A model-data fusion approach for quantifying the carbon budget in cotton agroecosystems across the United States

Abstract

Cotton (Gossypium hirsutum L.) cultivation contributes to economic development, particularly in the Cotton Belt of the Southern United States (U.S.). As one of the world's largest exporters of cotton, the U.S. cotton industry plays a pivotal role in both the domestic and international markets. Accurate quantification of carbon budgets and their responses to the environment is thus crucial for the sustainable production of cotton, but such quantification at the regional scale remains unclear. Here we use a framework that combines an advanced process-based model, ecosys, and a deep learning-based Model-Data Fusion (MDF) approach to quantify the magnitude and patterns of carbon flux and cotton lint yield under both rainfed and irrigated conditions in the U.S. We first evaluate the performance of the process-based model in simulating carbon budgets of cotton agroecosystems using eddy-covariance (EC) values at production-scale farm sites. We then apply MDF to use satellite-based gross primary production (GPP) and survey-based cotton lint yield data as constraints of the ecosys model to generate the holistic carbon budget of cotton cropland at the county level across the U.S. from 2008 to 2019. Validation at the three EC sites indicates that the ecosys model achieves R² values of 0.9 and 0.8 for the simulated versus the EC daily GPP and respiration, respectively, and 0.9 for the simulated versus the experimentally measured leaf area index. The R² at county level in our framework is 0.8 for both cotton lint yield and GPP: the simulated versus survey-based cotton lint yield, and the simulated versus satellite-based monthly GPP. The spatio-temporal patterns of the simulated cotton lint yield, GPP, and their responses to climate factors (average temperature, average vapor pressure deficit (VPD), and cumulative precipitation during the growing season) are consistent with the observations, indicating that our framework approach captures the underlying processes relating environmental conditions to cotton growth. Our analysis shows that cotton productivity (lint yield and GPP) decreased with increasing average VPD during the growing season, especially under rainfed conditions. It also shows that the carbon budget terms, including predicted net primary productivity, crop yield, and soil heterotrophic respiration, decreased as the VPD increased. Conversely, the predicted change in soil organic carbon was less influenced by climate, which decreased with increasing initial soil organic carbon content and cation exchange capacity, and increased with increasing soil bulk density. The variable impacts of crop management practices, climatic factors, and soil characteristics on carbon budgets highlight the intricate interactions among these factors that shape carbon dynamics in cotton agroecosystems, and further emphasize the necessity of accurately simulating the carbon budgets of cotton agroecosystems across temporal and spatial scales. This study has established a framework that utilizes advanced MDF to assess climate mitigation strategies for U.S. cotton agroecosystems.