This study presents a methodology for generating high-quality datasets of samples to train machine learning models for land use and land cover (LULC) classification, representing an initial step toward large-scale LULC mapping. The proposed approach generates a robust set of multi-pixel, multi-temporal, and multi-band samples by integrating label generation with remote sensing data processing across two phases. In the first phase, multidimensional data are collected from remote sensing sources, and label information is derived by comparing harmonized LULC maps, supplemented with additional processing to improve data quality. In the second phase, samples are extracted to produce a diverse, representative, and balanced dataset of reliably labeled multidimensional data for LULC classification. In this study, reference maps were sourced from the TerraClass and MapBiomas projects, and satellite imagery was obtained from the Brazil Data Cube platform.

The Upper Paraguay Basin (BAP, Portuguese acronym) is an ecosystem of strategic importance to Brazil because it includes the Pantanal, one of the world's largest floodplains. The wet plain is well preserved, but the plateau suffered the impact of anthropic activities on the natural resources, due to the rapid conversion of vegetation cover in recent decades. Therefore, it is essential to investigate land use and cover changes to define strategies that stimulate environmental conservation and regional development. The BAP plateau region is covered by the mappings of the TerraClass Project, which provides official data for monitoring land use and cover in the Amazon and Cerrado biomes. In this context, the objective of the work was to analyze changes in land use and land cover in municipalities of the BAP, using the modern GeoPortal TerraClass and the recent data collection from 2018, 2020, and 2022. The spatial analysis showed that the native forest loss occurred due to increased deforestation. Most of the forest regeneration areas persisted during this period, and the lost area was converted into pastures, where agriculture expanded, initially with one cycle production, followed by the intensification of crop and off-season production. The results demonstrate the potential of the data and tools offered by the Digital Platform TerraClass to support territorial management.

Integrated crop–livestock systems (ICLS) are among the main viable strategies for sustainable agricultural production. Mapping these systems is crucial for monitoring land use changes in Brazil, playing a significant role in promoting sustainable agricultural production. Due to the highly dynamic nature of ICLS management, mapping them is a challenging task. The main objective of this research was to develop a method for mapping ICLS using deep learning algorithms applied on Satellite Image Time Series (SITS) data cubes, which consist of Sentinel-2 (S2) and PlanetScope (PS) satellite images, as well as data fused (DF) from both sensors. This study focused on two Brazilian states with varying landscapes and field sizes. Targeting ICLS, field data were combined with S2 and PS data to build land use and land cover classification models for three sequential agricultural years (2018/2019, 2019/2020, and 2020/2021). We tested three experimental settings to assess the classification performance using S2, PS, and DF data cubes. The test classification algorithms included Random Forest (RF), Temporal Convolutional Neural Network (TempCNN), Residual Network (ResNet), and a Lightweight Temporal Attention Encoder (L-TAE), with the latter incorporating an attention-based model, fusing S2 and PS within the temporal encoders. Experimental results did not show statistically significant differences between the three data sources for both study areas. Nevertheless, the TempCNN outperformed the other classifiers with an overall accuracy above 90% and an F1-Score of 86.6% for the ICLS class. By selecting the best models, we generated annual ICLS maps, including their surrounding landscapes. This study demonstrated the potential of deep learning algorithms and SITS to successfully map dynamic agricultural systems.

This study examines how land tenure constrains Brazil’s ability to meet its deforestation control and forest restoration goals in its Amazonia biome. Our findings are based on an updated assessment of land tenure and land use in the region. Between 2019 and 2021, 44% of deforestation in Amazonia occurred in private lands, while forest removal in settlements ranged from 31% to 27% of the total. Deforestation in undesignated public lands increased from 11% in 2008 to 18% in 2021. Deforestation is highly concentrated, with 1% of properties accounting for 82.5% of forest cuts in 2021. In Amazonia, there is considerable non-compliance with the legal reserve provisions set by Brazil’s Forest Code. Legal reserve deficits in private lands sum up to 18.17 Mha (million hectares), compared with 12.49 Mha of legal reserve surpluses. Even if all forest surpluses are offered in the forest credits market set in the Forest Code, farmers still need to restore 5.67 Mha to comply with the law. Large-scale cattle ranchers have a legal reserve deficit of 10.35 Mha (34% of their area). Most crop farming occurs in medium and large properties (4.63 Mha) with a large proportion of legal reserve deficits (45%). Given the political power and financial resources of large ranchers and crop producers, Brazil faces major challenges in inducing these farmers to meet their legal obligations. Therefore, Brazil needs to combine robust command-and-control strategies with market-based policies to achieve its deforestation and forest restoration goals. The government should tailor forest protection and restoration policies to the needs of different landowners, considering their land use practices, technical capacity, and financial resources.

