The current land cover mapping paradigm relies on automatic classification of satellite images, with supervised methods being the most used, implying training data to have a crucial role. Aspects such as training sample size and quality should be carefully considered. This paper proposes assessing the use of a detailed class nomenclature to reinforce class diversity in the training sample. A Random Forest (RF) classification of Sentinel-2 multi-temporal data was conducted. Additionally, the effect of sample size and class distribution were evaluated. The results indicate that the use of a detailed nomenclature provided better results in terms of classification accuracy. With respect to sample distribution, adopting class sizes proportional to their occurrence in a reference land cover map exhibited superior performance in comparison to an equal size approach. The effect of sample size on classification performance was limited, as previous studies with RF suggested.

This paper presents an experimental crop classification of the 10 most abundant annual crop types in Portugal, using a study area located in Alentejo region. This region has great diversity of land uses as well as multiple crop types. Sentinel-2 2018 intra-annual time-series imagery is considered in the experiment. The Portuguese Land Parcel Identification System (LPIS) is used to extract automatic training samples. LPIS information is automatically processed with the help of auxiliary datasets to filter out crop areas more likely to have been mislabeled. Classification is obtained using random forest. Validation is performed using an independent dataset also based on LPIS. A global accuracy of 76% is obtained. The novelty of the methodology here presented shows that LPIS can be used together with auxiliary data for crop type mapping, helping to characterize the agriculture land diversity in Portugal.

Southern Portugal is characterized by disperse tree cover of Cork and Holm oaks in an agro-forestry system known as montado. Mapping these trees has been historically very difficult as they occur in isolation or in groups with different understory vegetation, including grass and shrubland. Automatic classification for binary tree/non-tree map production has been used elsewhere, but with limited success in the context of montado. Here, the potential of Sentinel-2 data was explored to map oaks using pure and mixed pixels to train a random forest. The output depicts a gradient of tree cover that can be transformed into a crisp map. The accuracy assessment of the latter shows commission and omission errors of 17% and 18%.

Pest damages in eucalyptus plantations cause significant economic losses for the pulp and paper industry. Longhorned borers (ELB) outbreaks induce mortality in eucalyptus stands. In this study, multispectral imagery was obtained from unmanned aerial vehicles. We attempt to improve the classification process done in previous work. The local maxima of sliding a window and the Large-Scale Mean-Shift segmentation (LSMS) were applied to extract tree crows. Subsequently, the mean of spectral bands and twelve vegetation indices were calculated to characterize each segment. To classify tree canopies into dead and healthy trees, supervised machine learning using Random Forest (RF) and Support Vector Machine (SVM) were applied. The overall accuracy of Random Forests was 98.35% and Support Vector Machine of 97.7%. We concluded that SVM did not perform better than RF. Moreover, adding new vegetation indices in the classification process did not increase accuracy.

Tree species classification with satellite data has become more and more popular since Sentinel-2 launch. We compared efficacy and effectiveness of Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LGBM) with widely used in remote sensing Random Forest (RF), Support Vector Machine (SVM) and K-Nearest Neighbour (KNN) algorithms. Analyses were performed over an area in Portugal with multi-temporal Sentinel-2 data registered in April, June, August and October 2018. The selected classes were: cork oak, holm oak, eucalyptus, other broadleaved, maritime pine, stone pine and other coniferous. Algorithm efficacy was measured through F1-score and accuracy while efficiency was measured through the median time needed for each fit. XGBoost and LGBM outperformed efficacy of other algorithms, which was already high (above 90% for the best variant of each algorithm). In terms of efficacy, LGBM overcame all algorithms, including XGBoost.

Classification accuracy of remote sensing images with supervised learning depends on the quality and characteristics of training samples. Size is a key aspect of a sample and its impact on classification depends on several factors, including the classifier employed, dimension on the feature space and land cover characteristics. Random Forest classifier is considered to be of low sensitivity to variations in sample size. However, further investigation is required when feature spaces are large and training is performed with spectral subclasses of the land cover classes to be mapped. This paper proposes to assess the impact of sample size in the classification accuracy of Random Forest using multitemporal Sentinel-2 data and a detailed set of training subclasses to produce a map with general land cover classes. The results revealed similar classification accuracies after major reductions in sample size.

