A spectral index for winter wheat mapping using multi-temporal Landsat NDVI data of key growth stages

https://doi.org/10.1016/j.isprsjprs.2021.03.015Get rights and content

Abstract

Winter wheat is one of the most important staple crops in the world. Accurate and timely information on the spatial distribution and temporal change of winter wheat is critical for food security and environmental sustainability. Multi-temporal images and time series data of medium resolution are widely used in winter wheat mapping. However, relatively long revisit times and image noise often result in a deficiency in full time series data. In this paper, a new spectral index, called the winter wheat index (WWI), using multi-temporal Landsat normalized difference vegetation index (NDVI) data of four key growth stages of winter wheat, was proposed to highlight and map winter wheat. Two distinctive NDVI contrasts, each consisting of an NDVI peak and trough, were identified and used in the WWI. The proposed index was evaluated through qualitative and quantitative analyses as well as winter wheat mapping, and compared with three state-of-the-art methods. To map winter wheat using WWI, a Monte Carlo cross validation procedure was adopted to determine the optimal thresholds of the WWI. Visual comparison showed that winter wheat was highlighted by higher WWI values, whereas other land cover types had lower WWI values. The experimental results from quantitative analysis indicated that WWI achieved better separability between winter wheat and other land cover types than the other comparative indices. The proposed WWI also produced more accurate winter wheat mapping results, compared with the state-of-art methods. Therefore, the proposed WWI provides a useful variable for winter wheat mapping, which reduced the dependence on full time series data and the use of noise images, and can be applied in other study areas.

Introduction

Wheat is a staple crop produced across the world, and winter wheat is a primary variety of wheat in many countries. With population growth, the steady production of winter wheat is necessary for food security (Zhou et al., 2017). However, winter wheat cultivation is changing owing to the climate change, farmland conversion and adjustment of the crop planting structure in some countries (Quan and Wang, 2013). Therefore, timely and accurate information on the spatial distribution and temporal change of winter wheat is essential for agricultural management, land planning and environmental sustainability (Qiu et al., 2017).

Remote sensing provides powerful spatial data to map and monitor the spatio-temporal distribution of winter wheat with lower labor and resource costs. Existing studies show that the multi-temporal images acquired at a few dates and the long-term dense time series data have been widely used in mapping winter wheat in recent decades (Belgiu and Csillik, 2018, Bullock et al., 2020, Gómez et al., 2016, Long et al., 2013, Pasquarella et al., 2018, Roy et al., 2008, Zhu et al., 2015a), because winter wheat has unique phenological features during its growth cycle, although it has similar spectral features with other land cover types (Pan et al., 2012).

Three main methods were developed to map winter wheat using multi-temporal images and time series images. First, winter wheat was distinguished from other crops using different classification methods, such as support vector machine (SVM) (Zheng et al., 2015), random forest (Tatsumi et al., 2015, Zhong et al., 2014) and decision trees (Tian et al., 2017, Tian et al., 2019). Different spectral features and vegetation indices were used in classification.

Second, different crops, including winter wheat, were mapped by comparing the similarity in time series profile between a target pixel and reference samples (Qin et al., 2015, Upadhyay et al., 2016, Zhang et al., 2019). Different similarity measures were used, such as time-weighted dynamic time warping (Maus et al., 2016, Dong et al., 2020). However, only the general similarity of the time series profile was measured in these studies, ignoring the similarity at some specific and key points on the time series profile, which may limit classification accuracy.

Third, key phenological features of the growth cycle of winter wheat were quantified to map winter wheat (Pan et al., 2012, Chu et al., 2016, Qiu et al., 2017, Tao et al., 2017). For example, Tao et al. (2017) used the unique phenological feature of winter wheat called peak before winter feature (PBWF) to map winter wheat from time series Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index (EVI) data. However, the PBWF is less significant in regions of considerably colder weather conditions owing to the relatively weak growth of winter wheat during winter (Wang et al., 2015). In Qiu et al. (2017), winter wheat was mapped using the combining variations before and after estimated heading dates (CBAH) from the MODIS 2-band enhanced vegetation index (EVI2) time series data, where two key phenological parameters and two phenology-based indices were used. In these studies, dense time series data were required to produce accurate winter wheat mapping.

