Potential of GPR data fusion with hyperspectral data for precision agriculture of the future

https://doi.org/10.1016/j.compag.2022.107109Get rights and content

Highlights

  • GPR outputs allowed characterizing the vertical soil profile surveyed.

  • Reflectance values were associated to clay content.

  • Geostatistical sensor data fusion led to a soil partitioning.

  • The challenge of data fusion is multidisciplinary.

Abstract

Precision Agriculture (PA) requires accurate spatial and temporal information of soil properties at a very fine scale. Traditional soil characterization methods are time consuming, laborious and invasive and do not allow long-term repeatability of measurements. Ground Penetrating Radar (GPR) appears to be a particularly suitable methodology for characterizing soil and subsurface from a physical property point of view. Visible-near infrared-shortwave infrared (VIS-NIR-SWIR) reflectance spectroscopy has now become a widespread technique in soil analysis. Information on soil variability can be improved by the integration of data from multiple sensors. The overall objective of this paper was to examine the potential of fusing GPR data with hyperspectral data using multivariate geostatistics for delineating the management zones in the soil of an olive grove of centuries-old trees in Italy.

A linear model of coregionalization (LMC) was individually fitted for the raw hyperspectral data and for GPR data including for each case a nugget effect and two spherical models at short scale and at longer scale. After that, one data set was obtained from the fusion of the two sensor data sets and a LMC was fitted for the combined data to be then used in factor cokriging. The application of this technique produced a delineation of the field into homogeneous zones, highlighting a wide southern-central zone, characterized by different granulometric and chemical properties. The proposed approach was then effective to discriminate areas with different properties by using multi-sensor data. It then has the potential to be used in PA.

Introduction

Acquiring accurate spatial and temporal information of soil properties at a very fine scale is crucial in Precision Agriculture (PA). Indeed, this technique aims to optimize crop quality and achieve high efficiencies in irrigation and fertilization, while minimizing the potential environmental impacts of farming, through informed management of spatial–temporal variability (ISPA-International Society of Precision Agriculture 2019 https://www.ispag.org/about/definition). Actually characterizing soil properties may be difficult, because they are often both spatially and temporally variable and then obtaining a sufficient number of measurement to describe their heterogeneity may often be prohibitively expensive.

Detailed 3D soil characterization requires knowledge of the spatial distribution and autocorrelation of both surface and subsurface soil properties (Lombardi and Lualdi, 2019, Castrignanò et al., 2018). Both lateral and vertical variations in soil properties have a significant impact on water movement and solute transport. Abrupt changes in soil texture and/or density create discontinuities in mechanical properties that in turn affect the circulation of chemical solution in the soil. Most physical and chemical characteristics actually are closely related to the dielectric properties of soil, which measure the capacity of a medium to allow itself to be conducted by electricity.

Because diverse results show a spatial relationship between crop yield and soil electrical characteristics (Viscarra Rossel et al., 2011, Cavallo et al., 2016, Falco et al., 2021), the ever-increasing diffusion of such measures in PA is now clearly evident. Traditional soil characterization methods, based on sampling using soil cores or augers and laboratory analysis, are accurate enough but have serious disadvantages. At the field scale they are time consuming, laborious and invasive and do not allow long-term repeatability of measurements.

Soil structure significantly affects subsurface water flow. Traditional sampling is in every way destructive of soil structure and is therefore inadequate to describe and predict soil solute flow. For agricultural management purposes, it is therefore necessary to use a non-invasive method that can produce maps of subsurface structural features. Ground Penetrating Radar (GPR) appears to be a particularly suitable methodology for characterizing soil and subsurface (in 3D space) from a physical property perspective at a fine spatial both lateral and vertical resolution and over relatively large areas. It is a non-invasive geophysical technique, which proved to efficiently assess and model surface and subsurface properties such as lithology, porosity, water content, hydraulic conductivity, and clay content (De Benedetto et al., 2012, Liu et al., 2019). However, sensor response depends on many variables, including texture, structure, soluble salts, water content, temperature, density, as well as measurement frequency (Daniels, 2004). In particular, the electrical permittivity of unsaturated soils strongly depends on the water content of the soil. The permittivity of water is of the order of 80, while that of dry mineral soils around 2–3, so even small amounts of water can cause significant increases in the permittivity of the material.

