International Turfgrass Society Research Journal Volume 11, 2009
EARLY DETECTION OF PYTHIUM BLIGHT AND BROWN PATCH THROUGH MULTISPECTRAL IMAGING OF CREEPING BENTGRASS FOLIAGE Zachary R. Anderson, and Thomas W. Fermanian* ABSTRACT Early detection, and eventually prevention of disease is vital to the success of a turfgrass stand. The turfgrass industry lacks accurate, reliable, and practical methods to detect the onset of symptoms of certain diseases in highly managed creeping bentgrass {Agrostis palustris Huds. [= A. stolonifera var. palustris (Huds.) Farw.]} turf. Direct sensing, where reflected light is measured from a turfgrass canopy, is capable of detecting and locating turfgrass diseases, with promise of detecting these areas prior to visual damage. Four controlled environment disease assays were conducted at the University of Illinois Urbana-Champaign, USA to evaluate the direct sensing of Pythium blight caused by (Pythium aphanidermatum (Edson) Fitzpatrick) and Brown patch caused by (Rhizoctonia solani Kuhn) on creeping bentgrass. Two imaging sensors were mounted on top of a Plexiglas container 38 cm above and nadir to the treated bentgrass. Pythium blight was first sensed 1.5 to 1.6 h before it was visuably perceptible and brown patch was detected 2.5 to 25 h before visible symptoms or signs. Additional research is needed to develop this technology into a reliable, practical and cost effective tool for predicting future disease problems over large areas of turf, such as golf courses. Keywords: Pythium blight, brown patch, creeping bentgrass, direct sensing, image sensors, disease prediction Zachary R. Anderson, First Assistant Golf Course Superintendant, Normandy Shores Golf Course 2401 Biarritz Dr., Miami Beach, FL 33141. (
[email protected]). Thomas W. Fermanian*, Associate Professor Emeritus, University of Illinois at Urbana-Champaign, 1102 S. Goodwin Ave., Urbana, IL 61801. * Corresponding author: (
[email protected]).
INTRODUCTION Native to Eurasia, creeping bentgrass {Agrostis palustris Huds. [= A. stolonifera var. palustris (Huds.) Farw.]}is maintained on golf courses throughout the world because it produces playing surfaces superior to other coolseason turfgrasses. In spite of their superior quality, creeping bentgrass cultivars remain susceptible to numerous diseases, which can ruin the aesthetic and playing quality of turf surfaces. Fungal pathogens cause the majority of turf diseases. Throughout the United States, turf managers anticipate Pythium blight (causal organism Pythium aphanidermatum (Edson) Fitzpatrick) and brown patch (causal organism Rhizoctonia solani Kuhn) fungal diseases on creeping bentgrass turf each year. Pythium is a genus that can cause root, crown, and foliar diseases in many grass species. Pythium blight is characterized by rapid leaf necrosis under conditions of high moisture, hot temperatures, and nitrogen rich, or lush foliage (Fermanian et al., 2003, Nutter et al., 1983). Symptoms of Pythium blight first appear as dark, water soaked foliage, with leaves eventually becoming light brown, dry and shriveled (Couch, 1995). Although many Pythium species can cause disease, P. aphanidermatum is the most prevalent causal agent of Pythium blight in creeping bentgrass in the Midwestern United States (Hall et al., 1980). Pythium aphanidermatum survives unfavorable growing conditions as oospores in the soil, thatch, and canopy of turfgrass (Nutter, 1980). Brown patch, caused by R. solani, primarily infects the foliage of turfgrasses during periods of high air temperatures and extended hours of leaf wetness or high humidity. Brown patch symptoms
can vary, but the initial symptoms generally appear as single light brown lesions with dark brown margins on leaf blades. Under favorable conditions mycelium spreads between adjacent leaf blades, resulting in a circular brown patch in the turf. When turf managers see the characteristic symptoms of fungal diseases, the pathogen(s) have already destroyed numerous turfgrass cells that can only be replaced through new growth. Even though fungi begin infecting turfgrass tissues before visible symptom expression, many turf managers prefer to wait for a chosen level of symptom expression to occur before applying fungicides. This method of disease control can be problematic because some diseases, such as Pythium blight, can kill extensive areas of seedling or established turf overnight. Detecting diseases at the earliest stages of development may help turf managers minimize the occurrence of visible disease symptoms; and indirectly help to prevent potential weed infestations and poor playing quality of the turf surface. To detect diseases soon enough to prevent extensive turf damage and improve the timing of all disease management practices, turf managers typically rely on disease modeling; enzyme-linked immunosorbent assays (ELISA) tests, and diligent scouting techniques. Disease warning models use actual weather data to predict current disease activity and disease forecast models use projected weather data to predict future disease activity. Neither method accounts for every variable that can affect pathogenhost interactions and subsequent disease development. For example, these models rarely take into account improved turfgrass
