Previous PageTable Of Contents

SatMaps pasture condition mapping by satellite: a farm management tool

M.C. Aubrey

Managing Director, Technical & Field Surveys Pty Ltd
TFS Rural Resources Applications Centre
1562 Limekilns Rd Bathurst, NSW
Tel: 02-63376660 Fax: 02-63376667
Email: tfs@lisp.com.au

Abstract

The SatMaps package uses Landsat TM satellite imagery to analyse pasture condition on a farm by farm and paddock by paddock basis. This method of pasture analysis was originally developed by the CSIRO Division of Animal Husbandry and later acquired by the Incitec Analysis Systems Division of Incitec Ltd. Incitec set up a joint venture with Technical & Field Surveys Pty Ltd to perform the image analysis and GIS operations.

The methodology involves applying a complex image analysis procedure to a Landsat image chosen to represent the peak growing period. This procedure produces 60 land cover classes. Statistical analysis then produces a set of rules that determine bonding strengths between classes which are used to aggregate the data into seven land cover categories, principal among which are fast, medium and slow growth pastures.

At the same time, selected farms in the area are digitized to paddock level. The farm boundary is used to extract the relevant information from the classified image and another statistical process is used to calculate the hectarage and percentage of each class in each paddock. A report containing this information plus a farm map and classified image is supplied to the farmer for management purposes such as fertilizer application.

The overall objective is to help landholders to manage their pastures more efficiently, cost effectively and environmentally soundly.

Introduction

The soils used for improved pastures in temperate Australia are usually deficient in phosphorous, sulphur and nitrogen. This has necessitated the application of large quantities of superphosphate which is a very significant input cost and may also represent an environmental threat. Even within a paddock, pasture condition and nutrient levels can vary substantially. Traditionally, paddocks have been treated as being spatially homogeneous for the purpose of determining fertilizer requirements. This has often led to a costly and wasteful application rate and a consequent detrimental impact on acid soils and water quality. It has been noted (Vickery & Hedges, 1987), that the traditional approach of soil testing to determine nutrient requirements, though effective for crops, often gives conflicting results for improved pastures. As well as being costly, soil samples do not provide spatial information. The same problems apply to plant tissue analysis. SatMaps uses satellite data to provide the spatial context.

Significant benefits can be obtained from analysing the spatial distribution of within-paddock pasture condition, thus enabling the most cost-effective rate of input application, such as fertilizers and lime, to be assessed and more selectively applied. This reduces the input cost of chemicals to the landholder, and also reduces the amount of excess nutrients that may be washed into the river systems to exacerbate algal bloom problem.

The initial objective of research was to provide a support tool for government agricultural extension officers however it was later recognized that individual farmers would like to use the data in their farm planning. This was prevented by the overall cost of the data and analytical process.

The SatMap program is aimed at using satellite imagery within the framework of a Geographic Information System (GIS), together with an innovative and cost-effective packaging and distribution method, to provide an economical basis for producing this assessment of pasture condition for individual farming enterprises.

History

The Armidale Research Station of the CSIRO Division of Animal Husbandry, under the direction of Dr Peter Vickery, conducted an eight-year research program (1987-1995) into the assessment of pasture condition using satellite imagery. This project, the “Growmax” program, generated a set of procedures and programs based on the use of the Disimp Image Analysis system and the EPPL7 GIS which defined 7 land cover classes and quantified them at paddock level. The work was supported by the WRDC of the Australian Wool Corporation. The NSW Dept of Agriculture and the Tasmanian Dept. of Agriculture also researched the agricultural extension applications of Vicery’s method for several years.

In 1995, Incitec Ltd acquired the results of this research and then entered into an agreement with TFS to carry out the technical and processing operations. Immediately after this Vicery’s group was disbanded. Unfortunately, this cut off any potential for follow-up research or collaboration.

It was soon discovered that the software and systems used to perform the analysis were virtually obselete and that most of the procedures were totally dependent on these systems. Substantial work was required to modify the methodology to suit more modern processing software and systems. This entailed translating the image processing algorithms from Disimp (now defunct) to a less system dependent regime, converting the programming language and converting the EPPL7 algorithms to MapInfo while retaining the basic methodology. The new program has been renamed, “SatMaps”.

Economic Considerations

Agriculture has been a principal target for satellite program design since the first ERTS-1 system in 1972 which led to major US agricultural programs, such as Agristars, in 1978. However, the cost base of applying satellite technology has largely dictated a national or regional approach to such applications. For instance, Australia set up an Agricultural Remote Sensing Committee, under the auspices of the Agricultural Council in 1978 of which the author was a founding member. This committee considered that almost all likely agricultural applications were at government level.

