Crop Classification Using Multi-Temporal Sentinel-2 Data in the Shiyang River Basin of China

Spatial distribution of pesticide use based on crop rotation data in La Plata River basin: a case study from an agricultural region of Uruguay

The intensive global use of pesticides presents an escalating threat to human health, ecosystems, and water quality. To develop national and local environmental management strategies for mitigating pollution caused by pesticides, it is essential to understand the quantities, timing, and location of their application. This study aims to estimate the spatial distribution of pesticide use in an agricultural region of La Plata River basin in Uruguay. Estimates of pesticide use were made by surveying doses applied to each crop. This information was spatialized through identifying agricultural rotations using remote sensing techniques. The study identified the 60 major agricultural rotations in the region and mapped the use and application amount of the nine most significant active ingredients (glyphosate, 2,4-dichlorophenoxyacetic acid, flumioxazin, S-metolachlor, clethodim, flumetsulam, triflumuron, chlorantraniliprole, and fipronil). The results reveal that glyphosate is the most extensively used pesticide (53.5% of the area) and highest amount of use (> 1.44 kg/ha). Moreover, in 19% of the area, at least seven active ingredients are applied in crop rotations. This study marks the initial step in identifying rotations and estimating pesticide applications with high spatial resolution at a regional scale in agricultural regions of La Plata River basin. The results improve the understanding of pesticide spatial distribution based on data obtained from agronomists, technicians, and producers and provide a replicable methodological approach for other geographic and productive contexts. Generating baseline information is key to environmental management and decision making, towards the design of more robust monitoring systems and human exposure assessment.

Advancing crop classification in smallholder agriculture: A multifaceted approach combining frequency-domain image co-registration, transformer-based parcel segmentation, and Bi-LSTM for crop classification

Agricultural Remote Sensing has the potential to enhance agricultural monitoring in smallholder economies to mitigate losses. However, its widespread adoption faces challenges, such as diminishing farm sizes, lack of reliable data-sets and high cost related to commercial satellite imagery. This research focuses on opportunities, practices and novel approaches for effective utilization of remote sensing in agriculture applications for smallholder economies. The work entails insights from experiments using datasets representative of major crops during different growing seasons. We propose an optimized solution for addressing challenges associated with remote sensing-based crop mapping in smallholder agriculture farms. Open source tools and data are used for inter and intra-sensor image registration, with a root mean square error of 0.3 or less. We also propose and emphasize on the use of delineated vegetation parcels through Segment Anything Model for Geospatial (SAM-GEOs). Furthermore a Bidirectional-Long Short-Term Memory-based (Bi-LSTM) deep learning model is developed and trained for crop classification, achieving results with accuracy of more than 94% and 96% for validation sets of two data sets collected in the field, during 2 growing seasons.

The 10-m cotton maps in Xinjiang, China during 2018-2021

Cotton maps (10 m) of Xinjiang (XJ_COTTON10), which is the largest cotton production region of China, were produced from 2018 to 2021 through supervised classification. A two-step mapping strategy, i.e., cropland mapping followed by cotton extraction, was employed to improve the accuracy and efficiency of cotton mapping for a large region of about 1.66 million km2 with high heterogeneity. Additionally, the time-series satellite data related to spectral, textural, structural, and phenological features were combined and used in a supervised random forest classifier. The cotton/non-cotton classification model achieved overall accuracies of about 95% and 90% on the test samples of the same and adjacent years, respectively. The proposed two-step cotton mapping strategy proved promising and effective in producing multi-year and consistent cotton maps. XJ_COTTON10 agreed well with the statistical areas of cotton at the county level (R2 = 0.84-0.94). This is the first cotton mapping for the entire Xinjiang at 10-meter resolution, which can provide a basis for high-precision cotton monitoring and policymaking in China.

