How much does multi-temporal Sentinel-2 data improve crop type classification?

Vuolo, Francesco; Neuwirth, Martin; Immitzer, Markus; Atzberger, Clement; Ng, Wai‐Tim

doi:10.1016/j.jag.2018.06.007

Cited by 228 publications

(157 citation statements)

References 34 publications

Supporting

Mentioning

113

Contrasting

Unclassified

Order By: Relevance

“…When considering all of the crops in the study area, the accuracy of crop separation is highest in the mid-season temporal window between 1362 and 2016 AGDD but accuracy reaches a maximum at 1556 AGDD. This is in line with other studies that report best accuracies for crop separation with data sets acquired in July [2,40,41]. At this time of the growing season, the winter crops are in a senescence stage and the summer crops in their most productive stage.…”

Section: Temporal Windows During the Growing Seasonsupporting

confidence: 92%

See 1 more Smart Citation

Optimal Timing Assessment for Crop Separation Using Multispectral Unmanned Aerial Vehicle (UAV) Data and Textural Features

2019

View full text Add to dashboard Cite

The separation of crop types is essential for many agricultural applications, particularly when within-season information is required. Generally, remote sensing may provide timely information with varying accuracy over the growing season, but in small structured agricultural areas, a very high spatial resolution may be needed that exceeds current satellite capabilities. This paper presents an experiment using spectral and textural features of NIR-red-green-blue (NIR-RGB) bands data sets acquired with an unmanned aerial vehicle (UAV). The study area is located in the Swiss Plateau, which has highly fragmented and small structured agricultural fields. The observations took place between May 5 and September 29, 2015 over 11 days. The analyses are based on a random forest (RF) approach, predicting crop separation metrics of all analyzed crops. Three temporal windows of observations based on accumulated growing degree days ( AGDD) were identified: an early temporal window (515-1232 AGDD, 5 May-17 June 2015) with an average accuracy (AA) of 70-75%; a mid-season window (1362 AGDD, 25 June-22 July 2015 with an AA of around 80%; and a late window (2626-3238 AGDD, 21 August-29 September 2015 with an AA of <65%. Therefore, crop separation is most promising in the mid-season window, and an additional NIR band increases the accuracy significantly. However, discrimination of winter crops is most effective in the early window, adding further observational requirements to the first window.

show abstract

Section: Temporal Windows During the Growing Seasonsupporting

confidence: 92%

“…As shown in a range of other studies, multitemporal data sets improve the accuracies of crop classification tasks [41,47]. Therefore, it would be interesting to investigate the potential of combining data from different acquisition dates to find the most promising temporal combinations for crop separation.…”

Section: Limitations and Outlookmentioning

confidence: 95%

Optimal Timing Assessment for Crop Separation Using Multispectral Unmanned Aerial Vehicle (UAV) Data and Textural Features

2019

View full text Add to dashboard Cite

show abstract

“…The Sentinel-2 satellite has a high temporal resolution, providing data more frequently than other medium resolution sensors. The frequent image revisit time increases the potential to capture well-timed images with spectral differences between the classes of interest [9,34]. Future studies in the boreo-nemoral region should focus on obtaining imagery from the first part of May (if possible), and also from mid-fall during senescence when there is a gradient in phenological activity.…”

Section: Discussionmentioning

confidence: 99%

Tree Species Classification with Multi-Temporal Sentinel-2 Data

2018

View full text Add to dashboard Cite

The Sentinel-2 program provides the opportunity to monitor terrestrial ecosystems with a high temporal and spectral resolution. In this study, a multi-temporal Sentinel-2 data set was used to classify common tree species over a mature forest in central Sweden. The tree species to be classified were Norway spruce (Picea abies), Scots pine (Pinus silvestris), Hybrid larch (Larix × marschlinsii), Birch (Betula sp.) and Pedunculate oak (Quercus robur). Four Sentinel-2 images from spring (7 April and 27 May), summer (9 July) and fall (19 October) of 2017 were used along with the Random Forest (RF) classifier. A variable selection approach was implemented to find fewer and uncorrelated bands resulting in the best model for tree species identification. The final model resulting in the highest overall accuracy (88.2%) came from using all bands from the four image dates. The single image that gave the most accurate classification result (80.5%) was the late spring image (27 May); the 27 May image was always included in subsequent image combinations that gave the highest overall accuracy. The five tree species were classified with a user's accuracy ranging from 70.9% to 95.6%. Thirteen of the 40 bands were selected in a variable selection procedure and resulted in a model with only slightly lower accuracy (86.3%) than that using all bands. Among the highest ranked bands were the red edge bands 2 and 3 as well as the narrow NIR (near-infrared) band 8a, all from the 27 May image, and SWIR (short-wave infrared) bands from all four image dates. This study shows that the red-edge bands and SWIR bands from Sentinel-2 are of importance, and confirms that spring and/or fall images capturing phenological differences between the species are most useful to tree species classification.