cultivars, varying fertility levels, different soil types, or pathogen variability. For this reason, disease models often over or under estimate disease epidemics (Burpee and Goulty, 1986, Nutter, et al., 1983, Schumann, et al., 1994). ELISA testing is typically used after disease becomes visible in order to identify the casual agent of disease. Although ELISA tests successfully detected Pythium sp. up to 24 hours prior to visible disease symptoms, more often ELISA detected Pythium sp. simultaneously with visual detection (Shane, 1991). Implementing ELISA over large areas of turf is expensive, time consuming, and labor intensive, making it a poor tool for prediction of diseases on golf courses (Schumann, et al., 1994). Visual detection of any turf disease is the physical observation of disease symptoms or signs. Symptoms include dead, discolored, or weakened turf, while signs include the vegetative structures (generally mycelium) of a fungal pathogen. Symptoms appear when enough grass cells are affected to cause visible damage to the turf. Signs generally appear early in the morning when dew or free moisture is present on the turf foliage, but can appear anytime if relative humidity (RH) is high enough for extended periods of time. The extensive acreage of turfgrass on golf courses coupled with the microscopic nature of fungi and delayed symptom expression make early detection of diseases extremely difficult. Disease prediction models, ELISA testing, and visual detection all require the application of fungicides at short notice, which may not be practical at some golf courses. Moreover, if disease symptoms occur, damaged tissues can remain in a turf even though the pathogen population may be decreasing or inactive.
The turfgrass industry lacks accurate, reliable, and practical methods to detect the onset of fungal disease symptoms in highly managed creeping bentgrass turf. Direct sensing, where reflected light is measured from a turfs canopy, is one approach capable of detecting and locating turfgrass diseases, with promise of detecting these areas prior to visual damage (Carter et al., 1996, Trenholm, et al., 2000). Plant stresses that alter leaf color (Trenholm, et al., 2000), leaf chlorophyll content (Bell et al., 2004, Lorenzen and Jensen, 1989), internal cellular structure of leaves (Knipling, 1970), and/or canopy biomass (Bell, et al., 2002, Fletcher, et al., 2001) all influence the amount of light reflected from plant foliage. All radiation is described in wavelengths. Individual wavelengths differ in energy, frequency, and length in nanometers (nm), but all wavelengths of radiation are absorbed, transmitted, or reflected by plant foliage. In the visible spectrum (400 - 700 nm), reflectance from vegetative foliage is low due to high absorption of light by photosynthetic pigments, primarily the chlorophylls, but also the carotenoids, xanthophylls, and anthocyanins (Knipling, 1970, Penuelas and Filella, 1998). Reflectance changes in the NIR region (approximately 700 - 900 nm) is believed to be influenced by the internal cellular structure in plant leaves (Hatfield, 1990, Penuelas and Filella, 1998). Canopy reflectance measured only at 600 and 800 nm from creeping bentgrass turf plots infected with dollar spot (/Sclerotinia homoeocarpa F. T. Bennett) were more accurate and had greater precision than visual estimates of disease severity (Nutter et al. (1993). Results from other studies indicated that reflectance indices, which are mathematical formulas
