Mon, Aug 15, 2022: 1:45 PM-2:00 PM
513A
Background/Question/Methods
There is a growing desire for accurate near-term forecasts of the carbon and water cycles. We used NOAA GEFS weather forecasts to predict carbon and water cycle pools and fluxes 35 days into the future using the Simplified Photosynthesis and Evapotranspiration Model (SIPNET). The forecast system then statistically assimilates net ecosystem exchange (NEE), latent heat (LE), and MODIS LAI, adjusting the model state variables used as initial conditions for the subsequent forecasts. We tested this forecast system at ten eddy-covariance sites that encompass a range of ecosystem types (forest, grassland, agriculture, semi-arid shrubland). Finally, we assessed the predictability of NEE, LE, and LAI as a function of forecast lead time, time of day, and season.
Results/Conclusions
Predictability was assessed by calculating forecast RMSE, Bias, and CRPS scores for NEE, LE, and LAI compared to observed values. The predictability of NEE, LE, and LAI all decreased as forecast lead time increases, but improved during the summer and winter months compared to fall and spring. Tower fluxes had lower RMSE and CRPS scores during the day compared to the night by a factor of 4 and 3 respectively. While bias oscillated from slightly positive during the night to slightly negative during the day. Overall error declines, as lead time increases, with increased numbers of data constraints. Building upon this initial analysis we will extend our prototype system to assimilate more data streams across a larger number of forecast locations.
There is a growing desire for accurate near-term forecasts of the carbon and water cycles. We used NOAA GEFS weather forecasts to predict carbon and water cycle pools and fluxes 35 days into the future using the Simplified Photosynthesis and Evapotranspiration Model (SIPNET). The forecast system then statistically assimilates net ecosystem exchange (NEE), latent heat (LE), and MODIS LAI, adjusting the model state variables used as initial conditions for the subsequent forecasts. We tested this forecast system at ten eddy-covariance sites that encompass a range of ecosystem types (forest, grassland, agriculture, semi-arid shrubland). Finally, we assessed the predictability of NEE, LE, and LAI as a function of forecast lead time, time of day, and season.
Results/Conclusions
Predictability was assessed by calculating forecast RMSE, Bias, and CRPS scores for NEE, LE, and LAI compared to observed values. The predictability of NEE, LE, and LAI all decreased as forecast lead time increases, but improved during the summer and winter months compared to fall and spring. Tower fluxes had lower RMSE and CRPS scores during the day compared to the night by a factor of 4 and 3 respectively. While bias oscillated from slightly positive during the night to slightly negative during the day. Overall error declines, as lead time increases, with increased numbers of data constraints. Building upon this initial analysis we will extend our prototype system to assimilate more data streams across a larger number of forecast locations.