Limits...
The Impact of Fine-Scale Disturbances on the Predictability of Vegetation Dynamics and Carbon Flux.

Hurtt GC, Thomas RQ, Fisk JP, Dubayah RO, Sheldon SL - PLoS ONE (2016)

Bottom Line: While large extreme events such as tropical cyclones, fires, or pest outbreaks can produce dramatic consequences, small fine-scale disturbance events are typically much more common and may be as or even more important.We found that predicted height change from a height-structured ecosystem model compared well to lidar measured height change at the domain scale (~150 ha), but that the model-data mismatch increased exponentially as the spatial scale of evaluation decreased below 20 ha.We demonstrate that such scale-dependent errors can be attributed to errors predicting the pattern of fine-scale forest disturbances.

View Article: PubMed Central - PubMed

Affiliation: Department of Geographical Sciences, University of Maryland, College Park, MD, United States of America.

ABSTRACT
Predictions from forest ecosystem models are limited in part by large uncertainties in the current state of the land surface, as previous disturbances have important and lasting influences on ecosystem structure and fluxes that can be difficult to detect. Likewise, future disturbances also present a challenge to prediction as their dynamics are episodic and complex and occur across a range of spatial and temporal scales. While large extreme events such as tropical cyclones, fires, or pest outbreaks can produce dramatic consequences, small fine-scale disturbance events are typically much more common and may be as or even more important. This study focuses on the impacts of these smaller disturbance events on the predictability of vegetation dynamics and carbon flux. Using data on vegetation structure collected for the same domain at two different times, i.e. "repeat lidar data", we test high-resolution model predictions of vegetation dynamics and carbon flux across a range of spatial scales at an important tropical forest site at La Selva Biological Station, Costa Rica. We found that predicted height change from a height-structured ecosystem model compared well to lidar measured height change at the domain scale (~150 ha), but that the model-data mismatch increased exponentially as the spatial scale of evaluation decreased below 20 ha. We demonstrate that such scale-dependent errors can be attributed to errors predicting the pattern of fine-scale forest disturbances. The results of this study illustrate the strong impact fine-scale forest disturbances have on forest dynamics, ultimately limiting the spatial resolution of accurate model predictions.

Show MeSH

Related in: MedlinePlus

Root mean squared error between the ΔCTH lidar measurement and the ED model prediction as a function of spatial scale.The model-data comparison at coarse scales (i.e. >50 ha) has relatively low error (RMSE). The error increases rapidly as the spatial resolution of comparison increases. Contours denote the additional expected average RMSE using the simulator at each scale assuming potential systematic bias errors in predicted growth or mortality. Only results with no systematic bias in growth or mortality produce results of similar magnitude as those predicted by ED.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4836756&req=5

pone.0152883.g005: Root mean squared error between the ΔCTH lidar measurement and the ED model prediction as a function of spatial scale.The model-data comparison at coarse scales (i.e. >50 ha) has relatively low error (RMSE). The error increases rapidly as the spatial resolution of comparison increases. Contours denote the additional expected average RMSE using the simulator at each scale assuming potential systematic bias errors in predicted growth or mortality. Only results with no systematic bias in growth or mortality produce results of similar magnitude as those predicted by ED.

Mentions: The spatial scale of comparison had a strong influence on the accuracy of model predictions (Fig 5). At the coarsest scale analyzed, 150 ha, the predicted canopy height change was in close agreement to observed (RMSE<0.25 m). Prediction error increased non-linearly with increasing spatial resolution. At 20 ha, prediction error was approximately twice the result at the coarsest scale, and at 1 ha the prediction error was nearly ten times higher (RMSE<2.69 m).


The Impact of Fine-Scale Disturbances on the Predictability of Vegetation Dynamics and Carbon Flux.

Hurtt GC, Thomas RQ, Fisk JP, Dubayah RO, Sheldon SL - PLoS ONE (2016)

Root mean squared error between the ΔCTH lidar measurement and the ED model prediction as a function of spatial scale.The model-data comparison at coarse scales (i.e. >50 ha) has relatively low error (RMSE). The error increases rapidly as the spatial resolution of comparison increases. Contours denote the additional expected average RMSE using the simulator at each scale assuming potential systematic bias errors in predicted growth or mortality. Only results with no systematic bias in growth or mortality produce results of similar magnitude as those predicted by ED.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4836756&req=5

pone.0152883.g005: Root mean squared error between the ΔCTH lidar measurement and the ED model prediction as a function of spatial scale.The model-data comparison at coarse scales (i.e. >50 ha) has relatively low error (RMSE). The error increases rapidly as the spatial resolution of comparison increases. Contours denote the additional expected average RMSE using the simulator at each scale assuming potential systematic bias errors in predicted growth or mortality. Only results with no systematic bias in growth or mortality produce results of similar magnitude as those predicted by ED.
Mentions: The spatial scale of comparison had a strong influence on the accuracy of model predictions (Fig 5). At the coarsest scale analyzed, 150 ha, the predicted canopy height change was in close agreement to observed (RMSE<0.25 m). Prediction error increased non-linearly with increasing spatial resolution. At 20 ha, prediction error was approximately twice the result at the coarsest scale, and at 1 ha the prediction error was nearly ten times higher (RMSE<2.69 m).

Bottom Line: While large extreme events such as tropical cyclones, fires, or pest outbreaks can produce dramatic consequences, small fine-scale disturbance events are typically much more common and may be as or even more important.We found that predicted height change from a height-structured ecosystem model compared well to lidar measured height change at the domain scale (~150 ha), but that the model-data mismatch increased exponentially as the spatial scale of evaluation decreased below 20 ha.We demonstrate that such scale-dependent errors can be attributed to errors predicting the pattern of fine-scale forest disturbances.

View Article: PubMed Central - PubMed

Affiliation: Department of Geographical Sciences, University of Maryland, College Park, MD, United States of America.

ABSTRACT
Predictions from forest ecosystem models are limited in part by large uncertainties in the current state of the land surface, as previous disturbances have important and lasting influences on ecosystem structure and fluxes that can be difficult to detect. Likewise, future disturbances also present a challenge to prediction as their dynamics are episodic and complex and occur across a range of spatial and temporal scales. While large extreme events such as tropical cyclones, fires, or pest outbreaks can produce dramatic consequences, small fine-scale disturbance events are typically much more common and may be as or even more important. This study focuses on the impacts of these smaller disturbance events on the predictability of vegetation dynamics and carbon flux. Using data on vegetation structure collected for the same domain at two different times, i.e. "repeat lidar data", we test high-resolution model predictions of vegetation dynamics and carbon flux across a range of spatial scales at an important tropical forest site at La Selva Biological Station, Costa Rica. We found that predicted height change from a height-structured ecosystem model compared well to lidar measured height change at the domain scale (~150 ha), but that the model-data mismatch increased exponentially as the spatial scale of evaluation decreased below 20 ha. We demonstrate that such scale-dependent errors can be attributed to errors predicting the pattern of fine-scale forest disturbances. The results of this study illustrate the strong impact fine-scale forest disturbances have on forest dynamics, ultimately limiting the spatial resolution of accurate model predictions.

Show MeSH
Related in: MedlinePlus