Yield Estimation Of Lycium Barbarum L.based On The WOFOST Model

As an important cash crop in Northwest China,the yield of Lycium barbarum L.（Chinese wolfberry）can be estimated by crop models.However,the crop model is a simplification of crop growth process.There are not only errors in model itself,but also errors in the initial conditions and boundary conditions when the model is running,which leads to the deviation between the simulated values and the real values.Assimilation method of adjusting model running trajectory according to actual element values in the crop growth stages is an effective way to improve the simulation accuracy.Compared with traditional methods such as report statistics and sampling,remote sensing technology provides an effective technical means for obtaining crop growth information in large areas.And it can quickly and intuitively obtain objective observations of crop spatial distribution and canopy elements.Jingyuan County is one of the main Chinese wolfberry producing areas in Gansu Province.Quickly and accurately getting the spatial distribution and yield of wolfberry has great significance to adjust local agricultural structure and sustainability of regional economy.Therefore,in this paper,crop planting structure in the study area was extracted to obtain the planting range of wolfberry,and model assimilation was conducting by combining the data of field experiments and Leaf Area Index（LAI）gotten from remote sensing images.Then in-depth studies and discussions from sensitivity analysis and other aspects were carried out.The main research work and preliminary conclusions of the paper were as follows:（1）Based on the Sentinel-2 data,this manuscript adopted an object-oriented classification method,used spectral features and texture features to construct a random forest classifier,and realized the extraction of wolfberry planting information.Among them,the classification method combining spectral features and texture features had the highest accuracy.The overall accuracy was 88.14%,the Kappa coefficient was 0.81,and the user accuracy of wolfberry was 81.03%,which could well extract wolfberry scattered in space.（2）During the annual growth period,wolfberry can blossom and fruit many times.Therefore,growing stages were divided into summer fruit（March–July）and autumn fruit（July–September）,and yield estimations based on the WOFOST model were carried out for these two periods,respectively.The relative error for annual yield of initial parameters was-20.95%,and the relative error for annual yield after assimilation was-5.51%.The yields of summer fruit and autumn fruit were 2588kg/hm² and 601 kg/hm²,respectively.At the same time,simulation accuracy of LAI was high.Coefficients of determination in summer fruit and autumn fruit simulation were both greater than 0.97,and the root mean square error were less than 0.02,which had significant improvement compared with those before assimilation.（3）Morris and Extended Fourier Amplitude Sensitivity Test（EFAST）methods were used to evaluate sensitivities of crop parameters and soil parameters in the WOFOST model.And the results showed that parameters related to CO₂ assimilation rate,leaf area expansion and thermal time during specific periods generally considerably affected simulated yield.