Identification of Sugarcane with NDVI Time Series Based on HJ-1 CCD and MODIS Fusion
- 101 Downloads
It is currently difficult to acquire the clear-sky data with high spatial resolutions in spring and summer in the southern region of China, making it impossible to carry out timely and fine monitoring of sugarcane planting information. Thus, Fusui, a sugarcane producing county in Guangxi, was selected in this paper to analyze the NDVI characteristics and change rules during the whole growth period of sugarcane based on MODIS and HJ-1 CCD remote sensing data, which were fused into 30 m resolution NDVI time series data with high accuracy by using the spatial and temporal fusion model of multi-source remote sensing data ESTRAFM. In addition, the NDVI change rate and sample automatic training threshold were used to construct the sugarcane planting information identification model. The results showed that the fused images showed a high similarity with the observed images, indicating good fusion quality. Moreover, the correlation coefficients in the sugarcane planting area reached 0.953, and AD, AAD and SD were 0.033, 0.019 and 0.007, respectively. The NDVI change rate model was used to identify the sugarcane planting information in different time phases of 113 d, 129 d, 145 d, 193 d and 209 d in spring and summer, and the overall accuracy was 92.17%, 92.58%, 91.78%, 90.52% and 91.17%, respectively. The established model also achieved good results in 2017 with the overall accuracies are 88.44%, 87.79%, 89.79%, 88.34% for 113 d, 145 d, 193 d and 209 d.
KeywordsSugarcane identification Space–time fusion NDVI time series HJ-CCD MODIS
The identification of crop planting information is of great significance to agricultural production management, sustainable agricultural development and national food security, and remote sensing technology has become an important means to quickly acquire information on the temporal and spatial distribution of crops. At present, optical remote sensing data is the main data source for crop information extraction. The identification methods mainly include computer supervised classification, multi-temporal analysis and object-oriented classification. Although microwave remote sensing data can overcome the effects of cloudy and rainy weather, it is difficult to promote and apply due to technical and cost restrictions. Sugarcane is an important raw material for sugar production. More than 50% of the world’s sugar comes from sugarcane. China is the third largest sugarcane producer in the world after Brazil and India. For China, more than 80% sugar comes from sugarcane from 2003, which is far higher than the world average level of 50%. Grasping sugarcane planting area in time is of important significance for the stability of sugar market, agricultural production and regional economic development (Tan et al. 2007). There are quite a lot reports on the current mainstream crop information identification methods which can identify sugarcane and the area estimation accuracy can reach up to 90%. In these reports, data sources selected include MODIS (Tan et al. 2007; Alexandre et al. 2006), HJ-1 (Ma et al. 2011; Ding et al. 2012; Wang et al. 2014; Zhou et al. 2016), GF-1 (Liu et al. 2014), Landsat-8 OLI (Chen et al. 2015), Landsat-7 ETM+ (Fortes and Demattê 2006; Vieira et al. 2012), SAR (Lin et al. 2009), and classification methods adopted include supervised classification (Tan et al. 2007), unsupervised classification (Alexandre et al. 2006), decision tree (Liu et al. 2014), object-oriented classification (Wang et al. 2014; Zhou et al. 2016; Vieira et al. 2012) or combination of multiple methods (Ma et al. 2011; Ding et al. 2012; Chen et al. 2015). The sugarcane planting areas in China are mainly distributed in Guangxi, Yunnan, Hainan and Guangzhou. Affected by cloudy and rainy weather and satellite revisiting cycles, it is difficult to obtain large-scale clear-sky images with high spatial resolution in spring and summer (here use Chinese lunar calendar division method, spring is from march to may and summer is from June to August) in these areas, so the existing research on optical remote sensing-based sugarcane classification mainly relies on clear-sky remote sensing images in October and November. Therefore, reconstructing the high spatial resolution data series of sugarcane planting area and constructing a sugarcane identification model based on new series can provide an effective way for timely and refined monitoring of sugarcane planting information in spring and summer.
