Detection of Diseases in Oil Palm Plantations in the Brazilian Amazon through Orbital Image

The detection of diseases in oil palm crops in the Brazilian Amazon represents a great challenge for the management of this crop in Brazil. The plantations in the State of Pará provide inputs for the food, cosmetics, agro-energy and biofuel industries, supplying Brazilian markets. In recent years, several factors such as pests, diseases and climate have interfered in the development of oil palm in the region, generating the need to adopt new techniques to detect and monitor such issues. In this work, spectral enhancements were carried out by simple reflectance and vegetation indices for four plots cropped on Companhia Palmares da Amazônia (CPA) farm, owned by Agropalma S.A. company in the municipality of Acará, in the state of Pará. The results allowed the identification of expressive patterns minimum and maximum reflectances of the studied plots, correlating with occurrences of diseases. The EVI index showed an excellent correlation with the occurrence of diseases. However, the NDVI and SAVI indexes showed adequate adjustments with the occurrence of diseases in 2017. The areas corresponding to the L36 and H27 plots showed higher occurrences of diseases, based on the analysis of reflectance through vegetation indices. It is concluded that the reflectance enhancements, NDVI, SAVI and EVI obtained by orbital sensors are efficient in the detection of diseases in the plots. The results allowed the identification of diagnostic anomalies of stresses in the plots, either by disease or other factor, allowing the decision making in an adequate time, therefore avoiding large scale eradication in the extensive areas in commercial palm oil plantations in Brazil.


Introduction
Oil palm (Elaeis guineenses Jacq.) is an oleaginous palm native to Africa, grown in large scale in the Brazilian Amazon, in areas that have already been deforested. In the region, approximately 207 thousand hectares of cropped area is totalized, about 88% of the Brazilian crop where the largest oil production is found (Abrapalma, 2018).
In Brazil and in the Brazilian Amazon, the state of Pará is the largest producer of oil palm, particularly in the municipalities of Acará, Baião, Bujaru, Concórdia do Pará, Igarapé-Açu, Moju, São Domingos do Capim, Tailândia, and Tomé-Açu. The region is favored by ideal agro-climatic conditions and governmental support provided by the Sustainable Palm Oil Production Program in Brazil (Alves et al., 2016).
Due to the increase in cropped areas and the recognized economic importance of the culture in the Brazilian Amazon, especially in the state of Pará (Brazil), there is a need for reliable, comprehensive, systematic and agile information on the oil palm, which has become even more crucial in the face of phytosanitary issues that strongly affect the productivity of these palms and decrease their expansion (Duarte et al., 2008;Pinho et al., 2016). Among these phytosanitary limitations, we highlight insects, drill pests (Rhynchophorus palmarum, Linnaeus, 1764;and Eupalamides cyparissias cyparissias Fabricius, 1776), defoliators (Opsiphanes invirae Hubner, 1818), and oil palm diseases, such Fatal Yellowing (FY), red ring disease (Bursaphelenchus cocophilus Cobb, 1919), sudden wilt (Phytomonas staheli McGhee and McGhee, 1979) and vascular wilt (Fusarium oxysporum f. sp. elaedis) (Flood, 2006;Maciel et al., 2015;Assis Costa et al., 2018;Nascimento et al., 2018;Martínez et al., 2019).
Most agronomic information related to the occurrence of diseases in Amazonian oil palm plantations is obtained manually and rudimentarily by means of field surveys performed by the so-called "pragueiros" or field agents, who range plant-to-plant, seeking the visual recognition of pests and diseases. As a result, work becomes expensive, time-consuming and subject to frequent mistakes. Such a model generally allows for late diagnosis, with no other curative action to be taken, resulting in destruction of the plant (Giese et al., 1975;Goodell, et al., 2014).
Among several techniques of aid for farm crop management, the use of remote sensing has shown promising results for rapid and efficient detection of diseased plants in small and large geographic areas (Naue et al., 2011;Zhang et al., 2019). It is a technology composed of sensors and platforms, capable of obtaining information from the surface without physical contact with it, and often at long distances, from satellites and manned aircraft. Thus, remote sensing has become an instrument of great potential, where its usefulness contributes in several studies such as vegetation, acting in precision agriculture through the monitoring and physiological and phenological studies of the plantation (Novo, 2010;Mahlein, 2016;Zhang et al., 2019).
For the evaluation and monitoring of agricultural crops, remote sensing allows spectral analyses through reflectance curves and vegetation indexes, qualifying this technique as relevant for agricultural management. Through the analysis of spectral parameters of the vegetation, with more attention to the visible and near infrared, the several vegetation indices allow the differentiation in several points of the agricultural and forest cultivation (Ponzoni & Shimabukuro, 2007).
The understanding of the spectral behavior allows to extract information of the vegetation distribution, canopy structure, phenological state, stress conditions, nutrient deficiency, among others (Ponzoni et al., 2012). The NDVI is an index that can be used in studies related to vegetation dynamics with more homogeneous values, besides being a good estimator of biomass. However, it can be influenced by atmospheric interferences (Souza & Silva, 2016).
By considering the successive occurrences of pests and diseases in the oil palm in recent years (Maciel et al., 2015;Torres et al., 2016;Murguía-González et al., 2018;Martínez et al., 2019), and the great difficulties of their containment in the current management routines, it is important to consolidate a technique that is efficient and shows a rapid response for the detection of diseases in the extensive plantations of the region, starting with the present experiment carried out in plots in the municipality of Acará. Thus, this paper evaluates the efficiency of the Sentinel-2A MSI orbital sensor for the detection of anomalies caused by diseases in the palm oil crop in the municipality of Acará, representative of the crops in the Brazilian Amazon.