used to combine reflectance data measured from different spectral regions, can further improve reflectance data for stress detection and quantification (Bell, et al., 2004, Blazquez and Edwards, 1986, Carter et al., 1996). Several reports review reflectance indices for plant stress detection (Carter, 1994, Hatfield, 1990, Metternicht, 2003). In turfgrass, a normalized difference vegetation index (NDVI = near infra-red (NIR) - red / NIR + red), and a green normalized difference vegetation index (NDVIg = NIR - green / NIR + green) on average regressed with tissue nitrogen (r2 = 0.76 and r2 = 0.81), and leaf chlorophyll content (r2 = 0.70 and r1 = 0.75) (Bell et al., 2004). These indices were as effective for estimating turfgrass responses to nitrogen fertilization as visual color assessments (r2 = 0.64) and clipping weight measurements (r2 = 0.81) (Bell et al., 2004). A ratio of the average reflectance measured between 500 - 640 nm and 660 900 nm provided reliable detection of three diseases on watermelon foliage (Blazquez and Edwards, 1986). This index was also accurate for determining early vs. advanced stages of disease development (Blazquez and Edwards, 1986). The ratio of reflectance measured at 694 ± 3 nm and 760 i 5 nm was successful at detecting herbicide induced stress in 5 year old loblolly pine (Pinus taeda L.) tree canopies 16 days prior to first visible detection of damage (Carter, et al., 1996a). Although not a primary objective, NDVI (computed as R935 - R661 / R935 + R66i), and the reflectance ratios IR/R (R935/R661) and (Rvoe/Rg 13) each detected nitrogen responses in Penncross creeping bentgrass turf 4 days prior to visual rating differences in turf quality, 8 days prior to visual rating differences in turf density, but 24 hours after visual
rating differences in turf color (Trenholm, et al., 2000). The same reflectance index and ratios, in addition to the another ratio (R706/R760), detected reflectance differences due to herbicide application on Tifway bermudagrass turf 24 hours prior to visual rating differences in turf quality and color, and 3 days prior to visual rating differences in turf density (Trenholm, et al., 2000). As for individual wavelengths, reflectance measured at wavelengths in the visible and far red spectrum (507, 559, 661, and 706 nm) detected nitrogen responses prior to visual rating differences in creeping bentgrass, but reflectance at wavelengths in the NIR spectrum (760, 813, 935 nm) did not (Trenholm, et al., 2000). NDVI and several reflectance ratios were as effective or better for earliest detection of the turfgrass responses to nitrogen and herbicide applications; however, reflectance ratios using data from wavelengths within the visible spectrum were not reported (Trenholm, et al., 2000). In addition to multispectral sensors designed to measure reflectance at specific wavelengths or over selected wavelengths ranges, recent data suggests that standard full color digital cameras that measure light only in the visible spectrum and subsequent digital image analysis (DIA) can successfully quantify certain turfgrass quality characteristics, such as color and percent cover (Karcher and Richardson, 2003, Richardson et al., 2001). DIA has not been examined for early detection of turf diseases. Continuous measurements of radiation reflected from turf surfaces at wavelengths sensitive to stress may signal plant health decline prior to visual assessments. No research has investigated canopy reflectance as a means to detect turfgrass diseases before initial detection
of disease by visual assessment. The objective of this study was to determine if continuous measurements of visible and NIR radiation reflected from turf foliage could signal subtle changes in creeping bentgrass health due to Pythium blight and brown patch diseases before visible detection of disease or signs of disease organisms. We further hypothesized that digital images captured with a standard digital camera and subsequent DIA may effectively signal the potential onset of disease in turf prior to visual assessments. MATERIALS AND METHODS Four disease assays were conducted in 2004 in a growth chamber on the campus of the University of Illinois at UrbanaChampaign: two Pythium blight experiments, classified as PB1 and PB2, and two Brown patch experiments, classified as BP3 and BP4. The experimental design for each assay was completely randomized with two treatments and fifteen replications. Continuous lighting was provided by four fluorescent (Philips 85 watt F72T12/CW/HO Alto Collection) and two soft white 60 watt incandescent lamps. Prior to each assay, the fluorescent lamps were cleaned and new incandescent lamps installed. A water-tight, colorless Plexiglas container (Illini Plastics Supply, Champaign, IL) equipped with a front door was put inside the growth chamber. The Plexiglas container measured 62 cm tall X 35 cm wide X 35 cm deep, and served to maintain high humidity. Materials put inside the Plexiglas container during disease assays included: the treated bentgrass, black and white tiles of known reflectance, water soaked paper towels for humidity, and an environmental Data Logger with external Quantum Light Sensor (WatchDog model #450, Spectrum
Technologies, Inc., www.specmeters.com).
