Characterisation of the Spectral-Temporal Pattern of the Crambe Crop Using Hyperspectral Sensors

Crambe is an oleaginous plant mainly cultivated in Brazil due to its oil characteristics that provide stability to oxidation, qualifying it for the use in a variety of products. Understanding the spectral-temporal pattern of the crambe crop is important to identify and quantify already cultivated areas via remote sensing. This study spectrally characterised the plant, seeking to relate the spectral pattern to the phenological stages of the crop throughout its development. The spectral information was obtained by passive terrestrial sensors in two harvests, thus generating a spectral-temporal pattern and the crambe temporal profile through the vegetation indices NDVI and SAVI. During the phenological stages of the seedling and the beginning of the vegetative growth, the red spectral band showed higher values of reflectance; this occurred because the crop had not yet completely covered the soil. Stages at the end of the vegetative growth and the beginning of the flowering, there was a higher reflectance in the near infrared and a lower reflectance in the mid-infrared. For the granulation and maturation stages, the reflectance in the mean and near infrared reduced due to leaf senescence and loss of cellular water content. The NDVI and SAVI temporal profiles demonstrate linear growth up to the vegetative peak, which occurs between the end of the phenological stage of the vegetative growth and the beginning of the flowering and highest amount of green biomass. At the beginning of grain formation and filling, yellowing of leaves and senescence, granulation and maturation stages, the values reduced.


Introduction
Crambe is a cruciferous plant originating from Ethiopia, with a seed oil content of up to 38% (Knights, 2002).Its development is composed of different phenological stages: seedling, vegetative growth, flowering, granulation and maturation.It has a short cycle, between 78 days and 125 days, depending on the region of cultivation and sowing time (Viana, 2013;Oliveira et al., 2013).Coutinho, Esquerdo, Oliveira, and Lanza (2012) state that in order to map and monitor areas of annual agriculture, spatial and temporal information of the national agricultural activity is required, synchronised with the phenological development of the crops.This characterises the spectral-temporal behaviour that supports the accurate assessment of the productive potential of the plants.
However, mapping agricultural crops by remote orbital sensing is still a challenging undertaking.Under full soil coverage conditions, for most current orbital sensors, different crops may appear to be spectrally similar (Yao, Tang, Wang, & Zhang, 2015).To change this, new technologies have been tested in terms of their potential for spectral differentiation.With the use of hyperspectral sensors, images with hundreds of narrow and continuous spectral bands are acquired.Thus, hyperspectral images have substantially improved the ability to distinguish multiple characteristics of agricultural crops by better differentiation and estimation of biophysical attributes (Mulla, 2013).
Field spectroradiometers perform in situ radiometric collections, thus providing not only detailed data on the spectral characteristics of targets, but also allowing the acquisition of physical values, such as radiance and reflectance, which spectrally characterise different objects without the interference of external factors (Martins & Galo, 2015).
Vegetation indices (VIs) are spectral measurements from mathematical combinations of spectral ranges from red (620 to 700 nm) and near infrared (NIR) (700 to 1.300 nm), which provide more than 90% of the spectral information of vegetation (Viña, Gitelson, Nguy-Robertson, & Peng, 2011).According to Motomiya et al. (2014), VIs are generally related to biomass, chlorophyll content and the productive potential of plants.
The normalised difference vegetation index (NDVI) and the soil-adjusted vegetation index (SAVI) are among the most used indices to monitor the biomass temporal evolution of agricultural crops (Ponzoni, Shimabukuro, & Kuplich, 2012).Research has shown that they are important tools in monitoring crop development (Risso et al., 2012) and correlate with grain yields (Bredemeier, Variani, Almeida, & Rosa, 2013;Monteiro, Angulo Filho, Xavier, & Monteiro, 2013).A time series of NDVI values of agricultural crops throughout their vegetative cycle can provide useful information on growth and crop state (Li et al., 2014).
In this context, this study aimed to characterise the spectral-temporal pattern of the crambe, obtained by a passive terrestrial sensor, and tried to relate the spectral pattern of the crambe crop with its phenological stages.Such an approach could provide a technical basis for the potential use of satellite images (SR orbital) in order to identify, delimit, map and monitor agricultural areas cultivated with crambe on a large scale.

Characterisation of the Study Area and Crop Implantation
The study was conducted during the 2014 and 2015 harvests in the experimental areas of the University Center FAG, Cascavel city of Paraná state, Brazil, latitude 24°56′1.21″S and longitude 53°30′41.63″W, at an altitude of 700 m (Figure 1).

The climat 18 ºC (me is infreque classified 2011).
In 2014 an sown in th Seeds of t nine rows basic fertil

Characterisation of the Spectral-Temporal Signature
The hyperspectral information of the 12 sample points was grouped by day of harvest in the software Excel, calculating the daily means; graphs of the spectral-temporal signature of each harvest were designed.

Calculation of Vegetation Indices
Daily averages of each sampling point were generated, totalling 12 mean samples per collection day.To calculate the Normalized Difference Vegetation Index (NDVI), spectral range information was selected from 620 to 1300 nm, classified into two spectral bands: V -Red (620 to 700 nm) and NIR -Near infrared (700 to 1300 nm), and applied in the formula proposed by Rouse , Hass, Schell & Deering (1974), according to Equation 1.
Where, ρIVP: Range of near infrared; ρV: Range of red.
Another index generated from the daily mean sample was the Soil-Adjusted Vegetation Index (SAVI), based on the formula proposed by Huete (1988), who introduced an adjustment factor (L) between -1 and +1, Equation 2.
Where, ρIVP: Range of near infrared; ρV: Range of red; L: Adjustment factor for the canopy substrate of the plant, which takes into account differential canopy extinction for red and near infrared (Huete, 1988).Huete, Justice and Liu (1994) found that by applying L = 0.5, the soil brightness variations were minimised, eliminating the need for additional calibration for different soil and canopy types.

