This paper evaluates GEDI's canopy height estimation accuracy in dense tropical forests located in Mayotte Island. It examines GEDI's ability to penetrate canopies and detect the ground, which is crucial for reliable estimates. The study tests the use of a single GEDI height metric (rh_95) in comparison with regression models using various GEDI metrics to enhance accuracy. Beam sensitivity plays a pivotal role, as it impacts significantly GEDI return waveforms and the subsequent derived height estimates. In the context of our study, GEDI tends to underestimate heights above 15 meters. Regression models outperform rh_95, mitigating the impact of beam sensitivity and canopy height (RMSE decreasing from 6.6 m to 5.5 m, bias going from -1.9 m to 0.0 m). They provide unbiased estimates, offering improved accuracies regardless of these factors. This study emphasizes GEDI's limitations and highlights regression models' potential to refine canopy height estimations in complex ecosystems where signal penetration is challenging.

This research presents a comprehensive canopy height map of French Guiana at 10 m spatial resolution, employing a data fusion approach integrating optical (Sentinel-2), radar (Sentinel-1 and ALOS), and ancillary data sources. The primary objective is to leverage a U-Net neural network model, trained and validated using Global Ecosystem Dynamics Investigation (GEDI) data as reference canopy height. We aim at understanding how canopy height prediction models can be improved through the integration of relevant remote sensing and environmental descriptors related to canopy structure. The accuracies of the generated canopy height maps are assessed against high-resolution airborne LiDAR (ALS) acquisitions conducted by the French National Forest Office. We observe that enriching input data with height above nearest drainage (HAND) as well as forest landscape information yielded improved accuracies for the prediction models. Moreover, accounting for GEDI database uncertainties, through filtering of usable waveforms and correction of geolocation errors, also resulted in a performance gain for canopy height estimation using a U-Net model.

This study explores the impact of Eucalyptus plantation characteristics and environmental factors on GEDI returned energy distribution. Random Forest (RF) regression was used to analyze the effect of a diverse parcel-scale Eucalyptus plantation characteristics including trees height, planting density, soil properties, understorey presence, environmental conditions and NDVI generated from Sentinel-2 as a proxy of leaf area index on the GEDI relative heights (RHn). According to the findings, as the vertical distance from the ground increases (from RH5 to RH100), the most important variable explaining a given relative height changes from "NDVI" to "Volume". Moreover, at higher quantiles of the returned energy, the behavior of the GEDI metrics becomes more dependent on a narrower set of forest and environmental characteristics. However, at low RH values, the interplay of complex canopy structures and environmental factors necessitates a combination of features to explain the observed variations.

The Global Ecosystem Dynamics Investigation (GEDI) instrument, as all FW systems, relies on very sophisticated pre-processing steps to generate a priori metrics in order to accurately estimate forest characteristics, such as forest heights and wood volume. The ever-expanding volume of acquired GEDI data, which to September 2020 comprised more than 25 billion shots, and requiring more than 90 TB of storage space, raises new challenges in terms of adapted preprocessing methods for the suitable exploitation of such a huge and complex amount of LiDAR data. Therefore, to avoid metric computation, we leveraged deep learning techniques in order to estimate canopy dominant heights (Hdom) and wood volume (V) of Eucalyptus plantations over five different regions in Brazil. Performance comparisons were conducted between a convolutional neural network based model that uses GEDI waveform data and a previously used, metric based, Random Forest regressor (RF). Cross-validated results showed that the CNN based model compared well against the RF counterpart for both Hdomand V. Indeed, the RMSE on the estimation of Hdomfrom the CNN based model was 1.61 m with a coefficient of determination R2of 0.90, while the RF model produced an accuracy on Hdomestimates of 1.45 m(R2=0.92). For V, CNN based estimates was 27.35 m3.ha-1(R2of 0.88), while for RF, the RMSE was 27.60 m3.ha-1 (R2=0.88).

Full waveform (FW) LiDAR systems have gained momentum to map forest biophysical variables in the last two decades, owing to their ability to accurately estimate canopy heights and aboveground biomass. Currently, the Global Ecosystem Dynamics Investigation (GEDI) system on board of the International Space Station (ISS) is the most recent FW spaceborne LiDAR instrument for the continuous observation of earth's forests. Here, we assess the accuracy of GEDI FW data for the estimation of stand-scale dominant heights ($H_{dom}$), and stand volume (V) using linear and nonlinear regression models based on several GEDI metrics. The models were calibrated and validated using in-situ data from Eucalyptus plantations in Brazil. Overall, the most accurate estimates of$H_{dom}$and V were obtained using the stepwise regression, with an RMSE of 1.44 m (R2 of 0.92) and 24.39 m3.ha−1(R2 of 0.90) respectively. The principal metric explaining more than 87% and 84% of the variability (R2) of$H_{dom}$and V was the metric representing the height above the ground at which 90% of the waveform energy occurs.

Land cover mapping is of great importance to provide reliable quantification of agricultural landscapes. However, one of the limitations in tropical regions is cloud and cloud shadow coverage, resulting in imagery gaps. In this study, we tested four methods of gap filling: Interpolation k = 1 and k = 2, Mean and Median using VENµS satellite time series. Further, we assessed these filled time series in an object-based classification using Random Forest algorithm in the center of São Paulo state, Brazil. We used a 10-day composite NDVI as input data for the gap-filling methods. The linear interpolation (k=1) showed good adaptation to high dynamics temporal profile crops over time, such as sugarcane and annual crops. This same database with interpolation (k=1) achieved high overall accuracy in the classification (0.81) allowing better discrimination on land use classes.

