Article Preview
Top1. Introduction
Wheat is a significant source of protein in low-income and middle-income countries. Approximately 2.5 billion people consume it in nearly 89 nations. India exports a considerable quantity of wheat to neighboring countries, including Bangladesh, Nepal, U. A. E. Rajasthan state of India accounts for 7.49% of the total wheat production and 7.24% of the wheat area in India. Thus, timely and precise estimation of the crop in India is critical for both local and worldwide food security.
In the past, crop yield prediction has been assessed using traditional models and statistical regression models. The researchers have been focusing on crop yield prediction to improve crop yield using many methods (Lobell & Burke, 2010), (Tao et al., 2009). The empirical model has the shortcoming in capturing landscape heterogeneity, whereas process oriented models require rigorous parameterization and validation before large scale predictions. Thus, these models are localized and not able to expand to a large area. Machine Learning has manifested its potent performance with data and agriculture exploration, considering crop classification and crop yield prediction (Cai et al., 2019) (Cai et al., 2018). Crop yield prediction accuracy relies on the multidimensional dataset with continuous spatial and temporal variations (Filippi et al., 2019) (Witten et al., 2011). Machine learning techniques are more effective for noisy data and have the ability to elucidate non-linear relations. Therefore, Machine Learning (ML) is more successful and adopted rapidly and extensively for crop yield prediction (Crane-Droesch, 2018) (Wang et al., 2019). Extensive study has been carried out on crop yield prediction. For consideration, Hunt et al. (Hunt et al., 2019) worked on a random forest model on high-resolution satellite imagery and mapped it within a wheat yield field with the 10m resolution in the UK. Cai et al. (Cai et al., 2019) estimated the wheat yield by implementing support vector regression, neural network, and random forest at the county level in Australia. The authors claimed that Machine learning models performed better than the traditional regression model.
Machine learning is dependent on several factors for crop yield prediction, including soil properties, climate variables, and satellite observations for monitoring crop growth and development (Kuwata & Shibasaki, 2015) (Jin et al., 2017) (Kern et al., 2018) (Azzari et al., 2017). Researchers mostly use climate data, satellite data, or both for the crop yield prediction at the smaller region. Climate data such as precipitation, temperature, and weather data provide environmental information of specified regions, although it can not directly detect crop growth. The satellite images can provide the crop growth status for a significant area context. The various spectral bands such as thermal, microwave wavelength, near infrared, and optical extend the input range for crop monitory. The vegetation indices (VI) provide predictive traits, for instance, leaf area, plant water content, biomass, or yield (Thorp et al., 2015). The extensively used VIs are normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI), which are derived from Near-infrared (NIR) and red color spectrum, and Near-infrared (NIR) and blue color spectrum, respectively (Becker-Reshef et al., 2010), (Azzari et al., 2017), (Dong et al., 2015). However, to our knowledge, there are very few studies conducted in Rajasthan state VIs with climate data outperforms climate data individually. Hence, we identified the wheat zones of Rajasthan state based on wheat production quantity for the crop yield estimation.