Farmland soil reaction kinetics process model modeling method

Abstract

The invention discloses a farmland soil reaction kinetics process model modeling method. A coupling model of a hydrodynamic model and a chemical reaction dynamic model is adopted to describe a farmland soil reaction dynamic process; solution concentration distribution is predicted, the parameters of the coupling model are corrected through the iterative set Kalman filtering technology, a machine learning algorithm based on Gaussian process regression is sequentially integrated into a data assimilation framework to estimate the structural error of the coupling model, and a more reasonable priorvalue is provided for data assimilation, so that the parameter compensation effect is reduced, and the prediction capability of the coupling model is improved. The method solves the problem that model structure errors which are difficult to solve and cannot be ignored in farmland soil reaction dynamics data assimilation calculation, does not need to make any substantive assumption for the prior distribution of the model errors, and avoids parameter overfitting in the assimilation process.

Description

technical field [0001] The invention belongs to the field of farmland water and soil environment, and in particular relates to a modeling method for a farmland soil reaction kinetic process model. Background technique [0002] Understanding the dynamics of farmland soil reactions is crucial to revealing the real evolution of farmland water and soil environments. For this, researchers have developed a variety of hydrological-geobiochemical process mechanism models. However, due to the complexity of the process and the complexity of the environment (soil type, stratum structure, biological factors and parameters, etc.), people's understanding of the dynamic process of farmland soil reaction is very limited, so the process mechanism model built inevitably exists. Multiple sources of uncertainty limit the credibility and usefulness of model results. How to quantify and reduce the uncertainty of model prediction and improve the accuracy of model prediction is a research hotspot ...

AI Technical Summary

Problems solved by technology

Among them, the sequential data assimilation algorithm represented by ensemble Kalman filter (Ensemble Kalman Filter, EnKF) has been widely used because of its low computational cost and strong adaptability. However, due to the inability to give a reasonable description of the model structure error , in the actual data assimilation calculation, the model structure error is often ignored, the model itself is considered correct, and the model deviation is attributed to the uncertainty of the parameters (and input), or the structural error is simply regarded as zero on the model output variable Mean Gaussian error addition, which ignores or underestimates uncertainty in the model structure, can lead to systematic bias and overconfident model predictions

[0004] In order to identify and quantify model structural errors, researchers have made a lot of attempts, such as amplifying the background error covariance matrix and multi-model methods, etc., but these methods are limited by their expensive calculation costs and the quality and quantity of candidate models. dependency

Images