ALL EYES fix on the computer screen as the vine of a kiwi fruit appears, puts out shoots and starts to grow. Leaves unfold, flowers bloom and fruit bud – a full year’s life cycle telescoped into less than a minute. The display is impressive, but that isn’t the point.
The developers of the virtual kiwi fruit are keen to demonstrate the practical value of the computer model, so they set their audience of fruit growers a practical problem. They give the growers details of the soil and climate conditions used in the model, and then ask: “Where would you expect to find the sweetest fruit – near the tips of the vines or closer to the centre of the plant?” Several experienced growers make tentative suggestions, but no one seems too sure. At the touch of a button, however, the computer scientists present the answer on screen. Slowly the fruit on the vine turn different shades of red, with the reddest – representing the sweetest of them all – glowing in the interior of the plant, under the densest canopy.
Such computer models are still a long way from being useful tools in a modern agricultural industry, but the demonstration does show off one of a number of projects around the world that are exploring three-dimensional computer simulations of plant growth and structure. Derived from the virtual reality technology used to create 3D graphics for movies, flight simulators and computer games, this latest development could change the way crops are cultivated and marketed, and even suggest the most desirable characteristics for the plants of the future. Or so say the technology’s proponents.
Advertisement
Industrial cool
Garth Smith, who helped to create the virtual kiwi fruit at Hort Research, a government-supported institute in New Zealand, foresees the day when computer models will be essential tools for keeping farmers competitive. “Visualisation is the way things are going to go,” he says. “It’s easy to get complex information across very simply.” The farmers themselves are still far from convinced, however. “Growers think the models are amazing,” says Monty Spencer, a kiwi fruit and persimmon farmer, and research and development manager for the New Zealand persimmon industry. “But the only way they will take things on is if it pays them.” For the moment, the New Zealand Kiwifruit Marketing Board is contributing just NZ$100 000 (£43 000) to Hort Research’s NZ$1.6 million project to develop virtual plants – and that is generous compared with the levels of industrial sponsorship elsewhere.
Most of the research is being done in New Zealand, Australia, France and Canada, with groups in the US, Britain and elsewhere occasionally dipping into the field. All are tackling problems whose answers are thought to lie in the 3D structures, or architecture, of plants. In New Zealand, for instance, where fruit growers are keen to increase exports, researchers are studying how the taste of fruit varies depending on where on the plant they are. In Australia, where the cotton industry is looking for environmentally friendly ways of dealing with pests, researchers are investigating how changing the shape of a plant influences infestation by destroying a pest’s favourite hideaway or foraging route.
Until four or five years ago, the powerful computers needed to handle the graphics were too expensive for such speculative work. But as computing power has become cheaper, plant modelling has really taken off. Equipment and software designed for virtual reality systems have helped too. Smith recalls that when he started to gather raw data for his early computer models, he and a colleague could spend two weeks with a surveyor’s theodolite recording the 3000 points needed to describe one mature kiwi fruit vine or apple tree. Now, using a tracker system similar to the kind found inside the helmets and gloves of virtual reality equipment, they can do it in less than a day (see: “How to store a plant on disc”).
But the hardware solves only part of the problem. “You can’t map every plant in the world,” says Smith, and no two plants are the same. So researchers set about creating a generic model, taking advantage of the fact that plants of the same species are essentially modular, composed of a series of repeating parts – buds, leaves, stems, flowers and fruits. In any species, each bud or leaf grows to the same basic pattern as any other bud or leaf. So, at the simplest level, the growth of a plant can be broken down into the sum of the growth of these repeating parts.
Reality does complicate matters, however, not least because one part of a plant often influences the development of another. For instance, the growth zone or meristem at the tip of a shoot produces hormones that suppress growth in nearby buds. Also, horticulturists know that a fruit’s sweetness depends on how close it is to the leaves, the source of its sugar. Smith’s group is looking at influences such as these, and at the way changes in environmental factors such as light and temperature affect the growth of different parts of a plant. By isolating and then modelling these effects, the researchers hope to be able to generate virtual plants that can offer real commercial benefits. A computer model could calculate in seconds, for instance, not only how much the nearest leaf contributes to the sweetness of a fruit, but also how this would be affected if, say, a certain percentage of that leaf were damaged by insects. Without a model, assessing interactions between fruit and leaf demands long, laborious experiments.
Another team, led by Philippe de Reffye at the International Centre of Agricultural Research for Development at Montpellier, in the south of France, has taken a different approach. Keen to explore branching patterns in their virtual plants, de Reffye’s group has developed statistical models based on software originally written for planning underground passages in the mining industry – the reseachers were struck by the similarity between the branching patterns of the passages and those of new shoots on plants, which in turn led to a link in the way the two sets of.models were constructed mathematically. The French researchers recorded shapes of plants grown under different environmental conditions. They then used this data to develop models of the plants, based on the probabilities of where and at what angle branches sprout under particular conditions, where leaves will grow, and where fruit will develop.
Though the models built up by Smith and de Reffye are based on real plants, they both have a major drawback – a separate model is needed for each species or group of species. Computer scientists in Canada are trying to get around this problem by building a universal model, using a notation known as L-systems to represent a plant’s development. L-systems were devised in 1968 by the Hungarian biologist Aristid Lindenmayer as a means of representing the interactions of cells, but he soon realised that it could also be used to describe other processes, most notably plant growth.
