Simulating soil organic carbon and greenhouse gas emissions in dry regions: A review of the scope and limitations of biogeochemical models

Abstract

Soil organic carbon (SOC) dynamics and greenhouse gas (GHG) emissions in dry (arid and semi‐arid) agroecosystems are regulated by low and variable moisture, episodic wet–dry cycles, and limited organic inputs. Although process‐based biogeochemical models are widely used to quantify these processes, a comprehensive assessment of their performance in dryland agroecosystems is lacking. Therefore, we evaluated four widely used models, Agricultural Production Systems sIMulator (APSIM), daily century model (DayCent), DeNitrification‐DeComposition (DNDC), and decision support system for agrotechnology transfer (DSSAT), across dry regions defined by the United Nations Environment Program aridity index (AI < 0.65). Peer‐reviewed studies published between 2010 and 2026 were assessed using reported quantitative metrics, including R ² , Nash–Sutcliffe efficiency (NSE), and root mean square error (RMSE), for SOC, CO _2, and N ₂ O emissions, and crop yields. DayCent and APSIM consistently simulate long‐term SOC dynamics under input‐constrained conditions (DayCent R ² ≤ 0.99; APSIM R ²  = 0.92, RMSE = 3.33 Mg C ha ^{− 1} ). DNDC performed well for SOC in residue‐incorporated systems (NSE ≤ 0.84) but underestimated surface‐applied residue SOC by 5%–12%. DSSAT reliably simulated yields (normalized root mean square error [nRMSE] ≤ 19%–22%) but underestimated SOC gains under organic amendments by 5%–22%. For GHGs, DayCent and APSIM captured cumulative N ₂ O emissions reasonably ( R ² ≈ 0.7–0.8), whereas DNDC resolved management‐driven contrasts. Shared limitations across models include difficulty in estimating emission pulses following wetting events and in representing yield–soil feedbacks. Overall, model performance in dryland systems can be improved by representing agroecosystem processes and improving calibration, irrespective of the parameters simulated or the model selection.