Elsevier

Field Crops Research

Volume 198, November 2016, Pages 131-139
Field Crops Research

Detection of rice phenology through time series analysis of ground-based spectral index data

https://doi.org/10.1016/j.fcr.2016.08.027Get rights and content

Highlights

  • Time-series of NDVI and CIred edge was obtained through ground-based spectrometers and smoothed by double logistic regression function.

  • The positions of the maximum, zero-crossing and minimum points in the first derivative of the smoothed CIred edge temporal profiles can be used to estimate the dates of jointing, middle booting and dough grain, respectively.

  • The positions of the maximum, zero-crossing and minimum points in the first derivative of the smoothed NDVI temporal profiles can be used to estimate the dates of active tillering, middle heading and maturity, respectively.

  • Between the NDVIs from passive and active sensors, NDVIGS yielded better performance than NDVIASD for estimating thephenological dates.

  • Rice cultivar and N rate exhibited more impact on the estimation accuracy than growing season and planting density.

Abstract

Monitoring crop phenology is of great importance for vegetation classification, yield estimation, and irrigation and fertilization management. To test the ability of ground-based remote sensing in detecting major phenological dates of rice, canopy spectra were collected by two portable spectrometers. The Red-edge Chlorophyll Index (CIred edge) and Normalized Difference Vegetation Index (NDVI) time-series derived from ground-based spectrometers were employed to detect the main specific phenological dates. CIred edge was obtained from ASD FieldSpec Pro spectrometer, while NDVI was from ASD FieldSpec Pro spectrometer (referred to as NDVIASD) and GreenSeeker RT 100 (referred to as NDVIGS). The phenology detection method consists of two procedures: (i) smoothing the temporal CIred edge and NDVI data with the double logistic regression function to represent intra-annual vegetation dynamics, (ii) determining the phenological dates through extracting the maximum, minimum and zero-crossing points (FDmax, FDmin and FDzero) from the first derivative value of the smoothed NDVI and CIred edge temporal profiles. A comparison of remote sensing-based estimates with field observations over three growing seasons with different cultivars, planting densities and nitrogen (N) rates showed that CIred edge can accurately estimate the dates of jointing, middle booting and dough grain. NDVI from both spectrometers can be used to detect the dates of active tillering, middle heading and maturity. Specifically, NDVIGS yielded better performance than NDVIASD for estimating the three phenological dates. Compared with growing season and planting density, rice cultivar and N rate exhibited more significant impact on the accuracy for phenology detection. This work has great potential to provide valuable support for assessing crop growth status and providing precise management strategy. The dates of active tillering, jointing and maturity detected from a combination of CIred edge and NDVI could be useful for irrigation and fertilization management, and harvest determination, respectively.

Introduction

Phenology information is essential for many applications, such as crop classification (Lloyd, 1990, Peña-Barragán et al., 2011, Siachalou et al., 2015, Son et al., 2013), estimation of net primary production (Kimball et al., 2004) and decision-making about water and fertilizer supply (Dingkuhn and Le Gal, 1996). Paddy rice (Oryza sativa L.) is one of major food crops in the world, especially in China (Xiao et al., 2002). According to Moldenhauer and Slaton (2001), rice phenology is generally divided into: (i) vegetative phase, including germination, seedling, tillering and jointing stages; (ii) reproductive phase, including booting, heading and flowering stages; (iii) maturation phase, including milk, dough grain and maturity stages. Within these stages, several dates are critical for precision farming management, such as active tillering date (date for field drying), jointing date (date for panicle fertilizer) and maturity date (date for harvesting) (Ling et al., 2007). Irrigation scheduling is critical in the rice growth period, especially in the active tillering stage. Yang et al. (2006) reported that field drying at active tillering stage can control the non-effective tillers and adjust the relationships between soil and water, and those between root and shoot, which are beneficial to rice growth. Panicle fertilizer has great impact on rice grain yield and quality. Suitable application stage of panicle fertilizer could increase the chlorophyll and nitrogen (N) contents of high photosynthetic-rate leaves and thus increase the rice yield (Ding et al., 2003). Harvesting needs to be prompt because late harvesting results in reduced milling quality and rice yield (Moldenhauer and Slaton, 2001). All the aforementioned studies demonstrated that the determination of crop phenology is vital for precision management of irrigation and fertilization and determination of harvesting.