It follows therefore that a complete characterization of an agricultural soil, to be used for effective delineation in homogeneous areas subject to differential management (management zones), cannot be based on the outputs of a single sensor (Adamchuk et al., 2011). Poor management of an agrarian soil may also be due to a lack of knowledge of properties such as mineralogy, which influence the physical and chemical attributes of soils (Camargo et al., 2015, Coblinski et al., 2021, Oliveira et al., 2020). Soil texture is directly influenced by clay minerals and in turn impacts numerous processes including water dynamics. Moreover mineral composition, particularly the content of iron oxides, determines how much certain nutrients and potential toxic elements are adsorbed by soils (de Oliveira et al., 2020; Sun and Zhang, 2017). Phosphorus adsorption and catio-exchange capacity are controlled by the nature and concentrations of minerals (Ramaroson et al., 2018; Soriano-Disla et al., 2014).

Visible-near infrared-shortwave infrared (VIS-NIR-SWIR) reflectance spectroscopy has now become a widespread technique in soil analysis due to its indisputable properties: it is indeed rapid, non-destructive, inexpensive, environmentally friendly and efficient (Viscarra Rossel et al., 2016, Bellino et al., 2015). As the radiometric sensor uses different extended portions of electromagnetic waves to assess the level of energy reflected / absorbed by soil, it provides considerably more information than that derived from a single spectral measurement (Coblinski et al., 2021, Gholizadeh et al., 2020). Peaks and valleys of spectra can be used to directly determine soil attributes including mineralogy (Fang et al, 2018) and texture (Coblinski et al., 2020), since certain chemical bonds and molecules absorb electromagnetic radiation at specific wavelengths (Ramaroson et al., 2018). In particular, in the VIS region (400–700 nm) there lies the spectral curve of chromophores, i.e. sets of atoms responsible for the color and structure of a molecule (Viscarra Rossel and Behrens, 2010), while in the NIR-SWIR region (700–2500 nm) there lie the spectral signatures of clay minerals and organic matter (Zhao et al., 2018).

From the above, it is therefore evident how information on spatial–temporal variability of soils can be improved by the integration of data from multiple sensors as compared to that from individual sensors. This process of combining different sensors, both proximal and remote sensors, known as sensor or data fusion, is becoming increasingly common in PA (Castrignanò et al., 2018, Grunwald et al., 2015, Schirrmann et al., 2017, Shaddad et al., 2016, Whattoff et al., 2017). The reason for this success is that it provides a more accurate, robust, and complete description of the agricultural environment and a better understanding of the processes at play of interest (Wenjun et al., 2019, Adamchuk et al., 2011) as well as extended both lateral and vertical coverage of attributes and increased dimensionality of the measurement space (Viscarra Rossel et al., 2011).

However, there no universal approach for data fusion, but several methods have been used including statistical methods, such as multiple linear regression, partial least squares regression (La et al., 2016, Mahmood et al., 2012, Piikki et al., 2013, Veum et al., 2017), machine learning/ data mining, such as classification and regression tree (Park et al., 2017), artificial neural network (Nawar and Mouazen, 2017), support vector machine (Khalil et al., 2005), random forest (Park et al., 2017) and multivariate adaptive regression splines (Nawar and Mouazen, 2017).

Alternatively, multivariate geostatistics can quantify the spatial dependence of the outcomes from different sensors and provide synthetic spatial indices, which integrate multi-source data in a physics-based and effective way (De Benedetto et al, 2013a). These indices can be used to partition soil into homogeneous zones to be submitted to site-specific management (Castrignanò et al., 2018). The overall objective of the work is to investigate the advantages of a joint use of GPR and hyperspectral data and to define a data fusion approach, in view of a possible application in precision agriculture for the delineation of management zones.

Section snippets

Study site

The study area is located in the South-East of Italy, in the province of Brindisi (40°31′10.57″ N, 17°39′42.86″ E), and is represented by an olive grove consisting of a single portion of century-old trees of the 'Ogliarola Salentina' cultivar. The trunk is large, twisted, with a thick and bushy canopy. The planting pattern is very wide and in some respects irregular; there is no irrigation system. The plot is flat, with loose, skeleton-free soil and is periodically mowed to control weeds. The

Hyperspectral data

The spectral raw data were preferred over the pretreated ones (Riefolo et al., 2020) because a lower number of PCs explained the same proportion of variance, probably due to the specific measurement modalities within the black box. On the other hand, the literature (Coblinski et al., 2020; 2021) also uses raw data without pretreatment, in addition to that already described in the methodology, to determine the chemical composition of soil samples from spectra.

