Continuous and operational monitoring of forest canopy structure plays an important role in assessing the global carbon budget, mapping forest disturbance, planning restoration activities, and informing decision-making. Several studies have taken advantage of synthetic aperture radar (SAR) for forest mapping and monitoring because of its regular reliable acquisitions and high sensitivity to the structural and dielectric properties of the forest. This work utilizes the Senitnel-1 C- and ALOS-2 PALSAR-2 L-band interferometric coherence for canopy height estimation in savanna woodlands. A simplified physics-based Random Volume over Ground (RVoG) model is used for the height estimation. This study uses datasets collected over two test sites, one in Injune, Australia, and the second in Kruger National Park (KNP), South Africa. The proposed method achieved an overall RMSE of 2.39m for canopy height with a Pearson coefficient, r = 0.83 by simultaneous use of both C- and L-band coherence.

The proliferation of machine learning models, architectures, and datasets for Earth observation (EO) continues to rise dramatically. This pattern is expected to continue growing bringing with it an increase in the generation of remote sensing derived data products powered by geospatial artificial intelligence (GeoAI) techniques. Rigorous quality assessments and accuracy analysis needs to be undertaken for the science community to adopt many of these data products for scientific discovery of changes of the Earth’s land surface. While there is existing literature supporting and documenting best practices for the validation of GeoAI data products, the software to support large-scale collaborative validation efforts is limited. In this study we present the design and software implementation of a flexible commodity-based framework for large-scale global to regional validation of GeoAI data products. This framework’s main purpose is to enable, speed up, and optimize the acquisition of validation data for large-scale science projects with support across multiple sensors and spatial resolutions with little to no code. In addition, we present several use cases where this framework has enabled and streamlined the validation of global to regional data products at different spatial resolutions and within different computational platforms.

Human-induced land cover land use (LCLU) changes such as agricultural extensification and forest degradation and loss have extensive negative impacts including biodiversity loss, land degradation, and a disruption to ecological services. In Senegal, where people are heavily reliant on dryland agricultural production, climate change and land degradation pose particularly significant threats especially as rapid population growth continues to fuel frequent LCLU change. Considering these challenges, approaches that facilitate increased insight into the spatial and temporal dynamics of land use are needed to implement sustainable land management practices and mitigation strategies. However, difficulties associated with Senegal’s highly variable phenology, sparse woody cover and small, irregular fields necessitate the use of Very High Resolution (VHR; < 3 m spatial resolution) data and modern techniques for modeling land use at sufficient scales.We take advantage of VHR data’s spatial resolution and Sentinel-1’s high temporal resolution by implementing an object-based data fusion strategy to model land use. By generating high resolution vector objects from single-date WorldView imagery and using the corresponding Synthetic Aperture Radar (SAR) time series to train a One-Dimensional Convolutional Neural Network (1D CNN), we can effectively leverage deep learning techniques to extract land use signals from multi-resolution and multi-temporal data in a near-autonomous manner.

Satellite image segmentation is crucial in various fields, but acquiring a large amount of labeled data can be challenging. Often, remote sensing images are unlabeled due to the high cost of manual labeling, which hampers training effectiveness and generalization. To address this issue, we propose a spatially and temporally informed graph-based semi-supervised learning approach for satellite image segmentation based on Poisson learning. The main difference to traditional Poisson learning is that our distance function that we use to compute similarity between pixels considers spectral, spatial, and temporal information. Experimental results on the Sentinel-2 time series demonstrate that our approach outperforms other traditional approaches, achieving robust performance in remote sensing image segmentation, especially at a very low label rate.

