Polyakov, Artem Nikolaevich

Stepanov A., Illarionova L., Aseeva T., Polyakov A.

Assessment of crop heterogeneity using remote sensing data is an urgent task of digital agriculture. NDVI values were calculated for arable lands of the Khabarovsk Territory from Sentinel-2 images with a resolution of 10 m. A cluster analysis of NDVI time series in 2022 was carried out to assess the heterogeneity of soybean, oat, corn and buckwheat crops at the regional level. Two clusters were described for soybeans, one of which represented clogged fields in 2022, and the average seasonal course of NDVI in this cluster in July–August corresponded to fallow lands. Extremes of the average seasonal NDVI course and time series for each cluster were calculated for agricultural crops of the Khabarovsk Territory. A comparative assessment of the heterogeneity of crop development within the same field was carried out using UAV and satellite data, while the NDVI distribution obtained from satellite images corresponds to the spatial distribution of NDVI according to the DJI Mavic3M UAV data. At the next stage of the research, it is planned to analyze the within field heterogeneity of crops for other crops of the Khabarovsk Territory using satellite monitoring, which will significantly reduce the cost of obtaining and analyzing UAV images.

Polyakov A.N., Stepanov A.S.

Methods of classification and mapping of the land cover using satellite monitoring data have recently been frequently applied to solve practical tasks in digital agriculture, including refining field boundaries and identifying unused lands. This paper discusses the recognition of arable lands using Sentinel-2 satellite images. Images with and without atmospheric correction were utilized for classifying five types of underlying surfaces in the Oktyabrsky and Leninsky districts of the Jewish Autonomous Region. Various machine learning methods and software tools were applied for image classification. It was determined that the overall classification accuracy for images with atmospheric correction exceeded 80%, which is significantly higher than the corresponding rate for uncorrected images. The obtained results were used to prepare shapefiles outlining agricultural fields in the Jewish Autonomous Region in 2022. The proposed approach can be applied to refine field boundaries at the regional level without the preparation and processing of time series of satellite images, which require substantial time and computational resources.

Prudnikova E.Y., Savin I.Y., Grubina P.G.

Satellite data have been used for a long time to assess various properties of arable soils. At the same time, there are certain difficulties associated with the fact that a number of agronomically important soil properties do not directly affect spectral reflectance of their surface, which complicates the remote assessment of such properties. In addition, to obtain reproducible models, it is necessary to take into account the state of the open soil surface during the survey. The aim of the study was to demonstrate a method for detecting agronomically important properties of arable soils based on Landsat 8-9 OLI satellite data and including information about the state of their open surface using the example of a test field in the Serebryano-Prudsky district of the Moscow region. Depending on the soil property, R2cv of the models developed based on Landsat 8-9 OLI satellite data varied from 0.57 to 0.91. The best models with R2cv>0.8 were obtained for organic matter and properties higly correlated with it such as the content of exchangeable calcium and magnesium cations, the content of total nitrogen, pH of water and salt extracts. The involvement of information on the state of the open surface of arable soils for most properties made it possible to obtain models of higher quality and predictive ability, regardless of the survey period. Based on the models obtained, maps of the spatial variation of agronomically important properties of arable soils were constructed as part of the demonstration of the method. The resulting models can be used for remote monitoring of the analyzed properties of arable soils of the test field. At the same time, for such properties as the content of exchangeable potassium and phosphorus compounds, it is necessary to search for the approaches that will take into account their high variability, as well as to perform a prior assessment of the informativity of the survey periods in which the open soil surface is not transformed.

Singh G., Singh S., Sethi G., Sood V.

Continuous observation and management of agriculture are essential to estimate crop yield and crop failure. Remote sensing is cost-effective, as well as being an efficient solution to monitor agriculture on a larger scale. With high-resolution satellite datasets, the monitoring and mapping of agricultural land are easier and more effective. Nowadays, the applicability of deep learning is continuously increasing in numerous scientific domains due to the availability of high-end computing facilities. In this study, deep learning (U-Net) has been implemented in the mapping of different agricultural land use types over a part of Punjab, India, using the Sentinel-2 data. As a comparative analysis, a well-known machine learning random forest (RF) has been tested. To assess the agricultural land, the major winter season crop types, i.e., wheat, berseem, mustard, and other vegetation have been considered. In the experimental outcomes, the U-Net deep learning and RF classifiers achieved 97.8% (kappa value: 0.9691) and 96.2% (Kappa value: 0.9469), respectively. Since little information exists on the vegetation cultivated by smallholders in the region, this study is particularly helpful in the assessment of the mustard (Brassica nigra), and berseem (Trifolium alexandrinum) acreage in the region. Deep learning on remote sensing data allows the object-level detection of the earth’s surface imagery.

