Crop Science 42:648-651 (2002)
© 2002 Crop Science Society of America
NOTES
Turf area mapping using vehicle-mounted optical sensors
G. E. Bell*,a,
D. L. Martina,
M. L. Stoneb,
J. B. Solieb and
G. V. Johnsonc
a Dep. of Horticulture and Landscape Architecture, Oklahoma State Univ., Stillwater, OK 74078
b Dep. of Biosystems and Agricultural Engineering, Oklahoma State Univ., Stillwater, OK 74078
c Dep. of Plant and Soil Sciences, Oklahoma State Univ., Stillwater, OK 74078
* Corresponding author (bgregor{at}okstate.edu)
 |
ABSTRACT
|
|---|
Environmental concerns require turf managers to minimize the amount of nutrients and pesticides used for turf maintenance. Vehicle-mounted optical sensing (VMOS) measures spectral reflectance from a turf canopy that can be converted to normalized difference vegetative indices (NDVI). Normalized difference vegetative index maps may provide opportunities for early detection of potential turf problems, economic savings for fertilizers and pesticides, and improvements in turf appearance and functional quality. The objective of this study was to evaluate the use of VMOS for mapping large turf areas. Sensor equipment was used to map a creeping bentgrass [Agrostis palustris Huds. = A. stolonifera var. palustris (Huds.) Farw.] putting green weekly for 8 wk. Normalized difference vegetative index maps constructed from VMOS measurements compared closely with turf response to N fertility and turf cover during grow-in. Sensor results were highly correlated (r2 = 0.98) with replicated plots fertilized with six different N rates. VMOS maps clearly indicated areas of poor nutrition, sparse turf cover, and some irrigation patterns.
Abbreviations: NDVI, normalized difference vegetative index • R, red (671 nm wavelength) • NIR, near infrared (780 nm wavelength) • VMOS, vehicle-mounted optical sensors/sensing
|
INTRODUCTION
|
|---|
THE USE OF FERTILIZERS and pesticides for maintaining turf and other agricultural commodities is a point of increasing economic and environmental concern. To enhance environmental stewardship, turf managers require methods for reducing fertilizer and pesticide use while maintaining adequate turf quality. Vehicle-mounted optical sensing is one method with potential for reducing the use of fertilizers and pesticides used in turf management. This procedure incorporates mobile instruments to measure specific solar spectra reflected from a turf canopy. These measurements can be used to construct a map of turf status. An effective VMOS system may provide opportunities for early detection of potential turf problems, realization of economic savings for fertilizers and pesticides, and improvement in turf appearance and functional quality.
Optical sensing has been used to evaluate plant biomass (Walburg et al., 1982; Kleman and Fagerlund, 1987; Wanjura and Hatfield, 1987), plant N content (Blackmer et al., 1994; Stone et al., 1996), turf chlorophyll content (Howell, 1999), and turf injury and quality (Trenholm et al., 1999; Bell et al., 2000). These reports suggest that a mobile optical sensing system may be used to provide maps of turf characteristics over large areas. These maps could be used to prompt visual inspections, enable variable rate nutrient applications, or signal a need for pest management. Maps could be archived to provide historical comparisons with current conditions or to justify chemical applications.
The purpose of this study was to evaluate the use of VMOS for mapping turf status over large areas.