Five PCs were extracted from

Conclusions

In this work, a combination of two apparently very different sensors was proposed: a GPR with a center frequency of 250 MHz and a spectroradiometer operating within the 350–2500 nm range of the electromagnetic spectrum, whose outputs were efficiently processed with multivariate geostatistical techniques. It was demonstrated that the approach used led to a partitioning of the soil into a large central-southern zone and neighboring zones with supposed different granulometric and chemical

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

Project “XylMap - Identification of CoDiRO diffusion dynamics after analysis of progression mechanisms and development of enhanced monitoring and mapping tools and methods” was financed by the Apulia Region (Italy) with reference to DD n. 494 of 14/10/2015 and n. 278 of 9/8/2016 (Cod. A)

References (66)

  • G. Dufrechou et al.

    Geometrical analysis of laboratory soil spectra in the short-wave infrared domain: clay composition and estimation of the swelling potential

    Geoderma

    (2015)
  • C. Gomez et al.

    Continuum removal versus PLSR method for clay and calcium carbonate content estimation from laboratory and airborne hyperspectral measurements

    Geoderma

    (2008)
  • F.A. Kruse et al.

    The spectral image processing system (SIPS) – interactive visualization and analysis of imaging spectrometer data

    Remote Sens. Environ.

    (1993)
  • J.S. de Oliveira et al.

    Soil properties governing phosphorus adsorption in soils of Southern Brazil

    Geoderma Reg.

    (2020)
  • S. Park et al.

    Drought monitoring using high resolution soil moisture through multi-sensor satellite data fusion over the Korean peninsula

    Agric. For. Meteorol.

    (2017)
  • K. Piikki et al.

    Sensor data fusion for topsoil clay mapping

    Geoderma

    (2013)
  • V.H. Ramaroson et al.

    Mineralogical analysis of ferralitic soils in Madagascar using NIR spectroscopy

    Catena

    (2018)
  • W. Sun et al.

    Estimating soil zinc concentrations using reflectance spectroscopy

    Int. J. Appl. Earth Obs. Geoinf.

    (2017)
  • F. Van Der Meer

    Analysis of spectral absorption features in hyperspectral imagery

    Int. J. Appl. Earth Obs. Geoinf.

    (2004)
  • F. Van Der Meer

    The effectiveness of spectral similarity measures for the analysis of hyperspectral imagery

    Int. J. Appl. Earth Obs. Geoinf.

    (2006)
  • K.S. Veum et al.

    Sensor data fusion for soil health assessment

    Geoderma

    (2017)
  • R.A. Viscarra Rossel et al.

    A global spectral library to characterize the world’s soil

    Earth-Sci Rev.

    (2016)
  • D. Whattoff et al.

    A multi sensor data fusion approach for creating variable depth tillage zones

    Adv. Anim. Biosci.

    (2017)
  • M.L. Whiting et al.

    Predicting water content using Gaussian model on soil spectra

    Remote Sens. Environ.

    (2004)
  • A.P. Annan
  • A. Bellino et al.

    Chemometric technique performances in predicting forest soil chemical and biological properties from UV-Vis-NIR reflectance spectra with small, high dimensional datasets

    iForest-Biogeosciences and Forestry

    (2015)
  • E. Ben-Dor et al.

    Near-infrared analysis as a rapid method to simultaneously evaluate several soil properties

    Soil Sci. Soc. Am. J.

    (1995)
  • E. Ben-Dor et al.

    Soil reflectance

  • Brigatti, M.F., Galán, E., Theng, B.K.G., 2013. Chapter 2-Structure and mineralogy of clay minerals. In: Bergaya, F.,...
  • R.B. Cattell

    Fixing the number of factors: The most practicable psychometric procedures

  • F.H. Chen

    Foundations on expansive soils

    (1988)
  • J.P. Chilès et al.

    Geostatistics: modeling spatial uncertainty

    (2012)
  • Clark, R.N, 1999. Spectroscopy of rocks and minerals, and principles of spectroscopy. Rencz, A.N. (Eds.). Remote...
  • Cited by (9)

    View all citing articles on Scopus
    View full text