Accurate mapping of crops with high spatiotemporal resolution plays a critical role in achieving the Sustainable Development Goals (SDGs), especially in the context of integrated crop-livestock systems (ICLS). Stakeholders can make informed decisions and implement targeted strategies to achieve multiple SDGs related to agriculture, rural development, and sustainable livelihoods by understanding the spatial dynamics of these systems. Accurate information on the extent of ICLS derived from multitemporal remote sensing and emerging map techniques such as deep learning can help in the implementation of sustainable agricultural practices. However, far too little attention has been paid to ICLS map accuracy because it may not be at the forefront of research agendas compared to those of other agricultural practices. This paper aims to map ICLS using high spatiotemporal resolution imagery and deep learning neural network classifiers at two different sites located in Brazil. The pipeline involves four interpretation approaches based on the ICLS class: evaluating deep neural network classifiers with different image composition intervals, explaining commission and omission errors, evaluating the temporal transferability of the method, and evaluating the influence of variables. The study area consists of two locations in Sa similar to o Paulo (study site 1, SS1) and Mato Grosso state (study site 2, SS2), Brazil. We derived nine spectral variables from PlanetScope (PS) images and four metrics through object-based image analysis (OBIA) using two time intervals, 10 and 15 days, to generate the image compositions. These input variables were used in three deep neural network classifiers: convolutional neural network in one dimension (Conv1D), long short-term memory (LSTM), and LSTM with a fully convolutional network (LSTM-FCN). Our results showed that mapping dynamic land use such as ICLS is possible by using high-spatiotemporal-resolution imagery and deep neural network classifiers. The 15-day LSTM-FCN classifier returned the highest map accuracies for both sites, with the following class-level accuracies: producer accuracy (PA) = 97.0% and user accuracy (UA) = 97.0% for SS1 and PA = 82.0% and UA = 96.5% for SS2. Meanwhile, we found map uncertainties arising from the diverse crop calendars and spectro-temporal similarities between ICLS and other land use. The best approaches revealed that temporal generalization was suitable for mapping ICLS, but some classifiers could not generalize due to the inherent characteristics of the class. Most variables were considered efficient for predicting ICLS, although spectral indices revealed better functional relationships, while the PS bands had a lower influence on the predictions. The accuracies achieved with the proposed method represent promising opportunities for the sufficiently accurate mapping of ICLS and other complex crop activities.

Regenerative agricultural practices are a suitable path to feed the global population. Integrated Crop–livestock systems (ICLSs) are key approaches once the area provides animal and crop production resources. In Brazil, the expectation is to increase the area of ICLS fields by 5 million hectares in the next five years. However, few methods have been tested regarding spatial and temporal scales to map and monitor ICLS fields, and none of these methods use SAR data. Therefore, in this work, we explored the potential of three machine and deep learning algorithms (random forest, long short-term memory, and transformer) to perform early-season (with three-time windows) mapping of ICLS fields. To explore the scalability of the proposed methods, we tested them in two regions with different latitudes, cloud cover rates, field sizes, landscapes, and crop types. Finally, the potential of SAR (Sentinel-1) and optical (Sentinel-2) data was tested. As a result, we found that all proposed algorithms and sensors could correctly map both study sites. For Study Site 1(SS1), we obtained an overall accuracy of 98% using the random forest classifier. For Study Site 2, we obtained an overall accuracy of 99% using the long short-term memory net and the random forest. Further, the early-season experiments were successful for both study sites (with an accuracy higher than 90% for all time windows), and no significant difference in accuracy was found among them. Thus, this study found that it is possible to map ICLSs in the early-season and in different latitudes by using diverse algorithms and sensors.

Land use and land cover (LULC) classification has long been an essential topic in Earth Observation research and plays a key role in the sustainable development of agriculture. This study evaluated the accuracy of LULC classification based on an initial clustering step in a heterogeneous agricultural landscape using PlanetScope imagery while checking for variability among their Normalized Difference Vegetation Index (NDVI) temporal signatures. We adopt an object-based image analysis to generate image-objects and then extract statistical information of PlanetScope spectral bands and vegetation indices as input information for classification. The exploratory analysis focused on the double crop class and calculated the distance between NDVI temporal signatures of paired land parcels. We applied an unsupervised clustering technique along with Random Forest algorithm based on multiple tests to classify and analyse gains and losses in accuracies produced by these approaches. Our results showed that the initial clustering method outperformed the non-clustered classification of LULC in overall accuracy measures. The exploratory analysis demonstrated that double crops might present high intra-class variability and diverse crop calendars for neighbour land parcels. The accuracies achieved represent promising opportunities for the sufficiently accurate classification of such areas, and the knowledge of the intra-class variability allows the analyst to infer the temporal dynamics of crop fields. We reinforce that further work could assess other types of classifiers, especially in areas with a large number of crop types and distinct management practices.