|
Materials and Methods
|
|---|
Sensor measurements were collected weekly for 2 mo on a putting green at the Oklahoma State University Turfgrass Research Center, Stillwater, OK in 1998. The putting green was established to ‘Penncross’ creeping bentgrass on a sand and peat mixture (85:15 w/w) in 1991. A portion of the green was reestablished to ‘SR1020’ creeping bentgrass in 1995 and another portion converted to ‘A4’ creeping bentgrass in 1998. The putting green covered 978 m2 and was categorized into five different sections determined by cultivar and research use (Fig. 1)
. Sections 1, 2, and 3 were SR1020 creeping bentgrass. Section 4 was being established to A4 creeping bentgrass, and Section 5 was a mixture of Penncross creeping bentgrass and indigenous annual bluegrass (Poa annua L.). Section 1 (205 m2) on the far west side of the putting green received no N fertilizer for several months before or during the study. Section 2 (140 m2) contained a N-fertility study and Section 3 (254 m2) was maintained as a normal putting surface and received 195 kg N ha-1 yr-1. The fertility study in Section 2 consisted of six fertility treatments: 6, 18, 30, 42, 54, and 67 kg N ha-1, replicated four times in a completely random design. Ammonium nitrate (34–0–0) in soluble form was applied to 1.5 x 3.1 m plots in this section on 14 May and again on 8 June 1998. Plant material in Section 4, an area 271 m2, was treated with glyphosate and reestablished with A4 creeping bentgrass during the study. The green, except for Section 4, was mowed at a height of 4 mm, and pesticides were applied on a curative basis. Section 5 measuring 108 m2 on the far east portion of the green was the only area of the original cultivar remaining and consisted of a mixture of Penncross and indigenous annual bluegrass.
View larger version (67K):
[in this window]
[in a new window]
|
Fig. 1. Normalized difference vegetative index (NDVI) maps of a creeping bentgrass putting green made weekly for an 8-wk period. A plot plan for variable N applications in Section 2 and a grayscale chart are included. Darker areas indicate high NDVI.
|
|
A sensor unit was designed and fabricated by the Biosystems and Agricultural Engineering Department at Oklahoma State University, Stillwater, OK. An optical sensor used to collect spectral data was mounted on the end of a tubular aluminum rod that extended 0.75 m in front of an aluminum pushcart. The sensor unit contained four photodiode detectors housed in an aluminum, rectangular box, attached to this rod and mounted 1.0 m above the ground. Two photodiode detectors, combined with interference filters (Andover Corp., Salem, NH), measured red (R) irradiance at 671 nm wavelengths, and two additional detectors and filters measured near infrared (NIR) at 780 nm. One R and one NIR detector faced upward to collect solar spectra, and one R and one NIR detector faced downward to collect reflected spectra. Down-looking detectors measured average reflectance in a 25- by 53-cm area of turf through collimation in the sensor housing. The long dimension of sensor view was oriented perpendicular to the direction of travel. Uplooking detectors captured incoming direct and diffuse (incident) radiation through Teflon diffusers that provided cosine correction. Red and NIR reflectance was calculated as a ratio of reflected radiation to incident radiation to account for changes in atmospheric conditions such as cloud cover (Lillesand and Kiefer, 1994).
The VMOS instrument was manually pushed across the turf toward the sun to avoid sensing reflectance in shadows originating from the pushcart and to minimize the effects of bidirectional reflectance. A shaft encoder, mounted on the front axle of the cart, monitored position and signaled a processor to collect spectral data according to user-specified spatial intervals. For this project, the spatial interval was 25 cm, a distance equal to sensor resolution in the direction of travel. This programming enabled spectral collection of reflectance over the entire research green without collection overlap or gaps between collections. Data collected by the processor was recorded by an onboard, portable computer through a serial interface using Hyperlink communications software. Power for the sensors and processors was supplied by a 12-V automobile battery. Spectral reflectance was collected in a manner similar to taking a photograph. Each data collection was called a frame. Each frame consisted of a recorded value based on electrical potential (a reading of 1000 
76.3 mV). This value represented the average R and NIR reflected from a turf area 25 x 53 cm. A collection of frames that bordered each other and completely covered the turfgrass area were collected at a medium walking pace (3.2 km h-1).
Once spectral data were collected, the reflectance values for each frame were used to determine NDVI. Normalized difference vegetative index is calculated by (NIR - R)/(NIR + R), and is a normalized measure of plant reflectance (Perry and Lautenschlager, 1984; Duncan et al., 1993) that may be used to evaluate plant N content (Stone et al., 1996), chlorophyll content and color (Howell, 1999), and turf injury (Bell et al., 2000; Trenholm et al., 1999). Normalized difference vegetative indices were adjusted for atmospheric conditions using the following equation:
 |
The adjusted NDVI values were matched with geographic location predetermined by area measurements and processed to produce maps using Surfer software (Golden Software, Golden, CO). Maps were generated weekly and compared with variable N applications and turf management practices. This procedure was used to determine VMOS accuracy for monitoring turf status. Regression and correlation were used to determine the relationship between VMOS and N fertility treatments in Section 2 (Fig. 1). These relationships were analyzed using Table Curve Software (SPSS, Chicago, IL).
