Purple moor grass is a major invasive grass species in heathlands across Western Europe. Due to its ability to quickly accumulate flammable biomass, purple moor grass could influence the fire risk in these open landscapes. In this study, we have demonstrated the potential of monthly NDVI and NDMI time series derived from the freely available Sentinel-2 data catalogue for distinguishing between areas with low and areas with high coverage of purple moor grass. While the results within a single year are excellent, temporal transferability remains a challenge. This could potentially be overcome by the use of phenological and/or bitemporal input features.

In recent years, machine learning has become crucial in remote sensing analysis, particularly in the domain of Land-use/Land-cover (LULC). The synergy of machine learning and satellite imagery analysis has demonstrated significant productivity in this field, as evidenced by several studies [1], [2], [3], [4]. A notable challenge within this area is the semantic segmentation mapping of land usage over extensive territories, where the accessibility of accurate land-use data and the reliability of ground truth land-use labels pose signifi-cant difficulties. For example, providing a detailed and accurate pixel-wise labeled dataset of the Flanders region, a first-level administrative division of Belgium, can be particularly insightful. Yet there is a notable lack of regulated, formalized datasets and workflows for such studies in many regions globally. This paper introduces a comprehensive approach to addressing these gaps. We present a densely labeled ground truth map of Flanders paired with Sentinel-2 satellite imagery. Our methodology includes a formalized dataset division and sampling method, utilizing the topographic map layout ’Kaartbladversnijdingen,’ and a detailed semantic segmentation model training pipeline. Preliminary benchmarking results are also provided to demonstrate the efficacy of our approach.

We evaluated a methodology, wherein airborne light detection and ranging (LiDAR) data with a point density of 15/m2and hyperspectral (HS) data with a spatial resolution of 2 m were used to identify nine dominant broad-leaved tree genera in the region of Brussels, Belgium. The classification was performed using a random forest classifier with the LiDAR-derived structural information and HS-derived spectral information at the individual tree crown scale. The highest overall accuracy (OA) and Kappa coefficient (Kappa) was achieved by using both HS and LiDAR features (OA = 70.2 ± 1.2%, Kappa = 0.64 ± 0.01) and HS features (OA = 65.7 ± 1.4%, Kappa = 0.58 ± 0.02) outperformed LiDAR features (OA = 61.1 ± 1.4%, Kappa = 0.53 ± 0.02) in classifying the nine tree genus types.

Data-driven environmental governance is gaining importance for tackling the current biodiversity and climate crises. Remote sensing can provide an efficient alternative to expensive and time-intensive in-situ monitoring. Deep learning is the current state-of-the art for knowledge extraction from remote sensing data. However, its practical, operational application for policy support remains limited. In this paper, we coupled the producer and user perspective to unravel the reasons behind this. We identified three main keys to success for an operational implementation of models on the interface between remote sensing and deep learning (technology) and environmental governance (policy): (i) a truly operational mindset, (ii) the use of generic model (components) and (iii) the availability of reference data. We argue that the road to success is paved with effective communication, a well-substantiated (prototype) use case selection and an optimal use of the scarce resource that is (the collection of) labeled reference data.

In the past five years, an international team has been working towards improved detection, characterization and modeling of abrupt changes in dryland ecosystem functioning, EF. This paper collects the recent advances and lessons learnt from the U-TURN project (Belspo SR/00/339, SR/00/366). Specifically abrupt changes in EF were mapped and categorized over global drylands; new 30m resolution time series of land cover maps and cover fractions were created, validated and released open access for the Sahel region; and new physically-based insights into dryland vegetation response to extreme rainfall were derived based on dryland optimized LPJ-GUESS simulations.

Optical remote sensing images of cities exhibit strong spectral variability, and characterising it remains a key challenge. Imaging spectroscopy is useful for this purpose, yet traditional endmember extraction algorithms are poorly suited for this type of imagery. Important issues include the non-spatiality, randomness and poor computational efficiency of existing methods, as well as the need for predefining the number of endmembers. We propose a novel algorithm, called Iterative Spectral Distancing addressing each of these issues. We show that ISD outperforms three established methods on a synthetic image, indicating its potential.

