohen's Kappa, sensitivity (true positive rate) and specificity (true negative rate) were examined in sunlit and shaded canopies separately. Bootstrapping was used for validation and to assess statistical differences between models. Our results show that shadows significantly affect the accuracies obtained with all types of variables. The predictions in shaded areas were generally inaccurate, leading to misclassification rates between 65% and 100% even when shadows were included during model calibration. The exclusion of shaded areas from model calibrations increased the predictive accuracies (especially in terms of sensitivity), decreasing false positives. Spectral and 2D textural information showed generally higher performances and improvements when excluding shadows from the analysis. Shadows significantly affected the model results obtained with any of the variables used, hence the exclusion of shadows is recommended prior to model calibration. This relatively easy pre‐processing step enhances models for classifying species occurrences using high‐resolution spectral imagery and derived products. Finally, a shadow simulation showed differences in the ideal acquisition window for each species, which is important to plan revisit campaigns.

Characterizing the spatial variability of the severity of wildfires is important to assess ecological and economic consequences and to coordinate mitigation strategies. Vegetation indices such as the differenced Normalized Burn Ratio (dNBR) have become a standard tool to assess burn or fire severity across larger areas and are being used operationally. Despite the frequent application of dNBR-like vegetation indices, it is not yet fully understood which variables exactly drive the variability in dNBR observed by multispectral satellites. One reason for this is the lack of high quality prefire information about vegetation structure and composition. Consequently, the influence of prefire vegetation composition and other potentially influential variables such as cast shadows has hardly been examined. Here, we use very high resolution Unmanned Aerial System (UAS) orthoimages collected briefly before and after the large wildfires in Central Chile in the fire season 2016/2017 to derive variables related to the pre- and postfire landscape composition and structure. The variables are used as predictors in Generalized Additive Models (GAM) to explain the spatial variability in dNBR and RdNBR pixel values as observed by Sentinel-2. Our models explain more than 80% and 75% of the variability in dNBR and RdNBR values, respectively, using a sparse set of five predictors. The results suggest that in our study area the largest fraction of variability in Sentinel-2 based dNBR and RdNBR values can be explained by variables related to the fraction of consumed canopy cover while the vegetation composition before the fire does not have a large influence on dNBR and RdNBR.Our results further suggest that cast-shadows of snags and standing dead trees with remaining crown structure have a notable influence on the dNBR signal which may have been underestimated so far. We conclude that spatially continuous, very high spatial resolution data from UAS can be a valuable data source for an improved understanding of the exact meaning of common vegetation index products, operationally used for monitoring the environment.

Plant functional traits play a key role in the assessment of ecosystem processes and properties. Optical remote sensing is ascribed a high potential in capturing those traits and their spatiotemporal patterns. In vegetation remote sensing, reflectance-based retrieval methods are either statistical (relying on empirical observations) or physically-based (based on inversions of a radiative transfer model, RTM). Both trait retrieval approaches remain poorly investigated regarding phenology. However, within the phenology of a plant, its leaf constituents, canopy structure, and the presence of phenology-related organs (i.e., flowers or inflorescence) vary considerably – and so does its reflectance. We, therefore, addressed the question of how plant phenology affects the predictive performance of both statistical and RTM-based methods and how this effect differs between traits. For a complete growing season, we weekly measured traits of 45 herbaceous plant species together with hyperspectral canopy reflectance (ASD FieldSpec III). Plants were grown in an experimental setup. The investigated traits comprised Leaf Area Index (LAI) and the leaf traits chlorophyll, anthocyanins, carotenoids, equivalent water thickness, and leaf mass per area. We compared the predictive performances of PLSR models and three variants of PROSAIL inversions based on (1) all observations and based on (2) a phenological subset where flowering plants were excluded and only those observations most suitable for modeling were kept. Our results show that both statistical and RTM-based trait retrievals were largely affected by phenology. For carotenoids for example, R2 decreased from 0.58 at non-flowering canopies to 0.25 at 100% flowering canopies. Temporal trends were diverse. LAI and equivalent water thickness were best estimated earlier in the growing season; chlorophyll and carotenoids towards senescence. PLSR models showed generally higher bias than the PROSAIL-based retrieval approaches. Lookup-table inversion of PROSAIL in combination with a continuous wavelet transformation of reflectance showed highest accuracies. We found RTM-based retrieval not to be as accurate and transferable as previously indicated. Our results suggest that phenology is essential for accurate retrieval of plant functional traits and varies depending on the studied species and functional traits, respectively.