当前位置:舍宁秘书网 > 专题范文 > 公文范文 > Mapping,winter,rapeseed,in,South,China,using,Sentinel-2,data,based,on,a,novel,separability,index

Mapping,winter,rapeseed,in,South,China,using,Sentinel-2,data,based,on,a,novel,separability,index

时间:2024-09-04 10:30:01 来源:网友投稿

TAO Jian-bin,ZHANG Xin-yue,WU Qi-fan,WANG Yun

Key Laboratory for Geographical Process Analysis &Simulation of Hubei Province/School of Urban and Environmental Sciences,Central China Normal University, Wuhan 430079, P.R.China

Abstract Large-scale crop mapping using remote sensing data is of great significance for agricultural production,food security and the sustainable development of human societies.Winter rapeseed is an important oil crop in China that is mainly distributed in the Yangtze River Valley.Traditional winter rapeseed mapping practices are insufficient since they only use the spectral characteristics during the critical phenological period of winter rapeseed,which are usually limited to a small region and cannot meet the needs of large-scale applications.In this study,a novel phenology-based winter rapeseed index (PWRI) was proposed to map winter rapeseed in the Yangtze River Valley.PWRI expands the date window for distinguishing winter rapeseed and winter wheat,and it has good separability throughout the flowering period of winter rapeseed.PWRI also improves the separability of winter rapeseed and winter wheat,which traditionally have been two easily confused winter crops.A PWRI-based method was applied to the Middle Reaches of the Yangtze River Valley to map winter rapeseed on the Google Earth Engine platform.Time series composited Sentinel-2 data were used to map winter rapeseed with 10 m resolution.The mapping achieved a good result with overall accuracy and kappa coefficients exceeding 92% and 0.85,respectively.The PWRI-based method provides a new solution for high spatial resolution winter rapeseed mapping at a large scale.

Keywords: phenology-based winter rapeseed index,winter rapeseed mapping,Sentinel-2,Google Earth Engine

Ending hunger,one of the most important Sustainable Development Goals of United Nations,is challenging for the sustainable development of human societies(Teklemariamet al.2016;Sadriet al.2020).Rapeseed(BrassicanapusL.) is one of the most essential oil crops in China,as well as the third largest oil crop in the world (Velosoet al.2017).Winter rapeseed plantings are mainly distributed in the Yangtze River Valley,which accounts for more than 90% of the total rapeseed planting area and yield in China (Taoet al.2019).In recent years,due to the rapid urbanization and growth in the global crop trade,the planting area of winter crops is shrinking and the associated cropping intensity is also decreasing continuously (Taoet al.2020).China used to be a net exporter of rapeseed,but now it has become the largest importer (Zhanget al.2020).Rapeseed and soybean are the most important raw materials for cooking oil in China,so the blocking of soybean imports caused by international trade issues may have created an opportunity for the domestic rapeseed industry.Information on the spatial distribution of rapeseed in major-producing regions is crucial for policy makers,farmers and relevant stakeholders in their decisionmaking processes.However,a generalized method to map rapeseed at regional or larger scales in China is still lacking,so the spatial-temporal dynamics of rapeseed in China remain unclear.Therefore,it is urgent to monitor the rapeseed planting area in an efficient way for the vigorous development of the Chinese rapeseed industry.

Identifying winter rapeseed in remote sensing images is challenging because it is easily confused with other winter crops,such as winter wheat,due to their similar phenological characteristics (Tianet al.2019a).In previous studies,considerable efforts were made to map rapeseed by capturing the crucial phenological characteristics based on multi-spectral remote sensing images.Three kinds of methods have been used in the existing studies: machine learning methods (Griffithset al.2019;Taoet al.2019;Yanget al.2019;Zhanget al.2022),phenology based methods (Ashourlooet al.2019;Tianet al.2019b;Taoet al.2020;Zanget al.2020;Hanet al.2021) and feature transformation based methods(Wanget al.2018).Machine learning methods are highly dependent on the training samples,and the high cost of collecting samples and variations in samples restrain its wide application (Donget al.2016;Ashourlooet al.2019;Zanget al.2020).Phenology-based methods are highly dependent on spectral characteristics in the flowering date of rapeseed (Wanet al.2018;Yinet al.2020).However,flowering data may vary because of phenology differences and cultivation habits (Wanget al.2018;Hanet al.2021).Furthermore,studies that depend on flowering characteristics of rapeseed are usually limited to local scales (Hanet al.2018;Wanget al.2018).

