Introduction

Observations made at the grass roots level form the basis for the majority of crop management decisions. Decisions based on such observations can be compromised by a lack of accurate data concerning spatial variability within any whole field. A brief field-by-field inspection in the form of a drive or walk, or spot-sampling at designated representative locations, provides the decision maker with an average picture of the status of a crop. Intrafield variability makes management difficult and often masks underlying environmental limitations to higher yield.

Information acquired from the air has the potential to provide a whole-paddock view of a crop with spatial resolution necessary to provide an indication of within-paddock variation in crop status (weeds, nutrient deficiency, disease and insect infestation) which may be overlooked or under-estimated from ground level observations at selected locations. Additional use of detectors which operate outside the range of human perception, including near and thermal infrared, often enhances the quality of this information, and the use of computer-based image processing based

on ground-truth data can establish signatures to indicate crop health and development. Measuring reflectance from crops cannot determine the cause of crop stress but is a useful indicator of stress and can provide information for early prediction of yield (Stutte et al., 1988).

The use of satellite imagery and, more recently, airborne photography and videography for assessing the status of crops and pastures and formulating responses to detected variability, is becoming widespread and the potential savings in both environmental and monetary terms are not trivial. For example, efficient targeting of pasture fertiliser by assigning priorities for treatments of particular parts of a property based on mapped responsiveness of pastures can provide annual savings for a typical 800 ha wool producing enterprise in the high rainfall zone of Australia of around $2000 pa (ABARE, 1993). The use of satellite imagery has been used to estimate DSE over a twenty year pasture cycle in selected holdings, and the maps used to maximise carrying capacity via a regime of selective pasture improvement (Vickery and Furnival, 1992). Results showed stocking rates on improved or native pastures could go from 9.4 DSE/ha to 10.0 DSE/ha if better targeting of fertiliser were used (approximately $3.25/ha per year extra return). It is most unlikely that these results still under-estimate the true worth of this technology. Likewise, effective control of weeds, diseases and pests in crops requires efficient application regimes. For example, the current efficiency of spraying weeds, that is the percentage of herbicide actually reaching the target, has been estimated at 2% while spot-spraying can be up to 60% efficient (Combellack, 1979). Significant savings can be achieved through a means of selectively targeting weed patches in crops for treatment. Aerial imagery is one means of obtaining spatial data necessary for effective targeting of potential trouble spots in crops.

The State of Play

A large portion of the development to date of remote sensing for agricultural land management has concentrated on utilising satellite imagery for large-scale crop and pasture assessment. Most of this initial work has been completed in the United States, although some has been completed in Australia. Most experimental work has concentrated on using ground-based radiometry for characterising the spectral signature of crop/pasture in the search for indicators of crop type and development stage (Kanemasu, 1974; Van der Rijt et al., 1992), plant stress (Stutte et al., 1988; Daughtry et al., 1980), nutrient status and estimation of yield (Ashcroft et al., 1990). Satellite imagery has been used to assess pasture nutrition to the extent where data were able to show differences between fertilised and unfertilised areas (Reid et al., 1993).

Because satellites orbit the earth at altitudes of many tens of kilometres, they have the ability to image large tracts of land and find many uses in broad-scale resource assessment. The large coverage of satellite images, in turn, limits image resolution. Coupled with the high cost of accessing information, satellite images are not widely used as a resource for managing smaller tracts of land on a small holding or individual paddock basis. For example, the Landsat Thematic Mapper (TM) has a smallest resolution (pixel) of 30m x 30m, Landsat Multi Spectral Scanner (MSS) 80m x 60m, and the French SPOT 20m x 20m. All satellites are confined to fixed orbital parameters and are limited by cloud, although the SPOT satellite has an off-nadir viewing capability which increases revisit frequency and the chance of acquiring cloud-free imagery.

An airborne video system (ABVS) comprises one or more downward-directed video cameras carried in a light aircraft. Images can be obtained using conventional colour video cameras, or in pre-selected spectral bands according to the signature of the intended target by using appropriate filters on individual cameras or a filter wheel comprising many filters on a single camera. Images can be recorded on video tape or frozen, captured and written as computer files using appropriate frame-grabber hardware.

Airborne video permits high resolution imagery. A video camera equipped with 12 mm lenses and flown at an altitude of 1000m gives approximately a 0.75m x 0.75m pixel with a medium to large coverage in a single image (approximately 20ha at 1000m altitude). Image coverage can be increased, at the expense of resolution, by increasing the aircraft altitude. In comparison to the satellite data providing 250 pixels/10ha, as used by Reid et al (1993), ABVS provides approximately 177000 pixels/10 ha at 1000m altitude.

The ABVS is not limited by cloud cover providing cloud is evenly dispersed higher than the intended mission altitude and only relative changes in the spectral signature of the target are required. Like satellite imagery, ABVS images are obtained in digital form allowing computer-based image processing to provide classifications based on ground truth information.