Cropformer: A new generalized deep learning classification approach for multi-scenario crop classification

Accurate and efficient crop classification using remotely sensed data can provide fundamental and important information for crop yield estimation. Existing crop classification approaches are usually designed to be strong in some specific scenarios but not for multi-scenario crop classification. In this study, we proposed a new deep learning approach for multi-scenario crop classification, named Cropformer. Cropformer can extract global features and local features, to solve the problem that current crop classification methods extract a single feature. Specifically, Cropformer is a two-step classification approach, where the first step is self-supervised pre-training to accumulate knowledge of crop growth, and the second step is a fine-tuned supervised classification based on the weights from the first step. The unlabeled time series and the labeled time series are used as input for the first and second steps respectively. Multi-scenario crop classification experiments including full-season crop classification, in-season crop classification, few-sample crop classification, and transfer of classification models were conducted in five study areas with complex crop types and compared with several existing competitive approaches. Experimental results showed that Cropformer can not only obtain a very significant accuracy advantage in crop classification, but also can obtain higher accuracy with fewer samples. Compared to other approaches, the classification performance of Cropformer during model transfer and the efficiency of the classification were outstanding. The results showed that Cropformer could build up a priori knowledge using unlabeled data and learn generalized features using labeled data, making it applicable to crop classification in multiple scenarios.

Long short term memory deep net performance on fused Planet-Scope and Sentinel-2 imagery for detection of agricultural crop

In view of the challenges faced by organizations and departments concerned with agricultural capacity observations, we collected In-Situ data consisting of diverse crops (More than 11 consumable vegetation types) in our pilot region of Harichand Charsadda, Khyber Pakhtunkhwa (KP), Pakistan. Our proposed Long Short-Term Memory based Deep Neural network model was trained for land cover land use statistics generation using the acquired ground truth data, for a synergy between Planet-Scope Dove and European Space Agency's Sentinel-2. Total of 4 bands from both sentinel-2 and planet scope including Red, Green, Near-Infrared (NIR) and Normalised Difference Vegetation Index (NDVI) were used for classification purpose. Using short temporal frame of Sentinel-2 comprising 5 date images, we propose an realistic and implementable procedure for generating accurate crop statistics using remote sensing. Our self collected data-set consists of a total number of 107,899 pixels which was further split into 70% and 30% for training and testing purpose of the model respectively. The collected data is in the shape of field parcels, which has been further split for training, validation and test sets, to avoid spatial auto-correlation. To ensure the quality and accuracy 15% of the training data was left out for validation purpose, and 15% for testing. Prediction was also performed on our trained model and visual analysis of the area from the image showed significant results. Further more a comparison between Sentinel-2 time series is performed separately from the fused Planet-Scope and Sentinel-2 time-series data sets. The results achieved shows a weighted average of 93% for Sentinel-2 time series and 97% for fused Planet-Scope and Sentinel-2 time series.

Rapid Identification of Main Vegetation Types in the Lingkong Mountain Nature Reserve Based on Multi-Temporal Modified Vegetation Indices

Nature reserves are among the most bio-diverse regions worldwide, and rapid and accurate identification is a requisite for their management. Based on the multi-temporal Sentinel-2 dataset, this study presents three multi-temporal modified vegetation indices (the multi-temporal modified normalized difference Quercus wutaishanica index (MTM-NDQI), the multi-temporal modified difference scrub grass index (MTM-DSI), and the multi-temporal modified ratio shaw index (MTM-RSI)) to improve the classification accuracy of the remote sensing of vegetation in the Lingkong Mountain Nature Reserve of China (LMNR). These three indices integrate the advantages of both the typical vegetation indices and the multi-temporal remote sensing data. By using the proposed indices with a uni-temporal modified vegetation index (the uni-temporal modified difference pine-oak mixed forest index (UTM-DMI)) and typical vegetation indices (e.g., the ratio vegetation index (RVI), the difference vegetation index (DVI), and the normalized difference vegetation index (NDVI)), an optimal feature set is obtained that includes the NDVI of December, the NDVI of April, and the UTM-DMI, MTM-NDQI, MTM-DSI, and MTM-RSI. The overall accuracy (OA) of the random forest classification (98.41%) and Kappa coefficient of the optimal feature set (0.98) were higher than those of the time series NDVI (OA = 96.03%, Kappa = 0.95), the time series RVI (OA = 95.56%, Kappa = 0.95), and the time series DVI (OA = 91.27%, Kappa = 0.90). The OAs of the rapid classification and the Kappa coefficient of the knowledge decision tree based on the optimal feature set were 95.56% and 0.95, respectively. Meanwhile, only three of the seven vegetation types were omitted or misclassified slightly. Overall, the proposed vegetation indices have advantages in identifying the vegetation types in protected areas.