Phenology-based methods can be improved when the flowering date information is obtained.There are some studies on detecting the flowering date using timeseries remote sensing data (D’Andrimontet al.2020;Hanet al.2020).Nevertheless,that detection itself is also dependent on the spectral characteristics of rapeseed.Microwave data,such as Sentinel-1 data,mitigate or eliminate cloud interference and facilitate rapeseed mapping (Zhanget al.2018;D’Andrimontet al.2020).However,the insufficient historical record of microwave data makes it impossible to conduct long-term rapeseed mapping applications (Zhanget al.2022).Furthermore,microwave data are usually used in combination with multi-spectral data in practice.

The methods based on spectral characteristics in the flowering date of rapeseed are popular at present.When it comes to flowering,the yellow flowers changed the value of the vegetation index of rapeseed significantly(Shenet al.2010).The existing studies usually take the normalized difference vegetation index (NDVI) (Wang and Zhang 2015;Taoet al.2020),normalized difference yellowness index (NDYI) (Sulik and Long 2016;Zanget al.2020) or other reflectance-based indexes such as Canola index (CI)=NIR×(Red+Green) (Ashourlooet al.2019) in the flowering date as proxies for mapping rapeseed.The flowering date usually lasts about 30 days(Hanet al.2021).The reflectance-derived vegetation indexes of rapeseed and wheat are similar when remote sensing observations are not taken at the peak flowering date (Wanget al.2018).Therefore,the date window for obtaining good-quality spectral images is narrow,restricting its application at a large scale (Zhanget al.2022).The application of phenology-based methods can be expanded when the difference between the two winter crops can be compared over a longer growing period.Most crops have a unique phenological patterns during the growing season (Ashourlooet al.2019) and examining this pattern at a wider date window may have the potential for developing new features for rapeseed mapping.For example,winter rapeseed and winter wheat have different leaf status conditions at different stages.Therefore,extending the observation window from the flowering date to a wider growing season may offer new possibilities for rapeseed mapping.

There is a trade-off between data coverage and the observation date window when multi-temporal remote sensing data are used.Mining the spectral characteristics during a longer period of winter rapeseed will extend the observation date window,expanding data coverage and making the application on a larger scale possible.There is an urgent need to develop a generalized method that can make full use of the spectral differences between winter rapeseed and winter wheat during the whole growing season in order to map winter rapeseed efficiently at a large scale.The launching of the Sentinel-2A&B satellite with 10 m spatial resolution and 5-day revisit cycles in 2015 and 2017 offers great potential for winter rapeseed mapping at a finer scale in South China.Sentinel-2 has four Red-edge bands which may be more capable of distinguishing the leaf states of the two winter crops at different stages than near-infrared.Moreover,the Google Earth Engine (GEE) provides an unprecedented opportunity for large-scale crop mapping due to its massive remote sensing dataset as well as numerous image processing algorithms.

The objective of this study was to present a novel separability index for mapping winter rapeseed in the Yangtze River Valley based on a Sentinel-2 time series.We proposed a phenology-based winter rapeseed index(PWRI) to efficiently separate winter rapeseed and winter wheat by analyzing the unique spectral signatures of winter rapeseed in different stages.The PWRI expanded the date window of separating winter rapeseed and winter wheat so that we could map winter rapeseed over a broader area.PWRI also enlarged the separability between winter rapeseed and winter wheat,which traditionally have been two easily confused winter crops.A dynamic threshold-based hierarchical classification method was used to identify the winter rapeseed by integrating multi-temporal Sentinel-2 images on the GEE platform.This method provides a new solution for high spatial resolution winter rapeseed mapping at a large scale.

