AgroFlux: A Spatial-Temporal Benchmark for Carbon and Nitrogen Flux Prediction in Agricultural Ecosystems

TL;DR

AgroFlux tackles the challenge of quantifying carbon and nitrogen fluxes in agricultural ecosystems by delivering a spatial-temporal benchmark that fuses physics-based simulations from Ecosys and DayCent with real-world observations. It defines standardized benchmark tasks (simulated prediction, observational prediction, and transfer learning) and uses common metrics such as $R^2$, RMSE, and MAE to enable fair comparisons. Baseline results show that LSTM excels in temporal extrapolation for ecosystem-like data, while Transformer variants perform best in spatial generalization, with $N_2O$ flux remaining the most challenging due to management-driven variability. By providing comprehensive data, evaluation protocols, and transfer-learning baselines, AgroFlux aims to catalyze robust AI approaches for agroecosystem–climate interactions and support data-driven decision-making in sustainable agriculture.

Abstract

Agroecosystem, which heavily influenced by human actions and accounts for a quarter of global greenhouse gas emissions (GHGs), plays a crucial role in mitigating global climate change and securing environmental sustainability. However, we can't manage what we can't measure. Accurately quantifying the pools and fluxes in the carbon, nutrient, and water nexus of the agroecosystem is therefore essential for understanding the underlying drivers of GHG and developing effective mitigation strategies. Conventional approaches like soil sampling, process-based models, and black-box machine learning models are facing challenges such as data sparsity, high spatiotemporal heterogeneity, and complex subsurface biogeochemical and physical processes. Developing new trustworthy approaches such as AI-empowered models, will require the AI-ready benchmark dataset and outlined protocols, which unfortunately do not exist. In this work, we introduce a first-of-its-kind spatial-temporal agroecosystem GHG benchmark dataset that integrates physics-based model simulations from Ecosys and DayCent with real-world observations from eddy covariance flux towers and controlled-environment facilities. We evaluate the performance of various sequential deep learning models on carbon and nitrogen flux prediction, including LSTM-based models, temporal CNN-based model, and Transformer-based models. Furthermore, we explored transfer learning to leverage simulated data to improve the generalization of deep learning models on real-world observations. Our benchmark dataset and evaluation framework contribute to the development of more accurate and scalable AI-driven agroecosystem models, advancing our understanding of ecosystem-climate interactions.

Figures (7)

• Figure 1: Temporal variations in N$_2$O flux (top), GPP (middle), and CO$_2$ flux (bottom) from both simulated and observation sets. The unit for all three variables is (g C m$^{-2}$ day$^{-1}$).
• Figure 2: Comparison of LSTM predictions: trained from scratch versus fine-tuned (FT).
• Figure 3: LSTM prediction performance on the Ecosys dataset averaged from 2016 to 2018.
• Figure 4: Feature distributions for the first five input features (TMAX, TMIN, PREC, FERTZR_N, PDOY) across four datasets (Ecosys, DayCent, N$_2$O Observations, CO$_2$/GPP Observations as the columns). For each subplot, the X-axis represents value intervals, and Y-axis represents frequencies. Values are standardized and clipped to the 1st--99th percentile, and legends report n, mean, and std.
• Figure 5: Feature distributions for the last six input features (PLANTT, TBKDS, TCSAND, TCSILT, TPH, TSOC) across four datasets (Ecosys, DayCent, N$_2$O Observations, CO$_2$/GPP Observations as the columns). For each subplot, the X-axis represent value intervals, and Y-axis represent frequencies. Values are standardized and clipped to the 1st-–99th percentile, and legends report n, mean, and std.