Land cover mapping has been a valuable tool in capturing changes in many developing regions in Africa. Senegal has been a hotspot of change where agricultural activity has rapidly increased. Agriculture in this region is often a complex mosaic of small fields which makes them difficult to classify using conventional land cover mapping methods and coarse-resolution satellite imagery. WorldView (WV) satellites provide very high-resolution imagery that is ideal for semantic segmentation using convolutional neural networks (CNN). In this study, we introduced training strategies that scale up the training data for the U-Net model using 2 m WV-2 and 3 imagery to overcome the challenges of regional mapping with a patchwork of hundreds of images. The proposed strategies increased the number of training data for the U-Net model in three main scenarios, (i) conventional training, (ii) model transfer, and (iii) transfer learning, and we evaluated model generalizability on test sets for two different regions in Senegal. The results showed that models rapidly reached a high level of performance with a limited increase in additional training in conventional and transfer learning strategies. In these two strategies, the U-Net consistently produced >87% average accuracy for trained images and >70% average accuracy for all test images at the final scale level. The research opens opportunities to produce regional land cover maps in West Africa without generating a prohibitively large amount of training data.

Validating land cover classification results from a machine learning model is a vital step in ensuring that further decisions are based on sound and robust results that can be trusted. Calculating pixel-wise validating statistics from a stack of land cover classification results, while computationally trivial for low-resolution imagery with a small spatial footprint, poses a significant challenge for very-high-resolution (VHR) imagery spanning a larger spatial footprint. Here we describe an open-source unified Python framework and workflow for the compositing of VHR imagery based on climatic and spatial information leveraging hardware acceleration. We additionally describe the implementation of per-pixel reduction algorithms which are used to reduce the stacked composite into a robust and accurate composite that is validated.

Multiresolution deep learning approaches, such as the U-Net architecture, have achieved high performance in classifying and segmenting images. Most traditional convolutional neural network (CNNs) architectures commonly use pooling to enlarge the receptive field, which usually results in irreversible information loss. The U-Net architecture avoids this information loss by introducing skip-connections that allow to reconstruct lost information. Leveraging this property of the U-Net, this study proposes to include a Riesz-Quincunx (RQ) wavelet transform, which combines 1) higher-order Riesz wavelet transform and 2) orthogonal Quincunx wavelets (commonly used to reduce blur in medical images) inside the U-Net to reduce noise in satellite images and their time-series. Combining both approaches, we introduce a hybrid Riesz-Quincunx-UNet Variational Auto-Encoder (RQUNet-VAE) scheme for image and time series decomposition used to reduce noise in satellite imagery. By including denoising capabilities directly inside the UNet architecture, we hypothesize that our RQUNet-VAE may improve downstream image processing tasks that use the traditional U-Net architecture. We present qualitative and quantitative experimental results that demonstrate that our proposed RQUNet-VAE is effective at reducing noise in satellite imagery yielding results similar to other state-of-the-art noise reduction methods. We further show that our RQUNet-VAE outperforms the U-Net architecture specifically in cases where images exhibit high levels of noise. We show this result in two down-stream applications for multi-band satellite images, including image time-series decomposition and image segmentation.

The remote sensing research community is grappling with methods to produce training data that are sufficiently representative of large areas to which they want to scale up their machine learning models for image classification. Effective generalization, will allow land cover classification models trained in data-rich regions to be applied to data-poor regions, with minimal increases in error. This study investigated cross-location generalization through model transfer of convolutional neural networks (CNN), in a series of experiments spread across eight counties within the Chesapeake Bay Catchment. The model transfer was effective (> 80% accuracy), even with as little training as 80/class, across distances up to 600 km. Classification accuracy and to a lesser extent, image similarity in CNN feature space, decreased with geographic distance, but are not the overriding factors governing model transfer.

