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.
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 São 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.
•A novel approach is proposed for integrated-crop livestock system (ICLS) mapping.•PlanetScope time series and deep learning effectively map ICLS.•Vegetation indices are powerful predictors for ICLS mapping.•ICLS mapping models can be transferred across years to avoid sample collection.
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.
Here we model and describe the wood volume of Cerrado Sensu Stricto, a highly heterogeneous vegetation type in the Savanna biome, in the state of Minas Gerais, Brazil, integrating forest inventory ...data with spatial-environmental variables, multivariate regression, and regression kriging. Our study contributes to a better understanding of the factors that affect the spatial distribution of the wood volume of this vegetation type as well as allowing better representation of the spatial heterogeneity of this biome. Wood volume estimates were obtained through regression models using different environmental variables as independent variables. Using the best fitted model, spatial analysis of the residuals was carried out by selecting a semivariogram model for generating an ordinary kriging map, which in turn was used with the fitted regression model in the regression kriging technique. Seasonality of both temperature and precipitation, along with the density of deforestation, explained the variations of wood volume throughout Minas Gerais. The spatial distribution of predicted wood volume of Cerrado Sensu Stricto in Minas Gerais revealed the high variability of this variable (15.32 to 98.38 m3 ha-1) and the decreasing gradient in the southeast-northwest direction.
The recent advances in unmanned aerial vehicle (UAV)-based remote sensing systems have broadened the remote sensing applications for agriculture. Despite the great possibilities of using UAVs to ...monitor agricultural fields, specific problems related to missing parts in UAV orthomosaics due to drone flight restrictions are common in agricultural monitoring, especially in large areas. In this study, we propose a methodological framework to impute missing parts of UAV orthomosaics using PlanetScope (PS) and Sentinel-2 (S2) data and the random forest (RF) algorithm of an integrated crop–livestock system (ICLS) covered by grass at the time. We validated the proposed framework by simulating and imputing artificial missing parts in a UAV orthomosaic and then comparing the original data with the model predictions. Spectral bands and the normalized difference vegetation index (NDVI) derived from PS, as well as S2 images (separately and combined), were used as predictor variables of the UAV spectral bands and NDVI in developing the RF-based imputation models. The proposed framework produces highly accurate results (RMSE = 6.77–17.33%) with a computationally efficient and robust machine-learning algorithm that leverages the wealth of empirical information present in optical satellite imagery (PS and S2) to impute up to 50% of missing parts in a UAV orthomosaic.
•UAV remote sensing is a feasible approach to estimate the aboveground biomass of herbaceous crops.•GLCM textures improved pasture estimates addressing internal image patterns.•Random Forest (RF) ...model solely with texture measures reached the best performance.•Selecting features by correlation improved the RF models' results and understanding.
Pasture production under integrated crop-livestock systems (ICLS) is a key element to support grazing management decisions. Therefore, further experimentation is required to produce reliable estimates of pasture productivity. An unmanned aerial vehicle (UAV) is a viable tool to obtain fast and accurate aboveground biomass (AGB) estimates in pastures under ICLS. We tested several datasets of variables composed of original spectral bands (RGB-NIR), vegetation indices, and gray-level cooccurrence matrix (GLCM) textures to estimate pasture AGB using the random forest (RF) algorithm and feature selection methods in a commercial ICLS farm in western São Paulo State, Brazil. Field measures of pasture AGB were carried out in three field campaigns over five months to capture the spatiotemporal variability of the pasture fields. Most tested models reached similar results on pasture AGB estimates (R2: 0.60 to 0.70), while the number of variables selected for the RF algorithm differed among models (from 6 to 160 variables). The three more accurate models used few variables, two of which used only texture measures, while the third also employed combined spectral bands and vegetation indices. The best model (R2 = 0.70) used only 10 texture measures. Texture measures that represented images’ inner patterns, calculated from NIR, red-edge, and triangular greenness index data, were the most relevant variables for the main models. The texture contribution indicates important information to be considered to estimate pasture AGB when using UAV imagery. The results achieved by UAV estimates allow the design of multitemporal AGB pasture maps, which may be useful for building more reliable spatiotemporal data.
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.
Detailed mapping of soil attributes is often not viable due to the high cost of wet‐chemical laboratory analysis, which requires a large number of samples. Thus, we evaluated whether the prediction ...of SOC contents through field‐specific diffuse reflectance spectroscopy (DRS) can increase the amount of samples available to SOC mapping through data interpolation. For such, we tested the performance of the partial least squares regression (PLSR), random forest (RF) and gradient boosting tree (GBT) algorithms to model and predict SOC. The field‐specific calibration approach proposed here proved to be suitable for predicting SOC content on soil samples, reducing the dependence on wet‐chemical soil laboratory analyses for mapping. With such SOC content prediction, the higher amount of samples to be used for spatial interpolation can be increased, leading to more accurate SOC maps that can be applied for site‐specific management.
Understanding the spatial pattern of a particular geographic phenomenon such as deforestation is a key issue to establish monitoring programs to prevent the depletion of natural resources. Thus, the ...goal of this study was to assess the spatial pattern of deforested areas in the Pardo and Jequitinhonha River basins using Ripley's K function. First, we mapped all deforested areas in these basins using Landsat multispectral imagery from 2007 to 2015. Then, we used the Ripley's K function to test for spatial interactions between deforestation events. Our results showed that deforestations predominantly occur in a clustering spatial pattern in these basins. Spatial statistical analyses as Ripley's K function may provide a baseline for deforestation monitoring, as well as allowing us to understand the spatial pattern of deforestation in different natural ecosystems, especially in countries like Brazil, where the territorial dimension presents a great difficulty for the effectiveness of deforestation monitoring.
In this study, we tested the effectiveness of stand age, multispectral optical imagery obtained from the Landsat 8 Operational Land Imager (OLI), synthetic aperture radar (SAR) data acquired by the ...Sentinel-1B satellite, and digital terrain attributes extracted from a digital elevation model (DEM), in estimating forest volume in 351 plots in a 1,498 ha Eucalyptus plantation in northern Minas Gerais state, Brazil. A Random Forest (RF) machine learning algorithm was used following the Principal Component Analysis (PCA) of various data combinations, including multispectr al and SAR texture variables and DEM-based geomorphometric derivatives. Using multispectral, SAR or DEM variables alone (i.e. Experiments (ii)-(iv)) did not provide accurate estimates of volume (RMSE (Root Mean Square Error) > 32.00 m
3
ha
−1
) compared to predictions based on age since planting of Eucalyptus stands (Experiment (i)). However, when these datasets were individually combined with stand age (i.e. Experiments (v)-(vii)), the RF models resulted in better volume estimates than those obtained when using the individual multispectral, SAR and DEM datasets (RMSE < 28.00 m
3
ha
−1
). Furthermore, a model that integrated the selected variables of these data with stand age (Experiment (viii)) improved volume estimation significantly (RMSE = 22.33 m
3
ha
−1
). The large and increasing area of Eucalyptus forest plantations in Brazil and elsewhere suggests that this new approach to volume estimation has the potential to support Eucalyptus plantation monitoring and forest management practices.