2.1.Study area,data and preprocessing

The study area is located in the Middle Reaches of the Yangtze River Valley.It includes parts of Hubei,Hunan,Anhui,and Jiangxi provinces,with latitudes ranging from 27.95°N to 32.45°N,and longitudes ranging from 111.16°E to 119.17°E.It is dominated by a subtropical monsoon climate and the major crop types include paddy rice,winter rapeseed,winter wheat,soybean,and others.This region has the largest winter rapeseed planting area,accounting for more than 50% of the total rapeseed area in China.

The data sources used in this study include the Sentinel-2 MSI data,MODIS vegetation index product (i.e.,MOD13Q1) and land cover products (i.e.,GlobeLand30)for 2020.We collected all available surface reflectance data of Sentinel-2 from November 1,2019 to May 30,2020,which covered the whole growing period of winter rapeseed.These Sentinel-2 images were orthographically rectified and atmospherically corrected.The QA60 bitmask band (quality flag) was used to detect and remove clouds and cirrus (Zhu and Woodcock 2012;Royet al.2014).Eight bands,including blue,green,red,nirinfrared and four red-edge bands (Bands 5,6,7,and 8A),were used in this study.The full-coverage high-quality monthly composite images were produced using the median composite method,which covered the growing season of winter rapeseed.MOD13Q1 time series were used to analyze the phenological information of different crop types because of its short revisit cycle and rich spectral bands.GlobeLand30 (Junet al.2014) was used to obtain the cropland extent,which is the base map for the winter rapeseed mapping.

All data except for GlobeLand30 were processed in the GEE cloud computing platform (https://earthengine.google.org/).GEE is a powerful cloud-based platform which provides a vast pool of satellite datasets.Since the latest version of GlobeLand30 was not provided in GEE,we downloaded it from the official website (http://www.globallandcover.com/) and uploaded it to GEE.Detailed information on the data is given in Table 1.

Table 1 The description of remote sensing data used in this study

Winter rapeseed mapping was conducted using winter crops map as the base map.Winter crops were obtained by masking out winter fallow field areas using a winter crop index (WCI) on the basis of cropland extent.As this study focused on distinguishing winter rapeseed from winter wheat,the two major winter crops in the study area,PWRI was designed to specifically improve the separability of winter rapeseed and winter wheat.Then,a localized threshold searching strategy was used to determine the optimal thresholds for clustering the PWRI.The flowchart of the method is presented in Fig.1.

Fig.1 Flowchart of the method.

2.2.Development of the winter crop index for identifying winter crops

A winter crop index was used to identify winter crops by masking out winter fallow fields on the basis of the GlobeLand30 land-cover product.According to the growing period of winter rapeseed (from November to May of the next year),croplands in the Yangtze River Valley in the winter season generally include three types,i.e.,winter rapeseed,winter wheat and winter fallow field.Winter crops share similar phenological characteristics,which are totally different from those of winter fallow fields.

To analyze the separability among these three types,time series MODIS NDVI of winter crops and winter fallow fields were drawn based on 5 000 samples,as shown in Fig.2.We found that the winter crops entered a stage of rapid growth as the temperature gradually increased in the mid February (named ‘T1’),while winter fallow fields remained in a fallow state.In mid to late May (named‘T2’),the winter crops were being harvested while sowing(usually paddy rice or soybean) started in the winter fallow fields.

Fig.2 MODIS normalized difference vegetation index (NDVI)curves for winter crops and winter fallow field.Mean NDVI values derived from about 5000 crop samples were used to draw this figure.DOY refers to the Julian day of a year.WR,WW and WFF represent winter rapeseed,winter wheat and winter fallow field,respectively.T1,the mid February;T2,mid to late May.

According to the distinct phenological difference between winter crops and winter fallow field,WCI was designed to separate them,which was calculated by eq.(1):

where T1 and T2 represent mid-February and mid-May to late May respectively,and NDVI was calculated from the red and nir-infrared bands.

The histograms of WCI for winter rapeseed,winter wheat and winter fallow fields are shown in Fig.3.Although there is some overlap between winter rapeseed and winter wheat,they presented significant differences from winter fallow field.