Earth observation data is collected by ever-expanding fleets of satellites including Landsat1-8, Sentinel1 & Sentinel2, SPOT1-7 and WorldView1-3. These satellites generate at spatial resolutions (pixel size) from 30m to 31cm and provide revisit rates of as frequent as every 5 days. This allows us not only to look at high-resolution images of every corner of the Earth, but also to track events and observe change over time. During the past 5 years, medium spatial resolution satellite data (30 − 10m pixels) have developed very high temporal revisit frequencies of 5-16 days and spatial-temporal structures have been developed to manage these vast data sets. However, high resolution satellite images and rapidly increasing revisit rates create major data management and mining challenges. This work discusses six challenges of integrating observations at different times, from different sensors, at different spatial resolutions and different temporal frequencies into a unified Earth Observation Data Cube, that is, a tensor of location, time, and spectral bands. Challenges include creating a unified data cube from heterogeneous sensors, scaling geo-registration (mapping pixel between images), accounting for uncertainty across observations, imputing missing observations, broad area event detection, and ultimately, predicting the future state of our planet. With such a unified Earth Observation Data Cube in place, we describe potential application areas such as detecting anthropogenic land cover change, early warning of natural hazards, tracing movement of animals, finding missing airplanes, and rapid detection of forest fires.

In this paper, we propose a new method for extracting features from time-series satellite data to detect land cover change. We propose to make use of the behavior of a deterministic nonlinear system driven by a time-dependent force. The driving force comprises a set of concatenated model parameters regressed from fitting a model to a Moderate Resolution Imaging Spectroradiometer time series. The goal is to create behavior in the nonlinear deterministic system, which appears predictable for the time series undergoing no change, while erratic for the time series undergoing land cover change. The differential equation used for the deterministic nonlinear system is that of a large-amplitude pendulum, where the displacement angle is observed over time. If there has been no change in the land cover, the mean driving force will approximate zero, and hence the pendulum will behave as if in free motion under the influence of gravity only. If, however, there has been a change in the land cover, this will for a brief initial period introduce a nonzero mean driving force, which does work on the pendulum, changing its energy and future evolution, which we demonstrate is observable. This we show is sufficient to introduce an observable change to the state of the pendulum, thus enabling change detection. We extend this method to a higher dimensional differential equation to improve the false alarm rate in our experiments. Numerical results show a change detection accuracy of nearly 96% when detecting new human settlements, with a corresponding false alarm rate of 0.2% (omission error rate of 4%). This compares very favorably with other published methods, which achieved less than 90% detection but with false alarm rates all above 9% (omission error rate of 66%).

The use of hyper-temporal MODIS time-series data for the detection of land cover change in South Africa has been an active research area the last few year. This paper expands on previous studies that show that this type of data can be effectively used in the detection of new informal settlements in South Africa. In this paper, the feasibility of using the temporal evolution of the distribution of MODIS reflectance values within a pixel neighborhood to detect land cover change is evaluated. More specifically, the covariance at each time point is evaluated for a specific pixel neighborhood and MODIS band combination and the temporal evolution of the Mahalanobis distance (between each pixel's reflectance value and the reflection distribution of the neighborhood) is calculated. The feasibility of using this derived time-series to detect land cover change was evaluated. Preliminary results indicate that using this derived time-series as opposed to the raw reflection time-series to do land cover change detection reduces false alarms in the order of 7% while maintaining above 90% accuracy.

Southern African savanna ecosystems and their woody resources are under pressure. Governments in the region need locally calibrated, cost effective, and regularly updated information on these resources in order to satisfy both national and international commitments to manage them. Using LiDAR data as a calibration dataset, this paper sets out to investigate the potential of hypertemporal C-band ASAR SAR data in mapping woody structural related parameters in a savanna environment. Images spanning three years where grouped by years (2007-2009), season (Wet or Dry) and polarization (HH or VV), and relationships were sought for the woody parameter total canopy cover (TCC). Results show that: Dry season combinations of images outperformed wet season images; HH co-polarised images outperformed VV images; temporally filtered images showed marked improvement on unfiltered images. While non-parametric random forest models achieved better validation accuracies than other models did. The single best result was achieved by combining all the temporally filtered images, from all of the various scenarios (R2=0.74; RMSE=8.52; SEP=35.27). The results show promise in delivering regional scale, locally calibrated, baseline products for the management of Southern Africa's woody resources.

