چكيده به لاتين
Estimating crop yield before the harvest season is one of the most essential tools for accurate agricultural planning. Self-sufficiency in wheat production is vital for every country, especially countries sanctioned, so estimating the performance of strategic products such as wheat is more important. This issue has become more prominent after the recent tension between the two major wheat-producing countries, namely Ukraine and Russia, and food security has taken on a new form. With proper forecasting of product performance, we will better manage the supply and distribution cycle and will improve export and import planning to a great extent. On the other hand, the traditional performance estimation methods are ineffective due to the high time spent, low accuracy, and high cost. So ground observations must combine with modern techniques such as remote sensing. The wide application of remote sensing facilitates precision agriculture and food security. In this study, dry wheat yield estimation has been done with two methods, the first method has been based on plant indices, climatic and physiological parameters, and the phenological growth cycle of winter wheat, and the second method was based on experimental equations. In both methods, the field data of wheat yield related to 103 agricultural fields with a total area of 331 hectares in 2 crop years 1398 and 1400 and according to the conditions, different satellites have been used. In the first method, by drawing the phenological diagrams of the indicators, the best statistical parameter, which is the area under the diagram, and the best growth stage to describe the performance, which was the flowering stage, were selected along with other factors and entered into the machine learning models, which is the random forest model. It gave the best results with a root mean square error (RMSE) of 162 kg/ha and a coefficient of determination (R2) of 0.78. Also, the correlation coefficient and the importance of the GLAI index for performance estimation were higher than other indices. In the second method, by calculating the biomass and estimating the final yield with experimental equations, RMSE was reported as 341 kg/ha. Also, due to the unreliability of ground evaporation and transpiration data, the pySEBAL algorithm was used in both methods. This study shows the superiority of the first method over methods based on empirical equations in the region. Consequently, combining experimental methods with remote sensing data is an applicable approach way to sensible improve the results.