Plainfield,
IL,
Two imaging sensors (A standard digital SLR camera, EOS 10D, Canon U.S.A., Inc., Lake Success, NY with a resolution of 3072 x 2048 pixels at 240 dpi and A Redlake 3100 multi-spectral sensor, Geospatial Systems, Inc.,West Henrietta, NY) were mounted on top of the Plexiglas container 38 cm above and nadir to the treated bentgrass. The Plexiglas container prevented moisture inside the container from forming on the sensor lenses. White cardboard display panels were positioned outside the Plexiglas to help defuse the light source evenly over the experimental units. A digital image of the entire experiment area (both healthy and inoculated grass) was captured and saved every 5 to 10 minutes during Pythium blight assays and every 15 to 30 minutes during brown patch assays until disease became clearly visible on the inoculated turfgrass. Both imaging sensors and the rack of treated bentgrass cones were fixed to prevent movement during experiments. Images from both cameras were saved to a laptop computer for later analysis. Turfgrass Culture Thirty plastic cones measuring 2.5 cm in diameter X 16 cm deep (Stuewe & Sons, Corvallis, OR, http://www.stuewe.com/ products/rayleach.shtml) were filled with sterile vermiculite media roughly 2.5 cm below the cone rim. Each cone was placed in a holding rack and sprinkled uniformly with approximately 0.055 g of bentgrass seed. Seedling turf was placed on a greenhouse mist bench that applied a mist of water for 15 s every 10 min. After 12 d, the established bentgrass was removed from the mist bench; foliage was trimmed level to the cone rim, and then inoculum
was applied. Only The turf was not trimmed during the BP3 experiment. There were no fertilizer or pesticide applications made to the bentgrass. Inoculum production and application P. aphanidermatum (isolate WF0291) and R. solani (isolate WF9910) were collected from diseased creeping bentgrass turf in Urbana, IL. Under sterile conditions, the pathogens were isolated and cultured in petri plates on 20% potato dextrose agar (PDA) media (BD Diagnostics, Sparks, MD) and stored at room temperature. To maintain pure isolates and pathogen virulence, petri plates were subcultured 2 -3 times weekly. Potato dextrose broth (PDB) was prepared to culture the pathogens in liquid media. To make PDB, one 1 L flask containing 200 g of fresh sliced potato and 500 mL of de-ionized water was autoclaved for 20 s. The potatoes were strained out and 20 g of dextrose was added to the remaining broth. Additional de-ionized water was added to the flask to make the total volume 1000 mL. This solution was autoclaved for 20 min. and then cooled to room temperature. Roughly 25 small agar squares (0.5 cm2) of pure isolate cultured on the 20% PDA media was transferred to the flasks of PDB using a clean, sterile spatula. Flasks were covered with tin foil and stored at room temperature. P. aphanidermatum was cultured in the PDB media for 7 d whereas R. solani was grown in PDB for 14 d. At this time the pathogen had grown into a mass of mycelium, which was strained from the PDB and blended for 30 s in 1000 ml of sterile de-ionized water. The blended solution of mycelium and water was filtered twice through cheesecloth and then equally divided. Half of the inoculum was autoclaved for 20 min. to kill the pathogen and serve as the control treatment.
Two flasks of PDB inoculum were used in the PB1 and BP4 experiments, while only one flask was used in PB2 and BP3. Dilution series attempts to quantify the number of colony forming units (CFU) in 1 ml of inoculum were unsuccessful for P. aphanidermatum inoculum. R. solani inoculum used in BP3 and BP4 averaged 9.08 X 102 CFU ml 1 and 5.05 X 103 CFU ml"1, respectively. All pathology techniques except the initial collection of diseased turfgrass and the actual inoculation of the bentgrass were conducted under a laminar flow hood to prevent fungal isolates from becoming contaminated by other organisms. The control treatment was applied by filling six 50 ml vials with the sterilized inoculum, and then sprayed each vial content as uniformly as possible to the bentgrass foliage in 15 cones. An equal amount of live inoculum was aspirated onto the foliage in the remaining 15 cones. All 30 cones were placed completely randomized in a holding rack, and immediately moved to the growth chamber environment for disease development. Growth chamber parameters The average temperature, RH, and photo synthetic active radiation (PAR) during Pythium blight assays was 34 °C, 93%, and 125 jxmol m"2 s"1, with standard deviations of 1.6, 10.5, and 31.4, respectively. The average temperature, RH, and PAR during brown patch assays was 29 °C, 90%, and 115 ^mol m"2 s"1, with standard deviations of 3.2, 7.2, and 31.3, respectively. Each day during the Brown patch assays a light mist of water was applied uniformly over the foliage of the bentgrass replications and ice was placed in the growth chamber to help lower the temperature.