Despite the relevance of historical land cover maps for scientists and policy makers, an accurate high resolution record is currently lacking over the Sudano-Sahel. In this study, 30m resolution historically consistent land cover and cover fraction maps are provided over the Sudano-Sahel for the period 1986–2015. These land cover/cover fraction maps are achieved based on the Landsat archive preprocessed on Google Earth Engine and a random forest classification/regression model, while historical consistency is achieved using the hidden Markov model. Using these historical maps, a multitude of variability in the dynamic Sudano-Sahel region over the past 30 years is revealed. These include cropland expansion and the re-greening of the Sahel, forest degradation & the detection of fine-scale changes, such as smallholder or subsistence farming. The historical land cover / cover fraction maps are made available via an open-access platform.

This study collected training samples of urban green space (UGS) area versus reflectance spectra from 52 city areas in Europe and trained a subpixel SVR model to quantitatively evaluate the level of urban green space development in different European countries on the Google Earth Engine platform as of 2015. The results show that RMSE values of the applied SVRmodel reach 0.09 to 0.13. Moldova, Norway, Belgium, and Ukraine are in the top tier of average urban green space density in Europe, while Greece, Monaco, Spain, and Albania are in the bottom quartile. Besides, the majority UGS resources concentrate in Western European countries.

Grasslands are of great ecological importance in Western Europe, mainly because they harbor a wide variety of (endangered) fauna and flora. However, not all grasslands are equally valuable. All grasslands in Flanders can be assigned to one of six (restoration phases' of increasing ecological value. In this paper, we assessed how well these different phases can be distinguished from an intra-seasonal time series of Sentinel 2 satellite images. We found that the phases at the low and high extremes are very distinctive in the spectro-temporal domain of our Sentinel 2 acquisitions. The intermediate phases however are more similar spectrally, making them more difficult to discern. Most phases can however be separated at minimum one acquisition date. This is an encouraging result as we aim at developing a monitoring system for grassland restoration at the parcel scale.

One of the major issues with spectral mixture analysis remains the lack of ability to properly account for the spectral variability of endmembers (EMs). EM variability is most often addressed using large spectral libraries incorporating the variability present in the image. We propose a new geometric-based methodology to efficiently evaluate different binary EM combinations. Our approach selects the best EM combination prior to unmixing, building upon the equivalence between the reconstruction error in least squares unmixing and spectral angle minimization in geometric unmixing. This geometric approach is tested on both a simulated data set based on field measurements and a HyMap image. It is demonstrated that selecting the best EM combination for a pixel based on the angle minimization provided identical results compared with using the projection distance or reconstruction error. It also has the additional benefit of reducing the computation time due to the simplicity of the angle calculations.

Changing climate conditions are expected to affect vegetation health globally, further emphasizing the importance to assess vegetation response to climate anomalies on a large scale and to understand its driving factors. Within this context, vegetation resilience and resistance against drought and temperature anomalies were assessed using an autoregressive model based on GIMMS NDVI and climate time series over Europe. The extracted vegetation stability metrics were subsequently related to Corine Land Cover classes. The model (i) allowed for exclusion of spurious results and (ii) provided vegetation resistance and resilience metrics normalized for climate variability. These ecosystem stability metrics further showed a clear correspondence with vegetation types in Europe.

We evaluated the potential of a multi-temporal MESMA for invasive species mapping in Hawaiian rainforests. EO-1 Hyperion data were compiled in a single image cube and ingested into MESMA. The temporal analysis provided a way to incorporate species phenology meanwhile a feature selection technique automatically identified the best time and best feature set to optimize the separability among the target tree species. Our analysis showed a systematic increase in the invasive species detection success when compared to the output of the traditional unitemporal approach (Kappa = 0.69 vs 0.78).

