Crop Mapping Research Articles

Integrating Sentinel-1 data and machine learning for effective paddy field monitoring in Cauvery Delta Zone, Tamil Nadu, India.

Paddy crop mapping is essential for agricultural monitoring, ensuring food security, and enhancing resource allocation. This study observes the Cauvery Delta Zone (CDZ), recognized as the rice bowl of Tamil Nadu and a crucial area for paddy farming in India. The study seeks to elucidate rice-growing trends over three years (2021-2023) by examining the regional variability of the Radar Vegetation Index (RVI) throughout a paddy crop growing season (June to September). A temporal examination of the RVI and the cross-polarization ratio (VH/VV) demonstrates a good correlation of 0.79, enhancing the comprehension of paddy crop dynamics. Additionally, machine learning algorithms such as random forest (RF), support vector machine (SVM), and decision tree (DT) are utilized on radar data in both VV and VH polarizations to improve the classification of paddy fields. The accuracy evaluation indicates that the RF algorithm exhibits superior performance, achieving accuracies of 86.72% in VH mode and 86.42% in VV mode. The results underscore the efficacy of integrating radar-based indices with machine learning methodologies for proficient agricultural surveillance. These findings offer essential assistance for enhancing crop yield, optimizing resource management, and enabling informed decision-making in the Cauvery Delta Zone.

Country-Scale Crop-Specific Phenology from Disaggregated PROBA-V

Large-scale crop phenology monitoring is essential for agro-ecosystem policy. Remote sensing helps track crop development but requires high-temporal and spatial resolutions. While datasets with both attributes are now available, their large-scale applications require significant resources. Medium-resolution data offer daily observations but lack detail for smaller plots. This study generated crop-specific phenomaps for mainland France (2016–2020) using PROBA-V data. A spatial disaggregation method reconstructed NDVI time series for individual crops within mixed pixels. Then, phenometrics were extracted from disaggregated PROBA-V and Sentinel-2 separately and compared to observed phenological stages. Results showed that PROBA-V-based phenomaps closely matched observations at regional level, with moderate accuracy at municipal level. PROBA-V demonstrated a higher detection rate than Sentinel-2, especially in cloudy periods, and successfully generated phenomaps before Sentinel-2B’s launch. The study highlights PROBA-V’s potential for operational crop monitoring, i.e., wheat heading and oilseed rape flowering, with performance comparable to Sentinel-2. PROBA-V outputs complement Sentinel-2: phenometrics cannot be generated at plot level but are efficiently produced at regional or national scales to study phenological gradients more easily than with Sentinel-2 and with similar accuracy. This approach could be extended to MODIS or SPOT-VGT, to generate historical phenological data, providing that a crop map is available.

Effects of the long-term rice expansion on ecosystem carbon budget in the typical agricultural area of Northeast China

Extensive rice expansion in northeast China has significantly altered land use and land cover (LULC) changes. However, the impact of long-term rice expansion on regional carbon budget dynamics remains unclear. A major obstacle in addressing this gap is the absence of agricultural LULC information with high spatiotemporal resolution. Here, we selected the Sanjiang Plain experienced dramatic rice expansion as the study area, presenting a framework that integrates high-resolution crop distribution data with a process-based model to quantify the impact of rice expansion on regional carbon budgets. Specifically, we employed a robust deep-learning network for crop mapping based on Landsat data, to reconstruct the spatiotemporal dynamics of rice expansion from 1985 to 2020. We then incorporated these long-term crop maps into the Denitrification-Decomposition (DNDC) model, to explore the effects of rice expansion on the regional carbon budget over this period. Analysis results showed that the rice planting area expanded by 708.64 km2/yr, resulting in more than a tenfold increase over the past 36 years. Spatially, rice planting expanded from the southwest to the northeast and from the interior to the exterior. This expansion has resulted in approximately 275 Mt. of CO2 and 6.92 Mt. of CH4 greenhouse gas emissions, altering the dynamics of the regional carbon budget and shifting the ecosystem from a carbon sink to a carbon source since 2016. Although the long-term expansion of rice increased soil respiration and CH4 emissions, it also enhanced soil carbon sequestration through agricultural management practices. These findings greatly enhance our understanding of the ecosystem carbon cycle's response to long-term agricultural LULC changes, providing more accurate data support and scientific evidence for developing low-carbon agricultural policies.