Traditionally, crop phenology studies relied on ground-based field visits that were limited by cost, labor and spatial coverage. Consequently, remote sensing (RS) techniques offer considerable benefits for detecting vegetation phenology in the past few decades (Table 1). For example, Tucker et al. (1979) used a hand-held radiometer to monitor corn and soybean growth and development successfully. Gallo and Flesch (1989) utilized the National Oceanographic and Atmospheric Administration’s Advanced Very High Resolution Radiometer (NOAA/AVHHR) data for monitoring the seasonal growth of maize at large scale and showed that the date of maximum normalized difference vegetation index (NDVI) agreed well with the silking stage. Meanwhile, several methods have been developed to determine vegetation phenological stages and the most commonly used methods are VI thresholding, inflection point and maximum slope. The VI thresholding method was used by many researchers (Delbart et al., 2006, Fischer, 1994, Guo et al., 2016, Markon et al., 1995, Motohka et al., 2010, Nagai et al., 2010). Motohka et al. (2010) defined “GRVI = 0” as a site-independent single threshold for detecting the early phase of leaf green-up and the middle phase of autumn coloring in four different vegetation types. With the inflection point method, the start of the growing season can be identified when the first derivative (FD) value of the time-series curve changes from negative to positive, while the end of the growth can be determined when it changes from positive to negative (Moulin et al., 2010, Sakamoto et al., 2005, Soudani et al., 2008, Zhang et al., 2003). With the maximum slope method, the growth stages are determined by the magnitude of variation in the VI time series (Wang et al., 2014a, Yu et al., 2003).

To date, most crop phenology studies relied on satellite RS data due to the widespread availability of high temporal resolution time series imagery from such instruments as the Moderate Resolution Imaging Spectroradiometer (MODIS) and the NOAA/AVHHR. Although these satellite imagery can cover large areas, they are affected by many factors, such as atmospheric disturbances, solar radiation effects and cloud cover duration (Ricotta and Avena, 2000, White et al., 2005). Therefore, the estimation error of phenology could be as high as seven days or even more (Sakamoto et al., 2005, Sun et al., 2009). Coarse resolution satellite imagery is not suitable for phenology detection in major rice growing regions such as south China because the rice fields in this region are relatively small, irregular and fragmented by well-developed roads and dense water networks (Wang et al., 2015). To the best of our knowledge, there was no report about phenology detection using airborne or unmanned aerial vehicle (UAV) imagery primarily due to the logistics complexity and resource limitation with frequent revisits over the growing season. Hence, ground-based RS has great advantages on phenology detection in south China due to its flexibility of spectral, spatial and temporal resolutions.

Recently, time series VIs from ground-based RS platforms have been applied to phenology studies. NDVI was commonly used in phenological studies and had good performance in phenology detection as shown in a summary of relevant references in Table 1. For instance, Wang et al. (2014a) used NDVI from a FieldSpec3 spectroradiometer and its slope curves to monitor rice development and demonstrated that they could be used as cultivar-independent phenological indicators. While handheld sensors with more bands become available in recent years for proximal monitoring of crops, more spectral indices could be obtained to detect crop vigor using red edge bands instead of the red band in NDVI. As one of the red edge indices, red-edge chlorophyll index (CIred edge) was originally proposed for chlorophyll estimation and later be found to be a better indicator for leaf area index (LAI) and biomass than NDVI (Gitelson et al., 2003a).

For rice phenology detection, several stages (e.g. transplanting, tillering, heading and harvesting stages) were determined in previous studies (Motohka et al., 2010, Sakamoto et al., 2005, Wang et al., 2014a). However, other dates (e.g. the dates of active tillering and jointing) that are considered to be more important for irrigation scheduling and fertilizer management, have received little attention. To fill this gap in the previous phenological studies, this study attempts to detect all those main phenological dates of paddy rice with CIred edge and NDVI. Our research objectives are: (1) evaluating the performance of a chlorophyll index (CIred edge) for rice phenology detection in comparison to the widely used NDVI; (2) determining rice critical phenological dates with the maximum, zero-crossing and minimum points from the first derivative (FDmax, FDmin and FDzero) of VI time series. We expected that our work would provide useful guidance for water and fertilizer management and harvesting with portable spectrometer devices.

Section snippets

Experiment design

Three field experiments were designed for this study, involving different rice cultivars, planting densities and N rates. All the experiments were conducted in the experimental station of National Engineering and Technology Center for Information Agriculture (NETCIA), which was located in Rugao city, Jiangsu province, China (120°45′ E, 32°16′ N). In 2013, one japonica rice cultivar Wuxiangjing14 (V1) and one indica rice cultivar Shanyou63 (V2) were seeded at DOY 138 and transplanted into the

Overall trend of NDVI

Time series of NDVI and its FD value under different conditions are shown in Fig. 1. The curve of the treatment with Wuyunjing24 under 200 kg ha−1 N rate and a planting density of 220,000 plants ha−1 (V1N2D1) in 2014 (black symbols in Fig. 1a) was taken as an example. NDVI increased rapidly from DOY 175 (initial tillering) to 211 (initial jointing) and its maximum increase rate (FDmax) was at DOY 192 (active tillering). NDVI reached the maximum around 0.93 at DOY 230 (middle booting) and then