Study Area
The study area is owned by Agropalma S.A., in municipality of Acará, located in the mesoregion of the Northeastearn Para and microregion of Tomé-Açu, about 114 km away from Belém, capital of the state of Pará (PA), in the Brazilian Amazon ( Figure 1). It has an estimated population of 55,513 thousand inhabitants and a territorial area of 4,343.550 km² (IBGE, 2018a). The climate of the municipality corresponds to the megathermal, Am type according to the classification of Köppen, that is, hot and wet equatorial, with propensity to agropastoral activities. Besides, the favorable climate and abundance of water allowed many projects aimed at the production of palm oil. Rainfall ranges from 2,000 to 3,000 mm annually, with higher occurrences between December and April. The soil type is predominantly Yellow Latosol and medium clayey texture. This study was carried out in an area of cropped plots, located by the geographic coordinates 02º18′10.1″ S and 48º39′40″ W.
The experiments were carried out in the L30, L36, H25 and H27 plots, with individual areas of approximately 36 hectares owned by Companhia Palmares da Amazônia (CPA) company belonging to Agropalma S.A. (Figure 1), The Normalized Difference Vegetation Index (NDVI) is a spectral enhancement resulting from the combination of reflectance levels of red and infrared spectral bands. The NDVI, proposed by Rouse et al. (1973) was calculated through the following formula: Where, ρNIR is the reflectance of the vegetation in the near infrared band and ρRED is the vegetation reflectance in the red band. Leaf reflectivity in the NIR stems is the result of the presence of intracellular spaces in its structure, and the larger the number of space within the leaf, the greater the radiation reflectivity (Jackson et al., 1983).
Soil-Adjusted Vegetation Index (SAVI) is an improvement on NDVI, proposed by Heute et al. (1988). It minimizes the effects of soil that partly influences NDVI, allowing the increase in the value of vegetation indices, especially in canopies with average levels of green cover. The constant "L" ranges from 0 to 1, according to the density of the vegetation. Thus, for low-density vegetation cover, the factor L is 1.0; medium coverage plant density, the factor will be 0.5; and for high densities, the L factor will be 0.25. The index proposed by Huete et al. (1988) is calculated through the formula: Where, ρNIR is the near infrared band, ρRED is the red band and L is the adjustment factor for the canopy substrate.
The Enhanced Vegetation Index (EVI), developed by Huete et al. (1997), seeks to improve vegetation sign by reducing soil and atmospheric interferences. It is very sensitive in regions with high biomass concentration. It is based on the following calculation: Where, ρNIR is the near infrared band, ρRED is the band of red, ρBlue is reflectance in the blue region, G is the gain factor, C1 and C2 are adjustment coefficients for the aerosol effect of the atmosphere and L is the adjustment factor for the canopy substrate.