Although coarse resolution time series data, such as MODIS data, are dense and regularly acquired with wide coverage (Ren et al., 2008), they are not applicable for winter wheat mapping at local and regional scales because mixed pixels are commonly present (Pan et al., 2012, Xu et al., 2014, Yang et al., 2015, Qiu et al., 2017, Huang et al., 2019). Medium spatial resolution images, such as Landsat Thematic Mapper (TM)/The Enhanced Thematic Mapper Plus (ETM+)/Operational Land Imager (OLI) and Sentinel-2 images, have been used in winter wheat mapping at finer scales in recent years (Belgiu and Csillik, 2018, Zhang et al., 2019). However, because of the relatively long revisit times and the influence of cloud, cloud shadow and snow, these medium resolution time series data are usually irregularly acquired and deficient, which may affect mapping accuracy when using the second and third methods mentioned above (e.g. Pan et al., 2012, Qiu et al., 2017, Tao et al., 2017, Tian et al., 2015).

To solve these problems, instead of using full time series data at medium resolution in winter wheat mapping as in the existing studies, this study used multi-temporal medium resolution Landsat normalized difference vegetation index (NDVI) data in winter wheat mapping. The main objective of this study was to propose a spectral index using Landsat NDVI data acquired at four key growth stages of winter wheat to highlight and map winter wheat.

Section snippets

Study area

Three areas located in North China were selected as study areas (Fig. 1). North China is a major region of winter wheat production and one of the most important grain production bases in China (Wu et al., 2011). The region is dominated by plains which are in the eastern warm temperate monsoon zone, with a semi-humid continental climate. The production of winter wheat in North China accounts for more than 50% of the production in China, and the acreage of winter wheat in the region accounts for

Methodology

A spectral index, called the winter wheat index (WWI), using multi-temporal NDVI data was proposed to highlight and map winter wheat. The development of the WWI was based on two distinctive contrasts from the NDVI time series profile of winter wheat. First, the NDVI time series profiles of reference samples from winter wheat and other land cover types were analyzed to identify the discriminative time series features of winter wheat. Second, a spectral index for winter wheat was developed.

Qualitative analysis

To illustrate the difference between different indices, image subsets from the three study areas and the corresponding four indices were shown in Fig. 4. In the first subset, in comparison with the reference data (Fig. 4(a)), the PBWF highlighted most winter wheat fields (patches) with low PBWF values (with green color in Fig. 4(b)). However, many pixels of other land cover types also had relatively low PBWF values (Fig. 4(b)). In Fig. 4(c) and (d), most winter wheat patches had high NVE and

General discussion

In this study, a new spectral index, called the WWI, was proposed for highlighting and mapping winter wheat using multi-temporal NDVI data from four winter wheat key growth stages. The WWI was evaluated in three study areas in North China. The experimental results demonstrated that the proposed WWI generally outperformed the existing indices (PBWF, NVE and NVL) and mapping methods (SVM classification, PBWF and CBAH).

By further analyzing separability between winter wheat and other land cover

Conclusions

A spectral index, the WWI, was proposed to highlight and map winter wheat in this study. The NDVI values from four key growth stages of winter wheat, constituting two distinctive NDVI contrasts, were used in the WWI. The use of NDVI values from four key growth stages reduced the dependence on full time series data and the use of noise images. Moreover, these four key growth stages represented stable and discriminative features. In contrast, the existing methods used variables from some of these

Author contributions

P.L. and C.Q. conceived and designed the paper. C.Q. performed the experiments, analyzed the data, and wrote the paper. P.L. analyzed the data and revised the paper. C.Z provided the winter wheat reference data of Dezhou area.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by Key Technologies Research and Development Program of China (Grant Number 2016YFD0300601) and National Science Foundation of China (Grant Number 42071307). The authors also thank Haidong Li from Peking University for his constructive comments and suggestions, which improved the quality of the manuscript.

References (62)

  • V.J. Pasquarella et al.

    Improved mapping of forest type using spectral-temporal Landsat features

    Remote Sens. Environ.

    (2018)
  • Y. Qin et al.

    Mapping paddy rice planting area in cold temperate climate region through analysis of time series Landsat 8 (OLI), Landsat 7 (ETM+) and MODIS imagery

    ISPRS J. Photogramm. Remote Sens.