Airborne video has been widely used since the mid-eighties for a variety of resource and agricultural land management applications, primarily in the United States. Like satellite imagery, detection and interpetation of particular phenomenon are based on observed spectral differences in the target. Examples of previous work include aerial detection of disease in potatoes (Manzer and Cooper, 1982), frost damage in citrus orchards (Escobar et al., 1983) nutrient stress in a range of horticultural crops (Stutte et al., 1990; Richardson et al., 1988), yield estimation of sorghum (Richardson et al., 1990) and weed detection in sorghum, cantaloupe and cotton (Richardson et al., 1985).

Specifically, in the work of Richardson et al. (1990) an ABVS was used to estimate sorghum yields at peak canopy development stage by calibrating obtained data to leaf area index, biomass and final harvest yield. When the signal was referenced to soil background the ABVS was found to be as accurate as ground radiometry. Other applications include improving the efficiency of irrigation scheduling in cotton crops based on within-field temperature variation of crop canopies using a thermal IR based video system (Button and Cull, 1990), delineation of soil types and slope based on spectral differences (Mausel et al., 1990) and rangeland monitoring (Grierson and Lewis, 1995).

Airborne video at Charles Sturt University - the way ahead

Our ABVS comprises 4 high resolution (CCD) video cameras, along with camera controlling and image acquisition hardware mounted in an equipment rack in place of the co-pilot seat in a Cessna 210 aircraft. The downward-looking camera acquires information in a preset spectral band determined by an interchangeable filter. General purpose narrow band interference filters at wavelengths of 450 nm (blue), 550 nm (green), 650 nm (red) and 770 nm (near infrared) are currently used.

Frame grabbing and image digitising operations are completed in-flight using a 486 computer to provide digitised, and spatially temporally registered, composite still-images (Louis et al., 1995). Camera control, image capture and recording is controlled through software (ABVCP) developed at CSU's Spatial Data Analysis Network (McKenzie et al., 1992). The ABVS is powered via 3 x 12 volt, 38 Ah, gel-cell batteries providing 3 hours continuous operation which can be extended to approximately 8 hours intermittent use.

We are currently investigating a number of potential applications which include assessment of water quality in inland rivers (Lamb and O'Donnell, 1995), irrigation scheduling in cotton crops without a thermal infrared capability (Louis, 1995) and detection and monitoring of weeds and nutrient status in selected crops. We are also building a mosaicing capability into the ABVS to allow high resolution imaging of large tracts of land. All potential applications are at the preliminary evaluation stage, but results are encouraging. By way of a suitable example, a brief treatment of the weed detection work is presented below.

High resolution airborne video imagery has been trialled for weed detection in sorghum, cantaloupe and cotton crops in the United States (Richardson et al., 1985). This feasibility test involved homogenous weed plots interspersed within uniform crop plots. As a sideline activity in one of our own recent airborne video missions, paddocks sown to wheat were imaged from 1500m in an attempt to evaluate the current filter configuration for general weed detection. Each field was at early tillering (7-week stage) and from each image in-field variability in vegetation was easily identifiable. A quick follow-up ground observation of selected fields identified regions under moderate to strong weed pressure from wild oats and wild radish. At this early stage of development the increased vigour of weeds provided the clearest indication of their presence against the background soil although some spectral discrimination was also evident.

Having identified the signature of a particular species from one location in a paddock, computer-based image processing was employed to classify the rest of that field. A true colour image of one such field is given in plate 1 and, based on ground-truth data, a classified image is given in plate 2. In this particular example spectral variations observed in this field were attributed to variations in crop density (poor germination, normal development and headlands), weeds (wild oats), standing stubble, and shadows. In another example, plate 3 shows a true colour image of an oat field at boot stage (12 weeks) taken at Charles Sturt University farm in 1994. At this stage of development the wild radish shows a clear spectral signature against the oats and allowed a classification as depicted in plate 4.

These examples demonstrate the potential of the ABVS for weed detection in two widely used crops. Future work is now required to optimise the filter configuration used in the ABVS for specific weed types and quantify the accuracy of detecting various weed types in particular crops under conditions of varying weed/crop densities and soil background.

Acknowledgements

The author would like to acknowledge the assistance provided by Ms T. Ellis in classifying and processing the images in plates 1-4; Mr E. Moloney for providing access to selected fields on his property "Tambea"; and Mr J. Medway for assistance throughout both missions during which the example weed data were obtained.

References

1. Ashcroft, P.M., Catt, J.A., Curran, P.J., Munden, J. and Webster, R. (1990). The relation between reflected radiation and yield on the Broadbank winter wheat experiment. Int.J.Remote Sensing 11 (10), 1821-1836.

2. Australian Bureau of Agriculture and Resource Economics (ABARE) Research Report (1993). An economic analysis of the use of remote sensing data for tree cover mapping and pasture management. W. Walker (ed), Australian Government Printing Service, Canberra.

