Previous PageTable Of ContentsNext Page

Satellite imagery as a data source for prescription and precision farming in Australia

Brian J. Button

Professor in Earth Observation Systems, University of Canberra
Managing Director, Agricultural Reconnaissance Technologies Pty Ltd

Important differences between prescription versus precision farming are highlighted.

Satellite sensors designed for earth resource monitoring offer an important source of data for utilisation in prescription and precision farming. The relative advantages of satellite imagery over aerial photography, airborne video, thermal and other forms of long wave energy sensing are presented by highlighting technical and data integrity problems of these alternative forms of data acquisition.

Satellite imagery is available at a range of different revisit frequencies, from a range of different providers, under a range of different conditions attaching to the use of this data, at a range of different spectral and spatial resolutions, at a range of prices that reflect the range of different product formats, accuracies and reliabilities. Selection of the appropriate imagery at a date of image acquisition that is appropriate to the task and stage of development of the crop under consideration are particularly important decisions affecting the utility and accuracy of the resulting analyses.

A tabular summary is provided of the orbital and imagery characteristics of the main forms of remotely sensed image data that are currently available (and soon to be available) on a commercial basis. This is followed by an introductory guide to critical issues that need to be considered in the selection of appropriate imagery.

Landsat ETM and SPOT satellite imagery has predominantly been used to date because of its ready availability, being acquired on a repetitive basis over the same location on earth within fixed acquisition cycles. The Australian Centre for Remote Sensing archives data from both sensors for most agricultural production areas in Australia. The Landsat satellite images an area of some 32400 km2 while a single sensor of the SPOT satellite images an area of some 3600 km2. The cost of satellite data per unit area is low – for Landsat the cost is $0.009/ha and for SPOT $0.024/ha. The spectral response and lower cost of Landsat mean that it is generally preferred for agricultural applications in general and prescription and precision agriculture specifically. However, the off-nadir viewing capability, higher resolution and guaranteed program continuity may mean that SPOT will become the preferred data source, despite its higher cost and fewer spectral bands compared with Landsat ETM.

The value of the data, particularly for prescription and precision farming, lies in the fact that the spatial resolution at which it is acquired (25m pixels for Landsat and 20m pixels for SPOT) is useful for detecting spatial variation in crop spectral response within the field sizes and at enterprise scales that are common in Australian agriculture.

Satellite imagery provides a cost effective way to facilitate pre-harvest yield estimation and yield variability mapping without the need for high cost machine and GPS based yield monitoring equipment. Early season imagery enables emerging problems to be identified and appropriate intervention and crop management strategies to be invoked to minimize within field yield variance. The synoptic viewpoint combined with detailed pixelisation makes satellite imagery superior to existing field reconnaissance techniques, especially where ground based instrumentation at strategic locations and targeted field observations by consulting agronomists at more representative sites is used to calibrate the results of image analysis for greater accuracy and precision.

Spectral response detected within the area occupied by any single pixel represents the integration of all factors such as crop phenology, soil water status and nutrient status that determine plant vigour, biomass and ultimately, yield potential. Instead of attempting to measure individual discrete parameters, satellite imagery reads the response of the plant as a single integrated instrument. Instead of seeking to build physical models of complex atmospheric/plant/soil interactions on a computer based on extrapolation of experimental observations from a few local sites, an empirical approach based on satellite imagery enables agricultural users to start from an understanding of spatial variability and use diagnostic spatial patterning, form or image morphology to explain the cause of these variations that are likely to result in reduced yield potential. Digital processing and interpretation of satellite data translates changes in spectral response into products that identify areas of variability within an individual field. Resulting products can take the form of crop growth, plant vigour or yield potential variability maps and resulting yield estimates, and can be presented in digital, image, map, tabular or chart formats.

Localised variations in atmospheric conditions, non branching plant architecture, low biomass, poor vegetation cover or lack of canopy closure (LAI < 3) means that aerosols, soil type and condition exert a stronger influence on spectral response, compounding or complicating imagery analysis and interpretation. However, an empirical approach using satellite imagery enables management impacts to be detected and discriminated from environmental or physical factors using diagnostic visual queues.

Pixel specific spectral response measured from satellite imagery can be used to develop empirical models that can be used in turn to estimate parameters such as early season vigour and biomass, yield on offer well before harvest, even protein level of a grain crop to facilitate product segregation at harvest to capture premium commodity prices. The estimated yield maps help growers to view variability in terms of economic return and to identify regions in fields which require some form of remedial action or to plan froward selling of the crop. These maps may be incorporated into variable rate technology (VRT) systems with the proviso that the data is acquired at the correct time, processed and delivered rapidly for utilisation in such systems. Aspects such as cloud cover at the time of data acquisition and the fixed acquisition schedule limit the use of these products as the primary source of data for VRT applications.

The advent of IKONOS and other ultra high resolution imagery opens the door to the use of satellite imagery for prescription and precision farming for the highest value horticultural crops such as grapes and orchards.