Spectral unmixing aims at finding the spectrally pure constituent materials (also called endmembers) and their respective fractional abundances in each pixel of a hyperspectral image scene. In recent years, sparse unmixing has been widely used as a reliable spectral unmixing methodology. In this approach, the observed spectral vectors are expressed as linear combinations of spectral signatures assumed to be known a priori and presented in a large collection, termed spectral library or dictionary, usually acquired in laboratory. Sparse unmixing has attracted much attention as it sidesteps two common limitations of classic spectral unmixing approaches, namely, the lack of pure pixels in hyperspectral scenes and the need to estimate the number of endmembers in a given scene, which are very difficult tasks. However, the high mutual coherence of spectral libraries, jointly with their ever-growing dimensionality, strongly limits the operational applicability of sparse unmixing. In this paper, we introduce a two-step algorithm aimed at mitigating the aforementioned limitations. The algorithm exploits the usual low dimensionality of the hyperspectral data sets. The first step, which is similar to the multiple signal classification array signal processing algorithm, identifies a subset of the library elements, which contains the endmember signatures. Because this subset has cardinality much smaller than the initial number of library elements, the sparse regression we are led to is much more well conditioned than the initial one using the complete library. The second step applies collaborative sparse regression, which is a form of structured sparse regression, exploiting the fact that only a few spectral signatures in the library are active. The effectiveness of the proposed approach, termed MUSIC-CSR, is extensively validated using both simulated and real hyperspectral data sets.

Recently, signal unmixing was proposed in remote sensing with the goal to infer physical parameters of materials of interest, such as vegetation, on the ground. The typical approach uses large collections of pure spectra, called spectral libraries, in which many possible states of the vegetation are modeled by simulated or on-site acquired spectra. Spectra randomly selected from these libraries are used as input to dedicated unmixing methods, such as Multiple Endmember Spectral Mixture Analysis (MESMA). The spectra leading to the lowest reconstruction error are considered to be representative for the materials present in the pixel, such that the physical parameters of the ground vegetation can be inferred. However, the large number of spectra in the library imposes limits to the performance of this combinatorial approach, mainly related to running time constraints. In this paper, we propose the inclusion of a pre-processing step in the processing chain, based on the group lasso, which has the goal of selecting groups of signatures likely to be present in the mixtures. In this sense, the Group Sparse Unmixing via variable Splitting and Augmented Lagrangian (GSUnSAL) algorithm is used. The signatures contained in the groups selected by GSUnSAL are then used as input for MESMA. Our experiments using a real dataset acquired by an ASD spectrometer in a South-African orchard show that the proposed approach introduces important improvements in the signal unmixing solutions.

Mixed satellite signals are traditionally modeled as linear combinations of the spectral signatures of its constituent components. Although nonlinearity has been shown to be significant for a variety of vegetation types, it is assumed to be negligible for most applications. We aim to assess the validity of the linear modeling assumption by making a quantitative analysis of the nature of multiple scattering effects in mixed forests. The effects of the spectral properties of the different species, structural differences and differences in tree height are evaluated. Virtual forest scenes and simulated hyperspectral satellite data were created through ray-tracing modeling using the Physically Based Ray-Tracer (PBRT) model. Results showed that both structure and the spectral properties influenced the nonlinear mixing behaviour, indicating that nonlinear unmixing models might be needed for forest cover mapping in heterogeneous forests.

Due to the subpixel contribution of background soils and shadows, hyperspectral image interpretation in agricultural management is often constrained. In this paper, the potential of multiple endmember spectral mixture analysis (MESMA) to simultaneously extract the subpixel cover fraction and pure spectral signature of the crop component from a mixed hyperspectral signal is evaluated. Radiative transfer models are used to build lookup tables (LUTs) for both the crop and the soil component, but the extensiveness of the LUTs will decrease the efficiency and operational implementation of MESMA. A clustering procedure is therefore presented, allowing a more efficient use of the LUTs in the MESMA model. The performance of MESMA, using clustered and nonclustered LUTs, to extract the cover fraction and the spectral signature of plant canopies was evaluated using 200 simulated mixtures generated from in situ measured hyperspectral data of soil and citrus canopies. Clustering of the LUT resulted in a more efficient and accurate estimation of the pure subpixel vegetation signal ( rmse = 0.097 stabilizing at 40 iterations) compared to a nonclustered LUT (rmse = 0.11 stabilizing at 200 iterations). The subpixel cover fraction estimations, on the other hand, stabilize for both methods around 100 iterations, with an rmse of 0.15 for both approaches. The clustering of the LUT will thus increase both the efficiency and the accuracy of MESMA for estimating the spectral signature of crops while, on average, maintaining the accuracy for the cover fraction estimates. This will enable a more accurate extraction of plant production parameters, which opens up new opportunities regarding precision farming.