Spectral Characterization Through Reflectance Enhancement
Values of average reflectance were extracted from the L30, L36, H25 and H27 plots in Sentinel-2A images (Figure 2) ( Table 1). The spectral band in the red band (RED) showed a lower reflectance average in the four plots, therefore, a higher absorption, indicating a RED radiation reception pattern for the plots due to the presence of photosynthetic pigments. However, the H25 plot displayed an anomalous pattern in relation to the other plots, with a greater reduction of the RED absorption, resulting in a strong collapse of chlorophyll and consequently an abnormal functioning of the plant, which is a response of the presence of diseases, pests or nutritional deficiency. The L30, L36 and H27 plots also showed low RED absorption.  The high reflectance in the NIR band refers to the strong interaction between radiation and internal structure of the leaves of the oil palm, common to the vegetation. It can be seen in the Figure 2 and Table 1 that the standard reflectance in NIR is very pronounced in L36 plot, an indicative of greater vegetative vigor. However, the H25, H27 and L30 plots showed a low reflectance, so it can be inferred that there was more deterioration of the cellular structure of the oil palm. When related to oil palm sanity, the short-wave infrared (SWIR) band shows similar behavior to the NIR (Figure 2 and Table 1), corroborating with the low reflectance observed in the H25 and L30 plots and better vegetative vigor in L36. These analyses are confirmed by the incidence of infected plants in the plots (Figure 4).
Thus, this spectral characterization represents a viable technique for the detection of spectral anomalies, indicative of diseases, which were only identified with long field and laboratory works (Amaral et al., 2009). Table 1 shows that the reflectances in the RED band, in the H25 and H27 plots, considered with a higher degree of infection, are higher than those with healthy plants. The reflectance in the NIR region for more infected plots is lower than for areas with healthy plants (L30 and L36). This evidences that the contrast between the leaf reflectance in the RED and NIR bands can characterize the unequivocal presence of anomalies or not in the oil palm.

Detection of Diseases in the Oil Palm Through Vegetation Indexes
The use of vegetation indices in the monitoring of oil palm plantations will allow the early detection of anomalies associated with a variety of causes, including infection caused by red ring and fatal yellowing diseases, pests, water stress or nutritional deficiency. From 2013 to 2017, the incidence of plants infected with Fatal Yellowing (AF) increased, and in 2017, the highest occurrences ( Figure 3) were observed. The application of the technique allowed to determine priority areas for emergency intervention in the crop management. The three indexes evaluated in this study allowed the identification of infected plants by any collapse factor, mainly caused by the diseases. In the four evaluated plots, anomalies referring to different symptoms or combination of symptoms were detected, thus, highlighting the occurrence of fatal yellowing. In this area of study, the width of canopies ranged from 8 to 16 meters for the adult plants, which generated a suitable pattern for area detection, compatible with the spatial resolution of the Sentinel-2A and the scale of this mapping. It was assumed that high index values (NDVI, SAVI and EVI) would be associated with increased photosynthetic activity and therefore, they could be used to discard the occurrence of infected plants (Wulder et al., 2006b). nd H27 in conv Vol. 12,No. 4; pattern of its l el and the abso f leaf reflectan on of the radi r the reflectan Visualizati disease, es these areas FA observ of occurren in the area the eradica Figure 6.
The analy representa images in documente SAVI, the improved palm in thi org ion of the SAV specially in the s ( Figure 6). It ved in Figure 4 nces. This SAV as of the plots ation of some p .  Vol. 12, No. 4;2020