Discrimination of drought occurrence for rainfed spring wheat in semi-arid area based on pattern recognition

Abstract: The mechanism of the damage process for agricultural drought is very complex, and many factors can affect it. Agricultural drought is the main limiting factor for crop yield in rainfed area. For defining drought occurrence during the crop growth, and predicting crop yield, we used pattern recognition based on meteorological data during growing season and yield data of spring wheat in semi-arid rainfed area in Dingxi, China from 1986 to 2011. Owing to the application of deviation for crop yield from its long-term mean to define agricultural drought, we divided the year pattern into two categories: drought series, and normal series on the basis of 30 percent deviation from the mean wheat yield. The iteration method was then applied in order to find a case wherein the drought could be linearly discriminated from normal category. According to our research, we find the spring wheat yield was affected by various factors. They can be categorized as 1) weather conditions, such as temperature, precipitation; 2) farm management factors and crop variety, such as soil tillage, soil depth, planting density, sowing date, crop protection against pests and diseases, and soil fertility level; 3) soil conditions, such as soil physical properties and soil water content. Measuring or estimating some of these factors was often not feasible, and the influence of some other factors may be considered insignificant or constant in an agrometeorological experimental station. It was therefore weather condition alone that can affect crop yield most significantly. However, it was found that no linear relation existed in any cases based on average temperature and precipitation during the main growing period without taking other factors into a consideration. After rejecting years in which the soil relative water content was more than 55%, we can predict if agricultural drought through establishing a linear equation with two parameters, the average temperature and precipitation during the main growth period for spring wheat. From the research, we also found the best parameter to predict the agricultural drought occurrence and factor that determined spring wheat yield was the precipitation in May. A Predictive Equation for spring wheat yield was also established by the least square method based on the precipitation in May. The predictive equation was simple but useful, and it can forecast spring wheat yield one and half month earlier before wheat harvest. Meanwhile, it should be noted that the predictive equation was established after rejecting the years in which the soil relative water content was more than 55%. We suggested that the agricultural drought differ from meteorological drought. As such, we should use the method much more carefully for quantitative prediction of agricultural drought occurrence and crop yield in future research.