Spatial Prediction Of Crop Growth Model Parameters Based On Environmental Similarity

Nowadays,with the global climate changes,environmental problems such as heavy precipitation,increasing temperatures,and desertification could greatly affect global food and energy security.The crop growth models as important tools are widely used in global climate change research to explore the impact of global climate change and support decisions for agricultural adaptation.The crop growth model was originally developed and applied on the field scale,and the natural factors such as climatic and soil conditions were assumed to be homogeneous.But when the crop model is applied in the global climate change research,the spatial scale would be raised form field scale to the regional or global scale.The key input parameters in the crop growth model will be spatial heterogeneous with the increasing space scale.Therefore,it is a necessary to get the large-scale and high-precision spatial distribution information of crop growth model parameters when the crop model is applied to large-scale.Now most studies focus on the spatialization of crop model parameters such as phenological parameters(e.g.,growing degree days of growing season(GDD),etc.)and management parameters(e.g.,planting date and harvest date,etc.).There are two main methods widely used for spatial prediction in this context.One is temperature threshold method and the other is regression analysis.These two methods have one key requirement which is the stationarity of the extracted relationship/threshold over the whole study area.However,actually,the requirement would not be fully met for spatial prediction with complex geographic processes.In order to overcome the requirement of the existing methods,in this thesis a method based on the similarity in geographic environment(The Third Law of geography)was developed for spatial prediction of crop model parameters.The basic idea based on environmental similarity estimation method can be simply summarized as "the more similar the geographic environment is between two locations,the more similar the geographic feature is between them".There are three steps for spatial prediction of crop model parameters based on environmental similarity.The first step is characterizing the environmental characteristics of the target variable and calculating the environmental similarity between the samples or the sample and the point of prediction.The second step is calculating the credibility of each sample based on environmental similarity and selecting the highly reliable samples for prediction.The third step is predicting the target variable value and calculating uncertainty value of the prediction points according to the sample environmental similarity.The proposed method was evaluated in a case study in maize planting area in China.Firstly,the phenological parameters(i.e.GDD)was predicted based on the environmental similarity.In order to compare with other method,the regression analysis was selected.For environmental similarity method,the Root mean Square Error(RMSE)of estimated GDD of spring maize and summer maize GDD were 105.8?days and 104.3? days,respectively,and the RMSEs derived from temperature threshold method were 182.7? days and 111.8? days,respectively.The results were showed that the predicting accuracy of the environmental similarity method is higher than the regression analysis method.Under the influence of global changes,this method can provide reference for farmers to choose crop cultivar.Secondly,the planting date in the management parameters was predicted based on the environmental similarity.The temperature threshold estimation method and multiple regression method were selected for comparative analysis.For environmental similarity method,the RMSEs of planting date of spring maize and summer maize were 9.8 days and 7.3 days,respectively.The RMSEs obtained by temperature threshold method were 15.9 days and 31.5 days,respectively.And the RMSEs obtained by multiple regression analysis method were 12.6 days and 11.0 days,respectively.The results were showed that the predicting accuracy of environmental similarity method was highest than other methods.Finally,the predictive scale and threshold settings in the environmental similarity estimation method were discussed.In the region where the target variable had large difference,the accuracy of the whole region was better than the partition prediction.For the threshold setting in the predictive process,the appropriate parameter values were chosen according to different needs and the acceptability of the error.In summary,the method based on environmental similarity for spatial prediction of crop model parameters can not only efficiently and accurately predict the multi-scale crop model growth model parameters,but also fully reflect the spatial heterogeneity of parameters.So it can provide reasonable parameter support for large-scale crop models.And it can provide a scientific basis and theoretical basis for regional and national adaptation of agricultural policies under global climate change.