Disease verification Each pathogen was re-isolated from infected grass tissues by placing the tissues in a 10% Clorox solution for 30 s to kill any organisms on the outside leaf surfaces. The surface sterilized tissues were placed in petri plates containing 20 % PDA media and rifampicin (RIF), an antibiotic product to help prevent bacterial contamination of the isolate. Through hyphal tip transfers a pure isolate was obtained and further identification of the fungus was made through petri plate observations. On water agar, P. aphanidermatum produced a fast growing white, coenocytic mycelium on water agar. On 20 % PDA, R. solarti produced buff colored mycelium that formed small bulbils that were brown to black in color and irregularly shaped. Imaging sensors Bitmap (.bmp) formatted digital images with a resolution of 1392(H) x 1040(V) pixels were acquired with a Redlake 3100 camera filtered to measure spectral energy in three separate 100 nm bands centered at 550, 650, and 800 nm, which represent the green, red, and NIR spectrum regions, respectively. Additional details on the imaging multi-spectral system can be found at (Schmidt, 2005). JPEG (joint photographic experts group, .jpg) formatted images with a resolution of 3078(H) X 2048(V) were acquired with a Table 1. Reflectance indices used to detect fungal diseases on A-4 creeping bentgrass with the Duncan Index
Index calculation as wavelengths measured NDVI (normalized difference vegetation index) R^o - R sso / Róso + R-850 NDVIg R550 - R 850 / R-550 + R-850 NIR/Red R850/RÓ50 NIR/Green R-85O/R-550 Green/Red R550/R650
Canon EOS 10D full color digital camera that quantified red, green, and blue light emitted from each pixel in an image. Image processing and data analysis Image Pro Plus (Media Cybernetics, Inc., Silver Spring, MD, http://www.mediacy.com/mediahm.htm) software ver. 3.0 was used to process the digital images. Each pixel of an image has an associated grey level value (0-255) that is unique to each channel that a sensor measured. Bentgrass foliage from each replication was selected as an independent region of interest, by which the software measured all pixel grey levels and provided one value, called the average grey level (AGL) for that region of interest, for each channel in the sensor. Measurements were taken from the same region of interest for each experimental unit over time. Organization, calibration, and analysis of the image data were performed using Microsoft Excel software 2003 version (Microsoft Corporation, Redmond, WA, www.microsoft.com). The following procedures were used to convert AGL data from the Redlake 3100 images into percent reflectance. One black and one white tile were measured with a LICOR 1800 spectroradiometer (LI-COR Biosciences, Lincoln, NE, www.licor.com) Table 2. Average grey level (AGL) indices used to detect fungal diseases on A-4 creeping bentgrass with the Canon EOS 1 OD standard digital imagine system. Index Index calculation f Green/Red green channel value / red channel value. Green/Blue green channel value / blue channel value. Red/Blue red channel value / blue channel value. | T h e exact wavelengths this sensor measured was not determined.
Table 3. Time of first detection of disease or signs on A-4 creeping bentgrass by human visual evaluation or a multispectral imaging system. Human visual detection f Duncan Tech 3100 multi-spectral imaging system* Experiment Disease assay Naked eye NDVI § (Vo f NDVIg # (%) NIR/R n (%) G/R1* (%) HAI PB1 Pythium blight 5.2 (87) 6.6 (111) 4.8 (82) 4.4 (75) PB2 Pythium blight 13.8 13.8 (100) no detection 13.8 (100) 12.2 (88) BP3 Brown patch 67.0 47.0 (70) no detection 47.0 (70) 42.0 (63) BP4 Brown patch 49^ 47.0 (95) 56.0 (113) 47.0 (95) 54.5 (110) |First visible detection on three replications of bentgrass exhibiting signs or symptoms of disease. ^Detection times based on consistent significant differences between treatments at an alpha level ^ 0.05. §Normalized difference vegetation index (NDVI) calculated as (near infrared (NIR) band - red band) / (NIR band + red band). ^Percent of time to detect initial symptoms or signs of disease compared to human detection. #Green NDVI calculated as (NIR band - green band) / (NIR band + green band). ttReflectance index calculated as the ratio of % reflectance at NIR band / red band. IJReflectance index calculated as the ratio of % reflectance at green band / red band.
and an integrating sphere with halogen white light source and BaSC>4 calibration reference to determine their mean percent reflectance over the corresponding wavelengths measured by the Redlake 3100 camera. Mean percent reflectance of the black tile was approximately 89, 90, and 88, for the NIR, red, and green channels respectively. Mean percent reflectance of the white tile was approximately 9, 6, and 5, for the NIR, red, and green channels respectively. The tiles of known reflectance were placed in each image. Percent reflectance of the bentgrass foliage was derived from a regression equation of the AGL measured from the black and white tiles against the known percent reflectance of the tiles. Tables 1 and 2 describe reflectance measurements used in this experiment. Data analysis Since there were only two treatments in each experiment, inoculated or control, the student T- test was used to determine differences between image data. In Microsoft Excel, the student T- test function was performed on the data processed through image analysis to determine the level of statistical significance between treatments.