    (2015)
  • B. Qiu et al.

    Winter wheat mapping combining variations before and after estimated heading dates

    ISPRS J. Photogramm. Remote Sens.

    (2017)
  • J. Ren et al.

    Regional yield estimation for winter wheat with MODIS-NDVI data in Shandong, China

    Int. J. Appl. Earth Obs. Geoinf.

    (2008)
  • D.P. Roy et al.

    Multi-temporal MODIS–Landsat data fusion for relative radiometric normalization, gap filling, and prediction of Landsat data

    Remote Sens. Environ.

    (2008)
  • J. Tao et al.

    Mapping winter wheat using phenological feature of peak before winter on the North China Plain based on time-series MODIS data

    J. Integr. Agric.

    (2017)
  • K. Tatsumi et al.

    Crop classification of upland fields using Random forest of time-series Landsat 7 ETM+ data

    Comput. Electron. Agric.

    (2015)
  • H. Tian et al.

    Mapping Winter Crops in China with multi-source satellite imagery and phenology-based algorithm

    Remote Sens.

    (2019)
  • X. Wang et al.

    Remote sensing monitoring of changes in winter wheat area in North China Plain from 2001 to 2011

    Trans. CSAE

    (2015)
  • Q. Xu et al.

    Monte Carlo cross validation

    Chemometrics Intell. Lab. Syst.

    (2001)
  • D. Yang et al.

    Mapping plastic greenhouse with medium spatial resolution satellite data: development of a new spectral index

    ISPRS J. Photogramm. Remote Sens.

    (2017)
  • X. Zhang et al.

    Identification and mapping of winter wheat by integrating temporal change information and Kullback-Leibler divergence

    Int. J. Appl. Earth Obs. Geoinf.

    (2019)
  • B. Zheng et al.

    A support vector machine to identify irrigated crop types using time-series Landsat NDVI data

    Int. J. Appl. Earth Obs. Geoinf.

    (2015)
  • L. Zhong et al.

    Efficient corn and soybean mapping with temporal extendability: a multi-year experiment using Landsat imagery

    Remote Sens. Environ.

    (2014)
  • L. Zhong et al.

    Deep learning based winter wheat mapping using statistical data as ground references in Kansas and northern Texas

    US. Remote Sens. Environ.

    (2019)
  • Z. Zhu et al.

    Improvement and expansion of the Fmask algorithm: cloud, cloud shadow, and snow detection for Landsats 4–7, 8, and Sentinel 2 images

    Remote Sens. Environ.

    (2015)
  • Z. Zhu et al.

    Generating synthetic Landsat images based on all available Landsat data: predicting Landsat surface reflectance at any given time

    Remote Sens. Environ.

    (2015)
  • C.A. Baldeck et al.

    Single-species detection with airborne imaging spectroscopy data: a comparison of support vector techniques

    IEEE J Sel. Top. Appl. Earth Observ. Remote Sens.

    (2014)
  • R. Bouhennache et al.

    A new spectral index for the extraction of built-up land features from Landsat 8 satellite imagery

    Geocarto Int.

    (2019)
  • G.C. Cawley et al.

    On over-fitting in model selection and subsequent selection bias in performance evaluation

    J. Mach. Learn. Res.

    (2010)
  • L. Chu et al.

    Monitoring of winter wheat distribution and phenological phases based on MODIS time-series: a case study in the Yellow River Delta

    China. J. Integr. Agric.

    (2016)
  • Cited by (27)

    • Global spatiotemporal completion of daily high-resolution TCCO from TROPOMI over land using a swath-based local ensemble learning method

      2022, ISPRS Journal of Photogrammetry and Remote Sensing
      Citation Excerpt :

      Geographic variates are relevant to the spatial distribution of atmospheric CO. In our study, population count, Normalized-Differential-Vegetation-Index (NDVI) (Chen et al., 2021; Qu et al., 2021), land cover proportions, and surface altitude are chosen as the auxiliary parameters. Specific information of the geographic variates is demonstrated in Table s1.

    • Automated soybean mapping based on canopy water content and chlorophyll content using Sentinel-2 images

      2022, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus
    View full text