Numerous efforts have been made to develop models to fit multispectral reflectance and vegetation index (VI) time series from satellite images for diverse land cover classes. The common objective of these models is to derive a set of measurable parameters that are able to characterize and to reproduce the land cover dynamics of natural- and human-induced ecosystems. Good-fitting models should therefore match different waveforms and be insensitive to sharp and localized variations, generally due to atmospheric disturbances. In this paper, we propose a model-based approach to identify and predict important dynamics for indiscriminate land cover classes. Our method relies on an original nonlinear harmonic model that remarkably matches intra-annual time series of multispectral reflectances and VIs obtained from satellite images. The proposed model is characterized by the following: 1) parsimonious, comprising only five parameters; 2) readily identifiable (in the maximum likelihood sense) from only few observations; 3) robust to noise; and 4) versatile, since it can reproduce a wide variety of intra-annual land cover dynamics as a deterministic function of time. To demonstrate the relevance of our approach, we use a time series of Moderate Resolution Imaging Spectroradiometer eight-day composite images acquired in Portugal over a one-year period at a 500-m nominal spatial resolution. For 13 different land cover classes, which are representatives of Mediterranean landscapes, we evaluate the data-model adequacy of our model and compare it with several other approaches. We then address a particularly interesting and promising application of our method using rice crops and shrublands as case studies. We not only show that phenological attributes can be accurately estimated from the fitted time series, but we also demonstrate that it is possible to make early predictions of phenological attribute dates and magnitudes from our expected model adjusted to only few anterior observations.

In this paper we present a study exploring the medium spatial resolution images of recently launched ENVISAT MERIS for land cover characterization in Portugal. The goal is to take advantage of enhanced spectral and temporal resolutions of images acquired by this sensor to discriminate properly between 19 land cover classes. We test both unitemporal and multitemporal classifications, trying to augment land cover classes' discrimination through the use of temporal variations of classes' characteristic spectral reflectances along one year period. We discuss about classifier's accuracy and map accuracy, showing the existence of noticeable differences between achieved values in both evaluations. The assessment of the final map resulted in an overall accuracy of 70%, considering a final set of 9 land cover classes; these classes were selected along the study as being the more adequate to map the Portuguese landscape at MERIS spatial resolution. Best classification results were attained by removing spectral bands 1, 2, 3, and 13 from input dataset for classification, and using the Maximum Likelihood as a supervised classifier.

This paper introduces a reference sample database that is being developed by the Remote Sensing Unit of the Portuguese Geographic Institute for the accuracy assessment of medium scale land cover products in Portugal. The goal is to provide the worldwide remote sensing community with sufficient data for the accurate estimation of overall and per class proportions of correctly classified area in regional and global land cover products at this part of the globe. This is a massive database that encloses various descriptive attributes for each sample observation, namely primary and alternate reference land cover labels, nominally scored interpretation ratings and location confidence ratings. We present in detail the attached land cover nomenclature and the database design, i.e. the sampling design used to collect sample observations, as well as the process used to identify the most pertinent reference land cover label for each observation. In addition, we briefly describe some statistics about attribute information that was recorded for each observation and that can be used as auxiliary information for the accuracy assessment of land cover products.

This is a preliminary study in the framework of an ongoing research work that aims at comparing the aptitude of MERIS and MODIS images for land cover mapping at regional scale. Overall and per class accuracies achieved with a Maximum Likelihood classification of MODIS and MERIS images acquired during August 2005, are used as a measurement of their adequacy for land cover characterization in Portugal. Attained results show that differences in spatial and spectral resolutions of used images do not produce overly disparities in land cover classes' discrimination. Still, the separation between such numerous land cover classes is hampered by landscape fragmentation of the territory at such spatial resolution.

This paper aims at presenting the usefulness of combining satellite optical data from the visible and infrared wavelengths with longer wavelength radar data for land cover mapping in Portugal. This is a ground-breaking study in a geographical region that does not experience continuous intra- annual dreadful atmospheric contamination that commonly justifies radar usage. In this study we exploit the ability of ASAR images as an extra input feature for land cover classification together with the most used satellite optical data, i.e. Landsat. The goal of this paper is three-fold: 1) compare single date classification of ASAR data with Landsat data for land cover mapping; 2) evaluate the usefulness of multi-temporal ASAR measurements for land cover classification improvement; and 3) to compare a final map accuracy assessment with the classification scores attained with training and test sample sets. We conclude that ASAR imagery does not individually improve overall classification accuracy, but their synergy with Landsat data or in a multi-temporal context show up specific advantages; statistically sound accuracy assessment of final map bends optimal classification accuracies attained with test sampling observations.

The geometric optical (GO) bidirectional reflectance model, combined with a new component spectral signature submodel, can be used to estimate the bidirectional reflectance distribution function (BRDF) of discontinuous canopies. This approach retains the GO approach of incorporating the effect of shadows cast by crowns on the background. The newly developed submodel uses an analytical approximation of the radiative transfer (RT) within the plant canopies to model the spectral properties of each scene component. A multiple scale-hotspot function that incorporates effects for smaller canopy objects like branches, stems and leaves was also well modeled. Comparison of model results with field measurements (ASAS, POLDER and PARABOLA) over an old black spruce forest in central Canada demonstrated that the model ran predict the basic features of the BRDF, i.e., bowl shape and the hotspot. The benefits of the model presented are simplicity, improved treatment of multiple scattering and a new method of estimating the component signatures.