Coupling optical and SAR imagery for automatic garlic mapping

Accurate garlic identification and mapping are vital for precise crop management and the optimization of yield models. However, previous understandings of garlic identification were limited. Here, we propose an automatic garlic mapping framework using optical and synthetic aperture radar (SAR) images on the Google Earth Engine. Specifically, we firstly mapped winter crops based on the phenology of winter crops derived from Sentinel-2 data. Then, the garlic was identified separately using Sentinel-1 and Sentinel-2 data based on the winter crops map. Additionally, multi-source validation data were used to evaluate our results. In garlic mapping, coupled optical and SAR images (OA 95.34% and kappa 0.91) outperformed the use of only optical images (OA 74.78% and kappa 0.50). The algorithm explored the potential of multi-source remote sensing data to identify target crops in mixed and fragmented planting regions. The garlic planting information from the resultant map is essential for optimizing the garlic planting structure, regulating garlic price fluctuations, and promoting a healthy and sustainable development of the garlic industry.

Surface biophysical features fusion in remote sensing for improving land crop/cover classification accuracy

Preparing up-to-date land crop/cover maps is important to study in order to achieve food security. Therefore, the aim of this study was to evaluate the impact of surface biophysical features in the land crop/cover classification accuracy and introduce a new fusion-based method with more accurate results for land crop/cover classification. For this purpose, multi-temporal images from Sentinel 1 and 2, and an actual land crop map prepared by Agriculture and Agri-Food Canada (AAFC) in 2019 were used for 3 test sites in Ontario, Canada. Firstly, surface biophysical features maps were prepared based on spectral indices from Sentinel 2 including Normalized Difference Vegetation Index (NDVI), Index-based Built-up Index (IBI), Wetness, Albedo, and Brightness and co-polarization (VV) and cross-polarization (VH) from Sentinel 1 for different dates. Then, different scenarios were generated; these included single surface biophysical features as well as a combination of several surface biophysical features. Secondly, land crop/cover maps were prepared for each scenario based on the Random Forest (RF). In the third step, based on the voting strategy, classification maps from different scenarios were combined. Finally, the accuracy of the land crop/cover maps obtained from each of the scenario was evaluated. The results showed that the average overall accuracy of land crop/cover maps obtained from individual scenario (one feature) including NDVI, IBI, Wetness, Albedo, Brightness, VV and VH were 66%, 68%, 63%, 60%, 57%, 62% and 58%, respectively, which by the surface biophysical features fusion, the overall accuracy of land crop/cover maps increased to 83%. Also, by combining the classification results obtained from different scenarios based on voting strategy, the overall accuracy increased to 89%. The results of this study indicate that the feature level-based fusion of surface biophysical features and decision level based fusion of land crop/cover maps obtained from various scenarios increases the accuracy of land crop/cover classification.

