Near real-time agriculture monitoring at national scale at parcel resolution: Performance assessment of the Sen2-Agri automated system in various cropping systems around the world

Defourny, Pierre; Bontemps, Sophie; Bellemans, Nicolas; Cara, Cosmin; Dedieu, Gérard; Guzzonato, Eric; Hagolle, Olivier; Inglada, Jordi; Nicola, Laurentiu; Rabaute, Thierry; Savinaud, Mickaël; Udroiu, Cosmin; Valero, Silvia; Bégué, Agnès; Dejoux, Jean-François; Harti, Abderrazak El; Ezzahar, Jamal; Kussul, Nataliia; Labbassi, Kamal; Lebourgeois, Valentine; Zhang, Miao; Newby, Terry; Nyamugama, Adolph; Salh, Norakhan; Shelestov, Andrii; Simonneaux, Vincent; Traoré, Pierre Sibiry; Traoré, S.; Koetz, Benjamin

doi:10.1016/j.rse.2018.11.007

Cited by 292 publications

(204 citation statements)

References 38 publications

Supporting

Mentioning

191

Contrasting

Unclassified

Order By: Relevance

“…Sentinel-2 images were obtained from the European Space Agency's Scientific Hub and were converted to surface reflectance using the Multisensor Atmospheric Correction and Cloud-Screening (MACCS; Hagolle et al, 2010) algorithm provided in the Sen2-Agri toolbox (Defourny et al, 2019). In the main site, the Sen2-Agri was used to generate a monthly cloud-free composite of March 2017, which corresponds to the middle of the growing season.…”

Section: Satellite Data and Preprocessingmentioning

confidence: 99%

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Waldner¹,

Diakogiannis²

2020

Preprint

View full text Add to dashboard Cite

Applications of digital agricultural services often require either farmers or their advisers to provide digital records of their field boundaries. Automatic extraction of field boundaries from satellite imagery would reduce the reliance on manual input of these records which is time consuming and error-prone, and would underpin the provision of remote products and services. The lack of current field boundary data sets seems to indicate low uptake of existing methods, presumably because of expensive image preprocessing requirements and local, often arbitrary, tuning. In this paper, we address the problem of field boundary extraction from satellite images as a multi-task semantic segmentation problem. We used ResUNet-a, a deep convolutional neural network with a fully connected UNet backbone that features dilated convolutions and conditioned inference, to assign three labels to each pixel: 1) the probability of belonging to a field; 2) the probability of being part of a boundary; and 3) the distance to the closest boundary. These labels can then be combined to obtain closed field boundaries. Using a single composite image from Sentinel-2, the model was highly accurate in mapping field extent, field boundaries, and, consequently, individual fields. Replacing the monthly composite with a single-date image close to the compositing period only marginally decreased accuracy. We then showed in a series of experiments that our model generalised well across resolutions, sensors, space and time without recalibration. Building consensus by averaging model predictions from at least four images acquired across the season is the key to coping with the temporal variations of accuracy. By minimising image preprocessing requirements and replacing local arbitrary decisions by datadriven ones, our approach is expected to facilitate the extraction of individual crop fields at scale.

show abstract

Section: Satellite Data and Preprocessingmentioning

confidence: 99%

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Waldner¹,

Diakogiannis²

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Future work could improve upon the approach, especially as new datasets and computational techniques emerge. For example, machine learning shows much promise to leverage satellite data for estimating crop-specific areas [59,60]. However, this approach would be restricted to the time domain of necessary satellite data.…”

Section: Lowmentioning

confidence: 99%

Probabilistic global maps of crop-specific areas from 1961 to 2014

Jackson

Konar

Debaere

et al. 2019

Environ. Res. Lett.