The limiting factor to applying satellite technology at individual farm level has been the cost of acquiring and processing digital imagery that usually puts it outside the reach of individual farming enterprises.

Organisations, such as the University of Canberra’s Agrecon, began to address the farm level usage of satellite imagery about a decade ago by designing and supplying low cost photo and software products. The TFS concept of a Rural Resources Applications Centre (TRRAC) to address the problem of practical, cost-effective rural applications and technology transfer was set up in 1996. This is the subject of another paper at this conference. The SatMaps project is part of this concept.

The SatMaps program has established an innovative way of making the program cost-effective at individual farm level by making use of existing agribusiness networks and group processing. Thus a whole satellite image can be processed and then the results for each farm are subset and supplied to each farmer in a pre-arranged group at a cost commensurate with the size of the farm. This has the additional benefit that the pasture condition results are expressed in relation to the region as a whole, thus providing a guide to how well the farm is progressing relative to its neighbours.

Methodology

The methodology involves applying a complex image analysis procedure to the data. This procedure produces 60 land cover classes. Multivariate analysis then produces a set of rules that determine bonding strengths between classes. These are then used to aggregate the data into seven land cover categories, principal among which are fast, medium and slow growth pastures.

This process involves the following steps.

GIS

Report

Image Processing

Data Compilation

The first activity is to set up a group of farmers, usually via a local stock & station franchise, and obtain from each farmer a topographic map showing their property boundary, a plan of their paddock layout and a paddock history. An information and data recording kit is provided to each farmer to assist in this process.

Based on the information provided, the property and paddock boundaries are digitised and corrected against topographic maps and images. This procedure is accomplished using MapInfo. Once the property is accurately positioned and digitised, other positional information such as roads and rivers are be overlaid from regional files. From this digital map, information, such as paddock size (hectares) can be accurately measured and placed into a database which also accommodates other information for paddock planning, such as pasture class percentages, paddock history, soil type etc.

Image Selection

The second step is to obtain a suitable satellite image representing the peak growth period for the region. Landsat Thematic Mapper data (30m resolution) is prefered in broad acre areas but SPOT imagery (20m resolution) is used in more intensively farmed areas with small paddocks. The potential use of Ikonos data (4m resolution) is being evaluated. Cloud cover is the main limiting factor since the peak growth period often coincides with high rainfall. Other considerations include differing growth conditions on one image, for instance, coastal plain, slopes and highland plateau regions.

Calibration

The pasture conditions for the region need to be calibrated to local conditions. It is also necessary to adjust for the impact of cropping areas that may distort the pasture signature. Calibration is guided by input from local agronomists and data provided by the farmer on paddocks with a known fertilizer history. This provides a means of predicting the likely responsiveness of particular pixels to the application of additional fertilizer. Once an initial calibration is established for a particular region, it can be applied to subsequent annual coverages.

Classification

First a large composite training area is identified which has a representative percentage of each class and no cloud cover. This area is used to derive reflectance statistics to seed the classification. An unsuperised hierarchic divisive (POLYDIV) classification algorithm is then used to produce 60 classes. A proprietary methodology, using contextural information, a PCA ordination plot of the classes, network analysis and a classification dendrogram, is used to analyse the bonds between the 60 classes. The bonding strengths together with the proximity of the class position on an XY plot are used to aggregate the 60 classes into 7 land cover classes – fast growth pasture (improved), medium growth and slow growth pasture (unimproved), water, open woodland, forest and bare ground. This operation is done using ILWIS (Integrated Land and Watershed Information System), designed by ITC in the Netherlands.

The final seven classes are:

Fast Growth

Medium Growth

Slow Growth

Sparse Vegetation/Bare Ground

Woodland

Forest

Water and shaded areas

Classes 1, 2, and 3 represent pasture condition relative to the district. Class 4 (sparse vegetation) includes any area covered by cloud, built up areas, and bare ground.Class 7 includes any large water bodies, cloud shadow and hill shading in forest areas. These classes apply to the entire region and therefore show an individual farm’s condition in a regional context.

Image Registration

The use of sub-paddock level information requires very precise positioning and rectification of the imagery so that the superimposed paddock boundaries are an exact fit. This is accomplished using PCI Orthoengine plus a DEM to correct the image to ground control points. At the same time the image is resampled to 25m pixels. This process is employed after the classification phase to avoid any modification of pixel values.