Statistical Analysis
The VIs were tabulated and graphs were generated with the temporal profiles of the daily means of NDVI and SAVI of the crambe in the 2014 and 2015 harvests.Both profiles were submitted to the normality test using the Shapiro-Wilk method and exploratory data analysis in the software Action Stat.

Results and Discussion
Crambe showed a growth cycle of around 120 days in the two harvests.In 2014 and 2015, cumulative precipitation was 832 and 827 mm, respectively.According to Pitol et al. (2010), precipitation excess during crop development favours the appearance of fungal diseases that affect grain yield.

2014 Harvest
Figure 3 represents the spectral signature at different crambe developmental stages.The curve that includes the spectral data mean of the phenological stages of PL and beginning of VG -21 DAS (Days After Seeding) shows the highest reflectance of the red spectral range (620 to 700 nm) when compared to curves 41, 56 and 88 DAS.This fact was related to the large soil area still exposed during data collection; in these phenological stages, the soil was only partially covered by the crambe canopy.The soil of the experimental area was rich in iron oxide, causing the red tonality, which may have contributed to this behaviour (Sousa Jr., Demattê, & Genú, 2008;Genú & Demattê, 2012).
The phenological stage curves of VG and FL, 41 and 56 DAS, show a gradual increase in the absorption of blue (400 to 500 nm) and red (620 to 700 nm) wavelengths.According to Ponzoni et al. (2012), the chlorophyll in the green leaves absorbs these spectral regions, converting heat and stored energy through photosynthesis.These curves, compared to the beginning of the cycle (21 DAS), show an increase in the reflectance of the green range due to leaf pigmentation.
The highest reflectances in the NIR range (near infrared; 700 to 1300 nm) are represented in the phenological stage curves of FL and FL/GR, 56 and 88 DAS, with the reflection peak in the stage that includes the beginning of flowering.Jensen (2009) emphasises that the energy incident on the structure of a green and healthy leaf generates scattering in the spongy mesophyll and increase in reflectance.Figure 3 shows examples of this occurrence in crambe leaves at 56 DAS.
The MT stages of the plant, 104 DAS curve, are characterised by the reflectance increase in the MIR range (1300 to 3200 nm).Leave senescence and water loss in the cell structure contribute to this behaviour, according to the field situation in 104 DAS, presented in the image of this day (Figure 3).The NDVI means of 2014 and 2015, submitted to the Shapiro-Wilk test, presented a p-value of 0.6385 and 0.2500, respectively.Values greater than the 0.05 significance level represent normality in the data distribution.
Table 2 shows the exploratory data analysis of NDVI of crambe, 2014 harvest.All characterised by low dispersion, the lowest standard deviation (SD) of 0.014 was collected at the beginning of the cycle, PL and early VG stages -24 DAS and higher during VG in 41 DAS (0.0402).According to Pimentel-Gomes (2002), the NDVI coefficients of variation ( 2014) are classified as low dispersion and high precision in the experiment.(2) NDVI Temporal Profile Figure 5 shows the time profile of NDVI of crambe in the 2014 and 2015 harvests.The indices obtained at the beginning of the cycle, 24 DAS (2014) and 21 DAS (2015), were the lowest in the whole crop cycle, 0.36 and 0.29, respectively.In both harvests, the plants were in the transition between the PL stages and early VG, unfurling the first true leaves.Plants with small size and on exposed soil contribute to electromagnetic energy absorption in the NIR range, justifying the lowest NDVI values (Poelking, Lauermann, & Dalmolin, 2007).
Figure 5 Risso   (2) SAVI Temporal Profile Figure 6 shows the SAVI time profile of crambe in the 2014 and 2015 harvests.The indices showed a behaviour similar to that of the NDVI time profile.Our results are in agreement with the findings obtained in a study comparing SAVI, NDVI and LAI values of vegetation coverage in Mato Grosso do Sul, Brasil (Braz, Águas, & Garcia, 2015).

Table 2 .
Exploratory analysis of the NDVI vegetation indices of crambe, 2014

Table 3
shows the NDVI exploratory data analysis of the crambe, 2015 harvest.Low dispersion and high precision are data characteristics, except the values generated during the VG -34 DAS.The lowest standard deviation and the lowest coefficient of variation were obtained in FL/early GR in 80 DAS, 0.005 and 0.58%, respectively.The highest SD (0.0556) and a CV of 12.51%, classified by Pimentel Gomes (2002) as mean dispersion, were found in 34 DAS in the VG stage.

Table 3 .
Exploratory analysis of the NDVI vegetation indices of crambe, 2015

Table 5
corresponds to the SAVI exploratory analysis -2015 harvest.Data obtained in 48, 80 and 114 DAS showed low dispersion of data.The other collection days classified by mean data dispersion (CV: 10 to 20%).

Table 5 .
Exploratory analysis of the SAVI vegetation indices of crambe, 2015 crop).We also thank the CNPq (National Council of Scientific and Technological Development) and CAPES (Coordination of Improvement of Higher Education Personnel) for financial assistance and the Laboratory of Geoprocessing and Topography -GEOLAB of the UNIOESTE for help with data collection in the field. jas.ccsenet.