MSCPUnet: A multi-task neural network for plot-level crop classification in complex agricultural areas

High-precision mapping of agricultural crops in complex planting areas is a prerequisite for precision agricultural management. This paper first proposes a novel multi-task neural network, Multi-task Multi-Scale Convolutional Pooling Unet (MSCPUnet), for extracting vector plots. MSCPUnet is based on the Unet model and enhances performance through the incorporation of attention mechanisms, multi-scale pooling layers, and a multi-task learning approach for parallel processing. A ZiYuan-1 02D (ZY1E) satellite image collected from the He-Tao irrigation district in China is selected for plot extraction experiments, where MSCPUnet is compared with three other deep learning network variants. The Inter section over Union (IoU) and Accuracy metrics for the MSCPUnet model achieve the highest values, converging at 0.837 and 0.928, respectively. Leveraging this capability, a crop classification framework is proposed, which first extracts crop attributes from Sentinel-2 (S2) time series data. The plot information obtained from the MSCPUnet model is then combined with the area dominance method to assign crop attributes to vector plots, facilitating crop structure identification at the plot scale. Results indicate that this method significantly improves classification accuracy for fragmented farmlands, with overall accuracy (OA) rising to 91 % and Kappa coefficient increasing to 0.84 compared to a random forest classifier. This integrated crop classification approach has been validated in this study for high-precision crop mapping in complex planting areas.

Mapping Crop Types for Beekeepers Using Sentinel-2 Satellite Image Time Series: Five Essential Crops in the Pollination Services

Driven by the widespread adoption of deep learning (DL) in crop mapping with satellite image time series (SITS), this study was motivated by the recent success of temporal attention-based approaches in crop mapping. To meet the needs of beekeepers, this study aimed to develop DL-based classification models for mapping five essential crops in pollination services in Quebec province, Canada, by using Sentinel-2 SITS. Due to the challenging task of crop mapping using SITS, this study employed three DL-based models, namely one-dimensional temporal convolutional neural networks (CNNs) (1DTempCNNs), one-dimensional spectral CNNs (1DSpecCNNs), and long short-term memory (LSTM). Accordingly, this study aimed to capture expert-free temporal and spectral features, specifically targeting temporal features using 1DTempCNN and LSTM models, and spectral features using the 1DSpecCNN model. Our findings indicated that the LSTM model (macro-averaged recall of 0.80, precision of 0.80, F1-score of 0.80, and ROC of 0.89) outperformed both 1DTempCNNs (macro-averaged recall of 0.73, precision of 0.74, F1-score of 0.73, and ROC of 0.85) and 1DSpecCNNs (macro-averaged recall of 0.78, precision of 0.77, F1-score of 0.77, and ROC of 0.88) models, underscoring its effectiveness in capturing temporal features and highlighting its suitability for crop mapping using Sentinel-2 SITS. Furthermore, applying one-dimensional convolution (Conv1D) across the spectral domain demonstrated greater potential in distinguishing land covers and crop types than applying it across the temporal domain. This study contributes to providing insights into the capabilities and limitations of various DL-based classification models for crop mapping using Sentinel-2 SITS.

