UAV-multispectral and vegetation indices in soybean grain yield prediction based on in situ observation

https://doi.org/10.1016/j.rsase.2020.100318Get rights and content

Abstract

Remote sensing techniques using vegetation indices (VI's) have successfully enabled the accurate and timely monitoring of several crops. The purpose of this monitoring is to provide an advantage in terms of on-farm management decision making, crop marketing planning, and support for policy decision-making. We identified here which VI's can be used to predict soybean grain yield (GY) by using UAV (Unmanned Aerial Vehicle) and remote multispectral sensor. For this purpose, experiments were carried out in three sites during the 2017/2018 and 2018/2019 crop seasons in Brazil. The VI's were measured together with the plant stand assessment at 25 days after emergency (DAE) when the plants were at V4 phenological stage. The processing of the VI models was performed based on the imaging reflectance factor data performed in the field. For statistical data analyses, a correlation network was used to express the relationship between GY and VI's graphically. Path analysis was performed for identifying the cause-and-effect relationship between VI's and GY. Subsequently, a decision tree algorithm was generated considering GY as a dependent variable. At last, the relative deviation coefficient was used to illustrate the differences between the VI's in the construction of the decision tree and GY. Our results showed potential for predicting soybean yield based on UAV and the multispectral sensor coupled. SAVI and NDVI indices stood out for predicting yield, where the regions with the highest values of these indices can obtain the highest yield observed in the field, providing an advantage in management at the property level.

Introduction

Soybean [Glycine max (L.) Merril] is one of the world's most important sources of protein and oil used in human and animal feed. The crop is the most cultivated in Brazil, which is the second-largest producer in the world and the largest producer in Latin America. In the 2018/2019 harvest, Brazilian soybean production accounted for 115 million tons of grains grown in an area of approximately 36 million hectares (Conab - National Supply Company, 2020). Only in the State of Goiás, Midwest Brazil, 3.73 Mha were cultivated with the crop according to a survey carried out by remote sensing (SojaMaps, 2020; Silva Junior et al., 2020).

Given the growing demand for food and energy, monitoring and predicting grain yield is essential for food security (Holzman et al., 2014; Zhao et al., 2020). Soybean yield predictions in Brazil are of great interest for market behavior, to drive governmental policies, and to increase global food security (Schwalbert et al., 2020). In this context, the development of techniques that allow the accurate and timely monitoring of crops and early prediction of yield is crucial for crop management to obtain high yields. One of the techniques that have stood out in monitoring and predicting crop yield is remote sensing (Gao et al., 2018), which provide low-cost, timely and accurate information on crop status.

Remote sensing techniques have been effective in predicting the yield of various crops, such as rice (Kanke et al., 2016; Rehman et al., 2019), wheat (Holzman et al., 2014; Pantazi et al., 2016; Hassan et al., 2019; Zhao et al., 2020), maize (Baez-Gonzalez et al., 2002; Bolton and Friedl, 2013; Dong et al., 2015), and soybean (Holzman et al., 2014; Dong et al., 2015; Christenson et al., 2016; Schwalbert et al., 2020). Among the remote sensing techniques, we can highlight the vegetation indices (VI's), which are mathematical models for different wavelengths. They are characterized as reliable algorithms for the evaluation of vegetation cover, vigor and growth dynamics, nutritional status, among other applications (Xue and Su, 2017; Baio et al., 2018b; Silva Junior et al., 2018).

The principle behind the relationship between VI's and yield is that crop yield is a function of canopy characteristics, such as chlorophyll content, biomass, and canopy architecture (Zhao et al., 2020). Several studies have been reported the correlation between VI's and yield. Baio et al. (2018a) found a positive linear relationship between the Normalized Difference Vegetation Index (NDVI) and yield in cotton. Zhao et al. (2007) also found a significant correlation between cotton yield and canopy reflectance at flowering measured by NDVI and Enhanced Vegetation Index (EVI). Although the NDVI is the most widely used vegetation index for monitoring the crop status and predicting yield, the use of multiple indices has demonstrated better predictive ability. Zhao et al. (2020) reported that a predicting model associating canopy structural-related indices using orbital data from the MSI/Sentinel-2 sensor system, such as NDVI, Optimized Soil-Adjusted Vegetation Index (OSAVI), and EVI, with chlorophyll-related indices, was more accurate than the model using a VI or more than one VI (structural and/or chlorophyll) for predicting field-scale wheat yield.