Qiu B., Lin D., Chen C., Yang P., Tang Z., Jin Z., Ye Z., Zhu X., Duan M., Huang H., Zhao Z., Xu W., Chen Z.

• Automatic method developed for mapping actively cropped fields without regional adjustments. • Combined use of Sentinel-1 and Sentinel-2 increases mapping accuracy. • High performances in correctly labelling sparse crops and separating grass from cropped fields. • The first 20-m national-scale map of cropped fields and abandoned cropland generated in China. • Around 77% cropland are cropped with field crops in China at national scale. Detailed and updated maps of actively cropped fields on a national scale are vital for global food security. Unfortunately, this information is not provided in existing land cover datasets, especially lacking in smallholder farmer systems. Mapping national-scale cropped fields remains challenging due to the spectral confusion with abandoned vegetated land, and their high heterogeneity over large areas. This study proposed a large-area mapping framework for automatically identifying actively cropped fields by fusing Vegetation-Soil-Pigment indices and Synthetic-aperture radar (SAR) time-series images (VSPS). Three temporal indicators were proposed and highlighted cropped fields by consistently higher values due to cropping activities. The proposed VSPS algorithm was exploited for national-scale mapping in China without regional adjustments using Sentinel-2 and Sentinel-1 images. Agriculture in China illustrated great heterogeneity and has experienced tremendous changes such as non-grain orientation and cropland abandonment. Yet, little is known about the locations and extents of cropped fields cultivated with field crops on a national scale. Here, we produced the first national-scale 20 m updated map of cropped and fallow/abandoned land in China and found that 77 % of national cropland (151.23 million hectares) was actively cropped in 2020. We found that fallow/abandoned cropland in mountainous and hilly regions were far more than we expected, which was significantly underestimated by the commonly applied VImax-based approach based on the MODIS images. The VSPS method illustrates robust generalization capabilities, which obtained an overall accuracy of 94 % based on 4,934 widely spread reference sites. The proposed mapping framework is capable of detecting cropped fields with a full consideration of a high diversity of cropping systems and complexity of fallow/abandoned cropland. The processing codes on Google Earth Engine were provided and hoped to stimulate operational agricultural mapping on cropped fields with finer resolution from the national to the global scale.

Pereira F.R., de Lima J.P., Freitas R.G., Dos Reis A.A., Amaral L.R., Figueiredo G.K., Lamparelli R.A., Magalhães P.S.

• Nitrogen variability within pasture fields was predicted using UAV and satellite data. • UAV multispectral data were outstanding in predicting N spatial distribution. • PlanetScope data resulted in more accurate N estimates than Sentinel-2A data did. • Combining visible light UAV data with orbital images improved N prediction. • The combination of high spatial and spectral resolutions enhanced N predictions. In agricultural production, nitrogen (N) deficiency reduces yield, while overapplication may have an unwanted impact on the natural environment and farm finances. Frequent field N monitoring is impractical due to the time and cost required for traditional laboratory analysis. However, remotely sensed data are an alternative to evaluate and monitor crop nutrition status throughout the growing season. This study evaluates the spatial distribution of N in pasture fields cultivated under an integrated crop-livestock system (ICLS) using unmanned aerial vehicle (UAV) and satellites data. We assessed the performance of UAV, PlanetScope, and Sentinel-2A platforms to predict the N parameters: plant N content (PNC), aboveground biomass (AGB), and nutritional nitrogen index (NNI). Moreover, we also simulated a UAV device with a visible light sensor (i.e., red–green-blue (RGB)) as a costly alternative to near-infrared (NIR) sensors to monitor N status. Finally, we assessed whether combining the information from these platforms would improve the N predictions in our study area. The UAV, PlanetScope, and Sentinel-2A data were able to estimate N parameters in the studied pasture fields using the random forest regression algorithm. The UAV multispectral data resulted in the best prediction accuracies (R 2 : 0.84-PNC, 0.70-AGB, and 0.84-NNI). The combination of UAV_RGB with either PlanetScope (R 2 : 0.79-PNC, 0.67-AGB, 0.77-NNI) or Sentinel-2A (R 2 : 0.76-PNC, 0.57-AGB, 0.69-NNI) improved the performance of the three platforms individually (UAV_RGB, PlanetScope or Sentinel-2A). The association between high spatial and spectral resolutions contributes to the highest prediction accuracy in estimating N variability in pasture fields using remote sensing data. Our results suggest that remote sensing techniques are a reliable approach for N monitoring in commercial pasture fields under ICLS.