Optimum Combination of Spectral Variables for Crop Mapping in Heterogeneous Landscapes based on Sentinel-2 Time Series and Machine Learning

Abstract. This article aimed to determine a workflow for more efficient large-scale crop mapping using a time series of images from the Sentinel-2 Satellite, statistical methods of attribute selection, and machine learning. The proposed methodology explores the best possible combination of spectral variables related to vegetation (16 vegetation indices in the RGB, NIR, SWIR, and Red Edge regions) to characterize different spectro-temporal profiles of Land Use and Land Cover (LULC) in spatially heterogeneous landscapes. First, we applied a data dimensionality reduction analysis using the PCA (Principal Component Analysis) method. Subsequently, the variables that showed the highest statistical correlation between each other were used in the spectro-temporal classification process, using the Random Forest, TempCNN, and LightTAE algorithms, following three different strategies: C1 (ALL), C2 (BE + IV (Red Edge)) and C3 (BE + IV (without Red Edge)), where ALL – All variables; BE – Spectral Bands; IV – Vegetation Indices. Given the results found, the C2 classification scenario (with bands B3, B4, B5, B6, B7, B8, and B8A and the NDRE1, RESI, and MSR indexes) demonstrated the best LULC classification accuracy at the crop pattern level, compared to the other scenarios, with average values of 0.91, 0.88, 0.91, 0.89, and 0.89 (Global Accuracy, Producer Accuracy, User Accuracy, Kappa index, and F1-Score, respectively, for the TempCNN model), the which emphasized the importance of both qualitative and quantitative variability of sampling data and variables based on the Red Edge region for improving LULC classification processes in large-scale heterogeneous landscapes.

An improved framework for mapping and assessment of dynamics in cropping pattern and crop calendar from NDVI time series across a heterogeneous agro-climatic region.

The absence of spatial and temporal cropping information in semi-arid regions poses a significant challenge in assessing the dynamics of agricultural systems at river basin scales. Satellite remote sensing provides qualitative and quantitative information to derive vegetation dynamics over extensive areas of inherent complexities due to limitations in the availability of field data and the diverse nature of agricultural cropping practices. Utilizing phenological information derived from MODIS Normalized Difference Vegetation Index (NDVI) time series data from 2003-2004 to 2021-2022, this study derives major crop types, and cropping calendar (sowing, maturity, and harvest dates) for each season and year at 250-m resolution. This study introduces an integrated function-fitting model based on the Fast Fourier Transform (FFT) and a Lagrangian 3-point derivative-based approach to extract phenological events from the NDVI time series automatically. The derived phenological information is used in a hybrid crop classification algorithm that combines a rule-based decision tree approach (using the phenological events/dates) and a random forest classifier (using the phenological metrics) to generate the classified crop map at a river basin scale for multiple seasons and years. The implemented approach successfully captured sowing and harvesting dates for every crop growing season over 19years, with an RMSE of 9days, as observed with the available field survey data from 2015 to 2021. The classification results from this hybrid approach demonstrate an overall 82% accuracy. The proposed method shows a substantial improvement in cropping area estimation with a 60% reduction in MAE and RMSE compared to the existing algorithm. From the long time series of derived seasonal crop data (19-year analysis period), the influence of monsoonal activities and shifts on the spatial and temporal dynamics of sowing time and cropping patterns are assessed for a large river basin, demonstrating the utility of this continuous crop calendar. Further, an extensive analysis of results highlights farmers' adaptive strategies in response to dry and wet years, providing foundational insights for future studies on assessing the impacts of climate change. Hence, the hybrid framework adopted in this study for deriving a continuous crop calendar holds immense relevance for parameterizing and utilizing such information for developing river basin scale hydrologic and crop growth models for water resources planning and assessment.

Optimizing crop monitoring: mapping cultivation stages and types with sentinel-1/2 and random forest algorithm

