ALBERT

Description: A technique was developed to obtain predictions of potential solar radiation and temperature for a prescribed, mostly unmonitored, area in the Cape Breton Highlands region of northeastern Nova Scotia (46°39′N 60°57′W to 46°40′N 60°24′W). Hourly predictions of incoming solar radiation are based on relations of sun-earth geometry, clear-sky atmospheric transmittance, and land-surface attributes resolved from digital terrain and vegetation models. The digital vegetation model characterizes vegetation cover and is used to define the average midday albedoes for the area in question. Hourly albedoes are calculated according to assigned mid-day albedo and sun-illumination angles. Land-surface characteristics (elevation, slope, aspect, horizon angles, terrain configuration factor, and view factor) affect total incident solar radiation by affecting the direct, diffused, and reflected energy components. Hourly spatial variability in above-ground daytime temperature is captured by way of a fully trained artificial neural network (ANN) that describes hourly fluctuations of interior highland temperatures according to i) reference temperatures taken at two lowland locations, one at Ingonish Beach and the other at Grande Anse; ii) distance from a north-south line representing the east coast of the study area and from the Grande Anse location; iii) time of day; and iv) land-surface attributes. Training the ANN involves supplying the network with actual data and having the network adjust its internal weights iteratively so that the output values are sufficiently close to the supplied target values. Comparison of predicted and observed hourly spring-summer (1997) temperatures revealed that the constructed ANN explained over 88% of the variability exhibited in the observed temperatures and that the standard error of estimate was 2.0 °C (mean absolute error = 1.5 °C). Key words: Sun-earth geometry, radiation laws, variable surface albedo, clear-sky atmospheric transmissivity, digital terrain, vegetation models, artificial neural networks