Research On Winter Wheat Yield Estimation At Multi-scale With Remote Sensing

The rapid and accurate grasp of crop yield information is not only of great value for the formulation of national food security policy,macro-control of market prices,rural economic development,and foreign grain trade,but also for the intelligent management of farmland production and agricultural insurance.Therefore,timely and accurate forecasting and estimation of crop yield information is a scientific and technological issue that is closely related to the national economy and people’s livelihood.In particular,the accurate grasp of its spatial pattern changes not only is conducive to the rapid development of precision agriculture,agricultural insurance,and other related industries,but also can promote the effective implementation of policies(e.g.“science and technology driven agriculture”),and the popularization and application of agricultural technology and its effect on increasing farmers’ production and income.Remote sensing technology has great potential in crop yield estimation because of its significant advantages such as rapidness,broad coverage,and non-contact nature.It has been gradually replacing traditional sampling survey methods.In terms of remote sensing based crop yield estimation,especially for large-area crop yield estimation,the following issues can be identified:(1)Although empirical statistics models can achieve high accuracy in a small range,the regional scalability and stability of the model are poor due to lack of agricultural knowledge;(2)The crop growth model carries out a full range of mathematical simulations of the crop growth process,but the excessive input parameters limit its wide application in different regions;(3)Existing models are highly dependent on field yield measurement samples,and it is difficult to realize intelligent and automatic estimation of regional yield.In order to solve the above issues,we focus on winter wheat in Hebei Province and proposed fast,stable,and accurate methods for county-level and pixel-level yield estimation.For county-level yield estimation,empirical models are used,based on multi-source data(such as remote sensing,meteorology,and statistics)to explore its influence on winter wheat yield estimation through combined analysis of different time phases(P1-P5)and agricultural conditions.A yield model was constructed based on the combination of optimal factors to realize the county-level yield estimation of winter wheat in Hebei Province.For pixel-level yield estimation,this paper proposes a yield estimation method based on multi-scenario growth simulation.First,by inputting all the scenarios(crop variety,weather,soil,management)that may occur in the area to the crop growth model,the production process of winter wheat under different conditions in the area can be simulated as much as possible to obtain a simulated data set.Second,the simulated data set was used to train the random forest regression model of different date combinations to form a model set.Finally,the LAI retrieved from Sentinel-2 images was feed into the model set to obtain regional yields.The main conclusions are as follows:(1)For the estimation of winter wheat yield at county-level in Hebei Province,P2,P3 and P4 are associated with higher accuracy than P1 and P5 in terms of single phases.The model accuracy using multi-phase features is higher than that of the single-phase,and the combination of P2 and P4 is the best.Among all features,crop growth features had the greatest impact on yield estimation accuracy,while the addition of environmental forcing factors(e.g.,water,light and temperature)did not significantly improve the accuracy.The addition of farmland landscape features can effectively improve the accuracy of yield estimation.(2)Five important features(PROP,NDVI＿P2,B2＿P2,ED,and B1＿P4)were selected,and a yield estimation model was established to obtain the county-level yield of winter wheat in Hebei Province.It was shown that the Mean Relative Error(MRE)can reach as low as 2.85%,and the Root Mean Square Error(RMSE)and R2 was respectively 253.25 kg/ha and 0.83.This study provides insights and new methods for nationwide estimation of winter wheat yield at county-level.(3)The WOFOST model can simulate the yield formation process of winter wheat in Hebei Province well under different scenarios(variety,climate,soil,management).The simulated data set generated on this basis can effectively solve the problem of dependence on the ground yield samples.(4)The comparison between the forecast results and the field survey results showed a high consistency.The results showed that R2 is 0.54,the root mean square error is 389.67 kg/ha,and the average relative error is 6.07%.The forecast results also shows good consistency with county-level statistical data.The mean relative error is 4.98%,R2 is 0.57,and the root mean square error is 345.53 kg/ha.This method does not rely on ground yield survey data,can be dynamically expanded,and is suitable for applications in different spatial scales,thus providing a new idea for crop yield estimation in large areas.