Investigating microclimatic influences on ozone injury in clover (Trifolium subterraneum) using artificial neural networks

Abstract

SUMMARYMicroclimatic factors interact during ozone episodes to influence the sensitivity of plants to ozone and thus arc likely to modify the amount of injury development. This paper investigates these interactions in an ozone-sensitive cultivar of clover {Trifolium subterraneum cv. Geraldton). Experiments were conducted using a glasshouse-based closed-chamber exposure system in which the plants were exposed for 7 h to either chareoal-filtered air {CF) or CF plus ozone at concentrations ranging from 40 to 160 ppb. The microclimatic conditions inside the chambers ranged from 16 to 36 °C, 0-9-3'6 kPa vapour pressure deficit (VPD), and 80--460 /imol m~'^ s"^ Photosynthetically Active Radiation (PAR). Seven days after ozone exposure, the extent of foliar ozone itijun.-was scored visually. The assessment scoring system was validated by pigment analysis. The data from these exposures were analysed using artificial neural networks (ANNs), the principles of which are described in tbe paper. Two ANNs were used, one to investigate the efTeets of microclimate on the threshold AOT40 (dose accumulated above a threshold of 40 ppb) above wbich injury developed, tbe other to determine the extent of visible injury development. Both networks used temperature, VPD, PAR and AOT40 as inputs. Testing with previously unseen data showed that the networks produced accurate predictions of the threshold and extent of injury for a range of ozone doses and microclimatic conditions. For example, the injury score network predicted that at 100 /(mol m"^ s"' PAR and 1 kPa VPD an AOT40 of 3. SO ppb h was required to produce an injury score of 1, wbereas in conditions of 400 /imol m"'' s"^ PAR and 3-5 kPa VPD, an AOT40 of 460 ppb h was required. Analysis of the weightings of components of the trained networks indicated that VPD and PAR had a stronger influence on the response to ozone than did temperature. Furthermore, this approach revealed that microclimate had a greater influence on the extent of ozone injury than on the threshold for injury.

Effects of ozone on wild plants

“…First, expression of symptoms is affected by environment, especially soil water deficit, vapour pressure deficit, light, temperature (Sand, 1960 ;Shinohara et al, 1973 ;Dunning & Heck, 1977 ;Keitel & Erhardt, 1987 ;Showman, 1991) and possibly u.v.-B radiation (Thalmair et al, 1995). In a recent study, Balls, Palmer-Brown & Sanders (1996) used artificial neural networks to examine the effects of microclimate on the threshold ozone dose for visible injury in Trifolium subterraneum. The analysis indicated that VPD and PAR had a stronger influence on the response to ozone than did temperature, and that microclimate had a greater influence on the extent of ozone injury than on the threshold for injury.…”

Leaf injury characteristics of grassland species exposed to ozone in relation to soil moisture condition and vapour pressure deficit

“…The use of artificial neural network (ANN) models to analyse the interactions between microclimate, AOT40 and injury was described in detail by Balls et al (1996). A number of ANN models were created using a back propagation algorithm within the Neuroshell 2 package (Ward Systems Group, Frederick, MD, USA).…”

“…A similar effect of increased air vapour pressure deficit (VPD) has long been recognized (MacDowall et al, 1964). In order to analyse the role of these modifying factors and their relative importance, Artificial Neural Networks (ANNs) have been used (Balls et al, 1995 ;Balls et al, 1996), and their use could be extended to include plant-specific factors. It was confirmed that VPD is one of the most important factors determining the occurrence of short-term visible injury under well watered conditions (Balls et al, 1996).…”

“…For each submodel used in the optimization process, the performance of the model was determined for the training data and for the test data, a randomly extracted subset of the data not directly involved in creating the model. Once the optimum model was determined, the relative importance of the model inputs was obtained by calculating the sum of the absolute values leading from each input (Balls et al, 1996).…”

Approach to Prediction of R/C Buildings’ Seismic Damage as Pattern Recognition Problem Using Artificial Neural Networks

“…Therefore, one type of input vectors was formed (with dimension (12x1)). (c) Stage 3: Using the two versions of "Stepwise" sensitivity analysis method [37,38,39] the number and the combination of seismic input parameters which lead to the optimum ANNs' predictions for the seismic damage state of r/c buildings was investigated. Thus, in this stage input vectors with dimension between (5x1) and (12x1) were formed.…”