This research investigates the spatiotemporal landscape characteristics within the Mediterranean Basin in the context of climate change in the South-Eastern Mediterranean. For this purpose, we used NDVI data from 21 sites spanning the Mediterranean Basin’s climatic gradient during a 35-year period, from 1986 to 2021. These sites were chosen to represent diverse desert-fringe ecosystems. The impact of climate change on long-term primary productivity and woody growth is examined by looking at NDVI time-series. We suggest that despite fluctuations in the vegetation patterns’ climatic conditions including draughts, no definitive northward migration of aridity could be identified.

Sarcopoterium Spinosum (SS) patterns cover wide Mediterranean landscapes. Following Global warming, SS is suspected to further spread into areas of disturbed forest and shrubland ecosystems and abandoned rain-fed agriculture areas. Despite their current extents and their potential expansion, mapping SS patterns and monitoring their spatio-temporal change received limited attention in the environmental remote sensing literature. Part of the explanation concern difficulties in detecting these plants during the winter due to their spectral similarity to other green plants and their spectral resemblance to bare soil and other dry plants in their vicinity during the summer. This paper reports on hyperspectral investigation of the spectral differences between SS plants and soil, rocks, herbaceous plants and an evergreen shrub Rhamnus lycioides which is also present in desert fringe environments. The results indicate the possibility of differentiating spectrally between them along and perpendicular to the soil line in RED and NIR reflectance combinations.

Shrubs are dominant life forms in arid and semi-arid climatic regions. Shrubs are carbon dioxide sinks; they prevent soil erosion and contribute to the water balance. However, there is limited information regarding the structure and biomass of these shrubs and their change across climatic gradients, partly due to limitations on the use of destructive methods. This work concentrates on the study of the shrubs` shape, and their volume which allows estimates of biomass. A field survey of 28 shrubs and dwarf shrubs was conducted at three sites ranging from sub-humid to arid climate areas in central Israel. Photogrammetric point clouds were generated from smart phone pictures for each shrub using PIX4D software. A comparison of the estimated volume of point clouds and the volume of 3 models: semi-ellipsoid, spherical cap, and cylinder was conducted. Using fine alignment between shrubs’ point clouds and the three model shapes, it was found that semi-ellipsoid is more suitable to describe the shape of most of the shrubs photographed.

Soil moisture plays an important role in agriculture, hydrology, plants’ seasonal growth cycle, and their sustainability under climate change. This letter presents a novel method for regional mapping of soil moisture with partial vegetation cover by using dual-polarized Sentinel-1 C-band SAR data. Model-based retrieval of soil moisture from radar backscatter under partial vegetation cover first necessitates the separation of the backscatter of the underlying soil from total radar backscatter and then the implementation of an inversion technique to estimate its soil moisture content from the backscattering coefficient of the underlying soil. The regional mapping of soil moisture was verified with in situ measurements along a climatic gradient (650–100 mm/year) in central Israel. A good agreement between the estimated soil moisture and the measured from ground samples was found, with an$R^{2}$of 0.793 with an RMSE of 0.047.

Deep neural networks (NNs) trained on hyperspectral images are employed typically for the classification of new images collected from the same sensor, assuming similar characteristics to those of the training images. Creating, however, high-quality ground truth (GT) for training is rather complex, especially when attempting to classify multi-temporal images over seasonal changes. To overcome this difficulty, we propose a novel method that utilizes an additional, one-time collection of hyperspectral FENIX images in the Spring along with ground observations from the end of the Fall. The hyperspectral data are then used for simulation of GT for training. At the same time, the field campaign allows for fine-tuning of the NN to achieve enhanced, multi-seasonal hyperspectral image classification. Indeed, we demonstrate how the proposed method successfully classifies new VEN μS images obtained during different seasons.

Training a deep neural network for classification constitutes a major problem in remote sensing due to the lack of adequate field data. Acquiring high-resolution ground truth (GT) by human interpretation is both cost-ineffective and inconsistent. We propose, instead, to utilize high-resolution, hyperspectral images for solving this problem, by unmixing these images to obtain reliable GT for training a deep network. Specifically, we simulate GT from high-resolution, hyperspectral FENIX images, and use it for training a convolutional neural network (CNN) for pixel-based classification. We show how the model can be transferred successfully to classify new mid-resolution VENμS imagery.