Visual assessments were made by opening each of the images collected with the standard digital camera in Photoshop 7.0 software (Adobe Systems Inc., San Jose, CA, www.adobe.com/aboutadobe/ contact.html) and using the zoom in function (to 100%) to look closely at each of the bentgrass foliar canopies. The initial appearance of disease was deemed the time at which 10% or 3 cones showed disease symptoms or signs. Statistical analysis of the reflectance data determined when the image sensors initially detected changes in the bentgrass treatments. RESULTS Pythium blight was first detected by human visual evaluation (HVE) 5.9 h after inoculation (HAI) in PB1, and 13.8 HAI in PB2 (Table 3). In BP3 and BP4, brown patch signs appeared 67.0 and 49.5 HAI, respectively (Table 4). In PB1, PB2, and BP3, disease was initially detected by HVE through symptom expression. In BP4, detection by HVE was based on the observation of aerial mycelium development, with no accompanying symptoms.
Table 4. Time of first detection of disease or signs on A-4 creeping bentgrass by human visual evaluation or a digital imaging system. Human visual detection Canon EOS 10D digital camera* Experiment Disease assay Naked eye G/R§ G/B# (%) R/B f t (%) HAI PB1 Pythium blight 5.9 5.2 (87) 5.4 (92) 5.7 (96) PB2 Pythium blight 13.8 11.5 (83) 14.8 (107) 15.7 (113) BP3 Brown patch 67.0 57.0 (85) 63.0 (94) no detection BP4 Brown patch 49^ 44.5 (90) 56.5 (114) no detection fFirst visible detection on three replications of bentgrass exhibiting signs or symptoms of disease. JDetection times based on consistent significant differences between treatments at an alpha level ^ 0.05. § Average grey level (AGL) index calculated as the ratio of the AGL at green band / red band. ^Percent of time to detect initial symptoms or signs of disease compared to human detection.
Detection of Pythium blight with the multispectral imaging sensor Disease was detected prior to HVE using the multispectral camera in PB1 and PB2 (Table 3). Analysis of the G/R reflectance index detected Pythium blight 25% and 12% sooner than HVE, in PB1 and PB2, respectively (Table 3). In PB1, NDVI and NIR/R signaled Pythium blight 13% and 18% sooner than HVE, respectively (Table 3). In PB2, these reflectance indices detected Pythium blight simultaneously with HVEs (Table 3). NDVIg failed to detect Pythium blight prior to human evaluations in PB1 and PB2 (Table 3). Detection of brown patch with the multispectral imaging sensor In BP3 and BP4, brown patch was detected prior to HVE using the multispectral camera (Table 3). NDVI and NIR/R detected brown patch 30% before HVE in BP3, and 5% earlier than HVE in BP4 (Table 3). In BP3, the G/R reflectance index detected brown patch 37% before HVE detected disease symptoms (Table 3). In BP4, the G/R index failed to detect aerial mycelium development prior to or simultaneously with HVE (Table 3). NDVIg failed to detect brown patch prior to HVE in BP3 and BP4 (Table 3).
