ABSTRACT
Seven years of winter survival data for winter wheat ( Triticum aestivum L.) were collected on a loam soil located on the Central Experimental Farm at Ottawa, Ontario (45 degrees 23'N, 75 degrees 43'W). The site was low-lying and subject to frequent winter flooding and ice-sheet formation. Two cultivars, a soft white and a hard red winter wheat, were planted in September. Crop establishment was measured in late fall and the percentage survival was measured in April of the following year. Meteorological data, which were available from the nearby weather site, were used to develop a large set of monthly weather indices that were felt to be important for winter survival. The objective of the study was to use genetic selection algorithms and artificial neural networks to select a subset of critical weather factors and topographic features and to model winter survival. The six weather indices selected were the total rain depth for December (mm), the total rain depth for February (mm), the number of days of the month with snow on the ground for January, the extreme minimum observed daily air temperature for March ( degrees C), the number of days of the month with snow on the ground for March, and the number of days of April with a daily maximum air temperature greater than 0 degrees C. It was found 89% of the variation in winter survival could be explained by these six weather indices, the cultivar, elevation and plot location.