Fig.3 Winter crop index (WCI) histograms of winter crops and winter follow field.WR,WW and WFF represent winter rapeseed,winter wheat and winter fallow field,respectively.

2.3.Development of the winter rapeseed index for identifying winter rapeseed

Winter rapeseed and winter wheat have similar growing periods,both of which are sown in late October and harvested in late May of the next year (Table 2).The key point for extracting winter rapeseed is to improve the separability between it and winter wheat (Hanet al.2020).

Table 2 Cropping calendars of the two major winter crops in the study area

The growing season of winter rapeseed generally includes two periods,i.e.,the period prior to budding (PPB,including the seedling stage and part of the budding stage) that lasts from late November to early February of the next year,and the period after budding (PAB,including part of the budding,flowering and pod stages) that lasts from mid-February to late April,according to the leaf status (Table 2).Roughly taking budding (mid-February)as turning point,winter rapeseed and winter wheat presented distinct leaf status features (chlorophyll content and canopy structure) at different stages (Herbertssonet al.2017).As illustrated in Fig.4,during PPB,winter rapeseed grew slowly due to the low temperatures and their leaves were in the form of blades,whereas those of winter wheat were in the form of needles.Compared to winter rapeseed,winter wheat had smaller leaves,so the plantings are more populated by bare soil,and it had a smaller leaf area index (LAI).During PAB,the winter crops grew rapidly as the temperature rises gradually.The canopy of winter rapeseed was gradually covered with flowers and the chlorophyll in the canopy of winter rapeseed started to decrease while winter wheat started tillering and jointing.Unlike in the previous stage,the leaves of winter wheat almost covered the ground completely,and it had a larger LAI than winter rapeseed(Fig.4).

Fig.4 Leaf status of winter rapeseed and winter wheat at different stages.PPB,period prior to budding;PAB,period after budding.

To characterize these phenological differences between winter rapeseed and winter wheat,we calculated the green chlorophyll index (GCVI) of winter rapeseed and winter wheat.Previous studies have shown that GCVI has an obvious linear relationship with LAI (Gitelsonet al.2003;Kanget al.2016),so it is a good indicator forreflecting the status of LAI.GCVI was calculated using eq.(2):

where NIR and Green represent the reflectance values of the near-infrared band and green bands,respectively.The time series GCVI for both crops during the growing season of winter rapeseed are shown in Fig.5-A.The GCVI value of winter rapeseed is significantly higher than that of winter wheat in PPB,but this phenomenon is reversed in PAB.

Fig.5 Sentinel-2 time series GCVI (A) and red-edge (B) of winter rapeseed and winter wheat within their lifecycles.PPB,period prior to budding;PAB,period after budding.

The red-edge is closely related to physicochemical parameters of the vegetation and plays an important role in agricultural remote sensing (Delegidoet al.2011).Contrary to the trend of time-series GCVI,the red-edge band of winter rapeseed is higher than that of winter wheat during the whole growing season (Fig.5-B).

Based on the differences between winter rapeseed and winter wheat in the time series GCVI and red-edge,PWRI was designed to increase the separability between winter rapeseed and winter wheat.The PWRI is expressed in the following eq.:

where RE is the reflectance of red-edge band 6,andais a scaling coefficient to balance the data ranges of the two items.It is an empirical parameter,and was set to 20 after several trials so that PWRI can stay within a reasonable value range.

2.4.Separability evaluation

The separability index (SI) was calculated to quantitatively evaluate the separability of winter rapeseed and winter wheat based on different features.SI is the ratio of interclass variability and intra-class variability (Somers and Asner 2012),and its eq.is as follows:

whereM1andM2represent the mean values of the samples,andV1andV2represent the standard deviations.The higher the SI value,the better the separability between two classes.