ABSTRACT Crop-type mapping is essential for agricultural monitoring, improving crop yield models, and supporting sustainable land management. Studying crop phenology, which includes the timing from germination to maturity, is equally important for precise agricultural practices. However, maps detailing crop phenology are scarce. This study addresses this gap by mapping both crop types and cultivation time, offering valuable insights for agriculture applications. The research integrates Sentinel-1/2 data, NDVI phenology curves, and machine learning (ML) algorithms through Google Earth Engine to map crops in El-Behera Governorate, Egypt. By stratifying the region into different soil types, it accounts for various agro-environmental conditions. Field visits in July 2022 provided crop-type information, and Sentinel-2 NDVI curves were used to determine cultivation time. The random forest (RF) classifier, known for its accuracy, was employed, achieving an overall accuracy (OA) of 97.4% for crop type and 95.6% for cultivation time, with high Kappa coefficients. Feature importance analysis highlighted the significance of specific spectral bands, particularly in the Red Edge region, for classification accuracy. Hyperparameter tuning across different zones demonstrated the need for customized tuning. This study enhances understanding of agro-environmental variability’s impact on crop classification and shows the potential of the RF algorithm in advancing crop mapping, suggesting further studies on how shifts in cultivation time could affect crop yield.

Precision crop mapping: within plant canopy discrimination of crop and soil using multi-sensor hyperspectral imagery

Leveraging diverse optomechanical and imaging technologies for precision agriculture applications is gaining attention in emerging economies. The precise spatial detection of plant objects in farms is crucial for optimizing plant-level nutrition and managing pests and diseases. High-resolution remote sensors mounted on drones have been increasingly deployed for large-scale crop mapping and field variability characterization. While field-level crop identification and crop-soil discrimination have been studied extensively, within-plant canopy discrimination of crop and soil has not been explored in real agricultural farms. The objectives of this study are: (i) adoption and assessment of spectral unmixing for discriminating crop and soil at within-plant canopy level, and (ii) generation of benchmark terrestrial and drone-based hyperspectral datasets for plant or sub-plant level discrimination using various spectral mixture modelling approaches and sources of endmembers. We acquired hyperspectral imagery of vegetable crops using a frame-based sensor mounted on a drone flying at different heights. Further, several linear, non-linear, and sparse-based spectral unmixing methods were used to discriminate plant and soil based on spectral signatures (endmembers) extracted from different spectral libraries prepared using in situ or field, ground-based, and drone-based hyperspectral imagery. The results, validated against pixel-to-pixel ground truth data, indicate an overall crop-soil discrimination accuracy of 99–100%, subject to a combination of endmember source and flying height. The influences of different endmember sources, spatial resolution as indicated by flying height, and inversion algorithms on the quality of estimated abundances are assessed from a verifiable and functionally relevant perspective. The generated hyperspectral datasets and ground truth data can be used for developing and testing new methods for sub-canopy level soil-crop discrimination in various agricultural applications of remote sensing.

Winter wheat identification in China through FY-3D NDVI&EVI satellite data and DNN Fusion

ABSTRACT Medium Resolution Spectral Imager-II (MERSI-II) on FengYun 3D (FY-3D) meteorological satellite is a visible and infrared spectral imaging instrument comparable with Moderate Resolution Imaging Spectroradiometer (MODIS), which provides ample opportunities for regional crop mapping. However, current research may have overlooked the potential of combining FY-3D MERSI-II data with deep neural network (DNN) algorithms for winter wheat identification. This research utilizes a DNN method to train a nonlinear model with the synthetic normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) at a 250 m resolution from FY-3D MERSI-II as the eigenvalues, enabling to accurately identify spatial distribution of winter wheat in China’s main production regions and filling the gap in the application of FY-3D MERSI-II data. The results indicate that the winter wheat identification with the combination of NDVI and EVI achieved an accuracy of 95.32%, with a Kappa coefficient of 0.87 and a determination coefficient (R 2 ) of 0.73, improving the overall accuracy by 1.27% and 1.07% compared to models trained solely using NDVI or EVI, respectively. In terms of spatial distribution, the overlap rate with the winter wheat dataset of China from MODIS at a 500 m resolution exceeded 95%, with an area bias of 2.67%. This study indicates that the medium resolution FY-3D MERSI-II remote sensing dataset combined with the DNN algorithm can accurately identify crop information over large areas.