Detection of Pythium blight with the standard digital camera Pythium blight was detected prior to HVE using the standard digital camera in PB1 and PB2 (Table 4). Analysis of G/R AGL index detected Pythium blight 13% and 17% earlier than HVE, in PB1 and PB2, respectively (Table 4). In PB1, AGL indices G/B and R/B signaled Pythium blight occurrence 8% and 4% before HVE detection, respectively (Table 4). These indices failed to detect Pythium blight prior to or simultaneously with HVEs in PB2 (Table 4). Detection of brown patch with the standard digital camera Using the standard digital camera, brown patch was detected prior to HVE with the G/R AGL index 15% and 10% before HVE, in BP3 and BP4, respectively (Table 4). The G/B AGL index detected brown patch 6% earlier than HVE in BP3, but failed to detect brown patch before mycelium was visible in BP4 (Table 4). The R/B index failed to detect brown patch prior to HVEs in BP3 and BP4 (Table 4). DISCUSSION Although the exact reasons are not clear, the rate of inoculum application and
subtle changes in temperature, light or humidity in the environment might explain differences in HAI before development of Pythium blight in PB1 and PB2, and brown patch in BP3 and BP4. The hypothesis that early detection of Pythium blight and brown patch in creeping bentgrass by HVE is equally effective as direct sensing must be rejected. Direct sensing of the creeping bentgrass foliar canopy with the multispectral camera and the standard digital camera provided a means to detect the onset of Pythium blight and brown patch in creeping bentgrass, before significant changes in the appearance of turfgrass health were detected. These results supported those of Carter, et al., (1996), and Trenholm, et al. (2000), who were able to detect plant stresses prior to HVE detection by monitoring canopy reflectance. With the multispectral camera, a G/R reflectance index was optimal for early detection of Pythium blight, while the NIR/R and NDVI indices were optimal for early detection of brown patch. Although the G/R index effectively detected brown patch prior to human evaluation in BP3, pre-visual detection was not achieved in BP4. In BP4, brown patch developed as signs and not symptoms, which might have affected the G/R index. The exact reason why the G/R ratio failed to detect brown patch before human evaluation in BP4 is not known. As in previous studies, the NDVI and NIR/R indices were useful at detecting changes in health prior to visual observations of stress (Carter, et al., 1996, Trenholm, et al., 2000). Although previous studies do not report the value of a G/R index, they do agree that wavelengths in the visible region detect the early stages of stress (Carter, 1993, Carter, et al., 1996, Carter
and Miller, 1994, Lorenzen and Jensen, 1989, Trenholm, et al., 2000). NDVIg was the least effective index for early detection in all 4 experiments. This in not in agreement with previous research conducted on bermudagrass and creeping bentgrass turf plots, where NDVIg (calculated as R780±6 nm ~ R550±6nm / R780±6 nm + R550±6nm), was a better index than NDVI (calculated as R780±6 nm ~ R670±6nm / R780±6 nm + R670±6nm), for predicting tissue nitrogen and leaf chlorophyll content (Bell et al., 2004). i
A high resolution standard digital camera is a useful tool for nondestructive detection of pythium blight and brown patch in creeping bentgrass turf prior to human evaluation. With this sensor, a G/R index provided the earliest and most consistent detection of disease in all the experiments. Unlike the G/R index data collected with the Duncan Tech. sensor, the G/R index measured with the standard digital camera did detect brown patch before human evaluation in BP4. It's likely that the wavelengths the two sensors were designed to measure differ, resulting in the differences in detection. The G/B and R/B indices with the standard digital camera were neither consistent nor reliable for detection of disease prior to human evaluation. Previous research looks at indices generated from reflectance measured from the visible and NIR regions of the spectrum, yet do not report the usefulness of indices generated from reflectance data measured in different regions of the visible spectrum for early disease detection. Early detection of fungal infections in creeping bentgrass turf through image analysis may be possible. Although turf
managers eagerly await new tools to improve disease detection, additional research is needed to develop this technology into a reliable, practical and cost effective tool for predicting future disease problems over large areas of turf, such as golf courses. As imaging systems continue to becoming simpler to operate and cheaper to purchase, we could see an increasing number of robot-operated sensors that monitor turfgrass health become an asset in 21st century precision turfgrass management. REFERENCES Bell, G. E., B. M. Howell, G. V. Johnson, W. R. Raun, J. B. Solie, and M. L. Stone. 2004. Optical Sensing of Turfgrass Chlorophyll Content and Tissue Nitrogen. Hort Science 39(5): 1130-1132. Blazquez, C. H. and G. Edwards, Jr. 1986. Spectral Reflectance of Healthy and Diseased Watermelon Leaves. Annals of Applied Biology 108:243-249. Burpee, L. L. and L. Goulty. 1986. Evaluation of Two Dollar Spot Forecasting Systems for Creeping Bentgrass. Canadian Journal of Plant Science 66:345-351. Carter, G. A. 1993. Responses of Leaf Spectral Reflectance to Plant Stress. American Journal of Botany 80(3):239-243. Carter,
G. A. 1994. Ratios of Leaf Reflectance in Narrow Wavebands as Indicators of Plant Stress. International Journal Remote Sensing 15(3):697-703.