3. Button, P.J. and Cull, P.O. (1990). An airborne video remote sensing system for operational management of irrigated crops. Proc. 5th Australasian Remote Sensing Conf., 680-687.

4. Combellack, J.H. (1979). A comparison of the drift potential of certain CDA and high volume spray equipment. M.Phil thesis, Brunel University, Uxbridge, UK.

5. Daughtry, C.S., Bauer, M.E., Crecelius, D.W. and Hixson, M.M. (1980). Effect of management practices on spring wheat canopies, Agronomy Journal 72, 1055-1060.

6. Escobar, D.E., Bowen, R.L., Gausman, H.W. and Cooper, G.R. (1983). Use of near infrared video recording system for the detection of freeze-damaged citrus trees. J. Rio Grande Valley Hort. Soc. 36, 61-66.

7. Grierson, I.T.* and Lewis, M.M.# (1995), personal communication, *Cooperative Centre for Soil and Land Management, Adelaide, SA, #Department of Environmental Science and Rangeland Management, University of Adelaide, SA.

8. Kanemasu, E.T. (1974). Seasonal canopy reflectance patterns of wheat, sorghum and soybean. Remote Sensing of Environment 3, 43-47.

9. Lamb, D.W. and O'Donnell, J. (1995). Preliminary evaluation of airborne-video for synoptic assessment of water quality in Australian inland rivers. Submitted to Aust.J.Soil and Water Cons.

10. Louis, J., Lamb, D.W., McKenzie, G., Chapman, G., Edirisinghe, A., McCloud, I. and Pratley, J. (1995). Operational use and calibration of airborne video for agricultural and environmental land management applications. Proc. 15th Biennial Workshop on Colour Photography and Air Videography, Terrahoute, Indiana, USA.

11. Louis, J. (1995), personal communication, Spatial Data Analysis Network, Charles Sturt University, NSW.

12. Manzer, F.E. and Cooper, G.R. (1982). Use of portable video taping for aerial infrared imaging of potato disease. Plant Disease 66, 665-667.

13. Mausel, P.W., Kalaluka, G.Y., Lee, J.K., Mudderman, D.E., Escobar, D.E. and Davis, M.R. (1990). Multispectral video analysis of micro-relief dominated poorly drained soils in western Indiana. J.Imaging Technology 16, 113-119.

14. McKenzie, G., Dare-Edwards, A., Louis, J., van der Rijt, V. and Pratley, J. (1992). Application of low-cost airborne video technology to Australian agriculture. Proc. 6th Australasian Remote Sensing Conference, Wellington, New Zealand.

15. Reid, R.N.D., Vickery, P.J., Hedges, D.A. and Williams, P.M. (1993). Measuring the response of pasture to superphosphate using aircraft and satellite remote sensing. Aust.J. Experimental Agriculture 33, 597-600.

16. Richardson, A.J., Menges, R.M. and Nixon, P.R. (1985). Distinguishing weed from crop plants using video remote sensing. Photogrammetric Engineering and Remote Sensing 51 (11), 1785-1790.

17. Richardson, A.J., Heilman, M.D., Escobar, D.E. and Everitt, J.H. (1988). Nitrogen and tillage treatment effects of sorghum as measured by aerial video and handheld ground reflectance. Proc. 1st Workshop on Videography, Amer. Soc. Photogram. and Remote Sensing, 175-187.

18. Richardson, A.J., Heilman, M.D. and Escobar, D.E. (1990). Estimating grain sorghum yield from video and reflectance based PVI (perpendicular vegetation indices) at peak canopy development. Journal of Imaging Technology 16, 104-109.

19. Stutte, G.W., Stutte, C.A. and Ashlock, B. (1988). Yield verification in field crops using image capture and analysis system (ICAS), ed. P. Mausel. Proc. 1st Workshop on Videography, Amer. Soc. Photogram. and Remote Sensing, 151-159.

20. Stutte, G.W., Bors, R. and Stutte, C.A. (1990). Quantification of nutrient stress in horticultural crops using videography. Journal of Imaging Technology 16, 124-127.

21. van der Rijt, V., Louis, J., Cregan, P., Dare-Edwards, A.J., Pratley, J.E. and McKenzie, G. (1992). Plant canopy spectral responses from field trials; a component of the airborne video imagery project. Data Report, Charles Sturt University.

22. Vickery, P.J. and Furnival, E.P. (1992). Development and commercial use of LANDSAT derived maps as an aid to more effective use of fertiliser. Proc. 6th Aust. Agronomy Conf., Armidale, NSW.

Plate 1. True colour image of a wheat field showing intra-field variations in vegetation appearance

Plate 2. Supervised classification of plate 1 distinguishing black oats, wheat and headlands, sparse vegetation, standing stubble and shadow.

Plate 3. True colour image of an oat field showing intra-field variations in vegetation appearance

Plate 4. Supervised classification of plate 3 distinguishing oats, headlands, no vegetation, and wild radish.