Evaluating Flood Damage to Paddy Rice Fields Using PlanetScope and Sentinel-1 Data in North-Western Nigeria: Towards Potential Climate Adaptation Strategies

Floods are significant global disasters, but their impact in developing countries is greater due to the lower shock tolerance, many subsistence farmers, land fragmentation, poor adaptation strategies, and low technical capacity, which worsen food security and livelihoods. Therefore, accurate and timely monitoring of flooded crop areas is crucial for both disaster impact assessments and adaptation strategies. However, most existing methods for monitoring flooded crops using remote sensing focus solely on estimating the flood damage, neglecting the need for adaptation decisions. To address these issues, we have developed an approach to mapping flooded rice fields using Earth observation and machine learning. This approach integrates high-resolution multispectral satellite images with Sentinel-1 data. We have demonstrated the reliability and applicability of this approach by using a manually labelled dataset related to a devastating flood event in north-western Nigeria. Additionally, we have developed a land suitability model to evaluate potential areas for paddy rice cultivation. Our crop extent and land use/land cover classifications achieved an overall accuracy of between 93% and 95%, while our flood mapping achieved an overall accuracy of 99%. Our findings indicate that the flood event caused damage to almost 60% of the paddy rice fields. Based on the land suitability assessment, our results indicate that more land is suitable for cultivation during natural floods than is currently being used. We propose several recommendations as adaptation measures for stakeholders to improve livelihoods and mitigate flood disasters. This study highlights the importance of integrating multispectral and synthetic aperture radar (SAR) data for flood crop mapping using machine learning. Decision-makers will benefit from the flood crop mapping framework developed in this study in a number of spatial planning applications.

European Union crop map 2022: Earth observation’s 10-meter dive into Europe’s crop tapestry

To provide the information needed for a detailed monitoring of crop types across the European Union (EU), we present an advanced 10-metre resolution map for the EU and Ukraine with 19 crop types for 2022, updating the 2018 version. Using Earth Observation (EO) and in-situ data from Eurostat’s Land Use and Coverage Area Frame Survey (LUCAS) 2022, the methodology included 134,684 LUCAS Copernicus polygons, Sentinel-1 and Sentinel-2 satellite imagery, land surface temperature and a digital elevation model. Based on this data, two classification layers were developed using a Random Forest machine learning approach: a primary map and a gap-filling map to address cloud-covered gaps. The combined maps, covering 27 EU countries, show an overall accuracy of 79.3% for seven major land cover classes and 70.6% for all 19 crop types. The trained model was used to derive the 2022 map for Ukraine, demonstrating its robustness even in regions without labelled samples for model training.

SSGAN: Cloud removal in satellite images using spatiospectral generative adversarial network

Satellite data’s reliability, uniformity, and global scanning capabilities have revolutionized agricultural monitoring and crop management. However, the presence of clouds in satellite images can obscure useful information, rendering them difficult to infer. Aiming at the problem of cloud cover, this study presents a SpatioSpectral Generative Adversarial Network (SSGAN) approach for effectively eliminating cloud cover from multispectral satellite images. It utilizes the Synthetic Aperture Radar (SAR) images as complementary information with the optical images from the Sentinel-2 satellite. The proposed model exploits feature extraction by sub-grouping the 13 channels of Sentinel-2 images based on their electromagnetic wavelength. Experimentally, we demonstrated that the proposed SSGAN model surpasses conventional and state-of-the-art (SOTA) methods and can reconstruct regions obscured by clouds. The subgrouping optimized the utilization of sensor information and improved the performance metrics for reconstructed images. Compared to the state-of-the-art (SOTA) approach, the SSGAN model demonstrates higher performance, achieving a mPSNR of 32.771, mSSIM of 0.880, and correlation coefficient (CC) of 0.889. The SSGAN model was further evaluated under varying conditions, including scenarios without the inclusion of SAR data, where it achieved a mPSNR of 26.825, mSSIM of 0.726, and CC of 0.615. Adding SAR images into the model significantly enhanced its performance, resulting in a mPSNR of 29.932, mSSIM of 0.857, and CC of 0.735. These results indicate that higher mPSNR, mSSIM, and CC values correspond to better image reconstruction quality. Our method enhances the usability of satellite data for crop mapping, crop health monitoring, and crop yield prediction.