As a matter of fact, time series data has become a hot topic in crop classification research. Jakubauskas et al. (2002) used the AVHRR NDVI time series data to classify corn, soybean and alfalfa crops by harmonic algorithm. Zhang et al. (2008) achieved the land cover classification in North China by using the decision tree algorithm based on MODIS EVI time series data. Miao et al. (2011) reconstructed the MODIS NDVI time series data by S-G filtering algorithm and completed the extraction of regional rice planting area information. Yang et al. (2015) remodeled the GF-1 NDVI time series data by using harmonic algorithm to study the effectiveness of various classification methods for winter wheat–summer corn, corn and rice. Li et al. (2013) rebuilded HJ-1 time series data by using the spline algorithm and carried out the classification and identification of soybean, corn, rice and cushaw. The spatial resolution of existing time-sensitive remote sensing data is generally low, and the acquisition period of high spatial resolution remote sensing data is long. Affected by topography and landform, there are few large-scale areas in the main sugarcane producing areas in southern China. Thus, the various factors lead to the fact that the single source of data is difficult to meet the demand for high-temporal resolution remote sensing data to obtain the sugarcane planting information under complex weather and terrain conditions in southern China. Therefore, there are few reports on sugarcane classification using NDVI time series data. In order to solve the problem of insufficient satellite remote sensing data acquisition ability, scholars have proposed a technology that can combine high-temporal resolution features of low spatial resolution remote sensing data with high spatial resolution features of medium and high spatial resolution remote sensing data, namely, the multi-source remote sensing data spatiotemporal fusion technology. Gao et al. (2006) proposed the spatial and temporal adaptive reflectance fusion model STARFM in 2006, and Zhu et al. (2010) proposed the enhanced spatial and temporal adaptive reflectance fusion model ESTARFM in 2010. Both models have been widely applied, and the application of the 2 spatiotemporal fusion models is focused on MODIS-Landsat remote sensing data (Gao et al. 2006; Zhu et al. 2010). So far, only Sun et al. (2016) have reestablished the NDVI series with STRAFM based on HJ-1 CCD and MODIS data and verified the effectiveness of STRAFM on HJ-1 CCD and MODIS data fusion. However, there is still no report on the application of the ESTRAFM model in HJ-1 CCD data. The fused NDVI series has made some progress in crop monitoring. Cai et al. (2012) studied the adaptability of MODIS and Landsat data fusion in crop monitoring, finding that the reflectance of fusion image of corn and cotton was similar to that of the observed image. In addition, the time series images based on MODIS-Landsat fusion have been successfully applied to the extraction of rice planting area (Zhang and Zeng 2015; Wu et al. 2010). However, there is no report on the application of this technology in sugarcane.
Through the review of the existing research, we know that in order to monitor the spring and summer sugarcane planting information in south China, it is necessary to generate a high-resolution NDVI time series. Therefore, in this study, the MODIS and HJ-1 CCD remote sensing data and fusion model ESTRAFM are used to obtain 30 m resolution NDVI time series. On the basis of this, the spectral characteristics and change rules of sugarcane in the whole growth period were analyzed, and the sugarcane identification model based on NDVI time series data was constructed, with the aim to provide scientific references for the refined monitoring of sugarcane by improving and enhancing the use of remote sensing technology.
Phenological calendar of main crops in Fusui County
Data and Methods
Data Sources and Pre-processing
HJ-1 CCD data: the systematically geometric corrected level-2 product data downloaded from China Center for Resources Satellite Data and Application. There were a total of 4 bands (spectrum range of 0.43–0.52 μm, 0.52–0.60 μm, 0.63–0.69 μm, 0.76–0.90 μm) with the spatial resolution of 30 m. The product data was sequentially subjected to radiometric calibration, atmospheric correction, geometric correction to obtain reflectance data, and NDVI corresponding to the date was calculated. Radiation calibration was performed using the 2011 HJ-1 A/B star absolute radiometric calibration coefficient provided by the China Center for Resources Satellite Data and Application; the geometric correction reference image was Landsat TM in 2008, and the rectification error was less than 0.5 pixels by using cubic convolution algorithm through selecting the ground control point GCP. The atmospheric correction adopted the FLAASH module. ENVI 5.0 was selected to perform the above processing, and the image coordinate system was UTM-WGS84.
MODIS data: the global vegetation index product MOD13Q1 was downloaded from NASA’s official website, with 250 m resolution and 16 d synthesis cycle, a total of 23 scenes of a year. The MRT (MODIS Reprojection Tool) software provided by USGS was used to perform projection conversion and resampling processing on the product data to ensure that it had the same coordinate system and spatial resolution as the HJ-1 CCD data.