Blickensdörfer L., Schwieder M., Pflugmacher D., Nendel C., Erasmi S., Hostert P.

Monitoring agricultural systems becomes increasingly important in the context of global challenges like climate change, biodiversity loss, population growth, and the rising demand for agricultural products. High-resolution, national-scale maps of agricultural land are needed to develop strategies for future sustainable agriculture. However, the characterization of agricultural land cover over large areas and for multiple years remains challenging due to the locally diverse and temporally variable characteristics of cultivated land. We here propose a workflow for generating national agricultural land cover maps on a yearly basis that accounts for varying environmental conditions. We tested the approach by mapping 24 agricultural land cover classes in Germany for the three years 2017, 2018, and 2019, in which the meteorological conditions strongly differed. We used a random forest classifier and dense time series data from Sentinel-2 and Landsat 8 in combination with monthly Sentinel-1 composites and environmental data and evaluated the relative importance of optical, radar, and environmental data. Our results show high overall accuracy and plausible class accuracies for the most dominant crop types across different years despite the strong inter-annual meteorological variability and the presence of drought and non-drought years. The maps show high spatial consistency and good delineation of field parcels. Combining optical, SAR, and environmental data increased overall accuracies by 6% to 10% compared to single sensor approaches, in which optical data outperformed SAR. Overall accuracy ranged between 78% and 80%, and the mapped areas aligned well with agricultural statistics at the regional and national level. Based on the multi-year dataset we mapped major crop sequences of cereals and leaf crops. Most crop sequences were dominated by winter cereals followed by summer cereals. Monocultures of summer cereals were mainly revealed in the Northwest of Germany. We showcased that high spatial and thematic detail in combination with annual mapping will stimulate research on crop cycles and studies to assess the impact of environmental policies on management decisions. Our results demonstrate the capabilities of integrated optical time series and SAR data in combination with variables describing local and seasonal environmental conditions for annual large-area crop type mapping. • Large-area crop type mapping without region−/class-specific feature selection. • Integration of data describing local and seasonal environmental conditions. • 24 agricultural land cover classes at national scale and for multiple years. • High accuracy despite strong inter-annual meteorological variability. • Combined crop type maps enable crop sequence analysis at national scale.

Pan L., Xia H., Yang J., Niu W., Wang R., Song H., Guo Y., Qin Y.

• Cropping intensity was mapping using the harmonized Landsat and Sentinel imageries. • Crop phenology was obtained using a newly developed approach. • Crop planting rules were clarified in fragmented and complicated planting conditions. With the decline of cultivated land quality and area in recent decades, the intensification of land use plays an important role in meeting the growing demand for food. Cropping intensity refers to the number of crop planting cycles in one year, which is important for improving food production and safety at the local, regional and national scales. Therefore, it is necessary to develop an accurate high spatial resolution dataset of cropping intensity. The existing datasets of cropping intensity were generally developed based on MODIS or Landsat images, both of which have defects in spatial and temporal resolutions. In this paper, we improved the quality of the dataset on the Google Earth Engine (GEE) platform, and developed a new algorithm incorporating crop phenology. The algorithm was based on the Landsat 7/8 and Sentine-2A/B time series imageries to map the 30 m cropping intensity in the Huaihe basin in 2018 by extracting complete growth cycle. Results show that single cropping, double cropping and triple cropping in the Huaihe basin accounted for 41.6%, 57.7% and 0.7% of the total cultivated area in 2018, respectively, and the proportion of multiple cropping reached 58.4%. The accuracy of single cropping, double cropping and triple cropping are 92.93%, 91.39%, and 72.78% respectively. The overall accuracy is 91.38% and the kappa coefficient is 0.84. This algorithm accurately captures the seasonal dynamics of planting patterns in arable land, which can be used to produce cropping intensity products with high-resolution and provide a reference for large-scale regional vegetation monitoring.