Remote Prediction of Soybean Yield Using UAV-Based Hyperspectral Imaging and Machine Learning Models

Early soybean yield estimation has become a fundamental tool for market policy and food security. Considering a heterogeneous crop, this study investigates the spatial and spectral variability in soybean canopy reflectance to achieve grain yield estimation. Besides allowing crop mapping, remote sensing data also provide spectral evidence that can be used as a priori knowledge to guide sample collection for prediction models. In this context, this study proposes a sampling design method that distributes sample plots based on the spatial and spectral variability in vegetation spectral indices observed in the field. Random forest (RF) and multiple linear regression (MLR) approaches were applied to a set of spectral bands and six vegetation indices to assess their contributions to the soybean yield estimates. Experiments were conducted with a hyperspectral sensor of 25 contiguous spectral bands, ranging from 500 to 900 nm, carried by an unmanned aerial vehicle (UAV) to collect images during the R5 soybean growth stage. The tests showed that spectral indices specially designed from some bands could be adopted instead of using multiple bands with MLR. However, the best result was obtained with RF using spectral bands and the height attribute extracted from the photogrammetric height model. In this case, Pearson’s correlation coefficient was 0.91. The difference between the grain yield productivity estimated with the RF model and the weight collected at harvest was 1.5%, indicating high accuracy for yield prediction.

Enhancing Pléiades-based crop mapping with multi-temporal and texture information

Accurate crop mapping using satellite imagery is crucial for improving the monitoring of agricultural landscapes. Very high resolution (VHR) satellite imagery offers unique capabilities in this respect, allowing for even small fields to be discerned and image texture analysis to be performed. Additionally, satellite imagery has greater efficiency than unmanned aerial vehicles due to its extensive coverage. Moreover, the operation flexibility of VHR satellites means that timely image acquisition is possible several times during the growing season. This study investigates the potential of VHR Pléiades images and the random forest classifier for accurate crop mapping. Four images acquired on April 9th, May 12th, May 31st, and June 20th were used to test 16 classification scenarios, including single-date and multi-temporal combinations of spectral bands, texture features, and vegetation Indices (VIs). The classification using the spectral bands from all four images achieved the highest overall accuracy, 93.9% and 96.3% at field and pixel levels, respectively. The bitemporal classifications had lower accuracy. Nevertheless, the combination of the May 12th and June 20th spectral bands had 90% accuracy, which indicated that two images may be sufficient for reliable mapping if the periods with phenological differences between crops are considered. Adding texture features to the spectral bands significantly enhanced the accuracy (up to 8%) of single-date classifications, making it highly recommended when only one image is available. However, the impact of texture was more pronounced on the later dates. It showed the most marked benefit for vineyards and alfalfa, with minimal or no improvement observed for other classes like winter barley. An additional increase in overall accuracy was achieved in three of the four dates by supplementing the spectral and texture bands with VIs. This study highlights the importance of considering image acquisition dates and crop types when designing satellite-based crop mapping strategies for optimal accuracy.

Hierarchical classification for improving parcel-scale crop mapping using time-series Sentinel-1 data

