Detection of Red Palm Weevil infected trees using thermal imaging O. Golomb1 2, V. Alchanatis1, Y. Cohen1, N. Levin2, Y. Cohen1, V. Soroker1 1Agricultural
research organization, the Volcani center, Israel, The Hebrew University in Jerusalem, Geography Department, Israel
[email protected] 2
Abstract The red palm weevil (RPW) is a palm borer insect that develops within the soft tissues of the trunk and crown, eventually leading to tree death. Early detection of RPW infestation is crucial. The RPW larvae developing inside the palm and direct visual detection of the infestation is quite difficult. The hypothesis was that the tunneling insects destroy the vascular system of the palm and create local conditions of water stress. The goal of this study was to examine the ability to detect infected trees using thermal images. By measurements, imaging and analyzing of infected and uninfected trees over multi-year experiments in quarantine and commercial orchards, results partially showed that the RPW creates water stress and affects canopy temperature. Analysis of aerial thermal image above date palm plantation successfully detected infected trees. Keywords: Palms, Palm Pests, Remote sensing, CWSI, Canopy temperature Introduction The red palm weevil (RPW), Rhynchophorus ferrugineus (Olivier), is a challenging palm borer insect that develops within the soft tissues of the palm trunk and crown, eventually leading to tree death (Gutiérrez et al. 2010). Since the mid-1980s the RPW has caused serious damage to palms worldwide (Abraham et al. 1998). The RPW originated from south Asia and Malaysia and attacks more than 20 palm species globally (Giblin-Davis 2001). In Israel and Jordan, infestation was first reported in 1999 (Soroker et al. 2005). The complexity of competing with this pest derives from the fact that the RPW is well hidden in the plant trunk and direct visual detection of the infestation is quite difficult. Early detection of RPW infestation is particularly crucial as palms at early infestation stage, with the palm heart not yet damaged and with the trunk still stable, can be treated and usually recover. Early detection methods are key for a successful fight against the pest and in any attempt to guarantee pest-free status in adult trees (Gutiérrez et al. 2010). The RPW spreading is a major risk to both ornamental palm trees and palm plantations. Thus, there is a need to develop a methodology that will enable detection of suspected infected trees over large areas. The major hypothesis of this study was that the tunneling insects destroy the vascular system of the palm, thus creating local conditions of water stress that can be detected by thermal imaging (Gates 1964; Tanner 1963). Aerial thermal images enables mapping of canopy temperature variability over large areas. Thermal imaging systems have been used to evaluate water status of various orchards such as olives (e.g. Ben-Gal et al. 2010), vineyards (e.g. Möller et al. 2007; Zarco-Tejada et al. 2012), almonds (GonzalezDugo et al. 2012), and citrus (Gonzalez-Dugo et al. 2013). A recent study showed that aerial thermal images can be used also to detect water stress in date palm trees on a
commercial scale (Cohen et al. 2011). The objective of this study was to examine the possibility of detecting RPW infection of palm trees by thermal imaging. Materials and methods Controlled Experiment in quarantine Experimental setup: Measurements were conducted from 2012 to 2014 at Eden Research Station, Ma'ayanot regional council, Israel (32.27°N, 35.29°E). During these three years, six experiments were conducted. In each, young palms were infested by RPW and compared to control trees. Three of the experiments were performed on Canary palm trees (P. canariensis) and three on date palm trees (P. dactylifera). Trees were planted in large (100 L) pots and grown in an isolated, insect-proof quarantine. Each experiment contained 4-5 replicates of control trees and 8-10 replicates of infested trees, which were intentionally infested by releasing a known number of adult RPW around every tree. In order to prevent secondary infection, all weevils were captured 1-2 weeks after they were released. After collecting the weevils, measurement campaigns were conducted as listed in Table 1. Table 1. Summary of experiments conducted from 2012 to 2014. Exp
Measurement times (weeks after infestation)
Species
Start date
End date
#1 #2 #3 #4 #5 #6
3, 5, 6, 8 2, 4, 6, 8, 12 3-7 (every week) 4-10 (every week) 1, 4-15 (every week) 6, 8, 9, 11
Canary Canary Canary Date Date Date
29/07/12 09/10/12 18/07/13 15/09/13 14/04/14 07/08/14
24/09/12 03/01/13 02/09/13 21/11/13 30/07/14 21/10/14
Image acquisition: Thermal infrared images (TIR) were acquired using an uncooled infrared thermal camera (ThermaCAM model SC655; FLIR Systems, Wilsonville, Oregon, United States) with a microbolometer sensor of 640 X 480 pixels operating in the spectral range of 7.5–13 µm, with thermal sensitivity of 0.1ºC, and temperature accuracy of ±2% of the reading. The camera was equipped with a 13-mm (45º) lens. In addition to the thermal images, RGB images were acquired by a digital camera (Power Shot A640, Canon Inc., Tokyo, Japan) for further identification and selecting sunlit leaves for analysis. The TIR and RGB cameras were mounted on a platform facing west, to prevent shaded areas in the frame. Cameras fields of view (FOVs) overlapped considerably. The cameras platform was elevated 3-m above the trees canopy, resulting in a spatial resolution of 38-mm per pixel in the thermal images. Image acquisition was carried out between 12:00-14:00 when solar radiation was at its maximum. Damage evaluation: The infected trees were physically dissected at the end of each experiment. Beginning from the outer perimeter and moving towards the palm heart, leaves were removed and examined for damage caused by the weevil larvae to the leaf bases and the trunk. This dissection allowed evaluation with high reliability of the damage caused by the RPW larvae and compare it to TIR data. After evaluating if the overall tree damage was enough to affect water transportation in the trunk, the trees were categorized as infected for further analysis based on od eessid eht observations .
