Abstract
Observations from the First ISLSCP (International Satellite Land Surface Climatology Project) Field Experiment (FIFE) showed that it is difficult to estimate the sensible heat flux from routinely observed environmental parameters. This study, therefore, explores the use of backpropagation neural networks to elucidate the link between sensible heat flux on the one hand and horizontal wind speed, air temperature, radiometric surface temperature, net radiation, and time on the other. Data collected over the FIFE site in 1987 and 1989 were used for network training and validation. Networks trained on part of the data from a narrow range of space–time coordinates performed well over the other part, with error (rms divided by mean of observations) values as low as 0.24. This indicates the potential in neural networks for linking sensible heat flux to routinely measured meteorological variables and variables amenable to remote sensing. When the networks were tested with data from other space–times, performance varied from good to poor (average error values around 1.27), depending on the degree of similarity between the training and validation datasets in terms of parameters not explicitly included in the training set. Poor predictive performance was primarily associated with the lack of input variables parameterizing canopy morphology and soil moisture, indicating that such variables should be incorporated in the design of future networks intended for large-scale applications. Observations also showed that an underparameterized network cannot be made more general by expanding the size of the training dataset. These findings have repercussions on the potential to derive energy and moisture balance estimates from standard meteorological and satellite-based remote sensing observations.
Corresponding author address: B. Abareshi, Department of Natural Resource Sciences, Macdonald Campus of McGill University, 21111 Lakeshore Rd., Ste-Anne-de-Bellevue, PQ H9X 3V9, Canada.
Email: abareshi@nrs.mcgill.ca
