Search

Multitemporal remote sensing images offer multiple views of Earth observation, enabling the integration of temporal information to effectively mitigate issues caused by clouds, shadows, seasonal changes, and atmospheric reflection. This integration significantly reduces the uncertainty and noise of images, facilitating a more comprehensive and accurate extraction of essential ground object features. However, most land cover classification models based on deep learning focus on interpreting single-temporal remote sensing images, and multitemporal remote sensing datasets are relatively scarce, leading to challenges in model training. To address these issues, this article proposes a cross-temporal self-supervised learning method with superpixel mask (CTS2L-sm) for multitemporal land cover classification. Specifically, we design a superpixel-based masking strategy to replace the fixed-size patch masking strategy used in masked image modeling (MIM), which prevents the complete occlusion of small-scale remote sensing scenes and more effectively preserves structural information. To further enhance the integration of multitemporal information, we introduce a cross-temporal attention module (CTAM) in both forward and backward to extract cross-temporal features. The resulting multitemporal denoised homogeneous features, when fused with each temporal feature, significantly improve classification accuracy. CTS2L-sm combines these denoised homogeneous features extracted by CTAM with single-temporal features and simultaneously reconstructs the original remote sensing images across all temporal, extending the MIM strategy from single-temporal to multitemporal self-supervised training. Experimental results on the WUSU and FUSU datasets demonstrate the effectiveness and superiority of the proposed CTS2L-sm compared to several state-of-the-art methods.

Questions: Can we map complex habitat mosaics from remote‐sensing data? In doing this, are measures of spectral heterogeneity useful to improve image classification performance? Which measures are the most important? How can multitemporal data be integrated in a robust framework? Location: Classical Karst (NE Italy). Methods: First, a habitat map was produced from field surveys. Then, a collection of 12 monthly Sentinel‐2 images was retrieved. Vegetation and spectral heterogeneity (SH) indices were computed and aggregated in four combinations: (1) monthly layers of vegetation and SH indices; (2) seasonal layers of vegetation and SH indices; (3) yearly layers of SH indices computed across the months; and (4) yearly layers of SH indices computed across the seasons. For each combination, a Random Forest classification was performed, first with the complete set of input layers and then with a subset obtained by recursive feature elimination. Training and validation points were independently extracted from field data. Results: The maximum overall accuracy (0.72) was achieved by using seasonally aggregated vegetation and SH indices, after the number of vegetation types was reduced by aggregation from 26 to 11. The use of SH measures significantly increased the overall accuracy of the classification. The spectral β‐diversity was the most important variable in most cases, while the spectral α‐diversity and Rao's Q had a low relative importance, possibly because some habitat patches were small compared to the window used to compute the indices. Conclusions: The results are promising and suggest that image classification frameworks could benefit from the inclusion of SH measures, rarely included before. Habitat mapping in complex landscapes can thus be improved in a cost‐ and time‐effective way, suitable for monitoring applications. [ABSTRACT FROM AUTHOR]

Significant areas of native forest in Kalimantan, on the island of Borneo, have been cleared for the expansion of plantations of oil palm and rubber. In this study multisource remote sensing was used to develop a time series of land cover maps that distinguish native forest from plantations. Using a study area in east Kalimantan, Landsat images were combined with either ALOS PALSAR or Sentinel-1 images to map four land cover classes (native forest, oil palm plantation, rubber plantation, non-forest). Bayesian multitemporal classification was applied to increase map accuracy and maps were validated using a confusion matrix; final map overall accuracy was >90%. Over 18 years from 2000 to 2018 nearly half the native forests in the study area were converted to either non-forest or plantations of either rubber or oil palm, with the highest losses between 2015 and 2016. Trending upwards from 2008 large areas of degraded or cleared forests, mapped as non-forest, were converted to oil palm plantation. Conversion of native forests to plantation mainly occurred in lowland and wetland forest, while significant forest regrowth was detected in degraded peatland. These maps will help Indonesia with strategies and policies for balancing economic growth and conservation.

