An evaluation of Guided Regularized Random Forest for classification and regression tasks in remote sensing

New Earth observation missions and technologies are delivering large amounts of data. Processing this data requires developing and evaluating novel dimensionality reduction approaches to identify the most informative features for classification and regression tasks. Here we present an exhaustive evaluation of Guided Regularized Random Forest (GRRF), a feature selection method based on Random Forest. GRRF does not require fixing a priori the number of features to be selected or setting a threshold of the feature importance. Moreover, the use of regularization ensures that features selected by GRRF are non-redundant and representative. Our experiments based on various kinds of remote sensing images, show that GRRF selected features provides similar results to those obtained when using all the available features. However, the comparison between GRRF and standard random forest features shows substantial differences: in classification, the mean overall accuracy increases by almost 6% and, in regression, the decrease in RMSE almost reaches 2%. These results demonstrate the potential of GRRF for remote sensing image classification and regression. Especially in the context of increasingly large geodatabases that challenge the application of traditional methods.     

A review on reflective remote sensing and data assimilation techniques for enhanced agroecosystem modeling

During the last 50 years, the management of agroecosystems has been undergoing major changes to meet the growing demand for food, timber, fibre and fuel. As a result of this intensified use, the ecological status of many agroecosystems has been severely deteriorated. Modeling the behavior of agroecosystems is, therefore, of great help since it allows the definition of management strategies that maximize (crop) production while minimizing the environmental impacts. Remote sensing can support such modeling by offering information on the spatial and temporal variation of important canopy state variables which would be very difficult to obtain otherwise.In this paper, we present an overview of different methods that can be used to derive biophysical and biochemical canopy state variables from optical remote sensing data in the VNIR-SWIR regions. The overview is based on an extensive literature review where both statistical–empirical and physically based methods are discussed. Subsequently, the prevailing techniques of assimilating remote sensing data into agroecosystem models are outlined. The increasing complexity of data assimilation methods and of models describing agroecosystem functioning has significantly increased computational demands. For this reason, we include a short section on the potential of parallel processing to deal with the complex and computationally intensive algorithms described in the preceding sections.The studied literature reveals that many valuable techniques have been developed both for the retrieval of canopy state variables from reflective remote sensing data as for assimilating the retrieved variables in agroecosystem models. However, for agroecosystem modeling and remote sensing data assimilation to be commonly employed on a global operational basis, emphasis will have to be put on bridging the mismatch between data availability and accuracy on one hand, and model and user requirements on the other. This could be achieved by integrating imagery with different spatial, temporal, spectral, and angular resolutions, and the fusion of optical data with data of different origin, such as LIDAR and radar/microwave.     

Reducing background effects in orchards through spectral vegetation index correction

Satellite remote sensing provides an alternative to time-consuming and labor intensive in situ measurements of biophysical variables in agricultural crops required for precision agriculture applications. In orchards, however, the spatial resolution causes mixtures of canopies and background (i.e. soil, grass and shadow), hampering the estimation of these biophysical variables. Furthermore, variable background mixtures obstruct meaningful comparisons between different orchard blocks, rows or within each row. Current correction methodologies use spectral differences between canopies and background, but struggle with a vegetated orchard floor. This background influence and the lack of a generic solution are addressed in this study.Firstly, the problem was demonstrated in a controlled environment for vegetation indices sensitive to chlorophyll content, water content and leaf area index. Afterwards, traditional background correction methods (i.e. soil-adjusted vegetation indices and signal unmixing) were compared to the proposed vegetation index correction. This correction was based on the mixing degree of each pixel (i.e. tree cover fraction) to rescale the vegetation indices accordingly and was applied to synthetic and WorldView-2 satellite imagery. Through the correction, the effect of background admixture for vegetation indices was reduced, and the estimation of biophysical variables was improved (ΔR² = 0.2–0.31).     

An interdisciplinary approach to depict landscape change drivers: A case study of the Ticuiz agrarian community in Michoacan, Mexico

Landscape changes are driven by a combination of physical, ecological and socio-cultural factors. Hence, a large amount of information is necessary to monitor these changes and to develop effective strategies for management and conservation. For this, novel strategies for combining social and environmental data need to be developed. The purpose of this study is to demonstrate the value of an innovative interdisciplinary approach to help in explaining landscape change. We integrated three main sources of information: biophysical landscape attributes, land-use/cover change analysis and social perceptions of land-use change, institutional and policy factors and environmental services. Multivariate statistical analysis was used to develop a weight for each variable described or quantified. Finally we identified proximate causes and underlying driving forces of land transformation in the study area. The study was undertaken in a typical community in Mexico.     

