Elsevier

Ecological Indicators

Volume 56, September 2015, Pages 79-86
Ecological Indicators

Mapping paddy rice areas based on vegetation phenology and surface moisture conditions

https://doi.org/10.1016/j.ecolind.2015.03.039Get rights and content

Highlights

  • Combined Consideration of Vegetation phenology and Surface water variations.

  • Dynamic relationship of LSWI and EVI2 during vegetation growing period.

  • Establishing decision rules before grouping or clustering.

  • Robust to intra-class variability and other related uncertainties.

  • Efficiency in dealing with data noise, rapid VI decrease when harvested and rainfall events.

Abstract

Accurate and timely rice mapping is important for food security and environmental sustainability. We developed a novel approach for rice mapping through Combined Consideration of Vegetation phenology and Surface water variations (CCVS). Variation of the Land Surface Water Index (LSWI) in rice fields was relatively smaller than that in other crops fields during the period from tillering to heading dates. Therefore, the ratios of change amplitude of LSWI to 2-band Enhanced Vegetation Index 2 (EVI2) during that period were utilized as the primary metric for paddy rice mapping. This algorithm was applied to map paddy rice fields in southern China using an 8-day composite Moderate Resolution Imaging Spectroradiometer (MODIS) in 2013. The resultant rice cropping map was consistent with the agricultural census data (r2 = 0.8258) and ground truth observations (overall accuracy = 93.4%). Validation with Landsat Thematic Mapper images in test regions also revealed its high accuracy (with overall accuracy of 94.3% and kappa coefficient of 0.86). The proposed CCVS method was more robust to intra-class variability and other related uncertainties compared with other related methods in rice mapping. Its successful application in southern China revealed its efficiency and great potential for further utilization.

Introduction

Rice is very important to food security since it provides a staple food for more than half of the world's population (Bouvet and Le Toan, 2011, Kuenzer and Knauer, 2013, Mosleh and Hassan, 2014, Tornos et al., 2015). Timely and accurate monitoring of rice cropping areas is essential for understanding changes in food production and environmental sustainability (Boschetti et al., 2014, Gumma et al., 2014). Significant research efforts have been directed towards mapping rice areas and deriving rice phenology using remote sensing time series (Xiao et al., 2002, Xiao et al., 2005, Sun et al., 2009, Bouvet and Le Toan, 2011, Chen et al., 2011, Peng et al., 2011, Bridhikitti and Overcamp, 2012, Jeong et al., 2012, Nuarsa et al., 2012, Kuenzer and Knauer, 2013, Gumma et al., 2014, Mosleh and Hassan, 2014, Gumma et al., 2015). Among them, the temporal profiles of Vegetation Indices (VIs) which represented vegetation greenness (e.g. Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI)) were generally applied. One of the most commonly utilized approaches was the spectral matching technique and other related methods through directly applying the annual VIs temporal profiles (Sakamoto et al., 2005, Biradar et al., 2009, Gumma et al., 2014, Gumma et al., 2015). For this kind of method, there would always be a degree of subjectivity in the grouping process (Gumma et al., 2014). Additionally, for a given specific crop, the VIs temporal profiles could exhibit considerable intra-class variability due to regional variations in climate and management practices (Wardlow et al., 2007). The intra-class variability of VIs temporal profiles was one major challenge associated with the annual vegetation indices-based categorization approach (Lunetta et al., 2010).