Pixelwise image classification based on time series profiles has been very effective in several applications. In this letter, we investigate recently proposed image-based time series encoding approaches [e.g., Gramian angular summation field/Gramian angular difference field (GASF/GADF) and Markov transition field (MTF)] to support the identification of eucalyptus regions in remote sensing images. We perform a comparative study concerning the combination of image-based representations suitable for encoding the most important time series patterns with the ability of state-of-the-art deep-learning-based approaches for characterizing image visual properties. The comparative study demonstrates that the evaluated image representations, combined with different deep learning feature extractors lead to highly effective classification results, which are superior to those of recently proposed methods for time-series-based eucalyptus plantation detection.

Pixelwise remote sensing image classification has benefited from temporal contextual information encoded in time series. In this paper, we investigate the use of data-driven features extracted from time series representations based on recurrence plots, with the goal of improving the effectiveness of classification systems. Performed experiments considered the classification of eucalyptus plantations based on time series profiles. Achieved results demonstrate that the combination of recurrence plot representations with deep-learning features are a promising research venue for addressing pixelwise classification problems.

Cloudy conditions impede and reduce the utility of optical imagery. With the launch of Sentinel-1A and B, the ongoing availability of RADARSAT-2 imagery, and the expected launch of the RADARSAT Constellation Mission (RCM), dense time series of C-band Synthetic Aperture Radar (SAR) data will now be readily available. For crop classification and mapping, SAR imagery has yet to be used to its full potential and has generally been combined with optical imagery. The JECAM SAR Inter-Comparison Experiment is a multi-year, multi-partner project that aims to compare global methods for SAR-based crop monitoring and inventory. Sets of dense time-series SAR imagery which include RADARSAT-2 and Sentinel-1 data were prepared for this experiment. AAFC's operational Decision Tree (DT) and newly implemented Random Forest (RF) classification methodologies were applied to these SAR only data-stacks, and to optimized, traditional data-stacks of optical/SAR combinations. This paper outlines the results of these dense time-series classifications and how these results were affected by changing numbers of agriculture classes, numbers of available SAR imagery and numbers of training and validation data points for individual crop types. In general, for the dense time-series SAR stacks, overall accuracies of greater than 85%, a typical operational goal, were obtained for 6 of 12 sites. These results have important operational implications for particularly cloudy regions where the availability of optical imagery is limited.

Suitable time series representations play an important role in classification tasks. In this letter, we investigate the use of recurrence-plot-(RP)-based representations in the classification of eucalyptus regions in remote sensing images. The proposed framework is composed of three steps. First, time series associated with image pixels are represented by RP images; next, RP images are characterized by means of visual description approaches; finally, we use a soft computing framework based on genetic programing to discover an effective combination of time series dissimilarity functions to combine extracted features. Performed experiments in a eucalyptus classification problem demonstrated that the proposed framework is effective when compared to approaches based on the use of time series itself.

The objective of this study was to analyze the L-band SAR backscatter sensitivity to forest biomass for Eucalyptus plantations. The results showed that the radar signal is highly dependent on biomass only for values lower than 50 t/ha, which corresponds to plantations of approximately three years of age. Next, Random Forest regressions were performed to evaluate the potential of PALSAR data to predict the Eucalyptus biomass. Regressions were constructed to link the biomass to both radar signal and age of plantations. Results showed that the age was the variable that best explained the biomass followed by the PALSAR HV polarized signal. For biomasses lower than 50 t/ha, HV signal and plantation age were found to have the same level of importance in predicting biomass. For biomasses higher than 50 t/ha, plantation age was the main variable in the random forest models. The use of PALSAR signal alone did not correctly predict the biomass of Eucalyptus plantations (R2lower than 0.5 and RMSE higher than 46.7 t/ha). The use of plantation age in addition to the PALSAR signal improved slightly the prediction results (R2increased from 0.88 to 0.92 and RMSE decreased from 22.7 to 18.9 t/ha).

The Geoscience Laser Altimeter System (GLAS) has provided a useful dataset for estimating forest height in many areas of the globe. Most of the studies on GLAS waveforms have focused on natural forests and only a few were conducted over forest plantations. The objective of this study was to test the best known models used for estimating canopy height and above ground biomass of intensively managed Eucalyptus plantations in Brazil using full waveform LiDAR data. Studies to estimate forest heights from LiDAR data have highlighted that the fitting coefficients of developed models are strongly dependent on environmental factors such as the region of the study site, terrain topography, and forest type. In this study, we evaluated the main models developed to predict canopy height using a combination of parameters extracted from GLAS waveforms and a digital elevation model, in order to explore which combination of parameters yields the best forest height estimates. In addition, a model to estimate above ground biomass from dominant height was calibrated.

The quantification of forest characteristics at different scales is key to understanding how forest ecosystems function. Hyperspectral imagery may provide information to quantitatively estimate some biochemical (e.g. chlorophyll or nitrogen content), structural (e.g. leaf area index) or physiological (e.g. light use efficiency) forest characteristics at different scales, but this information is not easily retrievable. The simple reflectance index concept has been inherited from a long history of multispectral data analysis, and is commonly transposed to hyperspectral data. The design and use of such hyperspectral indices is questioned in this study.

Very high spatial resolution (VHSR) satellite images provide interesting information for parameterizing tree-scale forest process-based models, and in particular their light absorption submodels, which is at the basis of photosynthesis calculation. Such tree-scale models require a large amount of field measurements to describe the forest ecosystems, i.e. all tree positions, their sizes and shapes, their leaf areas, etc. These data are generally measured directly in the field [1], which can be tedious for large areas like a forest stand. In this study, we explore the possibility to parameterize such tree-scale models directly or indirectly from panchromatic and multispectral very high resolution images.