Parcel-scale crop classification utilizing time-series satellite observations is of significant importance in precision agriculture. The prior knowledge that crop types can be organized in a hierarchical tree structure is beneficial for improving crop classification. Moreover, the crop hierarchy aligns with the coarse-to-fine cognitive process of geographic scenes. Based on the crop hierarchy, this study developed a general hierarchical classification framework for enhancing crop mapping using time-series Sentinel-1 data. Central to this method is a deep-learning-based hierarchical classification model that explores and makes use of crop hierarchical knowledge. First, preprocessed Sentinel-1 data were geometrically overlaid onto farmland parcel maps to derive parcel-scale time-series features. Second, we constructed a hierarchical crop type system for study areas based on the crop phenology of labeled crop-type samples. Third, we developed a deep-learning-based hierarchical classification model to identify crop types for each parcel, to generate final crop-type classification maps. The proposed approach was further discussed and verified through the implementation of parcel-scale time-series crop hierarchical classifications in a study area in France with farmland parcel maps and time-series Sentinel-1 data. The classification results, indicating significant improvements greater than 4.0% in overall accuracy and 5.0% in F1 score over comparative methods, demonstrated the effectiveness of the proposed method in learning multi-scale time-series features for hierarchical crop classification utilizing Sentinel-1 data sequences.

A lightweight CNN-Transformer network for pixel-based crop mapping using time-series Sentinel-2 imagery

Deep learning approaches have provided state-of-the-art performance in crop mapping. Recently, several studies have combined the strengths of two dominant deep learning architectures, Convolutional Neural Networks (CNNs) and Transformers, to classify crops using remote sensing images. Despite their success, many of these models utilize patch-based methods that require extensive data labeling, as each sample contains multiple pixels with corresponding labels. This leads to higher costs in data preparation and processing. Moreover, previous methods rarely considered the impact of missing values caused by clouds and no-observations in remote sensing data. Therefore, this study proposes a lightweight multi-stage CNN-Transformer network (MCTNet) for pixel-based crop mapping using time-series Sentinel-2 imagery. MCTNet consists of several successive modules, each containing a CNN sub-module and a Transformer sub-module to extract important features from the images, respectively. An attention-based learnable positional encoding (ALPE) module is designed in the Transformer sub-module to capture the complex temporal relations in the time-series data with different missing rates. Arkansas and California in the U.S. are selected to evaluate the model. Experimental results show that the MCTNet has a lightweight advantage with the fewest parameters and memory usage while achieving the superior performance compared to eight advanced models. Specifically, MCTNet obtained an overall accuracy (OA) of 0.968, a kappa coefficient (Kappa) of 0.951, and a macro-averaged F1 score (F1) of 0.933 in Arkansas, and an OA of 0.852, a Kappa of 0.806, and an F1 score of 0.829 in California. The results highlight the importance of each component of the model, particularly the ALPE module, which enhanced the Kappa of MCTNet by 4.2% in Arkansas and improved the model’s robustness to missing values in remote sensing data. Additionally, visualization results demonstrated that the features extracted from CNN and Transformer sub-modules are complementary, explaining the effectiveness of the MCTNet.

A novel soybean mapping index within the global optimal time window

Efficient soybean mapping is critical for agricultural production and yield prediction. However, current sample-driven soybean mapping methods heavily rely on large representative sample datasets, limiting the interpretability of physical mechanisms. Besides, sample-free methods failed to exploit key features that differentiate soybean from other crops, especially Chlorophyll content. Misclassification errors persist and spatiotemporal generalization remains limited. Therefore, this study develops a novel Soybean Mapping Composite Index (SMCI) within a precise Global Optimal Time Window (GOTW). It integrates unique features of soybean Chlorophyll content, canopy water content, and canopy greenness by coupling three red-edge bands (RE2, RE3, and RE4), one near-infrared band, one shortwave infrared band, and two feature indices (Enhanced Vegetation Index and Green Chlorophyll Vegetation Index). The novel index was applied to soybean mapping at six sites in four major soybean producing countries (China, Argentina, Brazil, and the United States) from 2019 to 2021, using an optimal threshold of 3.25. Within the GOTW, the index responds better to spectral features and improves soybean separability. The average overall accuracy (OA: 91%) and average Kappa coefficient (Kappa: 0.83) for the novel index at all sites outperformed the traditional sample-driven Random Forest (RF) method (OA: 84%, Kappa: 0.70) and the existing sample-free index-based Greenness and Water Content Composite Index (GWCCI) (OA: 81%, Kappa: 0.64). Furthermore, interannual transfer experiments consistently showed high accuracy, demonstrating robust spatiotemporal transferability. The proposed SMCI index meets the need for a lightweight and stable soybean mapping tool and serves as a valuable reference for efficient global crop mapping.