The woody component in African Savannahs provides essential ecosystem services such as fuel wood and construction timber to large populations of rural communities. Woody canopy cover (i.e. the percentage area occupied by woody canopy or CC) is a key parameter of the woody component. Synthetic Aperture Radar (SAR) is effective at assessing the woody component, because of its capacity to image within-canopy properties of the vegetation while offering an all-weather capacity to map relatively large extents of the woody component. This study compared the modelling accuracies of woody canopy cover (CC), in South African Savannahs, through the assessment of a set of modelling approaches (Linear Regression, Support Vector Machines, REPTree decision tree, Artificial Neural Network and Random Forest) with the use of X-band (TerraSAR-X), C-band (RADARSAT-2) and L-band (ALOS PALSAR) datasets. This study illustrated that the ANN, REPTree and RF non-parametric modelling algorithms were the most ideal with high CC prediction accuracies throughout the different scenarios. Results also illustrated that the acquisition of L-band data be prioritized due to the high accuracies achieved by the L-band dataset alone in comparison to the individual shorter wavelengths. The study provides promising results for developing regional savannah woody cover maps using limited LiDAR training data and SAR images.

It has been shown that time series containing reflectance values from the first two spectral bands of the MODerate-resolution Imaging Spectroradiometer (MODIS) land surface reflectance product can be modelled as a triply modulated cosine function. A meta-optimization approach has been proposed in the literature for setting the parameters of the non-linear Extended Kalman Filter (EKF) to rapidly and efficiently estimate the features for these triply modulated cosine functions using spatial information. In this paper we modify this approach to utilize temporal information instead of spatial information to greatly reduce the processing time and storage requirements to process each time series. The parameters derived from the newly proposed method is classified with a support vector machine and compared to the original approach. Performance of the methods is tested on the Limpopo province in South Africa.

We propose to modulate a deterministic non-linear system with state variables that were derived from an Extended Kalman Filter in order to detect change in a MODIS time series. The deterministic non-linear system used in this work is a gravity pendulum, where the displacement angle is observed over time. The analysis shows a change detection accuracy better than 99% obtained with this deterministic non-linear system when detecting new human settlements, with a corresponding false alarm rate lower than 3%.

The extraction of information on land cover classes using unsupervised methods has always been of relevance to the remote sensing community. In this paper, a novel criterion is proposed, which extracts the inherent information in an unsupervised fashion from a time series. The criterion is used to fit a parametric model to a time series, derive the corresponding covariance matrices of the parameters for the model, and estimate the additive noise on the time series. The proposed criterion uses both spatial and temporal information when estimating the covariance matrices and can be extended to incorporate spectral information. The algorithm used to estimate the parameters for the model is the extended Kalman filter (EKF). An unsupervised search algorithm, specifically designed for this criterion, is proposed in conjunction with the criterion that is used to rapidly and efficiently estimate the variables. The search algorithm attempts to satisfy the criterion by employing density adaptation to the current candidate system. The application in this paper is the use of an EKF to model Moderate Resolution Imaging Spectroradiometer time series with a triply modulated cosine function as the underlying model. The results show that the criterion improved the fit of the triply modulated cosine function by an order of magnitude on the time series over all seven spectral bands when compared with the other methods. The state space variables derived from the EKF are then used for both land cover classification and land cover change detection. The method was evaluated in the Gauteng province of South Africa where it was found to significantly improve on land cover classification and change detection accuracies when compared with other methods.

There has been recent developments in the use of hyper-temporal satellite time series data for land cover change detection and classification in South Africa and in particular, the monitoring of human settlement expansion is of relevance as it is the most pervasive form of land-cover change in the country. In this paper, a spatio-temporal change detection method is proposed that is applicable over large regions. This is achieved by adjusting the change threshold based on the change properties of a neighbourhood of pixels. Results indicate that the addition of spatial information increase change detection accuracy when compared to a pixel based approach.