In its Nationally Determined Contribution (NDC) to the United Nations Framework Convention on Climate Change, Brazil committed to reducing greenhouse gas emissions and restoring its forests. This study examines the challenges of fulfilling these commitments in Brazilian Amazonia. We carry out a detailed assessment of the current status of land tenure in the region and its relation to deforestation. After dealing with conflicts and overlaps between data from various sources, we produce a new map of public and private land tenure in Amazonia. Combining this map with Brazil's official data on deforestation, we find out how much natural vegetation has been preserved in each public or private area. The result is used to estimate how much deforestation is illegal. We also establish how much deforestation is associated with each land tenure type. Our results show that most deforestation inside rural properties is done by a few landowners, a finding that has important consequences for law enforcement. We then assess the challenges for reforestation in detail. To do so, we consider how much forest needs to be rehabilitated according to Brazil's Forest Code. Our analysis provides a comprehensive appraisal of the potential opportunity costs for forest restoration in the biome, considering farm size and land use. This analysis provides insights into targeted land use policies that can meet Brazil’s forest restoration goals.

With the recent evolution in the sensor's spatial resolution, such as the MultiSpectral Imager (MSI) of the Sentinel-2 mission, the need to use segmentation techniques in satellite images has increased. Although the advantages of image segmentation to delineate agricultural fields in images are already known, the literature shows that it is rarely used to consider temporal changes in highly managed regions with the intensification of agricultural activities. Therefore, this work aimed to evaluate a multitemporal segmentation method based on the coefficient of variation of spectral bands and vegetation indices obtained from Sentinel-2 images, considering two agricultural years (2018–2019 and 2019–2020) in an area with agricultural intensification. Images of the coefficient of variation represented the spectro-temporal dynamics within the study area. These images were also used to apply an edge detection filter (Sobel) to verify their performance. The region-based algorithm Watershed Segmentation (WS) was used in the segmentation process. Subsequently, to assess the quality of the segmentation results produced, the metrics Potential Segmentation Error (PSE), Number-of-Segments Ratio (NSR), and Euclidean Distance 2 (ED2) were calculated from manually delineated reference objects. The segmentation achieved its best performance when applied to the unfiltered coefficient of variation images of spectral bands with an ED2 equal to 7.289 and 2.529 for 2018–2019 and 2019–2020, respectively. There was a tendency for the WS algorithm to produce over-segmentation in the study area; however, its use proved to be effective in identifying objects in a dynamic area with the intensification of agricultural activities.

Various approaches were developed considering the need to increase agricultural productivity in cultivated areas without more deforestation, such as the Integrated Crop livestock systems (ICLS). The ICLS could be composed of annual crops followed by pastureland with the presence of cattle. Due to the high temporal dynamic of rotation between crops over the season, monitoring these areas is a big challenge. Also, agricultural organizations worldwide highlight the need for early-season maps for this kind of work. In this context, this study evaluated the potential of open data (Sentinel-2) data to map ICLS areas. The performance of two classifiers was evaluated: one of Machine Learning (random forest) and the other of Deep Learning (LSTM). Three different time windows of data were tested (Entire season, 180 days, and 120 days). Using the RF classifier, it was possible to achieve satisfactory results (Overall accuracy higher than 80%) for the early season (180 days). However, further studies are needed to explain better the lower(when compared to Random Forest) accuracy achieved by LSTM net (0.79 % for 180 days) and compare the results achieved here with results for a study area with different rates of cloud cover.

Mapping highly dynamic cropping systems using satellite image time series is still challenging even when robust approaches are used. We assessed the potential of using high spatial and temporal resolution PlanetScope time series and deep neural networks (Convolutional Neural Networks (CNN) in one dimension - Conv1D, Long Short-Term Memory (LSTM), and Multi-Layer Perceptron (MLP)) for mapping integrated crop-livestock systems (ICLS) and different land covers in the western region of São Paulo State, Brazil. We used 10-day and 15-day composite EVI and NDVI time series (both individually and combined) as input data in the neural network classifiers. Conv1D using both EVI and NDVI 10 day-composite time series outperformed the other classifiers evaluated in this study (LSTM and MLP), allowing improved discrimination of land parcels with ICLS in our study area.