The role of biophysical interactions within the ijzermonding tidal flat sediment dynamics

This paper focuses on the importance of biophysical interactions on short-term and long-term sediment dynamics. Therefore, various biological (macrobenthos, photopigments, colloidal EPS) and physical parameters (grain size, water content, sediment stability, bed level) were determined (bi)monthly in nine sampling plots on the IJzermonding tidal flat (Belgium, 51°08′N, 2°44′E) during three consecutive years (July 2005–June 2008). Results showed that sediment stability varied on the short timescale and was directly influenced by biota, while bed level varied mainly on the long-term due to interannual variability. The short-term dynamic relationships between mud content, water content, fucoxanthin and macrobenthos density resulted in a seasonal mud deposition and erosion cycle, and directly influenced sediment stability. Moreover, macrobenthos was proven to be the most important parameter determining sediment stability. On the long-term, a shift was observed from high fucoxanthin/chla concentration, high mud content and zero to moderate densities of Corophium volutator towards low fucoxanthin/chl a and mud content and high Corophium densities, which resulted in a transition from net accretion to net erosion. However, most measured variables proved to be poor predictors for these long-term bed level changes, indicating that external physical forces, such as waves and storminess, probably were the most important factors triggering long-term sediment dynamics. Nevertheless, biota indirectly influenced bed level changes by mediating short-term changes in sediment stability, thereby influencing the erodability of the sediment. The macrobenthos, and especially the mud shrimp Corophium, was suggested as the (indirect) driving destabilising factor for the sampling plots in the IIzermonding when considering the long-term evolution.     

A video-plankton and microstructure profiler for the exploration of in situ connections between zooplankton and turbulence

Turbulence and mixing are ubiquitous in the environment of planktonic organisms and critical in large-scale physical and chemical oceanographic budgets. Recent studies have shown conflicting results on the contribution of zooplankton to ocean mixing as well as the impact of turbulence on zooplankton feeding and swimming behavior. Some of the confusion arises from the lack of properly resolved, simultaneous and co-located observations of zooplankton and turbulence. This paper introduces and discusses results from preliminary deployments of a Video Plankton Recorder–Vertical Microstructure Profiler (VPR–VMP), which is shown to provide good quality, high-resolution, simultaneous observations of zooplankton and turbulence. Non-turbulent shear spectra associated with a zooplankton layer are discussed in terms of automating the rejection of unreliable dissipation estimates, the shear created by individual organisms and profiler avoidance. Comparing these fine-scale observations of zooplankton and turbulence with bulk measures and possible future improvements are also discussed.     

Estimation of vegetation LAI from hyperspectral reflectance data: Effects of soil type and plant architecture

The retrieval of canopy biophysical variables is known to be affected by confounding factors such as plant type and background reflectance. The effects of soil type and plant architecture on the retrieval of vegetation leaf area index (LAI) from hyperspectral data were assessed in this study. In situ measurements of LAI were related to reflectances in the red and near-infrared and also to five widely used spectral vegetation indices (VIs). The study confirmed that the spectral contrast between leaves and soil background determines the strength of the LAI–reflectance relationship. It was shown that within a given vegetation species, the optimum spectral regions for LAI estimation were similar across the investigated VIs, indicating that the various VIs are basically summarizing the same spectral information for a given vegetation species. Cross-validated results revealed that, narrow-band PVI was less influenced by soil background effects (0.15 ≤ RMSE_cv ≤ 0.56). The results suggest that, when using remote sensing VIs for LAI estimation, not only is the choice of VI of importance but also prior knowledge of plant architecture and soil background. Hence, some kind of landscape stratification is required before using hyperspectral imagery for large-scale mapping of vegetation biophysical variables.     

Cluster synchronisation: A mechanism for plankton patchiness and a simulation pitfall

We present a study on the emergence of spatial variability, or patchiness, in biophysical simulations of plankton ecosystems. Using a standard approach to modelling such ecosystems, we represent a distribution of plankton as a lattice of non-identical interacting oscillatory populations. Spatial variation is imposed in population parameters, such as maximum growth rate, leading to a spread in the natural (uncoupled) population properties. Using the methods of synchronisation theory, the emergent spatial structure of the coupled system is investigated as a function of the strength of interaction between populations. Surprisingly, a range of coupling strength is found to induce a tenfold increase in the spread in frequency of oscillation of populations in comparison with the uncoupled level of spatial variation. This apparent desynchronisation corresponds to the formation of temporally evolving clusters of local synchronisation: the interplay of grid-cell scale variability and dispersal between populations leads to patchiness at larger scales. However, the occurrence and length-scale of this patchiness is found to be sensitive to typical simulation parameters such as spatial resolution and strength of dispersal, with emergent spatial structure altering abruptly from patchy to homogeneous as these parameters are varied. These results indicate that whilst cluster synchronisation may be a genuine mechanism for the formation of spatial structure in plankton distributions, biophysical modellers should be aware of the possibility of artificial patchiness arising from the basic physical structure of their model.     