This study investigated the automated pre-processing and land cover classification of Landsat data. The Web-enabled Landsat Data (WELD) system was used to process large volumes of Landsat imagery to calibrated top of atmosphere reflectance and brightness temperature products which are composited temporally and mosaicked for the KwaZulu-Natal Province of South Africa. The usefulness of an Automatic Spectral Rule-base Classifier (ASRC) approach was evaluated by relating the produced spectral categories to land cover classes. The ASRC method uses a hierarchical rule set, which relies on universally set thresholds derived from the literature, to decide on the spectral category. To assess the performance, the spectral categories were treated as input features to supervised classifiers to optimally assign land cover labels. The land cover classes used in the experiments were obtained from the official map of the Kwazulu-Natal province in South Africa, which was generated by operators in 2008. This approach was compared to an experiment using the original 7 Landsat spectral bands and derived indices as input features. It was found that the ASRC spectral categories did not provide a useful translation to land cover classes (45.5% classification accuracy), while the experiments using the Landsat 7 spectral bands or indices did considerably better (82.7% classification accuracy).

There has been recent developments in the use of hyper-temporal satellite time series data for land cover change detection and classification. Recently, an Autocorrelation function (ACF) change detection method was proposed to detect the development of new human settlements in South Africa. In this paper, an extension to this change detection method is proposed that produces an estimate of the change date in addition to the change metric. Preliminary results indicate that comparable accuracy is achievable relative to the original formulation, with the added advantage of providing an estimate of the change date.

In this paper the Bias Variance Search Algorithm is proposed as an algorithm to optimize a candidate set of initial parameters for an Extended Kalman filter (EKF). The search algorithm operates on a Bias Variance Equilibrium Point criterion to determine how to set the initial parameters. The candidate set is then used by the EKF to estimate state parameters to fit a triply modulated cosine function to time series of the first two spectral bands of the MODerate-resolution Imaging Spectroradiometer (MODIS) land product. The state parameters are then used for land cover classification. The results of the search algorithm was tested on classifying land cover in the Limpopo province, South Africa. An improvement in land cover classification was observed when the method was compared to a robust regression method.

In this paper, the internal operations of an Extended Kalman Filter is investigated to see if any useful information can be derived to detect land cover change in a MODIS time series. The Extended Kalman Filter expands its internal covariance if a significant change in reflectance value is observed, followed by adapting the state parameters to compensate for this change. The analysis shows a change detection accuracy above 90% can be attained when evaluating the elements within the internal covariance matrix to detect new human settlements, with a corresponding false alarm rate below 11%.

A new quantitative method of evaluating fire danger index (FDI) performance is proposed, relying on a combination of modelled meteorological data and fire activity data obtained from earth observation satellites. The method simultaneously allows for localized FDI selection, as well as the analysis of spatial patterns in FDI performance. The Canadian Fire Weather Index (FWI) is shown, using well-known FDI performance metrics, to perform very well over most of southern Africa, compared to other FDIs historically employed in the region.

This study investigated the biomass patterns and sustainability of fuelwood extraction in the Lowveld of South Africa, where rural households are highly dependent on fuelwood from savannas. The objectives of this study were (i) to compare LiDAR-derived biomass between communal areas and references sites in conservation areas, and (ii) to investigate the sustainability of various future scenarios of fuelwood consumption, using a village-specific, supply-and-demand model based on biomass maps and socio-economic data. On granitic substrates the communal rangelands had an average of 12 ton/ha, which is less than half the biomass of the conservation sites. Under the current rate fuelwood consumption, i.e. 67% of households using fuelwood exclusively at an average of 3.5 ton per household per year, all biomass in the investigated site would be depleted within twelve years. Therefore, policies and interventions that promote the diversification of affordable energy alternatives and rural economic development are desperately needed.

Human settlement expansion is one of the most pervasive forms of land cover change in the Gauteng province of South Africa. A method for detecting new settlement developments in areas that are typically covered by natural vegetation using 500 m MODIS time-series satellite data is proposed. The method is a per pixel change alarm that uses the temporal autocorrelation to infer a change metric which yields a change or no-change decision after thresholding. Simulated change data was generated and used to determine a threshold during a preliminary off-line optimization phase. After optimization the method was evaluated on examples of known land cover change in the study area and experimental results indicate a 92% change detection accuracy with a 15% false alarm rate.