Fast and accurate quantification of the available pasture biomass is essential to support grazing management decisions in intensively managed fields. The increasing temporal and spatial resolutions offered by the new generation of orbital platforms, such as Planet CubeSat satellites, have improved the capability of monitoring pasture biomass using remotely sensed data. Here, we assessed the feasibility of using spectral and textural information derived from PlanetScope imagery for estimating pasture aboveground biomass (AGB) and canopy height (CH) in intensively managed fields and the potential for enhanced accuracy by applying the extreme gradient boosting (XGBoost) algorithm. Our results demonstrated that the texture measures enhanced AGB and CH estimations compared to the performance obtained using only spectral bands or vegetation indices. The best results were found by employing the XGBoost models based only on texture measures. These models achieved moderately high accuracy to predict pasture AGB and CH, explaining 65% and 89% of AGB (root mean square error (RMSE) = 26.52%) and CH (RMSE = 20.94%) variability, respectively. This study demonstrated the potential of using texture measures to improve the prediction accuracy of AGB and CH models based on high spatiotemporal resolution PlanetScope data in intensively managed mixed pastures.

Time series of satellite images have been widely used in a range of applications, specially involving the green biomass monitoring in the Earth surface. A set of land surface products, including images of vegetation indices (VI), is derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor since 2000. This paper reports the development of the Temporal Vegetation Analysis System (SATVeg), a free web-based tool designed to provide instantaneous access to temporal profiles of MODIS VI in South America. The system's architecture is based on a geospatial database specially modeled to store the time series of images and ensure instantaneous queries and efficient updating. The results of the geospatial database modelling and respective response time for queries and updates, as well as the main functionalities of the system, are presented and discussed. Instantaneous reconstruction of time series provided by a set of filtering procedures is also related. Application examples of VI temporal profiles generated by SATVeg are presented and the system's limitations are also discussed. Our results show the potential of the web-system for supporting a set of land use and land cover monitoring activities in South America through a simple and friendly user interface.

Brazil contains two-thirds of remaining Amazonian rainforests and is responsible for the most Amazon forest loss. Primary forest loss in the Brazilian Amazon has declined considerably since 2004 but secondary forest loss has never been quantified. We use a recently developed high-resolution land use/land cover dataset to track secondary forests in the Brazilian Amazon over 14 yr, providing the first estimates of secondary forest loss for the region. We find that secondary forest loss increased by (187 ± 48)% from 2008 to 2014. Moreover, the proportion of total forest loss accounted for by secondary forests rose from (37 ± 3)% in 2000 to (72 ± 5)% in 2014. The recent acceleration in secondary forests loss occurred across the entire region and was not driven simply by increasing secondary forest area but probably a conscious preferential shift towards clearance of a little-protected forest ecosystem (secondary forests). Our results suggest that secondary forests loss has eased deforestation pressure on primary forests. However, this has been at the expense of a lost carbon sequestration opportunity of 2.59–2.66 Pg C over our study period. Deforestation in the Brazilian Amazon affects both older (primary) and younger (secondary) forests. This study finds that most forest loss over the period 2008–2014 was from secondary forests and that the almost 190% rise in deforestation buffered losses from primary forests.

Resumo. As series temporais de dados multiespectrais do sensor MODerate Resolution Imaging Spectroradiometer (MODIS) permitem gerar perfis temporais de indices vegetativos para realizacao de analises. O Sistema de Analise Temporal da Vegetacao (SATVeg), desenvolvido pela Embrapa Informatica Agropecuaria, permite ao usuario final gerar esses perfis utilizando diferentes parâmetros, pre-filtros e filtros. A API SATVeg foi desenvolvida e incorporada a plataforma de APIs da Embrapa (AgroAPI) com o intuito de disponibilizar os dados fornecidos pelo SATVeg em um formato passivel de ser utilizado por empresas que desejam incorporar esses dados em seus sistemas proprietarios e agregar valor ao seu produto final.

