Votre recherche
Résultats 4 ressources
-
Abstract Digital leaf physiognomy (DLP) is considered as one of the most promising methods for estimating past climate. However, current models built using the DLP data set still lack precision, especially for mean annual precipitation (MAP). To improve predictive power, we developed five machine learning (ML) models for mean annual temperature (MAT) and MAP respectively, and then tested the precision of these models and some of their averaging compared with that obtained from other models. The precision of all models was assessed using a repeated stratified 10‐fold cross‐validation. For MAT, three combinations of models ( R 2 = .77) presented moderate improvements in precision over the multiple linear regression (MLR) model ( R 2 = .68). For log e (MAP), the averaging of the support vector machine (SVM) and boosting models improved the R 2 from .19 to .63 compared with that of the MLR model. For MAP, the R 2 of this model combination was 0.49, which was much better than that of the artificial neural network (ANN) model ( R 2 = .21). Even the bagging model, which had the lowest R 2 (.37) for log e (MAP), demonstrated better precision ( R 2 = .27) for MAP. Our palaeoclimate estimates for nine fossil floras were also more accurate, because they were in better agreement with independent paleoclimate evidence. Our study confirms that our ML models and their averaging can improve paleoclimatic reconstructions, providing a better understanding of the relationship between climate and leaf physiognomy.
-
Paleobotanists have long built leaf climate models based on site mean of leaf physiognomic characteristics of woody dicotyledons species (WDS) for estimating past climate. To explore the potential of the order Ericales in estimating paleoclimate, we developed two linear models for each climatic factor. One is based on WDS, and the other is based on both WDS and leaf physiognomic characters of the order Ericales (WDS-E). We found that, compared with WDS models, WDS-E models improved greatly in mean annual precipitation (MAP), growing season precipitation (GSP) and mean annual range in temperature (MART). When the minimum species number of the order Ericales is three per site, the WDS-E models improved the r2 from 0.64 to 0.78 for MART, from 0.23 to 0.61 for ln(MAP), and from 0.37 to 0.64 for ln(GSP) compared with the WDS models. For mean annual temperature (MAT), the WDS-E model (r2 = 0.86) also exhibited a moderate improvement in precision over the WDS model (r2 = 0.82). This study demonstrates that other patterns, such as those of the order Ericales, can contribute additional information towards building more precise paleoclimate models.
-
Summary For decades, researchers have thought it was difficult to remove the uncertainty from the estimates of forest carbon storage and its changes on national sales. This is not only because of stochasticity in the data but also the bias to overcome in the computations. Most studies of the estimation, however, ignore quantitative analyses for the latter uncertainty. This bias primarily results from the widely used volume‐biomass method via scaling up forest biomass from limited sample plots to large areas. This paper addresses (i) the mechanism of scaling‐up error occurrence, and (ii) the quantitative effects of the statistical factors on the error. The error compensators were derived, and expressed by ternary functions with three variables: expectation, variance and the power in the volume‐biomass equation. This is based on analysing the effect of power‐law function convexity on scaling‐up error by solving the difference of both sides of the weighted Jensen inequality. The simulated data and the national forest inventory of China were used for algorithm testing and application, respectively. Scaling‐up error occurrence stems primarily from an effect of the distribution heterogeneity of volume density on the total biomass amount, and secondarily from the extent of function nonlinearities. In our experiments, on average 94·2% of scaling‐up error can be reduced for the statistical populations of forest stands in a region. China's forest biomass carbon was estimated as approximately 6·0 PgC or less at the beginning of the 2010s after on average 1·1% error compensation. The results of both the simulated data experiment and national‐scale estimation suggest that the biomass is overestimated for young forests more than others. It implies a necessity to compensate scaling‐up error, especially for the areas going through extensive afforestation and reforestation in past decades. This study highlights the importance of understanding how both the function nonlinearity and the statistics of the variables quantitatively affect the scaling‐up error. Generally, the presented methods will help to translate fine‐scale ecological relationships to estimate coarser scale ecosystem properties by correcting aggregation errors.
-
Phosphorus (P) is a key and a limiting nutrient in ecosystems and plays an important role in many physiological and biochemical processes, affecting both terrestrial ecosystem productivity and soil carbon storage. However, only a few global land surface models have incorporated P cycle and used to investigate the interactions of C-N-P and its limitation on terrestrial ecosystems. The overall objective of this study was to integrate the P cycle and its interaction with carbon (C) and nitrogen (N) into new processes model of TRIPLEX-CNP. In this study, key processes of the P cycle, including P pool sizes and fluxes in plant, litter, and soil were integrated into a new model framework, TRIPLEX-CNP. We also added dynamic P:C ratios for different ecosystems. Based on sensitivity analysis results, we identified the phosphorus resorption coefficient of leaf (rpleaf) as the most influential parameter to gross primary productivity (GPP) and biomass, and determined optimal coefficients for different plant functional types (PFTs). TRIPLEX-CNP was calibrated with 49 sites and validated against 116 sites across eight biomes globally. The results suggested that TRIPLEX-CNP performed well on simulating the global GPP and soil organic carbon (SOC) with respective R2 values of 0.85 and 0.78 (both p < 0.01) between simulated and observed values. The R2 of simulation and observation of total biomass are 0.67 (p < 0.01) by TRIPLEX-CNP. The overall model performance had been improved in global GPP, total biomass and SOC after adding the P cycle comparing with the earlier version. Our work represents the promising step toward new coupled ecosystem process models for improving the quantifications of land carbon cycle and reducing uncertainty.