You are here
Harvard Forest REU Symposium Abstract 2022
- Title: Short-term iterative forecasting of carbon fluxes in Harvard Forest: validating and assessing soil respiration forecasts of the SIPNET model
- Author: Courtney J Leung (University of Chicago)
- Abstract:
Temperate forests account for the majority of terrestrial carbon dioxide storage, and much of this carbon is stored below ground in the soil. However, there is much uncertainty in the carbon budget of these forest ecosystems, driving modeling efforts to quantify the fluxes of various carbon and water processes. In the past decade, near-term forecasting has become a potential method for performing short-term forecasts that can be readily validated and improved using real-time data. Much work is needed to evaluate these models’ accuracy and sensitivity to the variables used to make sure predictions. In this study, I validated 35-day carbon-cycle forecasts generated using the SIPNET model against soil respiration data at Harvard Forest in 2021, comparing the predictions to NEON observations and assessing the model’s sensitivity to model parameters and climate variables. I found that uncertainty, measured using RMSE and MAE, remains relatively constant until its observed increase at a lead time of around 18 days, when the model begins to underpredict soil respiration. Using CRPS scores, uncertainty initially decreases with forecast lead time and is generally highest for mid-day predictions of soil respiration. The SIPNET model is most sensitive to soil temperature, base soil respiration, and the soil respiration Q10 value. This suggests that the SIPNET forecast could be improved through a recalibration that includes the respiration data and the soil respiration basal rate and Q10 parameter. Future steps should also seek to include soil respiration observations in the daily data assimilation process to help improve iterative predictions.
- Research Category: Ecological Informatics and Modelling; Forest-Atmosphere Exchange; Soil Carbon and Nitrogen Dynamics
