India is one of the highly populated countries, and its economy mainly depends on agriculture. The crop type classification is an essential requirement for ensuring food security, crop monitoring, and to understand the environmental consequences of cultivated ecosystems. This study exploits freely available multi-temporal SAR data for discriminating crop types, such as wheat, gram, and mustard, over Ashok Nagar district, Madhya Pradesh, India. Nine Sentinel-1 dual-polarized data acquired from January 2018 to April 2018 in interferometric wide swath mode are used. Class separability analysis using Bhattacharyya Distance (BD) has been performed for multi-temporal VV and VH backscatter, log-ratio, and Radar Vegetation Index (RVI) to quantify the ability to distinguish temporal profiles of crops. RVI has shown the significant result in class separability analysis in comparison with other parameters. Crop type classification map has been generated using a support vector machine classifier with overall accuracy and Kappa coefficient of 96.32% and 0.95, respectively.
TopIntroduction
According to the United Nations (United Nations, 2019), India is one of the top ten countries whose global population proliferate more than half of the projected growth between 2019 and 2050. It is also forecasted that India may be the highest populated country by 2027 based on the current growth rate. The steady increase in population will contemporaneously increase demand for food, which influence the future actions of mankind towards food security and nature conservation (FAO, 2009). Several studies, such as Yu et al. (2015) and Pingali et al. (2019) discussed the importance of agriculture and its need for a strategy in India to ensure future food security.
The spatial distribution of crop types and timely health status are significantly important for decision makers at regional, national and even global level (Brisco et al., 1998; Wu et al., 2015). In this regard, satellite Earth Observation (EO) data has become an essential data source. With the increase in EO satellites’ operating in optical and microwave regions of the electromagnetic spectrum enable continuous land monitoring/observations. Particularly, free and open satellite data provided from Landsat and Sentinel missions have been extensively used for these applications (Wulder et al., 2012; Veloso et al., 2017; Kamilaris et al., 2017). Although, several studies (Thenkabail et al., 2005; Dheeravath et al., 2010) have shown the potential capabilities of optical multi-spectral data for crop classification. As the data is highly sensitive to the weather conditions, these approaches may not provide acceptable results over areas having frequent cloud cover or foggy conditions. Since radar remote sensing can provide data during almost all-weather conditions, this may be effectively used for un-interrupted crop growth monitoring than optical. In addition, its unique sensitivity towards structural, geometrical and dielectric properties of various components of the crop makes possible for crop identification and biophysical parameters retrieval, concurrently assess the health status (Patel and Srivastava, 2013; Gao et al., 2018; Sivasankar et al. 2018). Furthermore, the ability of radar signal to penetrate through vegetation and interact with underneath soil enables to estimate soil characteristics like moisture (Srivastava et al., 2006; Baghdadi et al., 2006; Sharma et al., 2019), roughness (Baghdadi et al., 2002; Srivastava et al., 2008) and texture (Singh and Kathpalia, 2007). However, previous studies have also demonstrated the synergistic use of optical and SAR datasets for agricultural crop applications (Moran et al., 1997; McNairn et al., 2009; Gibril et al., 2017; Sun et al., 2019).
The distinct advantage of radar remote sensing techniques and the increasing number of space-borne SAR satellites has made researchers to widely use the data for these applications. The backscatter signal from the agricultural field is a complex function of both target (crop as well as soil) parameters and SAR sensor configuration (Henderson and Lewis, 1998). Thus, a proper chosen of sensor parameters is vital to enhance the sensitivity of SAR data for a specific application. The SAR sensor parameters such are frequency, polarization and incidence angle have a significant influence on scattering mechanisms in vegetation canopies. For more details, a brief review on radar remote sensing for agricultural applications is given in Sivasankar et al. (2018). Previous studies (Jia et al., 2012; Foody et al., 1989; Freeman et al., 1994; Patel and Srivastava, 2007; Tan et al., 2011; Kothapalli Venkata et al., 2017) have demonstrated that multi-configuration SAR datasets like multi-temporal, multi-polarization, multi-frequency and multi-incidence angle provide better accuracy for crop identification than single SAR data acquired at a given frequency, polarization and incidence angle. Among these, multi-temporal SAR data is a particularly attractive data source for crop type classification.