|
Results and Discussion
|
|---|
Sensor maps, made over a 2-mo period, suggested that changes in turf status could be observed using optical sensing and associated with causes. The system was geographically accurate within one pixel (25 x 53 cm). Location discrepancies were caused by operator error. Perfect geographic accuracy would require that the operator begin sensing on exactly the same point and continue along exactly the same route each time data were collected. Guide flags placed at the east and west ends of the research green indicated starting points and vectors that improved geographic sensitivity. A geographic positioning system and software capable of recording position simultaneously with spectral results were available. This system, however, was only geographically accurate to about ±1 m.
Section 1 was not fertilized before or during the study, resulting in poor turf color and low NDVI values throughout the study (Fig. 1). Turf color in Section 1 was not consistent, evidenced by mottled light and dark patches on VMOS maps. This mottled appearance was not readily apparent visually, but close inspection and linear measurements to specific areas revealed that light and dark green patches were present and were accurately represented on VMOS maps. The irrigation patterns clearly identified by VMOS in Section 4 on 3, 16, and 23 June and on 1 and 7 July could not be discerned by the human eye on July 1 or 7, but are apparent on VMOS maps (Fig. 1). These patterns could be seen, however, prior to July 1 before this seeded area reached nearly full cover. These irrigation patterns were caused by pop-up sprinkler heads sticking in the upright position at the end of the irrigation period. The patterns were not observed on June 9 because the area was covered with white seed cloth. These inspections of Sections 1 and 4 suggested that optical sensing was more sensitive than visual evaluation.
Sensor evaluation accurately assessed the N status of turf plants in Section 2 (Fig. 1). Fertilizer treatments were applied to the area on 14 May and 8 June. Sensor maps recorded a decline in turf color on 26 May through 9 June, and again on 1 and 7 July, as turf response to the quick-release N source declined. A significant (P = 0.05) curvilinear relationship occurred between NDVI measurements and turf response to the six fertilizer rates applied in Section 2 when results were averaged over all dates (NDVI = 0.66 + 0.0086 x fertility rate0.5). The strength of the relationship was r2 = 0.89 on a plot by NDVI basis and r2 = 0.98 on a treatment by NDVI basis. This particular relationship (y = a + bx0.05) produced a very strong correlation between NDVI and turf response to fertility rates on all dates of evaluation except 7 July. The coefficients of determination for treatment x NDVI comparisons were r2 = 0.99 on 19 May, 0.99 on 26 May, 0.95 on 3 June, 0.98 on 9 June, 0.95 on 16 June, 0.89 on 23 June, 0.89 on 1 July, and 0.51 on 7 July. Sensor correlation to N treatment was normally strong, but declined in late June and July as turf response to the quick-release N source declined. These results indicated a close relationship between turf response to N fertilization and VMOS mapping. For that reason, VMOS was considered an excellent signaling procedure for variable rate N applications.
Sections 3 and 5 were fertilized (6 kg N ha-1) with urea (46-0-0) on 20 May, 2 June, and 15 June. Increases in color were recorded by VMOS on 26 May, 9 June, and 23 June due to these applications. Sulfate of potash was applied to Sections 3 and 5 on 20 May (6 kg K ha-1). Any response to this K application was included in the response to the N application of 20 May.
The VMOS maps recorded for Section 4 indicated that turf color was not the only variable that affected optical sensing. Turf cover and density also influenced results. Section 4 was treated with glyphosate in early April 1998 to kill a bentgrass cultivar study. By 19 May, volunteer plants of many species had germinated and were growing in the area. Between 19 and 26 May, these plants increased in size and density. The increase in size and density was recorded as a darker color in this area on the VMOS map of 26 May. The area was treated with glyphosate again on 27 May and plant color declined slightly due to herbicide treatment by 3 June (Fig. 1). The sod was removed from Section 4 on 4 June. Following sod removal, the sand base was cultivated and smoothed, and A4 creeping bentgrass was seeded. After seeding, seed cloth was placed on the area. The consistent white color of the seed cloth was indicated in the VMOS map of 9 June. The cloth was removed on 12 June after the seed had germinated and seedling shoots extended through the cloth weave. Vehicle-mounted optical sensing maps from 16 June through 7 July recorded the increase in creeping bentgrass development in Section 4 during that period.