View full text Add to dashboard Cite

Agriculture has substantial socioeconomic and environmental impacts that vary between crops. However, information on how the spatial distribution of specific crops has changed over time across the globe is relatively sparse. We introduce the Probabilistic Cropland Allocation Model (PCAM), a novel algorithm to estimate where specific crops have likely been grown over time. Specifically, PCAM downscales annual and national-scale data on the crop-specific area harvested of 17 major crops to a global 0.5-degree grid from 1961 to 2014. To do this, pixels are assigned into probability clusters based upon crop-specific pixel suitability (based on mean climate and soil characteristics) and gridded historical agricultural areas. PCAM maps compare relatively well with an existing gridded dataset of crop-specific areas circa 2000 (simple matching coefficient value >0.8 for all crops). PCAM estimates compare less well with time series county-level agricultural census data for the United States. Importantly, deviations between census data and PCAM benchmark estimates (driven by soil and climate suitability) can be used to infer the importance of other factors of agricultural production (e.g. labor, agricultural policy, extreme climate) in future work. Our results provide new insights into the likely changes in the spatial distribution of major crops over the past half-century.

show abstract

“…Until a few years ago, scientists had access to imagery that was either free of charge but collected at coarse to medium spatial resolution (MODIS, Landsat), or at very high spatial resolution (a few metres or less) but costly, limiting land cover mapping to a low level of details or restricted spatial coverage. Since 2014, the availability of free satellite imagery combining both high spatial (10 m) and high temporal resolutions through the Copernicus Programme has dramatically changed what can be mapped from space, increasing opportunities to both detect small elements in the landscapes and capture their seasonal variation, thereby enhancing the definition and the classification of vegetation types (Defourny et al, ; Gómez, White, & Wulder, ; Lambin & Linderman, ; Wulder, Hall, Coops, & Franklin, ).…”

Section: Introductionmentioning

confidence: 99%

Improving the accuracy of land cover classification in cloud persistent areas using optical and radar satellite image time series

et al. 2020

View full text Add to dashboard Cite

The recent availability of high spatial and temporal resolution optical and radar satellite imagery has dramatically increased opportunities for mapping land cover at fine scales. Fusion of optical and radar images has been found useful in tropical areas affected by cloud cover because of their complementarity. However, the multitemporal dimension these data now offer is often neglected because these areas are primarily characterized by relatively low levels of seasonality and because the consideration of multitemporal data requires more processing time. Hence, land cover mapping in these regions is often based on imagery acquired for a single date or on an average of multiple dates. The aim of this work is to assess the added value brought by the temporal dimension of optical and radar time series when mapping land cover in tropical environments. Specifically, we compared the accuracies of classifications based on (a) optical time series, (b) their temporal average, (c) radar time series, (d) their temporal average, (e) a combination of optical and radar time series and (f) a combination of their temporal averages for mapping land cover in Jambi province, Indonesia, using Sentinel‐1 and Sentinel‐2 imagery. Using the full information contained in the time series resulted in significantly higher classification accuracies than using temporal averages (+14.7% for Sentinel‐1, +2.5% for Sentinel‐2 and +2% combining Sentinel‐1 and Sentinel‐2). Overall, combining Sentinel‐2 and Sentinel‐1 time series provided the highest accuracies (Kappa = 88.5%). Our study demonstrates that preserving the temporal information provided by satellite image time series can significantly improve land cover classifications in tropical biodiversity hotspots, improving our capacity to monitor ecosystems of high conservation relevance such as peatlands. The proposed method is reproducible, automated and based on open‐source tools satellite imagery.

show abstract

Near real-time agriculture monitoring at national scale at parcel resolution: Performance assessment of the Sen2-Agri automated system in various cropping systems around the world

Cited by 292 publications

References 38 publications

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Extracting field boundaries from satellite imagery with a convolutional neural network to enable smart farming at scale

Probabilistic global maps of crop-specific areas from 1961 to 2014

Improving the accuracy of land cover classification in cloud persistent areas using optical and radar satellite image time series

Contact Info

Product

Resources

About