Each part of a plant and its developmental state is represented by an alphabetic character, and a whole plant by a string of such characters. L- systems link the two, using a set of rules that describe how each part of a plant develops. For instance, suppose a meristem in state A transforms into a meristem in state B by growing a length of stem I and a branch on the left, attached to which is a young leaf K. In L-system notation, this would be represented as A →I[+K]B. The further development of B might be represented by another rule, as might the growth of the leaf K. Together, the rules plot out the growth plan of the plant.
Virtual lab
By 1990, Lindenmayer and Przemyslaw Prusinkiewicz of the University of Calgary in Canada, had produced a series of intricate and realistic computer-generated pictures of plants. In what they call their virtual laboratory in botany, or Vlab, Prusinkiewicz and his research group went on to develop software that can depict the growth of any virtual plant from a description written in Lindenmayer’s notation. Changes to characteristics such as the shape of a petal can be made to virtual plants on screen using a computer mouse. The software allows researchers to experiment with the architecture of plants, and ask questions of the “what if?” variety. For instance, if the branches were to grow at a different angle to the trunk, what difference would this make to the way the leaves are displayed or to how much sunshine reaches them. In this way, the Vlab can be used to design artificial plant scenery for films and, in the future, perhaps to produce fruit with a specific taste.
Lindenmayer’s original system for describing the rules for how plants develop was not designed to take account of the influence of genetic and environmental factors. But the system is now being extended by Jim Hanan, who used to be a member of the Calgary group and is currently at the Cooperative Research Centre for Tropical Pest Management in Brisbane, Queensland. Using measurements of the growth of real plants under different conditions, Hanan is adapting L-systems to take environmental factors such as light and moisture into account, and has grown virtual plants that develop accordingly. With Peter Room at the Brisbane research centre, Hanan has already used the modified L-systems to produce realistic models of two varieties of cotton by altering the values governing the leaf shape and orientation in a general model.
So far, almost all plant growth models have concentrated on the parts of the plant that are visible above ground. But Art Diggle from the Western Australian Department of Agriculture in Perth is taking a different tack, with software that simulates root growth. “Producing models of roots is even more desirable than shoots,” he says. “Using them, the computer can provide a window below the soil where you can’t see.” Despite the difficulties involved in gathering the necessary data, Diggle has produced a general model of root growth called Root map. It builds on simple rules devised from observations of how root tips move through the soil, and where and when the branches appear. He has now linked up with the Prusinkiewicz group in Calgary to adapt the L-system notation to modelling roots. The joint team hopes to combine root and shoot models into simulations of whole virtual plants.
Room predicts an exciting future for the technology. He foresees agricultural scientists designing plants on screen and then breeding or genetically engineering real plants to match the virtual plant with the most desirable characteristics. “All the pieces of the tool kit are there. We just have to find a way of putting them together efficiently” he says. “In a sense, we have already used the same philosophy … Researchers determined that what the world needed was rice plants with shorter stalks and bigger heads. We bred them, and hey presto!” In the future, says Room, if computer simulations indicated that bigger leaves lead to sweeter fruit and fewer hide-outs for pests, scientists could identify genes responsible for bigger leaves and transfer them into the plants.
Virtual plants could also help improve horticultural practice, insist the technology’s proponents. For instance, they could reveal the best way to prune plants to yield fruit with a particular taste. In Western Australia, where the soils are notoriously poor in nitrogen, Diggle is already experimenting with a model of how nitrogen moves in the soil and how it is taken up by roots. He aims to combine this nitrogen model with Rootmap to find out how to get the optimum amount of nitrogen into plants. Exercises like this could ultimately help farmers decide where to apply fertiliser, at what time and at what concentration.
Old story
But the farmers themselves have still to be convinced. Brian Hearn, a consultant to the Australian cotton industry, points out that growers have seen other computer applications come and go. “In terms of the cotton industry, [virtual reality] may be a solution in search of a problem,” he says. “Twenty-five years ago we had crop simulations, ten years ago it was expert systems, now it’s virtual reality.” Nevertheless, Hearn accepts that virtual plants are a useful research tool, and the scepticism on the ground has not dampened the enthusiasm of the technology’s proponents.
Room wants to combine CAT-scan technology with virtual plants to construct 3D pictures of crops. He foresees tractor-mounted scanners collecting data on the move, and feeding the information into computer models of the crop. The computer would use the model to calculate profits that would be expected to flow from different ways of dealing with the crop. At present, the cotton industry relies on crop scouts to survey crops by eye.
Smith has even greater ambitions for the technology. “I dream of a supermarket with a touch-screen computer at the entrance to the fruit section. You dial in the peach flavour of your choice and a near-infrared device scans across the peaches on display. It selects those peaches which most closely correspond to your taste, and gently sucks them into a tube. The tube then plunks them into a bag for you. All you have to do is collect your bag and pay for it.”
How to store a plant on disc
Using virtual reality technology, computer scientists at Hort Research in New Zealand have developed a system for recording the shape of real plants. The system has three components: a fixed electromagnet that generates a magnetic field around the plant, a sensitive detector that is moved to differeent points on the plant in turn and a portable computer programmed to record and assimilate the measurements.
The detector gauges the direction and strength of the magnetic field at each point. It relays this information to the computer that then calculates the position of each point. In this way, the computer gradually builds up a three-dimensional picture of stems, leaves and fruit, which it then stores for later use.
żěè¶ĚĘÓƵs at the Cooperative Research Centre for Tropical Pest Management in Brisbane use a similar device, but based on sound rather than electromagnetism. A pistol-shaped pointer is equipped with two separate sound emitters that give out a faint “crack”, one after the other, when a trigger is pulled. Three fixed microphones around the plant time the arrival of the cracks from the pointer. Software running on a portable computer translates these measurements into distances and angles, to build up a 3D picture of the plant.