Pasture biomass information is essential to monitor forage resources in grazed areas, as well as to support grazing management decisions. The increasing temporal and spatial resolutions offered by the new generation of orbital platforms, such as Planet CubeSat satellites, have improved the capability of monitoring pasture biomass using remotely-sensed data. In a preliminary study, we investigated the potential of spectral variables derived from PlanetScope imagery to predict pasture biomass in an area of Integrated Crop-Livestock System (ICLS) in Brazil. Satellite and field data were collected during the same period (May - August 2019) for calibration and validation of the relation between predictor variables and pasture biomass using the Random Forest (RF) regression algorithm. We used as predictor variables 24 vegetation indices derived from PlanetScope imagery, as well as the four PlanetScope bands, and field management information. Pasture biomass ranged from approximately 24 to 656 g.m -2 , with a coefficient of variation of 54.96%. Near Infrared Green Simple Ratio (NIR/Green), Green Leaf Algorithm (GLA) vegetation indices and days after sowing (DAS) are among the most important variables as measured by the RF Variable Importance metric in the best RF model predicting pasture biomass, which resulted in Root Mean Square Error (RMSE) of 52.04 g.m -2 (32.75%). Accurate estimates of pasture biomass using spectral variables derived from PlanetScope imagery are promising, providing new insights into the opportunities and limitations related to the use of PlanetScope imagery for pasture monitoring.

Agricultural expansion in Brazil is still intense for commodities (such soybeans and corn), mostly cultivated over large portions of the Cerrado biome. Therefore, the development and application of techniques based on remote sensing to map crop areas at a regional level, in a dynamic and more precise way is urgently necessary. In this context, the objective of this study is the improvement of techniques for mapping soybean crops in Brazil, through an analysis of the Centro Goiano mesoregion of Goiás state (a core area of Cerrado), using a time series of Enhanced Vegetation Index (EVI) images provided by TERRA/MODIS orbital sensor, in a test period between 2002 and 2010. Despite their proven quality, MODIS EVI images already contain atmospheric interferences inherent to the acquisition process, such as the presence of clouds. Thus, a set of methods to minimize such artifacts was applied to the data of this study. In general, the methodological procedures comprise of (1) the application of the pixel reliability band aiming to remove pixels contaminated by clouds; (2) the use of contaminated pixel estimates (excluded from the time series); (3) application of an interpolation filter to fill the void pixels in each scene, obtaining continuous and smoothed spectral-temporal profiles for each land use classes; and (4) the classification of agricultural areas using a specific algorithm for crops in the Cerrado region of Goiás. The areas reconstituted in the images matched neighboring pixels, maintaining good coherence with the original data. Likewise, areas mapped with soybeans had a high correlation with official IBGE census data, with a global accuracy value of 78%, and Pearson Correlation coefficient of 0.64. The application of this technique to other imagery sensors (such as RapidEye, Landsat 8 and Sentinel 2) is highly encouraged due a better spatial and temporal resolution (when applied together in a temporal image cube), ensuring more efficient crop monitoring in Brazil.

This paper presents a new approach to deal with agricultural crop recognition using SVM (Support Vector Machine), applied to time series of NDVI images. The presented method can be divided into two steps. First, the Timesat software package is used to extract a set of crop features from the NDVI time series. These features serve as descriptors that characterize each NDVI vegetation curve, i.e., the period comprised between sowing and harvesting dates. Then, it is used an SVM to learn the patterns that define each type of crop, and create a crop model that allows classifying new series. The authors present a set of experiments that show the effectiveness of this technique. They evaluated their algorithm with a collection of more than 3000 time series from the Brazilian State of Mato Grosso spanning 4 years (2009-2013). Such time series were annotated in the field by specialists from Embrapa (Brazilian Agricultural Research Corporation). This methodology is generic, and can be adapted to distinct regions and crop profiles.

RESUMO. O estado de Mato Grosso passou por um rapido processo de conversao da cobertura vegetal nas ultimas decadas, ainda pouco estudado. O Projeto TerraClass tem sido responsavel pelo mapeamento do uso e cobertura da terra na porcao desflorestada da Amazonia Legal. Neste contexto, o objetivo do trabalho foi analisar as mudancas do uso e cobertura da terra no estado de Mato Grosso por meio do GeoPortal TerraClass, de 2004 a 2014. A classe tematica que mais ganhou espaco foi a Cultura Agricola Temporaria, com 22.714,67 km², representando um crescimento de 151,08%, ocupando principalmente areas da classe tematica Pastagem Cultivada Herbacea, evidenciando o processo de transicao do sistema de producao da pecuaria para o sistema de producao agricola no estado. Nesse periodo de 10 anos o desmatamento apresentou uma reducao significativa de 92,48%. Os resultados mostraram o potencial do GeoPortal TerraClass no apoio a gestao territorial.