Advantages of CIred edge in phenology detection

N is the most important element which impacts yield and quality in crops (Cassman et al., 2002, Shanahan et al., 2008), and panicle fertilizer applied at jointing stage has a great impact on yield formation (Ding et al., 2003). However, this stage was not addressed in previous phenological studies. Our results suggest that CIred edge is a good indicator for phenology detection. The FDmax, FDzero and FDmin points of CIred edge temporal profile correspond well to the jointing, middle booting and

Conclusions

In this study, we used the temporal CIred edge and NDVI data smoothed with the double logistic regression function to represent intra-annual vegetation dynamics. Phenological dates were detected through extracting the maximum increase, inflection and maximum decrease points (FDmax, FDzero and FDmin) of smoothed VI temporal profiles. This method was applied to CIred edge and NDVI data over a 3-year period (2013–2015) for two different cultivars with different planting densities and N rates. We

Acknowledgements

This work was supported by grants from the National Key Research and Development Program of China (2016YFD0300601), Special Program for Agriculture Science and Technology from the Ministry of Agriculture in China (201303109), the National Natural Science Foundation of China (31470084), the Award for Jiangsu Distinguished Professor and the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD), China.

References (53)

  • N.M. Knox et al.

    Differentiation of plant age in grasses using remote sensing

    Int. J. Appl. Earth Obs. Geoinf.

    (2013)
  • S. Nagai et al.

    Field experiments to test the use of the normalized-difference vegetation index for phenology detection

    Agric. Forest Meteorol.

    (2010)
  • Z. Pan et al.

    Mapping crop phenology using NDVI time-series derived from HJ-1 A/B data

    Int. J. Appl. Earth Obs. Geoinf.

    (2015)
  • J.M. Peña-Barragán et al.

    Object-based crop identification using multiple vegetation indices: textural features and crop phenology

    Remote Sens. Environ.

    (2011)
  • B. Qiu et al.

    A simple smoother based on continuous wavelet transform: comparative evaluation based on the fidelity, smoothness and efficiency in phenological estimation

    Int. J. Appl. Earth Obs. Geoinf.

    (2016)
  • T. Sakamoto et al.

    A crop phenology detection method using time-series MODIS data

    Remote Sens. Environ.

    (2005)
  • T. Sakamoto et al.

    A two-step filtering approach for detecting maize and soybean phenology with time-series MODIS data

    Remote Sens. Environ.

    (2010)
  • J.F. Shanahan et al.

    Responsive in-season nitrogen management for cereals

    Comput. Electron. Agric.

    (2008)
  • K. Soudani et al.

    Evaluation of the onset of green-up in temperate deciduous broadleaf forests derived from moderate resolution imaging spectroradiometer (MODIS) data

    Remote Sens. Environ.

    (2008)
  • C.J. Tucker et al.

    Monitoring corn and soybean crop development with hand-held radiometer spectral data

    Remote Sens. Environ.

    (1979)
  • L. Wang et al.

    Multi-temporal detection of rice phenological stages using canopy spectrum

    Rice Sci.

    (2014)
  • F. Yu et al.

    Response of seasonal vegetation development to climatic variations in eastern central Asia

    Remote Sens. Environ.

    (2003)
  • X. Zhang et al.

    Monitoring vegetation phenology using MODIS

    Remote Sens. Environ.

    (2003)
  • M. Boschetti et al.

    Multi-year monitoring of rice crop phenology through time series analysis of MODIS images

    Int. J. Remote Sens.

    (2009)
  • K.G. Cassman et al.

    Agroecosystems, nitrogen-use efficiency: and nitrogen management

    AMBIO: J. Human Environ.

    (2002)
  • J. Chang et al.

    Clouds influence precision and accuracy of ground‐based spectroradiometers

    Commun. Soil Sci. Plant Anal.

    (2005)
  • Cited by (97)

    • Detecting winter canola (Brassica napus) phenological stages using an improved shape-model method based on time-series UAV spectral data

      2022, Crop Journal
      Citation Excerpt :

      Vegetation index (VI), an indicator by spectral calculation of two or more bands of light, is commonly used to highlight vegetative properties. The main optical remote-sensing-based phenology detection methods can be classified into three groups: 1) threshold methods, which assume that a phenological stage starts when the VI reaches a specific value, including “fixed” and “dynamic” thresholds (e.g., ratio, mean, and median of time-series VI) [18–20]; 2) change-detection methods, which detect the characteristic points on time-series VI curves, such as maximum, minimum, inflection points, and derivatives to determine phenological stage dates [21–23]; and 3) geometrical model fitting, which aims not only to smooth time-series VIs but to align the derived VI curves with ground-observed phenological stages. The typical method used is the shape-model method (SMM), a two-step filtering approach initially proposed by Sakamoto et al. [24].

    • Drones in agriculture: A review and bibliometric analysis

      2022, Computers and Electronics in Agriculture
    View all citing articles on Scopus
    View full text