On the other hand, the use of orbital data in yield prediction is still a challenge, since a previous mapping is necessary as an input to the crop mask (Zhang et al., 2019). On the other hand, the calibration of suborbital level sensors is still necessary for the soybean crop prediction to then perform the model above the atmosphere, requiring several tests in multiple environments. Maimaitijiang et al. (2020) presented a great potential in using UAV multisensor and deep learning for predicting soybean yield in the USA; however, they recommend tests for different crops and a larger number of genotypes at different development stages and environmental conditions.

The purpose of this monitoring is to provide an advantage in terms of on-farm management decision making, crop marketing planning, and support for policy decision-making. Therefore, it is necessary to perform tests to predict soybean yield in Brazil with sub-orbital data, associating different vegetation indices to assess the hypothesis of the high accuracy of these models. The objective of this paper was to identify which vegetation indices can be used to predict soybean grain yield by using UAV and remote multispectral sensor.

Section snippets

Study sites

The experiment was carried out in the municipality of Chapadão do Céu, State of Goiás, Brazil, during the 2017/2018 and 2018/2019 seasons, in three locations, as shown in Fig. 1 and Table 1. The soil in the region has a texture varying between sandy and clay, with clay content between 247 and 880 g kg−1 in the plots.

Relationship between yield grain and vegetation indices

Pearson's correlation network (Fig. 4) presents the interaction between yield and the evaluated vegetation indices (VI's). The network shows a similarity between the VI's using the same bands, such as NDVI and the relationship between the 790 nm and 660 nm bands. EVI and MTVI, which use the rgreen, rred e rnir bands were the VI's who distanced themselves from the others. According to the network built, the VI's that have the highest correlation with GY are SAVI and MSAVI, possibly due to the

Conclusion

The tests on spectral models expressed through vegetation indices collected via sub-orbital multispectral sensor showed satisfactory and promising results for predicting soybean crop yield. The SAVI and NDVI indices stood out for predicting yield, where the regions with the highest values of these indices can obtain the highest yield observed in the field, providing an advantage in management at the property level.

Our findings reveal the potential for predicting soybean yield based on UAV and

Ethical

All ethical practices have been followed in relation to the development, writing, and publication of the article entitled UAV-Multispectral and vegetation indices in soybean yield prediction based on in situ observation.

CRediT authorship contribution statement

Eder Eujácio da Silva: Conceptualization, Data curation, Methodology. Fabio Henrique Rojo Baio: Conceptualization, Methodology, Writing - review & editing, Visualization, Supervision. Larissa Pereira Ribeiro Teodoro: Data curation, Writing - original draft. Carlos Antonio da Silva Junior: Writing - review & editing, Visualization. Raisa Saraiva Borges: Visualization, Investigation. Paulo Eduardo Teodoro: Methodology, Software, Validation, Formal analysis, Writing - review & editing,

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors would like to thank the Federal University of Mato Grosso do Sul and CAPES. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001.

References (43)

  • X.E. Pantazi et al.

    Wheat yield prediction using machine learning and advanced sensing techniques

    Comput. Electron. Agric.

    (2016)
  • J. Qi et al.

    A modified soil adjusted vegetation index

    Remote Sens. Environ.

    (1994)
  • A. Ramoelo et al.

    Regional estimation of savanna grass nitrogen using the red-edge band of the spaceborne RapidEye sensor

    Int. J. Appl. Earth Obs. Geoinf.

    (2012)
  • R.A. Schwalbert et al.

    Satellite-based soybean yield forecast: integrating machine learning and weather data for improving crop yield prediction in southern Brazil

    Agric. For. Meteorol.

    (2020)
  • C.A. Silva Junior et al.

    Mapping soybean planting area in midwest Brazil with remotely sensed images and phenology-based algorithm using the Google Earth Engine platform

    Comput. Electron. Agric.

    (2020)
  • Y. Zhang et al.

    Effect of using crop specific masks on earth observation based crop yield forecasting across Remote

    Sens. Appl. Soc. Environ.

    (2019)
  • D. Zhao et al.

    Canopy reflectance in cotton for growth assessment and lint yield prediction

    Eur. J. Agron.

    (2007)
  • H. Ahmadi et al.

    Determination of soya plant population using NDVI in the dasht-e-naz agri-industry

    J. Agric. Sci.

    (2009)
  • A.D. Baez-Gonzalez et al.

    Using satellite and field data with crop growth modelling to monitor and estimate corn yield in Mexico

    Crop Sci.

    (2002)
  • F.H.R. Baio et al.

    Relationship between cotton productivity and variability of NDVI obtained by landsat images

    Biosci. J.

    (2018)
  • F.H.R. Baio et al.

    Vegetation indices to estimate spray application rates of crop protection products in corn

    Agron. J.

    (2018)
  • Cited by (50)

    View all citing articles on Scopus
    View full text