Carter, G. A. and R. L. Miller. 1994. Early Detection of Plant Stress by Digital Imaging with Narrow StressSensitive Wavebands. Remote Sensing of Environment 50:295-302. Carter, G. A., W. G. Cibula, and R. L. Miller. 1996a. Narrow-band Reflectance Imagery Compared with Thermal Imagery for Early Detection of Plant Stress. Journal of Plant Physiology 148:515-522. Carter, G. A., T. R. Dell, and W. G. Cibula. 1996b. Spectral Reflectance Characteristics and Digital Imagery of a Pine Needle Blight in the Southeastern United States. Forest Resources 26:402-407. Couch, H. B. 1995. Patch Diseases. Page 52 in: Diseases of Turfgrasses. Krieger Publishing Co., Malabar, FL. Fermanian, T. W., M. C. Shurtleff, R. Randell, H. T. Wilkinson, and P. L. Nixon. 2003. Controlling Turfgrass Pests. Third Ed. Prentice Hall: Upper Saddle, NJ. Fletcher, R. S., M. Skaria, D. E. Escobar, J. H. Everitt. 2001. Field Spectra and Airborne Digital Imagery for Detecting Pytophthora Foot Rot Infections in Citrus Trees. Horticulture Science 36(l):94-97. Green, D. E., L. L. Burpee, and K. L. Stevenson. 1998. Canopy Reflectance as a Measure of Disease in Tall Fescue. Crop Science 38:1603-1613.
Hall, T. J., P. O. Larsen, and A. F. Schmitthenner. 1980. Survival of Pythium aphanidermatum in Golf Course Turfs. Plant Disease 64(12):1100-1102. Karcher, D.E., and M. D. Richardson. 2003. Quantifying Turfgrass Color Using Digital Image Analysis. Crop Science 43:943-951.
Nutter, F. W., Jr., M. L. Gleason, J. H. Jenco, andN. C. Christians. 1993. Assessing the Accuracy, Intra-rater Repeatability, and Inter-rater Reliability of Disease Assessment Systems. Phytopathology 83:806-812. Penuelas, J., and L. Filella. 1998. Visible and Near-infrared Reflectance Techniques for Diagnosing Plant Physiological Status. Elsevier Science Ltd. 3(4): 151-155.
Knipling, E., B. 1970. Physical and Physiological Basis for the Reflectance of Visible and NearInfrared Radiation from Vegetation. Remote Sensing of Environment 1:155-159.
Richardson, M. D., D. E. Karcher, and L. C. Purcell. 2001. Quantifying Turfgrass Cover Using Digital Image Analysis. Crop Science 41:1884-1888.
Lorenzen, B. and A. Jensen. 1989. Changes in Leaf Spectral Properties Induced in Barley by Cereal Powdery Mildew. Remote Sensing of Environment 27:201-209.
Raikes, C. and L. L. Burpee. 1998. Use of Multispectral Radiometry for Assessment of Rhizoctonia Blight in Creeping Bentgrass. Phytopathology 88:446-449.
Metternicht, G. 2003. Vegetation Indices Derived from High-resolution Airborne Videography for Precision Crop Management. International Journal of Remote Sensing 24(14):2855-2877.
Schumann, G. L., B. B. Clarke, L. V. Rowley, and L. L. Burpee. 1994. Use of Environmental Parameters and Immuno-Assays to Predict Rhizoctonia Blight and Schedule Fungicide Applications on Creeping Bentgrass. Crop Protection 13:211-218.
Nutter, F., W. 1980. Forecasting Pythium. Golf Course Management. June 1980:18-25. Nutter, F. W., Jr., H. Cole, Jr, and R. D. Schein. 1983. Disease Forecasting System for Warm Weather Pythium Blight of Turfgrass. Plant Disease 67(10):! 126-1128.
Shane, W. W. 1991. Prospects for Early Detection of Pythium Blight Epidemics on Turfgrass by Antibody-Aided Monitoring. Plant Disease 75(9):921-925. Schmidt, M. A. 2005. Utilization and Configuration of a Multi-Spectral Image Sensor for Turfgrass Data Collection. Unpublished doctoral dissertation, University of Illinois, Champaign - Urbana, IL.
Trenholm, L. E., L. G. Schlossberg, and W. Parks. 2000. An Evaluation of MultiSpectral Responses on Selected Turfgrass Species. International Journal Remote Sensing 21(4):709-721.
Wilkinson, H. T. 1998. Interactive Turf: Cool-season Turf Diseases and Their Management. [CD ROM].