Multitemporal crop classification approaches have demonstrated high performance within a given season. However, cross-season and cross-region crop classification presents a unique transferability challenge. This study addresses this challenge by adopting a domain generalization approach, e.g., by training models on multiple seasons to improve generalization to new, unseen target years. We utilize a comprehensive five-year Sentinel-2 dataset over different agricultural regions in Slovakia and a diverse crop scheme (eight crop classes). We evaluate the performance of different machine learning classification algorithms, including random forests, support vector machines, quadratic discriminant analysis, and neural networks. Our main findings reveal that the transferability of models across years differs between regions, with the Danubian lowlands demonstrating better performance (overall accuracies ranging from 91.5% in 2022 to 94.3% in 2020) compared to eastern Slovakia (overall accuracies ranging from 85% in 2022 to 91.9% in 2020). Quadratic discriminant analysis, support vector machines, and neural networks consistently demonstrated high performance across diverse transferability scenarios. The random forest algorithm was less reliable in generalizing across different scenarios, particularly when there was a significant deviation in the distribution of unseen domains. This finding underscores the importance of employing a multi-classifier analysis. Rapeseed, grasslands, and sugar beet consistently show stable transferability across seasons. We observe that all periods play a crucial role in the classification process, with July being the most important and August the least important. Acceptable performance can be achieved as early as June, with only slight improvements towards the end of the season. Finally, employing a multi-classifier approach allows for parcel-level confidence determination, enhancing the reliability of crop distribution maps by assuming higher confidence when multiple classifiers yield similar results. To enhance spatiotemporal generalization, our study proposes a two-step approach: (1) determine the optimal spatial domain to accurately represent crop type distribution; and (2) apply interannual training to capture variability across years. This approach helps account for various factors, such as different crop rotation practices, diverse observational quality, and local climate-driven patterns, leading to more accurate and reliable crop classification models for nationwide agricultural monitoring. [ABSTRACT FROM AUTHOR]

Poplars are one of the most widespread fast-growing tree species used for forest plantations. Owing to their distinct features (fast growth and short rotation) and the dependency on the timber price market, poplar plantations are characterized by large inter-annual fluctuations in their extent and distribution. Therefore, monitoring poplar plantations requires a frequent update of information – not feasible by National Forest Inventories due to their periodicity – achievable by remote sensing systems applications. In particular, the new Sentinel-2 mission, with a revisiting period of 5 days, represents a potentially efficient tool for meeting this need. In this paper, we present a deep learning approach for mapping poplar plantations using Sentinel-2 time series. A reference dataset of poplar plantations was available for a large study area of more than 46,000 km2 in Northern Italy and served as training and testing data. Two classification methods were compared: (1) a fully connected neural network (also called multilayer perceptron), and (2) a traditional logistic regression. The performance of the two approaches was estimated through bootstrapping procedure with a confidence interval of 99%. Results indicated for deep learning an omission error rate of 2.77%±2.76%, showing improvements compared to logistic regression, omission error rate = 8.91%±4.79%.

The LCZ framework has become a widely applied approach to study urban climate. The standard LCZ typology is highly specific when applied to western urban areas but generic in some African cities. We tested the generic nature of the standard typology by taking a two-part approach. First, we applied a single-source WUDAPT-based training input across three urban areas that represent a gradient in South African urbanization (Cape Town, Thohoyandou and East London). Second, we applied a local customized training that accounts for the unique characteristics of the specific area. The LCZ classification was completed using a random forest classifier on a subset of single (SI) and multitemporal (MT) Sentinel 2 imagery. The results show an increase in overall classification accuracy between 17 and 30% for the locally calibrated over the generic standard LCZ framework. The spring season is the best classified of the single-date imagery with the accuracies 7% higher than the least classified season. The multi-date classification accuracy is 13% higher than spring but only 9% higher when a neighborhood function (NF) is applied. For acceptable performance of the LCZ classifier in an African context, the training must be local and customized to the uniqueness of that specific area. [ABSTRACT FROM AUTHOR]

Significant areas of native forest in Kalimantan, on the island of Borneo, have been cleared for the expansion of plantations of oil palm and rubber. In this study multisource remote sensing was used to develop a time series of land cover maps that distinguish native forest from plantations. Using a study area in east Kalimantan, Landsat images were combined with either ALOS PALSAR or Sentinel-1 images to map four land cover classes (native forest, oil palm plantation, rubber plantation, non-forest). Bayesian multitemporal classification was applied to increase map accuracy and maps were validated using a confusion matrix; final map overall accuracy was >90%. Over 18 years from 2000 to 2018 nearly half the native forests in the study area were converted to either non-forest or plantations of either rubber or oil palm, with the highest losses between 2015 and 2016. Trending upwards from 2008 large areas of degraded or cleared forests, mapped as non-forest, were converted to oil palm plantation. Conversion of native forests to plantation mainly occurred in lowland and wetland forest, while significant forest regrowth was detected in degraded peatland. These maps will help Indonesia with strategies and policies for balancing economic growth and conservation. [ABSTRACT FROM AUTHOR]