The advent of Hyperion and deployment of aircraft based systems such as CASI and HYMAP in Australia, coupled with the promise of other hyperspectral satellite imagery in the near future opens up the possibility of being able to detect and diagnose nutrient deficiencies and other site specific problems across broad areas before they are visible and without the need for extensive, regular field visitation. However, a significant amount of research in this regard and substantial reduction in cost of this imagery will be required before hyperspectral imagery analysis finds a market in prescription and precision farming.

In the same vein, post processing of high quality differential GPS data from precision steering systems opens up the opportunity that high quality digital terrain models can be generated to depict micro-topography that might explain localized differences in soil moisture status, waterlogging and even provide the basis for more detailed, more accurate floodplain mapping and modelling.

Capital costs and technical demands of machine based precision steering systems, yield monitors and VRT mean that the majority of growers will be slow to adopt these advanced forms of precision farming. The particular appeal of satellite imagery for the vast majority of growers is its use as a mapping base for property planning, identification of discrete portions of individual fields that can be farmed separately through a different prescription of inputs and management, and selection of more representative sites for field visitation, observation, soil and plant tissue sampling and instrumentation.

Empirical models are built by relating measured variations in some parameter within the field with spectral response from the satellite imagery. The importance of high quality collateral ground data, collected at a time that is meaningful relative to the date of image acquisition, cannot be over emphasised.

In this respect geo-positioning and geolinking of the imagery and collateral ground data, whether acquired through site specific observations in the field by human observers or from GPS based yield monitors, is crucial to both the development and refinement of empirical models. Satellite remote sensing and ground based observation are therefore complimentary to one another and not in competition with one another. One is used to calibrate and ground truth the other.

Users must pay careful attention to the reference datum, projection, spatial accuracy and degree of generalization inherent in each data set. This will ensure that field data and satellite imagery is geolinked for compatibility and comparability for use in the field or analysis via the growing range of geospatial farm mapping/management/precision agriculture software packages in the market place. Most of these packages are at a distinct disadavantage by being able to handle imagery only in pictorial format and without any capacity for registration or rectification.

The complexity of matters to be addressed in acquiring and handling satellite imagery and other forms of spatial and temporal imagery means that there are traps for would-be players in prescription and precision farming who blindly assume that they can incorporate these forms of data into their kit bag of tools. A sample listing of such matters is provided below. Image processing and GIS is not the core business of agriculturalists. It should be left for specialists such as Agrecon who is willing and able to support those who seek to build a business based on the application of their specialist knowledge in agriculture.

  • What platform, what sensor, what supplier, what product type based on scan line and pixel orientation, what level of accuracy, what level of geometric precision, whether or not to incorporate terrain correction?
  • How comparable is imagery from different sensors, the same sensor on different platforms, the same sensor on the same platform?
  • The need to understand FOV, IFOV, pixel specific spectral response, mixed pixel effects and sensor calibration.
  • Which spectral bands to use, what band combination or data transformation, whether to present imagery for visual display in natural or false colour, how selection of the appropriate band combination and mode of colour presentation changes according to geographic location and time of season, and how to interpret the raw imagery?
  • How to assess image quality and validate the integrity of digital data, what pre-processing should be undertaken prior to digital image analysis, what post-processing should be undertaken after digital image analysis?
  • How to derive relevant weighting factors and express spectral band combinations in the form of empirical algorithms, what spectral statistics to extract from the imagery, what parts of the imagery or target scene to extract spectral data from?
  • How to identify critical non image parameters that explain yield variance with a high degree of significance and reliability, how to handle and process paired data?
  • How to assess and express error, accuracy and reliability of yield estimates, how will it change with scale, the parameter being estimated, the time of year and the crop type?
  • How to translate experimentation and research in the use of remote sensing into a commercial fee for service operation for all clients?
  • How to deal with issues of timeliness in data acquisition, data supply, data processing, product output and service delivery?
  • How to price a satellite based prescription or precision farming service, and what is the value proposition of such a service?

Agrecon’s satellite based approach to prescription and precision farming in Australia comes after a decade of research and client servicing in this area, culminating in the development of an internet based system branded as FarmMAPS and YieldMAPS. The fundamental data sets underlying these systems, their existing and proposed new functionality, and their mode of service delivery to a national client base will be described. Particular focus will be on the ingredients of a business model that has seen Agrecon emerge from a cottage industry supplier for selected clients to a national and international leader in the provision of comprehensive crop forecasting and yield prediction information for all growers on an industry wide national basis.

Agrecon’s relationship with clients is based on continued improvement in the quality and reliability of advice arising from internet communication with and provision by clients of site specific observations that Agrecon uses to calibrate yield map products and yield estimates. The value of using multi-temporal imagery from previous years is highlighted to evaluate financial returns from previous management initiatives and assess progress towards increased yields and reduced yield variability.

The oral presentation of this paper will illustrated with examples of satellite based prescription and precision farming products, services and systems offered by Agrecon to illustrate the range of issues raised in this paper.

Previous PageTop Of PageNext Page