Irrigated Crop Types Mapping in Tashkent Province of Uzbekistan with Remote Sensing-Based Classification Methods

Appropriate crop type mapping to monitor and control land management is very important in developing countries. It can be very useful where digital cadaster maps are not available or usage of Remote Sensing (RS) data is not utilized in the process of monitoring and inventory. The main goal of the present research is to compare and assess the importance of optical RS data in crop type classification using medium and high spatial resolution RS imagery in 2018. With this goal, Landsat 8 (L8) and Sentinel-2 (S2) data were acquired over the Tashkent Province between the crop growth period of May and October. In addition, this period is the only possible time for having cloud-free satellite images. The following four indices "Normalized Difference Vegetation Index" (NDVI), "Enhanced Vegetation Index" (EVI), and "Normalized Difference Water Index" (NDWI1 and NDWI2) were calculated using blue, red, near-infrared, shortwave infrared 1, and shortwave infrared 2 bands. Support-Vector-Machine (SVM) and Random Forest (RF) classification methods were used to generate the main crop type maps. As a result, the Overall Accuracy (OA) of all indices was above 84% and the highest OA of 92% was achieved together with EVI-NDVI and the RF method of L8 sensor data. The highest Kappa Accuracy (KA) was found with the RF method of L8 data when EVI (KA of 88%) and EVI-NDVI (KA of 87%) indices were used. A comparison of the classified crop type area with Official State Statistics (OSS) data about sown crops area demonstrated that the smallest absolute weighted average (WA) value difference (0.2 thousand ha) was obtained using EVI-NDVI with RF method and NDVI with SVM method of L8 sensor data. For S2-sensor data, the smallest absolute value difference result (0.1 thousand ha) was obtained using EVI with RF method and 0.4 thousand ha using NDVI with SVM method. Therefore, it can be concluded that the results demonstrate new opportunities in the joint use of Landsat and Sentinel data in the future to capture high temporal resolution during the vegetation growth period for crop type mapping. We believe that the joint use of S2 and L8 data enables the separation of crop types and increases the classification accuracy.

Crop Identification Based on Multi-Temporal Active and Passive Remote Sensing Images

Although vegetation index time series from optical images are widely used for crop mapping, it remains difficult to obtain sufficient time-series data because of satellite revisit time and weather in some areas. To address this situation, this paper considered Wen County, Henan Province, Central China as the research area and fused multi-source features such as backscatter coefficient, vegetation index, and time series based on Sentinel-1 and -2 data to identify crops. Through comparative experiments, this paper studied the feasibility of identifying crops with multi-temporal data and fused data. The results showed that the accuracy of multi-temporal Sentinel-2 data increased by 9.2% compared with single-temporal Sentinel-2 data, and the accuracy of multi-temporal fusion data improved by 17.1% and 2.9%, respectively, compared with multi-temporal Sentinel-1 and Sentinel-2 data. Multi-temporal data well-characterizes the phenological stages of crop growth, thereby improving the classification accuracy. The fusion of Sentinel-1 synthetic aperture radar data and Sentinel-2 optical data provide sufficient time-series data for crop identification. This research can provide a reference for crop recognition in precision agriculture.

Investigating the Potential of Sentinel-2 MSI in Early Crop Identification in Northeast China

Early crop identification can provide timely and valuable information for agricultural planting management departments to make reasonable and correct decisions. At present, there is still a lack of systematic summary and analysis on how to obtain real-time samples in the early stage, what the optimal feature sets are, and what level of crop identification accuracy can be achieved at different stages. First, this study generated training samples with the help of historical crop maps in 2019 and remote sensing images in 2020. Then, a feature optimization method was used to obtain the optimal features in different stages. Finally, the differences of the four classifiers in identifying crops and the variation characteristics of crop identification accuracy at different stages were analyzed. These experiments were conducted at three sites in Heilongjiang Province to evaluate the reliability of the results. The results showed that the earliest identification time of corn can be obtained in early July (the seven leaves period) with an identification accuracy up to 86%. In the early stages, its accuracy was 40~79%, which was low, and could not reach the satisfied accuracy requirements. In the middle stages, a satisfactory recognition accuracy could be achieved, and its recognition accuracy was 79~100%. The late stage had a higher recognition accuracy, which was 90~100%. The accuracy of soybeans at each stage was similar to that of corn, and the earliest identification time of soybeans could also be obtained in early July (the blooming period) with an identification accuracy up to 87%. Its accuracy in the early growth stage was 35~71%; in the middle stage, it was 69~100%; and in the late stage, it was 92~100%. Unlike corn and soybeans, the earliest identification time of rice could be obtained at the end of April (the flooding period) with an identification accuracy up to 86%. In the early stage, its accuracy was 58~100%; in the middle stage, its accuracy was 93~100%; and in the late stage, its accuracy was 96~100%. In terms of crop identification accuracy in the whole growth stage, GBDT and RF performed better than other classifiers in our three study areas. This study systematically investigated the potential of early crop recognition in Northeast China, and the results are helpful for relevant applications and decision making of crop recognition in different crop growth stages.