During this study, specific R and NIR wavelengths were recorded using both incident and reflected radiation detectors simultaneously. These measurements were compared and converted to adjusted NDVI. This comparison was important for maintaining consistency of VMOS results despite differences in sky irradiance. Using these data, VMOS maps that accurately assessed turf status were constructed. Using known factors, a rough color chart was constructed for user assessment of turf characteristics from VMOS maps (Fig. 1). Maps such as these may provide early warning of plant decline. These maps may also be used to identify areas in need of N fertilization and the amount of fertilizer required. Irregularities in irrigation distribution may also be detected. If an optical sensing system and software can be economically produced, reasonably priced, and mounted effectively on normal maintenance equipment, a turf practitioner could save enough money in fertilizers and pesticides to pay for the equipment. This approach would be useful for reducing the amount of fertilizers and pesticides needed to adequately manage large turf areas. The use of optical sensing to determine fertilizer rate before or during application could increase turf uniformity and possibly, turf health. Sensor maps of large turf areas could be used to signal turf decline and provide an early warning system for turf managers.
|
NOTES
|
|---|
Research was conducted under Oklahoma Agricultural Experiment Station Project OKLO 2392.
Received for publication December 28, 2000.
|
REFERENCES
|
|---|
- Bell, G.E., D.L. Martin, R.M. Kuzmic, M.L. Stone, and J.B. Solie. 2000. Herbicide tolerance of two cold-resistant bermudagrass cultivars determined by visual assessmment and vehicle-mounted optical sensing. Weed Technol. 14:635–641.
- Blackmer,T.M., J.S. Schepers, and G.E. Varvel. 1994. Light reflectance compared with other nitrogen stress measurements in corn leaves. Agron. J. 86:934–938.[Abstract/Free Full Text]
- Duncan, J., D. Stow, J. Franklin, and A. Hope. 1993. Assessing the relationship between spectral vegetation indices and shrub cover in the Jornada Basin, New Mexico. Int. J. Rem. Sens. 14:3395–3416.
- Howell, B.M. 1999. The effectiveness of sensor based technology to determine nitrogen deficiencies in turfgrasses. M.S. Thesis. Oklahoma State Univ., Stillwater, OK.
- Kleman, J., and E. Fagerlund. 1987. Influence of different nitrogen and irrigation treatments on the spectral reflectance of barley. Remote Sens. Environ. 21:1–14.
- Lillesand, T.M., and R.W. Kiefer. 1994. Remote sensing and image interpretation. 3rd ed. John Wiley & Sons, New York.
- Perry, C.R., Jr., and L.F. Lautenschlager. 1984. Functional equivalence of spectral vegetation indices. Remote Sens. Environ. 14:169–182.
- Stone, M.L., J.B. Solie, W.R. Raun, R.W. Whitney, S.L. Taylor, and J.D. Ringer. 1996. Use of spectral radiance for correcting in-season fertilizer nitrogen deficiencies in winter wheat. Trans. ASAE 39: 1623–1631.
- Trenholm, L.E., R.N. Carrow, and R.R. Duncan. 1999. Relationship of mutispectral radiometry data to qualitative data in turfgrass research. Crop Sci. 39:763–769.[Abstract/Free Full Text]
- Walburg, G., M.E. Bauer, C.S.T. Daughtry, and T.L. Housley. 1982. Effects of nitrogen nutrition on the growth, yield, and reflectance characteristics of corn canopies. Agron. J. 74:677–683.[Abstract/Free Full Text]
- Wanjura, D.F., and J.L. Hatfield. 1987. Sensitivity of spectral vegetative indices to crop biomass. Trans. ASAE 30:810–816.
TOP
ABSTRACT
INTRODUCTION