In the past decades, spectral unmixing has been studied for deriving the fractions of spectrally pure materials in a mixed pixel. However, limited attention has been given to the collinearity problem in spectral mixture analysis. In this paper, quantitative analysis and detailed simulations are provided, which show that the high correlation between the endmembers, including the virtual endmembers introduced in a nonlinear model, has a strong impact on unmixing errors through inflating the Gaussian noise. While distinctive spectra with low correlations are often selected as true endmembers, the virtual endmembers formed by their product terms can be highly correlated. It is found that a virtual-endmember-based nonlinear model generally suffers more from collinearity problems compared to linear models and may not perform as expected when the Gaussian noise is high, despite its higher modeling power. Experiments were conducted on a set of in situ measured data, and the results show that the linear mixture model performs better in 61.5% of the cases.

A novel conceptual approach to address the mixture problem in agricultural remote sensing is presented in this study. The method, referred to as Signal Unmixing, combines both in situ and hyperspectral data in an adapted spectral mixture analysis algorithm and allows the extraction of pure and complete hyperspectral vegetation spectra (400–2400 nm) from mixed image pixels. The technique is evaluated using images generated from ray tracing simulations of a fully calibrated virtual orchard. Results show a proper extraction of pure vegetation spectra with a relative root mean square error < 0.075. As such, the undesired background effects in vegetation index calculations are reduced to a minimum, which in turn improved the monitoring of canopy water status and LAI.

BRDF effects present in dual field-of-view spectroscopy datasets were investigated. A data-driven normalization procedure was developed by decomposing the target BRDF into a target specific Lambertian component and a bi-directional component characterizing a group of similar targets,. The normalization method was used to convert reflectance factors obtained under cloud obscured conditions into clear sky conditions. An evaluation on four targets measured under different illumination conditions suggests that the normalization can reduce relative reflectance errors between 400 and 1800 nm from 15% to less than 5% even under full cloud obscuration. At higher wavelengths a decreased signal-to-noise ratio increases the error level.

Traditionally, spectral mixture analysis (SMA) fails to fully account for highly similar ground components or endmembers. The high similarity between weed and crop spectra hampers the implementation of SMA for steering weed control management practices. To address this problem, this paper presents an alternative SMA technique, referred to as Integrated Spectral Unmixing (InSU). InSU combines both magnitude (i.e., reflectance) and shape (i.e., derivative reflectance) related features in an automated waveband selection protocol. Analysis was performed on different simulated mixed pixel spectra sets compiled fromin situ-measured weed canopy,Citruscanopy, and soil spectra. Compared to traditional linear SMA, InSU significantly improved weed cover fraction estimations. An average decrease in fraction abundance error (Deltaf) of 0.09 was demonstrated for a signal-to-noise ratio (SNR) of 500 : 1, while for a SNR of 50 : 1, the decrease was 0.06.

Traditionally, Spectral Mixture Analysis (SMA) fails to fully account for highly similar ground components or endmembers. The high similarity between weed and crop spectra therefore hampers the implementation of SMA for steering weed control management practices. To address this problem, the current study presents an alternative SMA technique, referred to as Integrated Spectral Mixture Analysis (iSMA). iSMA combines both magnitude (~ reflectance) and shape (~ derivatives) related features in an automated waveband selection protocol and allows for an optimal separation between weed and crops, irrespective of the scenario considered. Compared to traditional approaches iSMA significantly improved weed cover fraction estimations (~ 17% increase). Analysis was performed on different simulated mixed pixel spectra sets compiled from in situ measured weed, Citrus canopy and soil spectra.