The use of artificial neural networks in ecological analysis: estimating microhabitat temperature

1. The thermal environment at the scale in which most species exist is largely unknown, and thus the majority of physiological models is based on meteorological measures of ambient temperature. 2. The use of artificial neural networks in ecological analysis is promoted by using them to model microhabitat temperature. 3. The performance of conventional multiple linear regression is compared with that of artificial neural networks in predicting the temperature profiles of two different microhabitats using ambient temperature, solar radiation, and wind speed as input (independent) variables. 4. In both cases, the artificial neural networks showed a lower mean absolute residual error than multiple linear regression (0.95degreesC compared with 1.41degreesC, and 0.29degreesC compared with 0.50degreesC) and a higher correlation (r(2)) between predicted and observed values (0.832 compared with 0.668, and 0.884 compared with 0.670). 5. An artificial neural network developed to include a microhabitat patch description based on height within patch, substrate, and four classes of per cent vegetation cover performed well (r(2) = 0.933, prediction error 95% confidence limits = +/- 2.5degreesC). 6. It is suggested that artificial neural networks are more appropriate than conventional regression-based approaches for estimating microhabitat temperature.