Biomass estimation of eastern Mediterranean shrublands was investigated using PALSAR full- and dual-polarization L-band and Sentinel-1 dual-polarization C-band data. First, we conducted an empirical assessment of single and multiple regressions between polarized backscattering coefficients and shrubland biomass distribution along the climatic gradient between semiarid and arid regions. We then found that the PALSAR L-band HV-polarized backscattering coefficient has higher biomass information content than Sentinel-1 C-band data. Based on a theoretical volume scattering model and a semiempirical model, we propose a new polarimetric radar vegetation index (PRVI) that utilizes the degree of polarization and the cross-polarized backscattering coefficient. The relationship between the new index and the biomass was assessed with reference to normalized difference vegetation index-based biomass estimates calculated using Landsat imagery. The PRVI was found to have higher correlation with biomass compared with other radar polarization parameters, in general, and an existing radar vegetation index (RVI), in particular. Assessment of PRVI-based biomass predictions compared with allometric data extracted from air photographs, Lidar, and field data for 67 sites across the desert fringe zone indicated moderate performance with an RMSE of 0.329 kg/m2, while an RVI-based biomass estimation had an RMSE of 0.439 kg/m2.

We present, in this paper, a new methodology for spectral unmixing, where a vector of fractions, corresponding to a set of endmembers (EMs), is estimated for each pixel in the image. The process first provides an initial estimate of the fraction vector, followed by an iterative procedure that converges to an optimal solution. Specifically, projected gradient descent (PGD) optimization is applied to (a variant of) the spectral angle mapper objective function, so as to significantly reduce the estimation error due to amplitude (i.e., magnitude) variations in EM spectra, caused by the illumination change effect. To improve the computational efficiency of our method over a commonly used gradient descent technique, we have analytically derived the objective function's gradient and the optimal step size (used in each iteration). To gain further improvement, we have implemented our unmixing module via code vectorization, where the entire process is “folded” into a single loop, and the fractions for all of the pixels are solved simultaneously. We call this new parallel scheme vectorized code PGD unmixing (VPGDU). VPGDU has the advantage of solving (simultaneously) an independent optimization problem per image pixel, exactly as other pixelwise algorithms, but significantly faster. Its performance was compared with the commonly used fully constrained least squares unmixing (FCLSU), the generalized bilinear model (GBM) method for hyperspectral unmixng, and the fast state-of-the-art methods, sparse unmixing by variable splitting and augmented Lagrangian (SUnSAL) and collaborative SUnSAL (CLSUnSAL) based on the alternating direction method of multipliers. Considering all of the prospective EMs of a scene at each pixel (i.e., without a priori knowledge which/how many EMs are actually present in a given pixel), we demonstrate that the accuracy due to VPGDU is considerably higher than that obtained by FCLSU, GBM, SUnSAL, and CLSUnSAL under varying illumination, and is, otherwise, comparable with respect to these methods. However, while our method is significantly faster than FCLSU and GBM, it is slower than SUnSAL and CLSUnSAL by roughly an order of magnitude.

Many vegetation indices have been developed for the estimation of green biomass over the last three decades. The Normalized Vegetation Index is the most well-known index; however, it has a saturation problem at moderate to high vegetation densities. The red-edge region (700-740nm) has been introduced to increase sensitivity at these moderate to high vegetation densities. We propose a new vegetation index for biomass estimation of short vegetation, to improve the saturation problem using the red-edge bands. By using the Hyper-spectral image data of Maize and Soybean, the nine well-known vegetation indices are evaluated and compared with the proposed index. For validation of the proposed model using Sentinel-2 data, Pereira allometric data is used (r-square is 0.765).

Potential for synergetic use of Sentinel- 1 and Sentinel-2 for mapping biomass of Mediterranean shrublands is investigated. As preliminary research, backscatter and its ratio from Sentinel-1 (C-band dual polarization SAR), and NDVI from Sentinel-2 (13 bands multi-spectral data) are assessed by using the NDVIR biomass model. Then the fusion biomass model is proposed based on shrub volume formations. The fusion model is verified by filed survey data which measured shrub height and diameter applied into the allometric model. The proposed fusion model shows around 14 % improvement of accuracy compared to the single sensor model (r-square: from 0.72 to 0.86, RMSE: from 0.158 to 0.109).

Performing standard unmixing of a hyperspectral image, while taking into account all of the potential endmembers (EMs) in a pixel, is known to be prone to error. Instead, determining first the set of EMs that actually reside in each pixel, leads to enhanced unmixing results. This important insight for achieving higher unmixing accuracy can be exploited efficiently by extracting relevant spatial information from a given image. In this work, we present a new method for spatially adaptive spectral unmixing, called the Gaussian based spatially adaptive unmixing (GBSAU) method. GBSAU takes advantage of the spatial arrangement of the image pixels and their spectral relations in order to determine an actual subset of EMs per pixel. It is based on spatial localization of the EMs by fitting, for each EM, the parameters of the series of spatial Gaussians whose sum represents the EM's fraction surface over the image.