2.5.Clustering based on dynamic threshold

Due to the vegetation phenological differences caused by regional differentiation,the optimal threshold for separating crop types could also be variable when applied to large-scale mapping.Therefore,the Otsu algorithm,also known as the maximum class square error method (Caoet al.2018;Srinivaset al.2019),was used to determine the optimal threshold automatically.The algorithm exhaustively searches for the threshold that maximizes the inter-class variance of different land cover classes,and is expressed as follow:

wherew1andw2are the proportions of target and background pixels;and m,m1,andm2correspond to the mean values of all pixels,target pixels and background pixels,respectively.

Once the thresholds were determined,the winter rapeseed was identified using a binary classification issue by clustering the PWRI image into target and background.A composited PWRI (which was calculated based on composited images during PAB) was used to cluster because every pixel may have a different composite date.In order to alleviate the intra-class spectral variances in different regions,the whole study area was divided into nine rectangular sub-regions,each spanning 1.5 degrees in latitude and 2.66 degrees in longitude.The Otsu algorithm was then used to determine the optimal thresholds of PWRI and a separate local threshold was used for each sub-region.

2.6.Validation

Nine patches (about 4 000 hectares each) from those regions with dense winter rapeseed distribution were selected as the validation patches.Three validation patches (D,E and F,in which each patch has more than 1 000 samples) along the Yangtze River were selected as the validation patches.Their central locations are: D(30.0693°N,112.3244°E),E (29.9932°N,115.4876°E) and F (30.1453°N,117.1514°E).The validation samples were collected from field surveys as well as visual interpretation on the Google Earth true-color composite image.The overall accuracies and Kappa coefficients were used for accuracy assessment.All land cover types except for winter rapeseed were combined and named as ‘others’.

3.1.Improvement of the separability

The true color composite image and PWRI image (a patch from Sentinel-2 scene T49RFP) are visually compared in Fig.6.In the true color composite image,winter rapeseed presents as obvious bright yellow while winter wheat presents as dark green.In the PWRI feature image,their levels of brightness are completely different and they can be easily distinguished.

Fig.6 Comparison of the true color composite image (A) (March 18) and the phenology-based winter rapeseed index (PWRI)feature image (B) (composite image from March 1 to March 31).

The histograms of GCVI,red-edge and PWRI (winter rapeseedvs.winter wheat) indicate that the statistical differences between winter rapeseed and winter wheat are obvious in PWRI (Fig.7).

Fig.7 Histograms of the three features (composite image from March 1 to March 31),green chlorophyll index (GCVI) (A),red edge (B),and phenology-based winter rapeseed index (PWRI) (C).About 4 000 samples from the Jianghan Plain were selected.WR,winter rapeseed;WW,winter wheat.

Several indices from existing studies (Table 3) were selected to visually compare their separabilities in the whole growing season with PWRI (Fig.8).

Fig.8 Comparing the separability of winter rapeseed and winter wheat on different features,normalized difference vegetation index (NDVI) (A),normalized difference yellowness index (NDYI) (B),normalized rapeseed flowering index (NRFIr) (C),winter rapeseed index (WRI) (D),Canola index (CI) (E) and phenology-based winter rapeseed index (PWRI) (F).WR,winter rapeseed;WW,winter wheat;WFF,winter fallow field.

Table 3 Reflectance-based spectral indices

Table 4 Confusion matrix of the three validation sub-regions with local thresholds

Table 5 Confusion matrix of the three validation sub-regions with a global threshold

The time series indices demonstrate that winter rapeseed and winter wheat have totally different separability from each other (Fig.8).The confusion between winter rapeseed and winter wheat on the time series NDVI exists throughout the growing season,except for significant separability in mid-March (peak flowering date).The separabilities of the other indices (NDYI,NRFIr,WRI and Canola index) were higher than that of NDVI,but they were still prominent only on the flowering date.However,for the time series PWRI,the separability continues to run high in PAB,with winter rapeseed on the top and winter wheat on the bottom of the feature space.Compared with conventional features such as NDVI,NDYI,etc.,PWRI greatly improved the separability of the winter crops in PAB.

3.2.Winter rapeseed map

The spatial distribution of winter rapeseed in the Middle Reaches of Yangtze River Valley can be obtained by using the PWRI.Winter rapeseed was mainly distributed on the Jianghan Plain in Hubei Province,the Dongting Lake in Hunan Province,and regions along the Yangtze River in Anhui Province.