Individual tree crown approach for predicting site index in boreal forests using airborne laser scanning and hyperspectral data

Site productivity is essential information for sustainable forest management and site index (SI) is the most common quantitative measure of it. The SI is usually determined for individual tree species based on tree height and the age of the 100 largest trees per hectare according to stem diameter. The present study aimed to demonstrate and validate a methodology for the determination of SI using remotely sensed data, in particular fused airborne laser scanning (ALS) and airborne hyperspectral data in a forest site in Norway. The applied approach was based on individual tree crown (ITC) delineation: tree species, tree height, diameter at breast height (DBH), and age were modelled and predicted at ITC level using 10-fold cross validation. Four dominant ITCs per 400 m² plot were selected as input to predict SI at plot level for Norway spruce (Picea abies (L.) Karst.) and Scots pine (Pinus sylvestris L.). We applied an experimental setup with different subsets of dominant ITCs with different combinations of attributes (predicted or field-derived) for SI predictions. The results revealed that the selection of the dominant ITCs based on the largest DBH independent of tree species, predicted the SI with similar accuracy as ITCs matched with field-derived dominant trees (RMSE: 27.6% vs 23.3%). The SI accuracies were at the same level when dominant species were determined from the remotely sensed or field data (RMSE: 27.6% vs 27.8%). However, when the predicted tree age was used the SI accuracy decreased compared to field-derived age (RMSE: 27.6% vs 7.6%). In general, SI was overpredicted for both tree species in the mature forest, while there was an underprediction in the young forest. In conclusion, the proposed approach for SI determination based on ITC delineation and a combination of ALS and hyperspectral data is an efficient and stable procedure, which has the potential to predict SI in forest areas at various spatial scales and additionally to improve existing SI maps in Norway.     

Applications of ALOS PALSAR for monitoring biophysical parameters of a degraded black mangrove (Avicennia germinans) forest

Within the last few decades mangrove forests worldwide have been experiencing high annual rates of loss and many of those that remain have undergone considerable degradation. To understand the condition of these forests, various optical remote sensing platforms have been used to map and monitor these wetlands, including the use of these data for biophysical parameter mapping. For many mangrove forests a reliable source of optical imagery is not possible given their location in quasi-permanent cloud cover or smoke covered regions. In such cases it is recommended that Synthetic Aperture Radar (SAR) be considered. The purpose of this investigation was to examine the relationships between various ALOS-PALSAR modes, acquired from eight images, and mangrove biophysical parameter data collected from a black mangrove (Avicennia germinans) dominated forest that has experienced considerable degradation. In total, structural data were collected from 61 plots representing the four common stand types found in this degraded forest of the Mexican Pacific: tall healthy mangrove (n = 17), dwarf healthy mangrove (n = 15), poor condition mangrove (n = 13), and predominantly dead mangrove (n = 16).Based on backscatter coefficients, significant negative correlation coefficients were observed between filtered single polarization ALOS PALSAR (6.25 m) HH backscatter and Leaf Area Index (LAI). When the dead stands were excluded (n = 45) the strength of these relationships increased. Moreover, significant negative correlation coefficients were observed with stand height, Basal Area (BA) and to a lesser degree with stem density and mean DBH. With the coarser spatial resolution dual-polarization and quad polarization data (12.5 m) only a few, and weaker, correlation coefficients were calculated between the mangrove parameters and the filtered HH backscatter. However, significant negative values were once again calculated for the HH when the 16 dead mangrove stands were removed from the sample. Conversely, strong positive significant correlation coefficients were calculated between the cross-polarization HV backscatter and LAI when the dead mangrove stands were considered. Although fewer in comparison to the HH correlations, a number of VV backscatter based relationships with mangrove parameters were observed from the quad polarization mode and, to a lesser extent, with the one single VV polarization data.In addition to backscatter coefficients, stepwise multiple regression models of the mangrove biophysical parameter data were developed based on texture parameters derived from the grey level co-occurrence matrix (GLCM) of the ALOS data. A similar pattern to the backscatter relationships was observed for models based on the single polarization unfiltered data, with fairly strong coefficients of determination calculated for LAI and stem height when the dead stands were excluded. In contrast, similar coefficients of determination with biophysical parameters were observed for the dual and quad polarization multiple regression models when the dead stands were both included and excluded from the analyses. An estimated mangrove LAI map of the study area, derived from a multiple regression model of the quad polarization texture parameters, showed comparable spatial patterns of degradation to a map derived from higher spatial resolution optical satellite data.