An improved method of interpolating annual crop yield data using wavelet transform
Background: Annual crop yield data is a typical time series data in Precision Agriculture. Most data is constituted by some observations at irregularly scattered points in time due to the finite cost of data acquisition and its difficulties, so the data is corresponding to a subset of the whole study period. In this study, we aim to present an improved interpolation method for the annual crop yield data by introducing wavelet filtering to auto-regressive integrated moving average (ARIMA) model.
Methods: With the proposed method, wavelet transforms were conducted to decompose the original time series data into approximation coefficient and detail coefficients, and each coefficient was fitted by ARIMA model. Then, the interpolated data was obtained by re-aggregating the fitted approximation coefficient and detail coefficients, and a cross-validation was applied to evaluate the interpolating accuracy.
Results: In the experiment, annual wheat yield data got from 27 agrometeorological stations distributed throughout main winter wheat producing areas of China over a 30-year period (1981 to 2010) was used to evaluate the proposed method. The results showed that the proposed method yielded more accurate results than conventional ARIMA and linear interpolation. Moreover, the proposed method is not sensitive to the length of the time series, because the wavelets have the advantage of handling multi scale features of the original data.
Discussions: Besides the greater interpolation accuracy, the proposed method concerns of multi-scale feature of the time series data, and brings about multi-scale decomposition of the data, which will reveal more inherent information of the data.
Conclusion: The method introduced in this paper increases the accuracy of annual crop yield data interpolation, and also will be an effective method of analysing the data.