In this paper, a change detection accuracy comparison is made between a recently proposed EKF method and a sliding window Fast Fourier Transform (FFT) alternative within a spatio temporal change detection framework. Both methods produce a mean and amplitude parameter sequence which is then used to determine a change metric which yield a change of no-change decision after thresholding. The objective is to determine which of these methods produces a change metric value that is able to best discriminate between change and no-change.

In this paper a neural network is used to perform a mapping between Synthetic Aperture Radar (SAR) backscatter information and LiDAR measurements, and the performance of the neural network model is evaluated against that of a multiple linear regression model. Our aim is to find a relationship between SAR backscatter information and the LiDAR tree volume measurements on a number of land uses in South Africa's Kruger National Park, using a linear as well as a non-linear model. We also seek to find the optimal grid cell size as well as the best combination of SAR polarisation-and decomposition parameters. Our findings suggest that there exists a linear or at least a near-linear relationship between the SAR backscatter information and the LiDAR measurements in South African savannas and that the addition of polarisation-and decomposition parameters to the input of the models aid in improving the Root Mean Squared Error (RMSE) performance.

Time series derived from the first two spectral bands of the MODerate-resolution Imaging Spectroradiometer (MODIS) land surface reflectance product can be modelled as a pair of triply (mean, phase and amplitude) modulated cosine functions. This paper proposes a meta-optimization approach for setting the parameters of the non-linear Extended Kalman Filter to rapidly and efficiently estimate the features for the pair of triply modulated cosine functions. The approach is based on a unsupervised search algorithm over an appropriately defined manifold using spatial and temporal information. Performance of the new method is compared to other applicable methods and is tested on the Gauteng province which is South Africa's province with the fastest growing economy.

A method for detecting land cover change using NDVI time-series data derived from 500-m MODIS satellite data is proposed. The algorithm acts as a per-pixel change alarm and takes the NDVI time series of a 3 × 3 grid of MODIS pixels as the input. The NDVI time series for each of these pixels was modeled as a triply (mean, phase, and amplitude) modulated cosine function, and an extended Kalman filter was used to estimate the parameters of the modulated cosine function through time. A spatial comparison between the center pixel of the 3 × 3 grid and each of its neighboring pixel's mean and amplitude parameter sequence was done to calculate a change metric which yields a change or no-change decision after thresholding. Although the development of new settlements is the most prevalent form of land cover change in South Africa, it is rarely mapped, and known examples amount to a limited number of changed MODIS pixels. Therefore, simulated change data were generated and used for the preliminary optimization of the change detection method. After optimization, the method was evaluated on examples of known land cover change in the study area, and experimental results indicate an 89% change detection accuracy while a traditional annual NDVI differencing method could only achieve a 63% change detection accuracy.

A method for detecting land cover change using NDVI timeseries data derived from MODerate-resolution Imaging Spectroradiometer (MODIS) satellite data is proposed. The algorithm acts as a per pixel change alarm and takes as input the NDVI time-series of a 3 × 3 grid of MODIS pixels. An Extended Kalman Filter was used to estimate a series of parameters related to each NDVI signal. A spatial comparison between the center pixel of the the 3 × 3 grid and each of its neighboring pixels' parameters was done to calculate a change metric which compared to a threshold yielded a change or no-change decision. The method was tested on real change examples in the study area and results indicate 90% detection of new settlements occurring in naturally vegetated areas.

Previous work has shown the ability of waveform LiDAR sensors to accurately describe various land cover types [1] and biomass estimates made in the field [2]. What is lacking, however, is a way to describe the different structural components that are embedded in the digitized backscattered energy from the LiDAR pulse. This study aims to extract structural components from waveform LiDAR data in terms of woody, herbaceous, and bare ground components from data collected over a savanna environment in and around Kruger National Park (KNP), South Africa. These components are comprised of metrics extracted from the waveforms and validated using biomass measurements made in field plots. Different size windows around plot centers, 3 × 3 pixels and 9 × 9 pixels (resulting in 1.5m and 4.5 m footprint, respectively), were used to examine scale effects of larger footprints. It was found that composite waveforms resembling plot sizes (9 × 9) most often are able to describe more than 80% of the woody biomass variability across the entire study site, and individually for two of the three land uses within the area. However, the herbaceous component of the waveform did not correlate well with the field measurements, while the bare ground component was verified visually in a side-by-side comparison with optical imagery.

