HYBRID MODELING FOR CARBON SEQUESTRATION IN AGROFORESTRY

Abstract

This paper highlights the process of combining processes based biogeochemical modelling with machine learning technique in order to develop a hybrid model that has the capability of quantifying the amount of carbon stored in Agroforestry systems.  Based on field measurements of soil organic carbon (SOC), biomass and canopy structure, we acquired a powerful predictive model using multispectral remote sensing data.  The process model demonstrated the dynamics of carbon flows in time and the machine learning model was more accurate in forecast because it considered spatial disparities and it eradicated the structural bias.  In the case of having calibrated and validation datasets, the model was very accurate in making prediction as it had a smaller RMSE and R2R2 than when it was projected alone.  Simulations of scenarios revealed that the key difference between how much carbon could be stored and depending on not-so-good agroforestry practices substantially increased when the more-good agroforestry practices (such as planting more trees, growing different crops, and taking better care of the soil) were included.  Farmers interviews provided us with qualitative data which aided in deciphering the quantitative findings and related the output of the model to what can happen and what will potentially occur in the real environment.  The findings indicate that agroforestry is a climate smart land use and it is beneficial to the environment and rural livelihoods.  This research provides policymakers and land managers with a decision-making support framework that can be applied at a larger scale to ensure agroforestry works better towards carbon-reduction and sustainable development objectives.