Optimal Bands Combination Selection for Extracting Garlic Planting Area with Multi-Temporal Sentinel-2 Imagery

Garlic is one of the main economic crops in China. Accurate and timely extraction of the garlic planting area is critical for adjusting the agricultural planting structure and implementing rural policy actions. Crop extraction methods based on remote sensing usually use spectral-temporal features. Still, for garlic extraction, most methods simply combine all multi-temporal images. There has been a lack of research on each band's function in each multi-temporal image and optimal bands combination. To systematically explore the potential of the multi-temporal method for garlic extraction, we obtained a series of Sentinel-2 images in the whole garlic growth cycle. The importance of each band in all these images was ranked by the random forest (RF) method. According to the importance score of each band, eight different multi-temporal combination schemes were designed. The RF classifier was employed to extract garlic planting area, and the accuracy of the eight schemes was compared. The results show that (1) the Scheme VI (the top 39 bands in importance score) achieved the best accuracy of 98.65%, which is 6% higher than the optimal mono-temporal (February, wintering period) result, and (2) the red-edge band and the shortwave-infrared band played an essential role in accurate garlic extraction. This study gives inspiration in selecting the remotely sensed data source, the band, and phenology for accurately extracting garlic planting area, which could be transferred to other sites with larger areas and similar agriculture structures.

Feature Fusion Approach for Temporal Land Use Mapping in Complex Agricultural Areas

Accurate temporal land use mapping provides important and timely information for decision making for large-scale management of land and crop production. At present, temporal land cover and crop classifications within a study area have neglected the differences between subregions. In this paper, we propose a classification rule by integrating the terrain, time series characteristics, priority, and seasonality (TTPSR) with Sentinel-2 satellite imagery. Based on the time series of Normalized Difference Water Index (NDWI) and Vegetation Index (NDVI), a dynamic decision tree for forests, cultivation, urban, and water was created in Google Earth Engine (GEE) for each subregion to extract cultivated land. Then, with or without this cultivated land mask data, the original classification results for each subregion were completed based on composite image acquisition with five vegetation indices using Random Forest. During the post-reclassification process, a 4-bit coding rule based on terrain, type, seasonal rhythm, and priority was generated by analyzing the characteristics of the original results. Finally, statistical results and temporal mapping were processed. The results showed that feature importance was dominated by B2, NDWI, RENDVI, B11, and B12 over winter, and B11, B12, NDBI, B2, and B8A over summer. Meanwhile, the cultivated land mask improved the overall accuracy for multicategories (seven to eight and nine to 13 during winter and summer, respectively) in each subregion, with average ranges in the overall accuracy for winter and summer of 0.857–0.935 and 0.873–0.963, respectively, and kappa coefficients of 0.803–0.902 and 0.835–0.950, respectively. The analysis of the above results and the comparison with resampling plots identified various sources of error for classification accuracy, including spectral differences, degree of field fragmentation, and planting complexity. The results demonstrated the capability of the TTPSR rule in temporal land use mapping, especially with regard to complex crops classification and automated post-processing, thereby providing a viable option for large-scale land use mapping.