Nine validation patches were selected for comparing the winter rapeseed map with the true color images(Fig.9).The winter rapeseed maps had good consistency with the true color images and reflected the overall winter rapeseed pattern in those croplands with dense winter rapeseed distributions.

Fig.9 Comparisons of the true color images (left column) and winter rapeseed maps (right column) of the nine validation patches(A–I) in the research area.The central geographical locations of A–I are as follows: A (31.3546°N,112.1483°E),B (31.0563°N,113.8827°E),C (31.2045°N,118.4772°E),D (30.0693°N,112.3244°E),E (29.9932°N,115.4876°E),F (30.1453°N,117.1514°E),G (29.1517°N,112.1486°E),H (28.7673°N,115.5757°E),and I (29.0751°N,117.1581°E).

3.3.Accuracies

The confusion matrix,overall accuracies and Kappa coefficients of the three validation patches with local thresholds were compared to those with a global threshold (Tables 4 and 5).The overall accuracies of the three validation patches all exceed 92%,and the Kappa coefficients are all greater than 0.85,demonstrating the reliability of the method.Therefore,the dynamic threshold method was more flexible when it came to large-scale winter rapeseed mapping.

PWRI was also compared with other published indices(NDVI,NDYI,NRFIr,WRI and Canola index) regarding the mapping accuracies (Table 6).The results show that only PWRI could guarantee high accuracies in all three validation regions.

Table 6 Comparing the accuracies of the indices in the three validation sub-regions with local thresholds

4.1.PWRI meets the requirements for large-scale winter rapeseed mapping

Traditional winter rapeseed mapping methods usually take spectral bands in the peak flowering date of winter rapeseed as proxies,and are heavily dependent on data availability.The peak flowering date of winter rapeseed in the Yangtze River Valley usually lasts from the middle of March to late March,within which winter rapeseed and winter wheat have good separability on all features (Fig.10).If conventional features (NDVI,NDYI,etc.) are used,only a narrow date window can be used.On the other hand,PWRI has good separability throughout the period of mid-February to late April.Obviously,PWRI can separate winter rapeseed and winter wheat across a broader date window.

Fig.10 Comparisons of the separability index (SI) values of winter rapeseed and winter wheat on different features at different stages.PWRI,phenology-based winter rapeseed index;WRI,winter rapeseed index;NDVI,normalized difference vegetation index;NDYI,normalized difference yellowness index;NRFIr,normalized rapeseed flowering index.

Conventional spectrum-based methods use remote sensing images only at the peak flowering date of winter rapeseed,so the mapping is heavily dependent on data availability and usually limited to small areas.In our method,the spectral features of winter rapeseed during the whole flowering period were analyzed,so the multi-temporal information of winter rapeseedvs.winter wheat throughout the growing season of winter rapeseed was fully utilized.This expansion of the date window facilitates the collection of remote sensing observations and nearly full coverage of the entire Yangtze River Valley can be obtained.

The very brief peak flowering date of winter rapeseed leads to the lack of available remote sensing images in many regions.There are no effective observations in many regions (Fig.11-A) during the peak flowering date,which limits the application of methods based on timeseries NDVI or NDYI to large-scale winter rapeseed mapping.Compared with other features,PWRI expands the date window of image acquisition,and has the advantage of distinguishing winter rapeseed from winter wheat in a wider date window (from mid-February to late April).Since at least one observation was recorded in most regions of the study area during PAB (Fig.11-B),the coverage is almost full so it can meet the requirement of large-scale winter rapeseed mapping.

Fig.11 Number of high-quality observations of Sentinel-2 images per pixel during the growing season of winter rapeseed.A,peak flowering date.B,period after budding (PAB).

4.2.Comparison of different red-edge bands

To explore the performance of different red-edge bands of Sentinel-2 for calculating PWRI,their SI values were visually compared for four red-edge bands (Bands 5,6,7,and 8A) and the near-infrared band (Band 8) (Fig.12).The separabilities are generally acceptable on Bands 6,7,8 and 8A during PAB.However,Band 6 has the highest separability,so it is the band most sensitive to the leaf status of winter crops.