Properties of HJ-1 CCD and MODIS
Passing territory time
0.62–0.67 μm (250 m)
0.84–0.88 μm (250 m)
Field sampling data: combined with the digital elevation model DEM in the test area, field sampling of sugarcane, rice, corn, forest, towns, water bodies and other ground objects was carried out on September 5, October 23, 2011, May 14, 2017. In order to improve the reliability of the verification data, the sugarcane planting plot with an area larger than 60 m × 60 m was selected as the sample area, and a hand-held GPS (GPSMap 629sc) was used to record the position and measure the area of the sample area. A total of 55 representative sugarcane sample areas were selected.
ESTRAFM Spatial and Temporal Fusion Model
Construction Method of Sugarcane Identification Model
Input, output and validation data of ESTRAFM
Output data (predicted HJ-1 CCD images)
Validation data (real HJ-1 CCD images)
Real HJ-1CCD images
Real MODIS images
A total of 23 scenes of a year
A total of 23 scenes of a year
Results and Analysis
NDVI fused Image Evaluation
Accuracy evaluation of fused images
Sugarcane sampling plot
Sugarcane Extraction Model
Analysis on Spectral Characteristics of Sugarcane
Construction of Sugarcane Identification Model
Threshold value of NDVI change rate for sugarcane identification in the test area
− 0.0005 to 0.0015
Evaluation of Sugarcane Identification Accuracy
Accuracy evaluation of classification results in 2011
Data time phase
Total accuracy (%)
The sugarcane sampling points for accuracy verification were classified by combining with DEM to clarify the sugarcane identification accuracies in plains, hills and mountains. The average identification accuracies of the 5 time phases in plains, hills and mountains were 95.61%, 91.84% and 83.80%, respectively, in which the identification accuracy of sugarcane in plains was the highest (Table 6).
Accuracy evaluation of classification results in 2017
Data time phase
Total accuracy (%)
It can be seen that the sugarcane identification model and indicators established in this study have also achieved good results in other years, but the overall recognition accuracy is declining. If the accuracy is to be improved, the sugarcane samples must be reanalyzed and the index threshold adjusted. This is because the growth of sugarcane in 2011 may be different from that of other years, and some areas are no longer sugarcane areas due to the change of planting structure.
In this study, we found that ESTRAFM can generate high-quality and high-resolution data. The ESTRAFM model is used to carry out the fusion of MODIS and HJ-1 CCD data, producing the 30 m resolution NDVI series images of the whole growth period of sugarcane, which provides reliable data for the timely and refined identification of sugarcane planting information in spring and summer. The fused images show high similarity to the observed images, indication good fusion quality. The correlation coefficient in the sugarcane planting area is 0.953, and AD, AAD and SD are 0.033, 0.019 and 0.007, respectively. According to the NDVI characteristics and change rules of sugarcane, the identification model for sugarcane in spring and summer is constructed by using the NDVI change rates of sugarcane from the emergence to the stem elongation stages, and the threshold of automatic training model is collected in the field. The model is used to identify the sugarcane planting information in different time phases of 113 d, 129 d, 145 d, 193 d and 209 d in spring and summer, and the overall accuracies are 92.17%, 92.58%, 91.78%, 90.52% and 91.17%, respectively. The established model also achieved good results in 2017 with the overall accuracies are 88.44%, 87.79%, 89.79%, 88.34% for 113 d, 145 d, 193 d and 209 d. Therefore, we concluded that the established sugarcane recognition model are ideal, which can provide a new way to achieve the timely and refined monitoring of sugarcane using optical remote sensing data in spring and summer.
The reflectance of the central pixel of high-resolution image fused from ESTRAFM model is determined by the similar pixel in the low-resolution image. Thus, the accuracy of the similar pixel selection determines the accuracy of the fused image. However, due to the coarse scale of low-resolution images and the scattered planting of sugarcane in some regions, it is easy to have “the same spectrum with different objects” at the junctions of different vegetation types caused by adjacent terrains. Therefore, it is necessary to introduce other auxiliary information to improve the fusion effect of the fused image on the sugarcane area in further research.
For the identification model for spring and summer sugarcane planting information constructed from NDVI change rates and sample automatic training threshold, the setting of the change rate threshold is the key to the identification accuracy. The analysis on the identification accuracy of sugarcane at different landforms of plains, hills and mountain shows that the selection and distribution of samples have an important effect on the identification accuracy. However, the limited samples selected in this study have limitations. Therefore, it is necessary to conduct threshold training for more sugarcane samples with different growth vigor at different landforms in the test area for the sugarcane identification in much wider ranges. Moreover, due to the effects of revisiting cycle and cloudy and rainy weather, the spring and summer HJ-1 CCD data in the test area is seriously deficient. For example, there are only clear-sky images for April 19 and May 28 in 2011 for the test area, making it hard to carry out the systematic analysis on the sugarcane identification accuracy of the images with corresponding time before and after fusion. In the further research, the difference between the fusion images and original images of sugarcane recognition accuracy can be further analyzed, which will lay a foundation for improving the fusion model and obtain the fusion image of sugarcane area with higher accuracy.