A novel unmixing methodology is presented, searching for a fraction combination of end-members (EMs) that reconstructs the integrated source signal. The search starts with computing an initially estimated unmixing solution and then assesses combinations selected at random within an envelope surrounding this estimated solution. From each of these combinations, it then progresses iteratively along a path of neighboring combinations, so as to minimize the spectral angle between the corresponding (integrated) signatures and the source signal, until reaching a satisfactory solution. The new iterative fraction combination search (IFCS) was compared to the standard least squares unmixing (LSU). An assessment of both methods was conducted with a real Airborne Visible/Infrared Imaging Spectrometer image and nine synthetic images generated by randomly selecting fractions for two up to ten EMs derived from this real image. Considering all these EMs for the unmixing solution (not knowing specifically which or how many of them are actually mixed at each pixel), the IFCS method performed considerably better than LSU.

The inaccurate estimation of the incoming solar flux density, owing to the lack of adequate representation of the surface orientation, causes the high variability of the spectral reflectance of a given surface material and, consequently, confusion between different materials. A new generic solution is presented in this letter, based on spectral parameterization techniques derived from wavelet analysis. Significant classification improvements were obtained for both synthetic data and the data acquired by hyperspectral imaging of Mediterranean vegetation.

Information extraction from hyperspectral imagery is highly affected by difficulties in accounting for flux density variation and bidirectional reflectance effects. Calculation of flux density requires digital description of the surface structure at the pixel level, which is frequently not available at the accuracy required (if exists). The result of these shortcomings in achieving accurate radiometric image calibration is reduced separability of surface types: limiting the performance of spectral classification schemes. In this study an alternative approach is presented: application of features of the spectral signature which mainly represent the shape of the spectral curve. This is achieved by applying features calculated based on Wavelet decomposition.

Recent developments in sensing and tracking technologies have enabled large geographical databases to be established that represent spatial dynamics of ‘behavioral entities’. Within this type of dynamics there are several levels and modes of organization that need to be revealed. Clusters are high‐level groupings of entities, where change in their location and form, including split and merge events, represents self‐organization and functioning patterns. Such information may contribute for better understanding spatially complex dynamic patterns. The main objective of this article is to develop an adaptable methodology that facilitates exploration of spatial order and processes in point pattern dynamics. The approach presented here utilizes data‐clustering at each snapshot of the moving pattern, and then involves pairwise linking between the clusters identified at each snapshot and those identified in the following snapshot. Such linking is based on a new methodology that defines well globally optimized solutions for numerous possible linking combinations based on Linear Programming. A preliminary assessment of the approach was conducted with an existing Ants' simulation tool, capable of creating data sets covering in detail a substantial portion of the nest's life cycle.

In this paper, we investigate how to further exploit the various characteristics of mean shift, in an attempt to achieve a robust and efficient clustering module for remotely sensed data. A mean shift algorithm has shown o be promising in various image-processing applications, specifically in cluster analysis.

A remote sensing methodology for herbaceous areal above-ground biomass (AAB) estimation in a heterogeneous Mediterranean environment is presented. The methodology is based on an adaptation of the semiempirical water-cloud backscatter model to complex vegetation canopies combined with shrubs, dwarf shrubs, and herbaceous plants. The model included usage of the green leaf biomass volumetric density as a canopy descriptor and of cover fractions derived from unmixing Landsat Thematic Mapper image data for the three vegetation formations. The inclusion of the unmixed cover fractions improves modeling synthetic aperture radar backscatter, as it allows separation between the different radiation interaction mechanisms. The method was first assessed with reference to the reproduction of the backscatter from the vegetation formations. In the next phase, the accuracy of AAB retrievals from the backscatter data was evaluated. Results of testing the methodology in a region of climatic gradient in central Israel have shown a good correspondence between observed and predicted AAB values (R/sup 2/=0.82). This indicates that the methodology developed may lay a basis for mapping important and more advanced ecological information such as primary production and contribute to better understanding of processes in Mediterranean and semiarid regions.

An artificial neural network (ANN) for change detection from multi-temporal satellite images, which was reported in I. Feldberg (2001), has been further developed and tested, as part of a study of an area of high spatio-temporal heterogeneity along a climatic gradient between humid and and climate regions. Four recognition classes, "positive change", "negative change", "false change", and "no change" were learned by a backpropagation feedforward ANN and then applied to Landsat images that were acquired over the study area in 1992 and 1997. A comparison with existing classification techniques indicates, in many instances, significantly improved performance due to the ANN developed.

A new technique utilizing combination of feature extraction by change vector analysis and analysis of distances between features allows improvement in change detection of linear features such as roads and water channels. The technique reduces false detection of changes due to image calibration differences, illumination differences and misregistration. The method was applied to areas of steep climatic gradient between Mediterranean and extreme desert regions.