Thermal images processing: Thermal images were processed using ThermaCAM Researcher software (FLIR Systems, Inc.). Local environmental conditions and leaves emissivity (0.98; López et al. 2012) were entered into the software to produce reliable leaf temperature maps. Images were exported to Matlab (Mathworks Inc., Natick, MA, USA) and ArcMap (ESRI, Redlands, CA, USA) files for further analysis. The thermal images were processed to extract the canopy representative temperature. In order to extract the temperature of pure canopy pixels, a Graphical User Interface (GUI) was developed using Matlab. First, background pixels were separated and canopy pixels were identified using the histogram for every image. Next, the user defined interactively the minimum and maximum temperatures that represent the canopy and the average temperature was calculated. Furthermore, the crop water stress index (CWSI) (Jackson et al. 1981) was used as a tool for assessing the water stress induced by the RPW: CWSI
Tc Twet Tdry Twet
(1)
Where Tc is the average canopy temperature; Twet and Tdry are the lower and higher boundaries for canopy temperature, respectively. Tdry was set as air temperature + 50C and Twet was derived from the energy balance following Jones (1999). For Twet calculation, air temperature, solar radiation, relative humidity, and wind speed were monitored using a meteorological station within the experimental location with sensors at canopy level. These estimates of Tdry and Twet were found adequate for CWSI calculations for palm trees (Prigojin 2010). Statistics: comparisons were of mean CWSI values for each campaign day. First, compare between treatments. Second, compare every infested tree CWSI value to the mean upper bound of the confidence interval around the mean CWSI of control trees. Aerial thermal imaging of commercial plantation At the beginning of 2013, a few RPW infected foci were detected in large commercial date plantations in Israel. After ground validation of the infection, aerial thermal images were acquired above a commercial Hayani date palm orchard in Ma'ale Gamla (32°53'20.13"N, 35°41'16.66"E) on September 11, 2013. The same thermal camera was used, as described in the ground experiments. The camera was equipped with a 24.4mm lens and acquired images from a height of around 600-m above the ground, enabling a spatial resolution of approximately 0.5 m. Environmental conditions were monitored using a meteorological station that was placed at the plantation. In order to separate the palm canopy from soil background, an image-processing algorithm was applied to the TIR images. The entire processing was based on the algorithm used by Cohen et al. (2011), where a watershed image-processing algorithm was used to outline the palm canopy pixels. Application of the watershed algorithm to a palm orchard resulted in basins that could represent either palm trees or soil. Potential palm tree basins can be isolated by identifying cooler basins. In the present study, this was done by selecting basins that contained at least one pixel below a certain temperature threshold, Tth, and excluding all other basins from further analysis. The value of Tth, was set automatically by the Otsu (1979) method for thresholding. Selection of pure-canopy pixels within the basins was done by searching Nmax pixels within each of the basins with the lowest temperatures. The value Nmax = 100 pixels
(~25 m2) was set according to the aerial thermal image resolution and the canopy size. Figure 1 shows trees' boundaries in the plantation as detected by the algorithm.
a
b
c
Figure 1. RGB (a) and thermal image (b) of the date palm plantation and the boundaries of palm trees canopy detected by the watershed algorithm (c). Statistics: Statistical analysis was performed using JMP (SAS Institute Inc., Cary, NC, USA). Significant differences between the treated and control plants were determined using ANOVA test (α