The authors thank the China centre for resources satellite data and application for providing HJ-1 CCD remote sensing data and technical support.
Part of this research was jointly supported by the Guangxi Natural Science Foundation (No. 2018GXNSFAA281338); Drought Meteorological Science Research Foundation (No. IAM201707); Guangxi Natural Science Foundation (No. 2017GXNSFBA198153); National Key Basic Research Development Plan (No. 2013CB430205).
- Cai, D. W., Niu, Z., Wang, L., & Li, Wang. (2012). Adaptability research of spatial and temporal remote sensing data fusion technology in crop monitoring. Remote Sensing Technology and Application,27(06), 927–993.Google Scholar
- Chen, L. F., Lin, K. P., Hu, B. Q., Li, J. J., & Ning, W. Y. (2015). Monitoring of sugarcane planting area based on landsat8_OLI data. Journal of Southern Agriculture,46(11), 2068–2072.Google Scholar
- Ding, M. H., Tan, Z. K., Li, H., Yang, Y. H., Zhang, H. Q., Mo, J. F., et al. (2012). Survey methods of sugarcane plant area based on HJ-1 CCD data. Chinese Journal of Agrometeorology,33(2), 265–270.Google Scholar
- Li, X. C., Xu, X. G., Wang, J. H., Jin, L. X., Li, C. J., & Bao, Y. S. (2013). Crop classification recognition based on time-series images from HJ satellite. Transactions of the Chinese Society of Agricultural Engineering,29(2), 169–176.Google Scholar
- Liu, J. K., Zhong, S. Q., Xu, Y., & Chen, Y. L. (2014). Sugarcane extraction in the southern hills using multi-temporal GF-1 WFV data. Guangdong Agricultural Sciences,41(18), 149–154.Google Scholar
- Ma, S. J., Pei, Z. Y., Wang, Q. F., Guo, L., Liang, Z. L., & Teng, D. J. (2011). Research of sugarcane harvest process monitoring with multi-temporal HJ—a satellite data. Transactions of the Chinese Society of Agricultural Engineering,27(3), 215–219.Google Scholar
- Sun, R., Rong, Y., Su, H. B., & Chen, S. H. (2016). NDVI time-series reconstruction based on MODIS and HJ-1 CCD data spatial–temporal fusion. Journal of Remote Sensing,20(3), 361–373.Google Scholar
- Tan, Z. K., Wu, L. L., Ding, M. H., Yang, X., Ou, Z. R., He, Y., et al. (2007). Study on the extraction of sugarcane planting area from EOS/MODIS data. Meteorological Monthly,33(11), 76–81.Google Scholar
- Wang, J. L., Huang, J. L., Wang, L. H., Hu, Y. X., Han, P. P., & Huang, W. (2014). Identification of sugarcane based on object-oriented analysis using time-series HJ CCD data. Transactions of the Chinese Society of Agricultural Engineering,30(11), 145–151.Google Scholar
- Wu, M. Q., Niu, Z., & Wang, C. Y. (2010). Mapping paddy fields by using spatial and temporal remote sensing data fusion technology. Transactions of the Chinese Society of Agricultural Engineering,26(7), 240–244.Google Scholar
- Yang, Y. J., Zhan, Y. L., Tian, Q. J., Gu, X. F., Yu, T., & Wang, L. (2015). Crop classification based on GF-1/WFV NDVI time series. Transactions of the Chinese Society of Agricultural Engineering,31(24), 155–161.Google Scholar
- Zhang, M., & Zeng, Y. N. (2015). Mapping paddy fields of Dongting Lake area by fusing Landsat and MODIS data. Transactions of the Chinese Society of Agricultural Engineering,31(13), 178–185.Google Scholar
- Zhou, Z., Huang, J. F., Wang, J., Zhang, K. Y., Kuang, Z. M., Zhong, S. Q., et al. (2016). Object-Orented classification of sugarcane using time-series middle-resolution remote sensing data based on adaboost. PLoS ONE,10(11), e0142069. https://doi.org/10.1371/journal.pine.0142069.CrossRefGoogle Scholar
This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.