Fig.12 Comparison of the separability index (SI) values of winter rapeseed and winter wheat on phenology-based winter rapeseed index (PWRI) using different red-edge bands at different stages.

4.3.Determining the scaling coefficient

Empirical methods have proven to be effective for

1)PWRI,phenology-based winter rapeseed index;NDVI,normalized difference vegetation index;NDYI,normalized difference yellowness index;NRFIr,normalized rapeseed flowering index;WRI,winter rapeseed index;CI,Canola index.

2)D,Jianghan Plain;E,Qichun County;F,Wuhu City.

3)OA,overall accuracy.threshold and parameter determination (Qiuet al.2015;Ashourlooet al.2019).The empirical method was used here to compare the overall accuracies by using different scaling coefficients,with a step of 5 and a range from 0 to 50 (taking the validation area D as an example).As Fig.13 shows,the accuracy increased greatly then decreased slightly,and the highest accuracy of 93.9% was reached when the coefficient was set to 20.Therefore,20 was considered as a suitable scaling coefficient for separating winter rapeseed from the others in this study.

Fig.13 The accuracies obtained by using different scaling coefficients.

4.4.Uncertainties of PWRI

Although the method developed here has achieved promising performance,it inevitably includes some uncertainties.First is the uncertainty derived from the cropland mask and winter crop mask.The mask products are from remote sensing images and have some inherent errors (e.g.,the overall accuracy of the GlobeLand30 V2020 product is 85.7%).Secondly,many non-grain crops (such as cauliflower and herbal plants etc.) have phenologies that are similar to winter rapeseed,which would also lead to confusion between them.In addition,the mixed pixel problem caused by the small field size in the Yangtze River Valley will also affect the accuracy.

The composite image itself inevitably introduces more errors.For example,by analyzing the pixel acquisition date (Julian day) distribution of the PWRI image (taking the composite image of March 1 to April 1 as an example),we found that vast majority regions had similar pixel acquisition dates ranging from DOY 75 to 81.However,there were still a few pixels with broader acquisition dates (ranging from DOY 60 to 67,67 to 75,81 to 88),resulting in the dispersion of the acquisition dates of PWRI,thus affecting the accuracy of the threshold-based segmentation method.

High-spatial-resolution and large-scale winter rapeseed mapping is of great significance for food security and agricultural regionalization.This paper reports our work on mapping winter rapeseed in the Yangtze River Valley using time-series remote sensing images on the GEE platform.The temporal differences between winter rapeseed and winter wheat,two major winter crops that were traditionally difficult to distinguish,were explored,and a novel PWRI was proposed to improve their separability.Moreover,a hierarchical classification method was used to map winter rapeseed following the process of “cropland–winter crops–winter rapeseed”.This method can map winter rapeseed in the Yangtze River Valley with nearly full coverage in space,and with the overall accuracy exceeding 90% and a Kappa coefficient larger than 0.85.

This study addressed the problems of high-spatialresolution winter rapeseed mapping at a large-scale and made progress in two ways.Firstly,PWRI expanded the spatial and temporal scales of winter rapeseed monitoring.PWRI is effective throughout stage 2,rather than the fullbloom stage.Therefore,the mapping can be conducted throughout a wider date window and over a broader area.Secondly,PWRI increased the separability of these two winter crops,and showed good separability throughout the flowering period of winter rapeseed.PWRI provided a solution to the confusion of these two winter crops and the lack of data in large-scale winter rapeseed mapping.The results demonstrate the ability of the proposed method to map winter rapeseed in the Yangtze River Valley,thereby providing a solution for winter rapeseed mapping in South China.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (41971371),the National Key Research and Development Program of China (2022YFB3903504) and the Fundamental Research Funds for the Central Universities,China(CCNU22JC022).The authors appreciate the comments and suggestions from anonymous reviewers.

Declaration of competing interest

The authors declare that they have no conflict of interest.

推荐访问:South China rapeseed

猜你喜欢