Paddock Information Computation and Presentation

The digital paddock map is used as a mask to extract the classified data for each farm and another proprietary process is used to calculate the class statistics for each paddock.

The paddocks are numbered and labeled on the map. For each paddock, the area covered by each of the classes is calculated. For example,

Paddock 7: Total area = 104ha

Class

Type

Area (ha)

% of paddock

1

Fast growth

21.2

20.4

2

Medium growth

11.5

11.1

3

Slow growth

16.4

15.8

4

Sparse veg

22.8

21.9

5

Woodland

18.4

17.7

6

Forest

13.6

13.1

7

Water/shadow

0.0

0.0

Final Product

A summary report is provided to the farmer which includes:

  • a digitised farm map with local road and creeks.
  • a satellite image for the farm superimposed with the digitised map,
  • a summary report of the paddock sizes and the percentage/hectarage of each class in each paddock.
  • a database of information on each paddock
  • a ranking of the condition of each paddock

The data can be supplied in digital format for input into computer farm management programs, such as PinPoint for Agriculture.

Example of a final SatMap

Interpretation of results

Class 1 (dark green) represents the most actively growing pastures that require only low levels of fertilizers for maintenance purposes. Under normal rainfall conditions in the Northern Tablelands of NSW, these pastures could be expected to produce 10 000kgs/ha of dry matter annually. Class 2 (light green) represent more slowly growing pastures which should be highly responsive to the application of additional fertilizers. Class 3 (brown-yellow) represents slow growth pastures, usually degraded or unimproved, which are unlikely to respond to fertilizer application unless improved pasture seeds are introduced with the fertilizer. These pastures generally provide only 2500-300- kgs/ha of dry matter per annum.

The other classes are unsuitable for fertilizer application.

Applications and Benefits

The first and most basic benefit is an accurate assessment of paddock size that can be employed in many management activities.

Fertilisers represent approximately 20 percent of the variable costs incurred by graziers in temperate Australia. This method of monitoring and assessing the pasture within a paddock provides a means of farming more strategically and sustainably. The landholder is now able to assess the amount of fertiliser required on various areas of the paddock, resulting in a more economical application of the fertiliser. The fertiliser will be applied more heavily on the medium growth areas, where it is required, and less heavily on the fast growth areas. This method provides useful information with which to cost effectively manage pasture improvement. It also ensures that the excess fertiliser does not end up in the environment.

The spatial context of the data can add value to the interpretation of soil and plant tissue tests and provide a means of determining the most representative sample locations.

Further Development

This methodology is being modified to incorporate cropping as well as grazing by using aggregate data from several scenes in different seasons.

Conclusions

This procedure provides a standardized and repeatable means of assessing pasture condition relative to the region as a whole. The overall purpose is to offer a cost effective mechanism for managing fertiliser and other soil improvement inputs. Its success is practically demonstrated by the fact that a number of farmers have continued to commission coverages over several years.

Acknowledgements

The author would like to acknowledge the dedicated research performed by Peter Vickery’s team at the CSIRO Division of Animal Husbandry, Pastoral Research Laboratory in Armidale.

The foresight of Incitec Ltd in choosing to adopt the technology is also acknowledged.

References

Aubrey M.C. (1991), Applications of remote sensing - a commercial perspective. ABARE, Nat. Agric. & Resources Outlook Conf.

Hedges D.A. & Vickery P J. (1987), Use of a Principal Components strategy as the basis for an unsupervised classification of routine to examine Landsat data from grassland vegetation, Proc. 4th Australas. Rem. Sens. Conf.

Thorburn L.J. & Tilll M.R. (1988), An Overview of Agricultural Remote Sensing in Australia, CSIRO COSSA Report 023

Vickery P J, (1984), The use of Landsat Technology to determine the fertilizer status of improved pastures, Proc. Aust. Soc. Anim, 15

Vickery P J. & Hedges D.A. (1987), Use of Landsat MSS data to determine the fertilizer status of improved grasslands, Proc. 4th Australas. Rem. Sens. Conf.

Vickery P.J. & Furnival E.P. (1992) Development and Commercial Use of Landsat derived maps as an aid to more effective use of fertilizer, Proc 6th Aust. Soc of Agronomy Conf.

Weissel J. & Aubrey M.C. (1987), Methodology for combining Landsat MSS imagery and SPOT data for crop type and area assessment at regional level, 4th Australasian Remote Sensing Conf.

Williams W.T. & Lance G.N. (1975), POLYDIV: a divisive classificatory system for all numeric data, Aust. Computer Journal, 7

Previous PageTop Of Page