An automated land cover change detection method is proposed that uses coarse resolution hyper-temporal satellite time series data. The study compared two different unsupervised clustering approaches that operate on the short term Fourier transform coefficients of subsequences of 8-day composite MODerate-resolution Imaging Spectroradiometer (MODIS) surface reflectance data that were extracted with a temporal sliding window. The method uses a feature extraction process that creates meaningful sequential time series that can be analyzed and processed for change detection. The method was evaluated on real and simulated land cover change examples and obtained a change detection accuracy higher than 76% on real land cover conversion and more than 70% on simulated land cover conversion.

The Moderate Resolution Imaging Spectroradiometer (MODIS) time-series data afford the remote sensing community a unique opportunity to investigate the frequency and distribution of fires. Previous research that validated the MODIS burned area product (MCD45A1) in South Africa was only limited to two Landsat 7 Enhanced Thematic Mapper plus (ETM+) scenes in savanna vegetation, which is not adequate for robust assessment of fire distribution across diverse environments. In this study, validation of the MCD45A1 and the Backup MODIS burned area product (hereafter BMBAP) was extended over different South African vegetation types by quantifying their burned area detection and estimation accuracy using Landsat 5 Thematic Mapper (TM) imagery. Results from the four validation sites reveal that there are subtle differences in the accuracy of the two products. These differences could be influenced for example by, vegetation type, spectral characteristics, and size distribution of the burned areas. These results have significant implications for fire monitoring in Southern Africa.

It is proposed that the normalized difference vegetation index time series derived from Moderate Resolution Imaging Spectroradiometer satellite data can be modeled as a triply (mean, phase, and amplitude) modulated cosine function. Second, a nonlinear extended Kalman filter is developed to estimate the parameters of the modulated cosine function as a function of time. It is shown that the maximum separability of the parameters for natural vegetation and settlement land cover types is better than that of methods based on the fast Fourier transform using data from two study areas in South Africa.

Factors influencing woody savanna vegetation structure across a land-use gradient of intensity (highly and lightly utilized communal rangeland) and type (national protected area, private game reserve and communal rangelands) were investigated. Small-footprint discrete return LiDAR data (1.12 m point spacing) from the Carnegie Airborne Observatory (CAO) `Alpha system' were used to measure three-dimensional vegetation structure across the different treatments. A volumetric pixel (voxel) approach was used to characterise the vertical distribution of LiDAR returns, i.e., vegetation density, in one metre increments for comparison using descriptive statistics across the land-use type and intensity gradient. Vegetation structure in the national protected area was most similar to the lightly utilized rangelands, and the private game reserve was most similar to the highly utilized rangelands with low levels of structural diversity present. Current trends in structural diversity can be related to harvesting, regeneration, herbivory and fire.

It is proposed that the NDVI time series derived from MODIS multitemporal remote sensing data can be modelled as a triply (mean, phase and amplitude) modulated cosine function. A non-linear Extended Kalman Filter was developed to estimate the parameters of the modulated cosine function as a function of time. It was shown that the maximum separability of the parameters for different vegetation land cover was better than that of a spectral method based on the Fast Fourier Transform (FFT). Thus it is theorized that the cosine function parameters estimated using the EKF is superior for both classifying land cover and detecting change over time when compared to methods based on the FFT. Results from two study areas in Southern Africa are provided to show the improved separability using MODIS data.

