Predicting Soil Microbial Biomass Carbon Using Stacking Machine Learning Techniques to Enhance Soil Health

Abstract

Soil microbial biomass carbon (SMBC) is an important factor that affects soil fertility ,biogeochemical cycling and also carbon elimination from atmosphere .Traditionally, the determination of SMBC has been labor - intensive and complex, relying on methods such as the chloroform fumigation-extraction technique, which are both time-consuming and error-prone. The most recent breakthroughs in artificial intelligence have revealed a promising application of the use of AI for the automation and enhancement of the precision of SMBC estimation, especially by the application of deep learning models, such as ANN. Yet, the development of AI applications in this field is relatively underdeveloped and less explored in terms of its practical application and performance. This paper presents an innovative approach for SMBC values using machine learning techniques for improved precision. We use a stacking method, which utilizes advantages of a two machine learning models (lightgbm for numerical features and catboost for categorical features) with a meta-learner , such as Random Forest. The results are promising, as our method yields an R-squared value of 0.75 and an MSE of 0.23, thus making it a useful tool for SMBC estimation. Such approach shows significant steps forward on the challenges faced by other classical approaches; thus, they will represent more efficient, reliable alternatives in large scale assessments for soil health along with environmental monitoring.