Net ecosystem carbon exchange prediction and insightful data mining with an optimized data-matching algorithm
Net ecosystem carbon exchange (NEE) measures the carbon interchanges between the Earth’s biosphere and atmosphere. NEE datasets for two northern European sites (730 and 413 data records) incorporating twenty-two meteorological and environmental data influencing variables, collected on a daily basis...
Guardado en:
| Autor principal: | |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
Elsevier
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/7618048a60d64d9f8334dd27d71da931 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Net ecosystem carbon exchange (NEE) measures the carbon interchanges between the Earth’s biosphere and atmosphere. NEE datasets for two northern European sites (730 and 413 data records) incorporating twenty-two meteorological and environmental data influencing variables, collected on a daily basis for years spread across the period 1997 to 2013, are evaluated by an optimized data matching machine learning algorithm to predict NEE and data mine the datasets. The model’s transparency and avoidance of regressions/hidden correlations facilitates detailed data mining of the dataset exploiting two distinct objective functions. This reveals useful insights concerning similarities and influences between specific data records. Cumulative absolute error and squared error trends of predictions enable areas of the NEE distribution, predicted to different degrees of accuracy, to be identified. Such trends also facilitate detailed comparisons of the prediction calculation of each data record. The prediction accuracy achieved by the algorithm for the UK-Gri dataset (MAE = 0.6898 gC m−2 d−1; RMSE = 0.9558; R2 = 0.8903) and the hybrid UK-Gri plus NL-Loo dataset (MAE = 0.5072 gC m−2 d−1; RMSE = 0.7746; R2 = 0.9149) substantially outperform the NEE prediction accuracies achieved by four regression-based machine learning algorithms applied to the exact set of data records. |
|---|