A new transferable deep learning approach for crop mapping

ABSTRACT Accurate crop mapping provides important information for government decision-making and agricultural management. Recent advancements in deep learning have significantly enhanced the capabilities of remote sensing-based crop mapping. In this study, we developed a new deep learning approach, DSH, which comprises three modules, DeepLabV3+ (D), channel self-attention (S), and histogram matching (H). The DeepLabV3+ module learns the spatial distribution of crops in the spatial and spectral dimensions. The channel self-attention module was used to enhance the weights of the important features. The histogram-matching module addresses the domain gap between the training and testing images and enhances the transferability of the DSH. In addition, the input data of the DSH are monthly synthesized Sentinel-2 data, thus relaxing the data collection requirements. The performance of the DSH was evaluated and benchmarked against HRNetV2, DeepLabV3+, and random forest (RF) at eight sites in the U.S. China, and France with different farming structures, climate conditions, and landscape complexities. Temporal transfer revealed that the classification performance of DSH improved as the growth season progressed, peaked in the peak growth season (August), and then declined at each study site. The DSH model trained during the peak growth season demonstrated superior temporal transferability. Spatial transfer experiments demonstrated that transferring DSH to sites with similar planting structures achieved higher precision than full spatial transfers. Spatiotemporal transfer experiments conducted at eight sites over three years (2020–2022) demonstrated an average overall accuracy (OA) of 88.1% for DSH, highlighting the importance of similar planting structures for effective spatiotemporal transfer. DSH outperformed HRNetV2, DeepLabV3+, and RF in OA, producer’s accuracy, and user’s accuracy at all eight study sites. Notably, DSH exhibited comparable performance during both the peak and full growth seasons, with an average OA of 93.9%. The classification accuracy of test images after histogram matching remained stable from 2020 to 2022, whereas the accuracy of raw test images fluctuated annually. This indicates that histogram matching is effective in improving the transferability of DSH. Embedding the channel self-attention module into the position of the high-level features had a more significant impact on improving DSH’s performance than other configurations. Visualizing features from the channel self-attention and high-level layers revealed that integrating the channel self-attention module with DeepLabV3+‘s high-level features significantly improved crop mapping by enhancing relevant semantic information. Overall, DSH exhibits robust spatiotemporal generalizability and requires minimal input data, making it suitable for large-scale crop mapping.

Bibliometric Computation Mapping Analysis of Publication Machine and Deep Learning for Food Crops Mapping using VOSviewer

Machine learning and deep learning are currently widely used in various fields, including remote sensing for food security. However, there is no research that specifically examines the interests, developments, and trends of this research in the future. This study aims to examine the development of machine and deep learning research for mapping food crops through a bibliometric approach with computational mapping analysis using VOSviewer. Article data was obtained from the Google Scholar database using the publish or perish reference manager application. The title and abstract of the article were used to guide the search process by referring to the keyword “Machine and Deep Learning Mapping Food Crops”. 114 relevant articles were discovered. Google Scholar-indexed articles over the last ten years, from 2014 to 2023, were used as study material. The results show that machine research and deep learning for mapping food crops can be separated into three terms: machine learning, deep learning, and plant mapping. The term “Crop Mapping” has 57 links for a total of 199 links. The term "machine learning" has 41 links for a total of 79 links, and the term "deep learning" has 26 links for a total of 41 links. The results of the analysis of machine development and deep learning publications for mapping food crops in the last 10 years show a constant increase. The peak of the increase occurred in 2021 and 2022, namely 25 articles published per year, respectively. This means that this research topic is still relatively new in terms of interest and exploration, therefore there is still room further research. We examine numerous articles that have been published on machine and deep learning for crop mapping and their relation to the field studied with VOSviewer. This review can serve as a starting point for further research in different domains