Recognition of invasive species and their distribution is key for managing and protecting native species within both natural and man-made ecosystems. Small woody features (SWF) represent fragmented patches or narrow linear tree features that are of high importance in intensively utilized agricultural landscapes. Simultaneously, they frequently serve as expansion pathways for invasive species such as black locust. In this study, Sentinel-2 products, combined with spatiotemporal compositing approaches, are used to address the challenge of broad area black locust mapping at a high granularity. This is accomplished by conducting a comprehensive analysis of the classification performance of various compositing approaches and multitemporal classification settings throughout four vegetation seasons. The annual, seasonal (bi-monthly), and monthly median values of cloud-masked Sentinel-2 reflectance products are aggregated and stacked into varied time-series datasets per given year. The random forest algorithm is trained and output classification maps validated based on field-based reference datasets across Danubian lowlands (Slovakia). The main results of the study proved the usefulness of spatiotemporal compositing of Sentinel-2 products for mapping black locust in small woody features across wide area. In particular, temporally aggregated monthly composites stacked to seasonal time series datasets yielded consistently high overall accuracies ranging from 89.10% to 91.47% with balanced producer’s and user’s accuracies for each year’s annual series. We presume that a similar approach could be used for a broader scale species distribution mapping, assuming they are spectrally or phenologically distinctive, as is often the case for many invasive species.

The assumption of the European Union Common Agricultural Policy is to maintain good agricultural practices for sustainability in the environment. A number of requirements are imposed on farmers, including the maintenance of permanent grassland, fallow land or crop diversification. To meet these requirements, the European Union guarantees subsidies, but at the same time fields must be monitored focusing on crop identification. The limitation of field inspection and substituting it with crop recognition using satellite images could increase the effectiveness of this procedure. The application of satellite imagery in automatic detection and identification of dominant crops over a large area seems to be technically and economically sound. The paper discusses the concept and the results of automatic classification based on a Random Forests classifier performed on multitemporal images of Sentinel-2 and Landsat-8. A test site was established in a complex agricultural structure with long and narrow parcels in the south-eastern part of Poland. Time-series images acquired during the growing season 2016 were used for multispectral classification in different configurations: for Sentinel-2 and Landsat-8 separately and for both sensors integrated. Different Random Forests approaches and post-processing methods were examined based on independent data from farmers' declarations records, reaching the best accuracy of over 90% for crops like winter or spring cereals. Overall accuracy of the classification ranged from 72% to 91% depending on the classification variant. The elaborated scheme is novel in the context of Polish complex agricultural structure and smallholders. [ABSTRACT FROM AUTHOR]

Sown Biodiverse Pastures (SBP) are the basis of a high-yield grazing system tailored for Mediterranean ecosystems and widely implemented in Southern Portugal. The application of precision farming methods in SBP requires cost-effective monitoring using remote sensing (RS). The main hurdle for the remote monitoring of SBP is the fact that the bulk of the pastures are installed in open Montado agroforestry systems. Sparsely distributed trees cast shadows that hinder the identification of the underlaying pasture using Unmanned Aerial Vehicles (UAV) imagery. Image acquisition in the Spring is made difficult by the presence of flowers that mislead the classification algorithms. Here, we tested multiple procedures for the geographical, object-based image classification (GEOBIA) of SBP, aiming to reduce the effects of tree shadows and flowers in open Montado systems. We used remotely sensed data acquired between November 2017 and May 2018 in three Portuguese farms. We used three machine learning supervised classification algorithms: Random Forests (RF), Support Vector Machine (SVM) and Artificial Neural Networks (ANN). We classified SBP based on: (1) a single-period image for the maximum Normalized Difference Vegetation Index (NDVI) epoch in each of the three farms, and (2) multi-temporal image stacking. RF, SVM and ANN were trained using some visible (red, green and blue bands) and near-infrared (NIR) reflectance bands, plus NDVI and a Digital Surface Model (DSM). We obtained high overall accuracy and kappa index (higher than 79% and 0.60, respectively). The RF algorithm had the highest overall accuracy (more than 92%) for all farms. Multitemporal image classification increased the accuracy of the algorithms. as it helped to correctly identify as SBP the areas covered by tree shadows and flower patches, which would be misclassified using single image classification. This study thus established the first workflow for SBP monitoring based on remotely sensed data, suggesting an operational approach for SBP identification. The workflow can be applied to other types of pastures in agroforestry regions to reduce the effects of shadows and flowering in classification problems.