High resolution airborne hyperspectral and discrete return LiDAR data were used to assess bare soil and tree cover differences along a land use transect consisting of state-owned, privately-owned conservation areas, and communal areas in South African savannas. The results show that tree cover is higher in conservation areas as compared to communal areas where local people use fuel wood for personal consumption. Low impact communal sites (limited use) tend to have higher tree cover than higher impacted communal sites. Generally communal areas have altered tree height distribution but in diverse way depending on the geology or the level of human utilization. Bare soil cover was generally found to be quite low (< 10%) in all different land uses, suggesting that the degradation level in communal areas might not be as high as generally perceived.

This paper shows that a feedforward Multilayer Perceptron (MLP) operating over a temporal sliding window of multi-spectral time series MODerate-resolution Imaging Spectroradiometer (MODIS) satellite data is able to detect land cover change that was artificially introduced by concatenating time series belonging to different types of land cover. The method employs an iteratively retrained MLP that is a supervised method, and thus captures all local environmental patterns. Depending on the length of the temporal sliding window used in the short-term Fourier transform, an overall change detection accuracy of between 87.62% and 97.02% was achieved. It is shown that for this type of simulated land cover change, where land cover change was abrupt, a short-term FFT window of 18 months or less, using only the two NDVI spectral bands of MODIS data was sufficient to detect change reliably.

This study used remotely-sensed phenology data derived from Advanced Very High Resolution Radiometer (AVHRR), in a fully supervised decision-tree classification based on the new biome map of South Africa. The objectives were: (i) to investigate the long-term spatial patterns and inter-annual variability in satellite-derived vegetation phenology in relation to the recently revised biome map and (ii) to identify the phenological attributes that distinguishes between the different biomes. The long term phenometrics gave ecologically-meaningful results which reflect our current understanding of the spatial patterns of production and seasonality of vegetation growth in southern Africa. Regression tree analysis based on remotely-sensed phenometrics performed as good as, or better than, previous climate-based predictors of biome distribution.

Measurement and management of vegetation biomass accumulation in ecosystems typically involves extensive field data collection, which can be expensive and time consuming, while leaving the user with relatively crude inputs to intricate biomass models. Light detection and ranging (LiDAR) remote sensing, which provides extensive height measurements of terrain and vegetation, has become an effective alternative to characterize vegetation structure. In this paper, we report on ongoing efforts at developing signal processing approaches to model herbaceous biomass using a new generation of airborne laser scanners, namely full-waveform LiDAR systems. Structural and statistic-based feature metrics are directly derived from LiDAR waveforms at the pixel level and related to plot-level field data. Initial results reveal a definite correlation between the LiDAR waveform and herbaceous biomass. Ongoing research focuses on the links between fractional cover estimated from imaging spectroscopy and woody biomass.

The paper introduces an Artificial Neural Network (ANN) operating on a sliding window over time for detecting land cover class changes. If a class change is detected, then change detection has been accomplished. The ANN is presented as well as experimental results usingMODIS time-series data derived from 8-day 500m MODIS [MOD09] composite images in two study areas in Southern Africa. The method is supervised, i.e. it assumes training data is available, but after training it is able to detect change on testing data not seen before.

Satellite-derived phenology allows monitoring of terrestrial vegetation on a global scale and provides an integrative view at the landscape level . Understanding these seasonal phenological patterns is essential to (i) the characterisation and classification of vegetation, (ii) studying the impact of climate change , and influence of rainfall variability (iii) monitoring desertification and (iv) detecting changes in land use/land cover. This study analyzed vegetation phenology across Southern Africa in order to investigate which phenometrics (and their inter-annual variability) distinguish biomes based on functional patterns. A second objective was to quantify the inter-annual variability of phenometrics during a 15-year period (1985 to 2000).

Land degradation involves conditions of reduced net primary production (NPP). Our objective was to develop and evaluate a land degradation monitoring approach that is based on NPP. NPP was modeled using the Global Production Efficiency Model (GLO-PEM) and 1km AVHRR data. Degraded areas in the Northern Province of South Africa were successfully detected using NPP, Rain Use Efficiency (RUE) and Local NPP Scaling (LNS) as measures of land condition.