Machine Learning in Agriculture: A Comprehensive Updated Review

The digital transformation of agriculture has evolved various aspects of management into artificial intelligent systems for the sake of making value from the ever-increasing data originated from numerous sources. A subset of artificial intelligence, namely machine learning, has a considerable potential to handle numerous challenges in the establishment of knowledge-based farming systems. The present study aims at shedding light on machine learning in agriculture by thoroughly reviewing the recent scholarly literature based on keywords' combinations of "machine learning" along with "crop management", "water management", "soil management", and "livestock management", and in accordance with PRISMA guidelines. Only journal papers were considered eligible that were published within 2018-2020. The results indicated that this topic pertains to different disciplines that favour convergence research at the international level. Furthermore, crop management was observed to be at the centre of attention. A plethora of machine learning algorithms were used, with those belonging to Artificial Neural Networks being more efficient. In addition, maize and wheat as well as cattle and sheep were the most investigated crops and animals, respectively. Finally, a variety of sensors, attached on satellites and unmanned ground and aerial vehicles, have been utilized as a means of getting reliable input data for the data analyses. It is anticipated that this study will constitute a beneficial guide to all stakeholders towards enhancing awareness of the potential advantages of using machine learning in agriculture and contributing to a more systematic research on this topic.

Delineating Smallholder Maize Farms from Sentinel-1 Coupled with Sentinel-2 Data Using Machine Learning

Rural communities rely on smallholder maize farms for subsistence agriculture, the main driver of local economic activity and food security. However, their planted area estimates are unknown in most developing countries. This study explores the use of Sentinel-1 and Sentinel-2 data to map smallholder maize farms. The random forest (RF), support vector (SVM) machine learning algorithms and model stacking (ST) were applied. Results show that the classification of combined Sentinel-1 and Sentinel-2 data improved the RF, SVM and ST algorithms by 24.2%, 8.7%, and 9.1%, respectively, compared to the classification of Sentinel-1 data individually. Similarities in the estimated areas (7001.35 ± 1.2 ha for RF, 7926.03 ± 0.7 ha for SVM and 7099.59 ± 0.8 ha for ST) show that machine learning can estimate smallholder maize areas with high accuracies. The study concludes that the single-date Sentinel-1 data were insufficient to map smallholder maize farms. However, single-date Sentinel-1 combined with Sentinel-2 data were sufficient in mapping smallholder farms. These results can be used to support the generation and validation of national crop statistics, thus contributing to food security.

Potential of Hybrid CNN-RF Model for Early Crop Mapping with Limited Input Data

When sufficient time-series images and training data are unavailable for crop classification, features extracted from convolutional neural network (CNN)-based representative learning may not provide useful information to discriminate crops with similar spectral characteristics, leading to poor classification accuracy. In particular, limited input data are the main obstacles to obtain reliable classification results for early crop mapping. This study investigates the potential of a hybrid classification approach, i.e., CNN-random forest (CNN-RF), in the context of early crop mapping, that combines the automatic feature extraction capability of CNN with the superior discrimination capability of an RF classifier. Two experiments on incremental crop classification with unmanned aerial vehicle images were conducted to compare the performance of CNN-RF with that of CNN and RF with respect to the length of the time-series and training data sizes. When sufficient time-series images and training data were used for the classification, the accuracy of CNN-RF was slightly higher or comparable with that of CNN. In contrast, when fewer images and the smallest training data were used at the early crop growth stage, CNN-RF was substantially beneficial and the overall accuracy increased by maximum 6.7%p and 4.6%p in the two study areas, respectively, compared to CNN. This is attributed to its ability to discriminate crops from features with insufficient information using a more sophisticated classifier. The experimental results demonstrate that CNN-RF is an effective classifier for early crop mapping when only limited input images and training samples are available.