In this paper, we propose a novel method for the joint classification of both multidate and multiresolution remote sensing imagery, which represents an important and relatively unexplored classification problem. The proposed classifier is based on an explicit hierarchical graph-based model that is sufficiently flexible to address a coregistered time series of images collected at different spatial resolutions. Within this framework, a novel element of the proposed approach is the use of multiple quadtrees in cascade, each associated with the images available at each observation date in the considered time series. For each date, the input images are inserted in a hierarchical structure on the basis of their resolutions, whereas missing levels are filled in with wavelet transforms of the images embedded in finer-resolution levels. This approach is aimed at both exploiting multiscale information, which is known to play a crucial role in high-resolution image analysis, and supporting input images acquired at different resolutions in the input time series. The experimental results are shown for multitemporal and multiresolution optical data. [ABSTRACT FROM PUBLISHER]

In this paper, we present a method for the multitemporal and contextual classification of georeferenced optical remote sensing images acquired at different epochs and having different geometrical resolutions. The method is based on Conditional Random Fields (CRFs) for contextual classification. The CRF model is expanded by temporal interaction terms that link neighboring epochs via transition probabilities between different classes. In order to be able to deal with data of different resolution, the class structure at different epochs may vary with the resolution. The goal of the multitemporal classification is an improved classification performance at all individual epochs, but also the detection of land-cover changes, possibly using lower resolution data. This paper also contains a comparison of the performance of different models for the interaction potentials. Results are given for two different test sites in Germany, where Ikonos, RapidEye, and Landsat images are available. Our results show that the multitemporal classification does indeed increase the overall accuracy of all epochs compared to a monotemporal classification and to a state-of-the-art multitemporal classification method, and that it is feasible to detect changes in lower resolution images. [ABSTRACT FROM PUBLISHER]

The amount of images used in multitemporal classification studies has greatly increased along with enhanced temporal sensor capacities. Handling large intra-annual time series leads to the question of how the selection of image acquisition dates could be optimized. In this study, an empirical approach for evaluating the relative classification power of single acquisition dates is introduced for the differentiation of seminatural grassland vegetation. The main question is how many acquisitions from which phenological origins are preferable to achieve a certain classification accuracy target. The tested time series contains 24 single RapidEye scenes from 2009 to 2011. The vegetation index composites of these images were iteratively classified into different combinations of acquisition dates using the support vector machine (SVM) algorithm. The subsequent results were tested for significant accuracy improvements over single acquisition dates. These acquisition dates are subsumed under phenological seasons to evaluate adequate temporal acquisition windows. The results show that a three-scene composite reaches more than 0.8 overall accuracy (OAA). The best tradeoff amount between number of acquisition dates and classification accuracy is achieved using a seven-scene NDVI composite. The most important season for the differentiation of seminatural grassland is early summer (ESu). Full spring (FuS), late summer (LSu), and midsummer (MSu) can also be identified as influential temporal windows for data acquisition. [ABSTRACT FROM PUBLISHER]

We present an adaptation algorithm focused on the description of the data changes under different acquisition conditions. When considering a source and a destination domain, the adaptation is carried out by transforming one data set to the other using an appropriate nonlinear deformation. The eventually nonlinear transform is based on vector quantization and graph matching. The transfer learning mapping is defined in an unsupervised manner. Once this mapping has been defined, the samples in one domain are projected onto the other, thus allowing the application of any classifier or regressor in the transformed domain. Experiments on challenging remote sensing scenarios, such as multitemporal very high resolution image classification and angular effects compensation, show the validity of the proposed method to match-related domains and enhance the application of cross-domains image processing techniques. [ABSTRACT FROM AUTHOR]

