Comparative analysis of three fitting methods of rice trend yield

It is a prerequisite to have accurate meteorological data for correct assessment of the effects of meteorological factors on grain yield. As such, it is important to explore rational methods to fit trend yields of crops to guide future crop production efforts. In this study, 62 years of historical data on rice yield from 17 sites in 9 regions in Liaoning Province were used to rationalize the separation of the trend yield of rice in relation to meteorological factors. The separated trend yields with three fitting methods-HP filter method, exponential smoothing method and logistic method, were compared and discussed. The rationality of the above three methods were evaluated based on the following preconditions:1) the agricultural productivity and technological development in similar regions was consistent, 2) the yield series change reflected promoting effects of state favorable policies on crop yield, and 3) uniformity of the regions with similar heat condition caused same changes in crop yield. The results showed that:1) the trend yield series fitted by the three methods were consistent with the average trend yield series of Liaoning Province, suggesting that the three methods properly reflected the regional consistency of development. Results analysis indicated good consistencies of trend yields of Shenyang, Tieling, Anshan, Liaoyang, Dandong, Jinzhou with that of Liaoning Province, and coefficients of correlation for the regions was as high as 0.91.2) The trend yield series fitted by HP filter method truly reflected the actual trend yield affected by the changes in the levels of national productivity and national policy. This suggested that the goodness of fit for the HP filter method with actual trend yield was the best, followed by the exponential smoothing method. The extraction of the trend yield also basically reflected actual social development, whereas the goodness of fit for the logistic method had the worst performance. 3) Average regional meteorological yield series obtained with different methods had similar inter-annual and inter-decadal variation characteristics, with no significant difference among the methods (P > 0.05). Compared with the two other methods, the meteorological yield series obtained by the HP filter method matched well with climatic variations, followed by the exponential smoothing method. In conclusion, among the three methods of fitting, the HP filter method was the best for fitting the trend yield of rice in Liaoning Province, followed by the exponential smoothing method and then the Logistic method (which was actually not suitable for fitting the trend yield of rice in Liaoning Province). The research results provided the needed reference for trend yield trend fit analysis.