|
(88-02) Neural network models as tool for recognizing the source of ecotoxicological stress effects in plants. Soja, Gerhard*,1, 1 ARC Seibersdorf, Seibersdorf, Austria ABSTRACT- Wheat (Triticum aestivum cv. Extradur) and bean (Phaseolus vulgaris cv. Maxi) were analysed for their leaf chlorophyll fluorescence responses to three types of stresses: enhanced soil moisture deficit (35 % W.C.), elevated ozone concentrations (90-100 nl.l-1), and prolonged flooding. By measuring the fast kinetics of chlorophyll fluorescence a dataset was created for developing neural network classification models. Potted plants of both species were exposed to the three stress treatments (plus one non-stressed control) for three weeks at flowering stage. Chlorophyll fluorescence measurements were taken at three consecutive days after this stress period. Measurements were only taken at visually undamaged leaves. Finally leaves were harvested for chlorophyll concentration analyses and the above-ground biomass for dry matter determinations. Based on the results of a principal component analyses, 7 inputs were selected which were not dependent on one another. These inputs were used to develop neural network classification models for recognizing the source of stress ? the four treatments of the experiment. The correct classification rate of the species-specific models were 78 % for wheat and 87 % for bean. The wheat model had problems in distinguishing flooding stress from the other treatments whereas the bean model was weakest in recognizing ozone stress. For improving model performance, the classification task was split in two steps. At first models were developed to distinguish between stressed and control plants. Then for the stressed plants separate models were developed which had the task to discern between ozone, drought and flooding effects. These models achieved a correct classification rate of 77 ? 97 %. Key words: neural network model, photosynthesis, crop, stress recognition |