Many remote sensing data processing problems are inherently constituted by several tasks that can be solved either individually or jointly. For instance, each image in a multitemporal classification setting could be taken as an individual task. Here, the relation to previous acquisitions should be properly considered because of the nonstationary behavior of temporal, spatial, and angular image features which gives rise to distribution changes. This phenomenon is known as covariate shift. Additionally, when labeled data are scarce or expensive to obtain, the small sample-set problem arises, which makes solving the problems independently in each domain difficult. Multitask learning (MTL) aims at jointly solving a set of prediction problems by sharing information across tasks. This paper introduces MTL in remote sensing data classification. The proposed methods alleviate the data set shift by imposing cross-information in the classifiers through matrix regularization. We consider the support vector machine (SVM) as the core learner and two different regularization schemes: 1) the inclusion of relational operators between tasks and 2) the pairwise Euclidean distance of the predictors in the Hilbert space. These methods rely on simple and intuitive modifications of the kernel used in the standard SVM. Experiments are conducted in three challenging remote sensing problems: cloud screening from multispectral images, land-mine detection using radar data, and multitemporal and multisource image classification. The pairwise method consistently outperforms standard independent and aggregate approaches by about +2% to 4% in all problems at no additional cost. Also, the solutions found give us information about the distribution shift among tasks. [ABSTRACT FROM AUTHOR]

Abstract: Soil surface characteristics (SSC) constitute an important land surface property that drives the partitioning between infiltration and runoff. Therefore, knowledge of SSC is crucial for runoff-forecasting in hydrology. However, the difficulties in measuring spatial variabilities and temporal dynamics of SSC have limited the use of this property in operational hydrology at the catchment extent. Recent progresses have permitted to characterize hydrological SSC classes (H-SSC) with distinct infiltration rates, by implementing monotemporal classifications along with aerial photos. However, when dealing with Mediterranean vineyards, some classes still are difficult to discriminate on the basis of remotely sensed spectral and spatial information only. The objective of the current study was to propose a multitemporal classification that integrates a priori information about possible H-SSC evolutions, such as it is possible improving their characterization. H-SSC evolutions could be either natural, depending on rainfall events, or anthropogenic, driven by soil management practices. Knowledge of possible H-SSC evolutions was translated in the form of decision rules. It was applied to a time series of H-SSC class maps derived from a monotemporal classification of monthly aerial photos. As compared to the monotemporal classification, the multitemporal classification had two advantages for the identification of H-SSC classes. First, it allowed improving the discrimination of classes related to crusting processes, with increased performances between 35 and 48% relative. Second, it was able to detect H-SSC temporal evolutions in relation to soil management practices. [Copyright &y& Elsevier]

Abstract: This paper proposes a new fuzzy cascade multitemporal classification method based on Fuzzy Markov Chains. This method differs from prior fuzzy multitemporal approaches proposed thus far, as the method does not require the knowledge of the true class at an earlier date; instead it uses the attributes of the image object being classified at the earlier date. This method combines the fuzzy, non-temporal, classification of a geographical region at two points in time to provide a single unified result. A transformation law based on class transition possibilities projects the earlier classification onto the later date before combining both results. Performance analysis was conducted upon a sequence of three LANDSAT images from the central region of Brazil using a Genetic Algorithm to estimate transition possibilities. The analysis showed that the increase in performance is highly dependent on whether or not a significant correlation exists between the temporal data sets, as well as on the accuracy of the monotemporal classifier at the earlier date. While the monotemporal approach used in the experiments attained an average class accuracy of approximately 55%, the multitemporal scheme achieved between 65% and 95%. Similar results in terms of overall accuracy were also observed. Furthermore, compared to two alternative cascade multitemporal classification approaches, the proposed method consistently showed better results. [Copyright &y& Elsevier]

Recognition of invasive species and their distribution is key for managing and protecting native species within both natural and man-made ecosystems. Small woody features (SWF) represent fragmented patches or narrow linear tree features that are of high importance in intensively utilized agricultural landscapes. Simultaneously, they frequently serve as expansion pathways for invasive species such as black locust. In this study, Sentinel-2 products, combined with spatiotemporal compositing approaches, are used to address the challenge of broad area black locust mapping at a high granularity. This is accomplished by conducting a comprehensive analysis of the classification performance of various compositing approaches and multitemporal classification settings throughout four vegetation seasons. The annual, seasonal (bi-monthly), and monthly median values of cloud-masked Sentinel-2 reflectance products are aggregated and stacked into varied time-series datasets per given year. The random forest algorithm is trained and output classification maps validated based on field-based reference datasets across Danubian lowlands (Slovakia). The main results of the study proved the usefulness of spatiotemporal compositing of Sentinel-2 products for mapping black locust in small woody features across wide area. In particular, temporally aggregated monthly composites stacked to seasonal time series datasets yielded consistently high overall accuracies ranging from 89.10% to 91.47% with balanced producer's and user's accuracies for each year's annual series. We presume that a similar approach could be used for a broader scale species distribution mapping, assuming they are spectrally or phenologically distinctive, as is often the case for many invasive species. [ABSTRACT FROM AUTHOR]