show abstract

“…Better results were achieved by incorporating radar data from Sentinel-1. Vuolo et al [27], also using random forest, classified multi-date Sentinel-2 datasets. They achieved overall accuracy of~95% in mid-August based on stacks of five and eight cloud-free images in 2016 and 2017 growing seasons respectively.…”

Section: Discussionmentioning

confidence: 99%

“…To construct the second dataset, consisting of multi-date and multi-spectral bands, quick-looks of all available Sentinel-2 images for 2017-2018 agricultural year were visually examined and three cloud-free images (less than 5% cloud cover over the test sites) were selected in each of the two test sites. The approach was analogous of what was reported by Vuolo et al [27]. The images registered on 9 April, 29 April, and 8 June 2018 for "Knezha" test site and on 12 December 2017, 1 May, and 31 May 2018 for "Belozem" test site were assembled.…”

Section: Sentinel-2 Data and Pre-processingmentioning

confidence: 93%

Sub-Pixel Crop Type Classification Using PROBA-V 100 m NDVI Time Series and Reference Data from Sentinel-2 Classifications

et al. 2019

View full text Add to dashboard Cite

This paper presents the results of a sub-pixel classification of crop types in Bulgaria from PROBA-V 100 m normalized difference vegetation index (NDVI) time series. Two sub-pixel classification methods, artificial neural network (ANN) and support vector regression (SVR) were used where the output was a set of area fraction images (AFIs) at 100 m resolution with pixels containing estimated area fractions of each class. High-resolution maps of two test sites derived from Sentinel-2 classifications were used to obtain training data for the sub-pixel classifications. The estimated area fractions have a good correspondence with the true area fractions when aggregated to regions of 10 × 10 km 2 , especially when the SVR method was used. For the five dominant classes in the test sites the R 2 obtained after the aggregation was 86% (winter cereals), 81% (sunflower), 92% (broad-leaved forest), 89% (maize), and 67% (grasslands) when the SVR method was used.including the linear mixture model (LMM) [4], artificial neural network (ANN) [9-11], regression tree [5], fuzzy classification [6,12], and support vector machine (SVM) [13].Liu and Wu [11] argued that the non-linear models, especially neural network-based models, outperformed the traditional linear unmixing models. Support for such a conclusion is given in Verbeiren et al.[1] who compared ANN and LMM in an attempt to generate a sub-pixel map from SPOT-VEGETATION 1 km data across Belgium. The authors showed that the ANN approach outperformed LMM and that, for the major classes, the acreage estimates obtained via ANN, when aggregated to the level of the administrative regions, were in good agreement with the true values. Also, a multilayer perceptron (MLP) neural network regression algorithm has been shown to outperform the regression tree algorithm [5]. Atkinson et al. [9] also obtained better results with ANN than with the other tested methods but pointed out that its successful implementation depends on accurate co-registration and the availability of a training data set. Liu et al.[14] compared a linear spectral unmixing model, a supervised fully-fuzzy classification method and a SVM to generate a fraction map and achieved the most accurate fraction result using SVM. Six machine learning methods were compared in a recent study [15] based on multiple criteria, where the authors found that, in general, no method performs best for all evaluation criteria. However, when both time available for preprocessing and the magnitude of the training data set are unconstrained, support vector regression (SVR) and least-squares SVM for regression clearly outperform the other methods.Regarding the satellite imagery widely used for agricultural monitoring, SPOT-VEGETATION (SPOT-VGT) sensors provided one of the longest time series of multispectral reflectance since 1998. The mission was succeeded in 2013 by PROBA-V (PRoject for On-Board Autonomy-Vegetation), a small satellite commissioned by the European Space Agency. The sensor on-board PROBA-V generates products at three different ...

show abstract

How much does multi-temporal Sentinel-2 data improve crop type classification?

Cited by 228 publications

References 34 publications

Optimal Timing Assessment for Crop Separation Using Multispectral Unmanned Aerial Vehicle (UAV) Data and Textural Features

Optimal Timing Assessment for Crop Separation Using Multispectral Unmanned Aerial Vehicle (UAV) Data and Textural Features

Tree Species Classification with Multi-Temporal Sentinel-2 Data

Sub-Pixel Crop Type Classification Using PROBA-V 100 m NDVI Time Series and Reference Data from Sentinel-2 Classifications

Contact Info

Product

Resources

About