Another generally applied approach was the Subtraction-Based Algorithm (SBA) based on particular data composite. It was proposed through evaluating the differences between the NDVI (or EVI) and the Land Surface Water Index (LSWI) during the rice transplanting period (Xiao et al., 2005). The LSWI values were generally high during the period of irrigation or paddy rice transplanting since it represented the condition of leaf water and soil moisture. The pixel was identified as a paddy rice field if the condition LSWI+TNDVI was satisfied (T is a threshold). This algorithm has been widely applied to Asia, South America and the Mediterranean since then (Sakamoto et al., 2007, Sakamoto et al., 2009, Manfron et al., 2012, Handisyde et al., 2014). However, there were some uncertainties that could be introduced by rainfall or irrigation events in other crops, cloud contamination or snow disturbance in data composites, and other related factors (Xiao et al., 2005, Jeong et al., 2012). Recent studies have focused on reducing these uncertainties, particularly through modifying the threshold. For example, this algorithm was modified to accommodate different growth calendar regionalization by configuring specific thresholds for single, early, and late rice cultivation (Sun et al., 2009, Peng et al., 2011). Variable threshold models were also developed for detection of irrigated paddy rice fields and irrigation time by evaluating sub-pixel land cover heterogeneity from the Synthetic Aperture Radar (SAR) microwave images (Jeong et al., 2012). Although the variable threshold models demonstrated significant improvements over the fixed threshold model, a common threshold across all sites was more appropriate than a locally adaptive one (Boschetti et al., 2014). Till now, very limited progress has been made in relieving these above uncertainties.

This paper proposed a new method which could account for intra-class variability of VIs temporal profiles and other possible uncertainties introduced by climatic and management conditions. Instead of directly applying the annual VIs temporal profiles or one particular data composite, we took advantage of the sustained vegetation growth and its corresponding soil-leaf moisture condition during the period from tillering to heading dates. In the following sections, we gave a detailed description of our rice mapping method and presented its application in southern China using the Moderate Resolution Imaging Spectroradiometer (MODIS) 2-band Enhance Vegetation Index (EVI2) time series datasets.

Section snippets

Study area

Our study area included 15 provincial-level administrative units (13 provinces plus 2 direct-controlled municipalities) in southern China (Fig. 1). In Huanghuai plain, a rotation of double dry crops (e.g., winter wheat plus maize) was primarily cultivated. In the Yangtze Plain, there generally occurred a rotation of rice plus other dry crops (primarily in southern Jiangsu and Anhui provinces) and double rice (particularly the Dongting and Poyang Lake plains). Single cropping of rice mainly

Methodology

A new method for mapping paddy rice through Combined Consideration of Vegetation phenology and Surface water variations (CCVS) was proposed (Fig. 2). It included the following procedures: data preprocessing, designing phenology-based parameters, mapping rice area and accuracy evaluation. The entire procedure was executed using the Matlab software package (The MathWorks, Natick, MA, USA).

Spatial distribution of paddy rice areas

Through some experiments, the constants for discriminating rice in function (2) were obtained. The θ1 for LSWImin was 0.1, and the θ2 for RCLE was 0.6. The spatial distribution map of paddy rice areas was then obtained (Fig. 5). Paddy rice areas dominated in the plains near the Yangtze River. It was worthwhile to note that only a few areas in Henan Province were cultivated with rice. As the food base of China, Henan Province was famous for its cultivation of double dry crops, particularly a

Merit and challenges of phenology-based approaches

Land cover/crop mapping methods based on VIs time series could be roughly classified into two types: one type is based on long-term trends and seasonal patterns; and the other type is based on land surface phenological stages. The phenology-based approaches have been widely applied since they provide more detailed information (Potgieter et al., 2013, Zhang et al., 2013). One major challenge associated with the phenology-based approaches and other VIs-based categorization was the intra-class

Conclusions

This study proposed a new rice mapping method by taking advantage of the dynamic relationship between EVI2 and LSWI during specific phenological stages. The ratios of change amplitude of LSWI to EVI2 during the period from tillering to heading dates were applied as the primary metric to identify seasonal rice areas. Its application in southern China revealed its efficiency, with an overall accuracy of 93.4% with ground truth data and r2 = 0.8258 with agricultural statistical datasets. The

Acknowledgements

This work was supported by the National Natural Science Foundation of China (grant no. 41471362, 41071267). We are very grateful for the thorough and helpful comments from the reviewers of the manuscript.

References (39)

Cited by (0)

View full text