The multitemporal classification of remote sensing images is a challenging problem, in which the efficient combination of different sources of information (e.g., temporal, contextual, or multisensor) can improve the results. In this paper, we present a general framework based on kernel methods for the integration of heterogeneous sources of information. Using the theoretical principles in this framework, three main contributions are presented. First, a novel family of kernel-based methods for multitemporal classification of remote sensing images is presented. The second contribution is the development of nonlinear kernel classifiers for the well-known difference and ratioing change detection methods by formulating them in an adequate high-dimensional feature space. Finally, the presented methodology allows the integration of contextual information and multisensor images with different levels of nonlinear sophistication. The binary support vector (SV) classifier and the one-class SV domain description classifier are evaluated by using both linear and nonlinear kernel functions. Good performance on synthetic and real multitemporal classification scenarios illustrates the generalization of the framework and the capabilities of the proposed algorithms. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of detecting land-cover transitions by analysing multitemporal remote-sensing images. In order to develop an effective system for the detection of land-cover transitions, an ensemble of non-parametric multitemporal classifiers is defined and integrated in the context of a multiple classifier system (MCS). Each multitemporal classifier is developed in the framework of the compound classification (CC) decision rule. To develop as uncorrelated as possible classification procedures, the estimates of statistical parameters of classifiers are carried out according to different approaches (i.e., multilayer perceptron neural networks, radial basis functions neural networks, and k-nearest neighbour technique). The outputs provided by different classifiers are combined according to three standard strategies extended to the compound classification case (i.e., Majority voting, Bayesian average, and Bayesian weighted average). Experiments, carried out on a multitemporal remote-sensing data set, confirm the effectiveness of the proposed system. [Copyright &y& Elsevier]

A partially unsupervised approach to the classification of multitemporal remote-sensing images is presented. Such an approach allows the automatic classification of a remote-sensing image for which training data are not available, drawing on the information derived from an image acquired in the same area at a previous time. In particular, the proposed technique is based on a cascade-classifier approach and on a specific formulation of the expectation-maximization (EM) algorithm used for the unsupervised estimation of the statistical parameters of the image to be classified. The results of experiments carried out on a multitemporal data set confirm the validity of the proposed approach. [Copyright &y& Elsevier]

Sown Biodiverse Pastures (SBP) are the basis of a high-yield grazing system tailored for Mediterranean ecosystems and widely implemented in Southern Portugal. The application of precision farming methods in SBP requires cost-effective monitoring using remote sensing (RS). The main hurdle for the remote monitoring of SBP is the fact that the bulk of the pastures are installed in open Montado agroforestry systems. Sparsely distributed trees cast shadows that hinder the identification of the underlaying pasture using Unmanned Aerial Vehicles (UAV) imagery. Image acquisition in the Spring is made difficult by the presence of flowers that mislead the classification algorithms. Here, we tested multiple procedures for the geographical, object-based image classification (GEOBIA) of SBP, aiming to reduce the effects of tree shadows and flowers in open Montado systems. We used remotely sensed data acquired between November 2017 and May 2018 in three Portuguese farms. We used three machine learning supervised classification algorithms: Random Forests (RF), Support Vector Machine (SVM) and Artificial Neural Networks (ANN). We classified SBP based on: (1) a single-period image for the maximum Normalized Difference Vegetation Index (NDVI) epoch in each of the three farms, and (2) multi-temporal image stacking. RF, SVM and ANN were trained using some visible (red, green and blue bands) and near-infrared (NIR) reflectance bands, plus NDVI and a Digital Surface Model (DSM). We obtained high overall accuracy and kappa index (higher than 79% and 0.60, respectively). The RF algorithm had the highest overall accuracy (more than 92%) for all farms. Multitemporal image classification increased the accuracy of the algorithms. as it helped to correctly identify as SBP the areas covered by tree shadows and flower patches, which would be misclassified using single image classification. This study thus established the first workflow for SBP monitoring based on remotely sensed data, suggesting an operational approach for SBP identification. The workflow can be applied to other types of pastures in agroforestry regions to reduce the effects of shadows and flowering in classification problems. [ABSTRACT FROM AUTHOR]