Письман Т.И., Ерунова М.Г., Ботвич И.Ю., Емельянов Д.В., Кононова Н.А., Бобровский А.В., Крючков А.А., Шпедт А.А., Шевырногов А.П.

Hashiev A.B., Babakov V.P.

Tarasov A.V.

The article discusses the experience of using machine learning methods (gradient boosting, deep neural networks) for cloud detection and mapping on the example of the Perm Region using high spatial resolution images in visible range (Sentinel-2). With existing cloud detection algorithms (Fmask, Sen2Cor) Comparisons are made, as well as with the basic cloud mask provided with Sentinel-2 images. For automatic production of cartographic data Python script in the ArcGIS environment was cones fad.

Grüner E., Wachendorf M., Astor T.

Organic farmers, who rely on legumes as an external nitrogen (N) source, need a fast and easy on-the-go measurement technique to determine harvestable biomass and the amount of fixed N (NFix) for numerous farm management decisions. Especially clover- and lucerne-grass mixtures play an important role in the organic crop rotation under temperate European climate conditions. Multispectral sensors mounted on unmanned aerial vehicles (UAVs) are new promising tools for a non-destructive assessment of crop and grassland traits on large and remote areas. One disadvantage of multispectral information and derived vegetations indices is, that both ignore spatial relationships of pixels to each other in the image. This gap can be filled by texture features from a grey level co-occurrence matrix. The aim of this multi-temporal field study was to provide aboveground biomass and NFix estimation models for two legume-grass mixtures through a whole vegetation period based on UAV multispectral information. The prediction models covered different proportions of legumes (0–100% legumes) to represent the variable conditions in practical farming. Furthermore, the study compared prediction models with and without the inclusion of texture features. As multispectral data usually suffers from multicollinearity, two machine learning algorithms, Partial Least Square and Random Forest (RF) regression, were used. The results showed, that biomass prediction accuracy for the whole dataset as well as for crop-specific models were substantially improved by the inclusion of texture features. The best model was generated for the whole dataset by RF with an rRMSE of 10%. For NFix prediction accuracy of the best model was based on RF including texture (rRMSEP = 18%), which was not consistent with crop specific models.

Yakushev V.P., Kanash E.V., Yakushev V.V., Matveenko D.A., Rusakov D.V., Blokhina S.Y., Petrushin A.F., Mitrofanov E.P.

Maes W.H., Steppe K.

Remote sensing with unmanned aerial vehicles (UAVs) is a game-changer in precision agriculture. It offers unprecedented spectral, spatial, and temporal resolution, but can also provide detailed vegetation height data and multiangular observations. In this article, we review the progress of remote sensing with UAVs in drought stress, in weed and pathogen detection, in nutrient status and growth vigor assessment, and in yield prediction. To transfer this knowledge to everyday practice of precision agriculture, future research should focus on exploiting the complementarity of hyperspectral or multispectral data with thermal data, on integrating observations into robust transfer or growth models rather than linear regression models, and on combining UAV products with other spatially explicit information.

Miklashevich T.S., Bartalev S.А., Plotnikov D.E.

Dwyer J.L., Roy D.P., Sauer B., Jenkerson C.B., Zhang H.K., Lymburner L.

Data that have been processed to allow analysis with a minimum of additional user effort are often referred to as Analysis Ready Data (ARD). The ability to perform large scale Landsat analysis relies on the ability to access observations that are geometrically and radiometrically consistent, and have had non-target features (clouds) and poor quality observations flagged so that they can be excluded. The United States Geological Survey (USGS) has processed all of the Landsat 4 and 5 Thematic Mapper (TM), Landsat 7 Enhanced Thematic Mapper Plus (ETM+), Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) archive over the conterminous United States (CONUS), Alaska, and Hawaii, into Landsat ARD. The ARD are available to significantly reduce the burden of pre-processing on users of Landsat data. Provision of pre-prepared ARD is intended to make it easier for users to produce Landsat-based maps of land cover and land-cover change and other derived geophysical and biophysical products. The ARD are provided as tiled, georegistered, top of atmosphere and atmospherically corrected products defined in a common equal area projection, accompanied by spatially explicit quality assessment information, and appropriate metadata to enable further processing while retaining traceability of data provenance.

Ivanov M.A., Prishchepov A.V., Golosov V.N., Zalyaliev R.R., Efimov K.V., Kondrat’eva A.A., Kinyashova A.D., Ionova Y.K.