Rodolphe Palm

Testing coexistence of extinction debt and colonization credit in fragmented calcareous grasslands with complex historical dynamics

Calcareous grasslands are among the most species-rich ecosystems in temperate countries. However, these ecosystems have suffered from fragmentation and destruction during the last century. We studied the response of calcareous grassland plant diversity to landscape changes in Belgium. Results indicated that high area loss (since 1965) old habitat patches exhibited an extinction debt inverse to low area loss old habitat patches, little depending on the area loss threshold (60%, 70%, 80% or 90%) considered for the distinction between the high and low area loss patches. However, human activities also created new habitat patches in the landscape and therefore provided opportunities for calcareous grassland plant species to colonize new habitats. This also provided opportunities to study species colonization abilities in the context of habitat restoration. We analyzed species richness in new patches compared to old patches in order to detect colonization credit. We detected the presence of a colonization credit in new patches when using high loss old patches (area loss >80%, exhibiting an extinction debt) or all old patches as a reference. However, when the reference was low loss old patches alone (area loss <80%, less likely to exhibit an extinction debt), no colonization credit was detected. In addition, species composition was similar between new patches and old patches. These results are encouraging for restoration programs. However, the results indicated that the presence of an extinction debt in reference habitats could lead to inaccurate conclusions in restoration monitoring. Therefore, extinction debt should be considered when choosing reference habitats to evaluate restoration success.

Canonical variables of aquatic bryophyte combinations for predicting water trophic level

A method for predicting water quality by linear regression using aquatic bryophyte canonical variables as predictors is presented. An example of application in the Alsatian Rhine valley is developed. The r-squared obtained before and after cross-validation reached 0.68 and 0.58 for the standard deviation of temperatures, 0.55 and 0.46 for the logarithm of the mean concentrations of N-NH4+, 0.52 and 0.43 for the logarithm of the mean concentrations of N-NO3− and 0.38 and 0.31 for the logarithm of the mean concentrations of P-PO43−. Higher r-squared were not expected due to the broad physico-chemical ranges and the low diversity of genuine aquatic bryophytes. The predicted values of the mean concentrations of N-NH4+, P-PO43− and of the standard deviation of temperatures were often greater than the measured ones. The aquatic bryophytes integrate the sudden increase of the trophic level in oligotrophic streams during the floods of the main eutrophic river and testify to a higher trophic level than expected from regular physico-chemical analyses outside the flood period. Aquatic bryophytes are also affected by water quality in the long term and indicate pollutants other than those measured in current water quality. Other factors besides trophic level might influence the aquatic bryophyte assemblages and should be monitored in order to find the precise relationships between water quality and aquatic macrophytes and in order to create a more accurate model of the effects of the flooding of disconnected streams by the Rhine waters (currently in progress in the Upper Rhine) on the aquatic macrophyte assemblages.

A sorting optimization curve with quality and yield requirements

Binary classifiers used for sorting can be compared and optimized using receiver-operating characteristic (ROC) curves which describe the trade-off between the false positive rate and true positive rate of the classifiers. This approach is well suited for the diagnosis of human diseases where individual costs of misclassification are of great concern. While it can be applied to the sorting of merchandise or other materials, the variables described by the ROC curve and its existing alternatives are less relevant for that range of applications and another approach is needed. In this paper, quality and yield factors are introduced into a sorting optimization curve (SOC) for the choice of the operating point of the classifier, associated with the prediction of output quantity and quality. Given examples are the sorting of seeds and apples with specific requirements. In both cases the operating point of the classifier is easily chosen on the SOC, while the output characteristics of the sorted product are accurately predicted.

Compared Regression Methods For Inferring Ammonium Nitrogen Concentrations In Running Freshwaters From Aquatic Bryophyte Assemblages

The cross-validated r-squared (hereafter, r2cv) was used to compare the performance of weighted averaging (WA) of species indicator values, ordinary least squares regression without variable selection (OLS) and stepwise regression (STEPWISE), principal components regression (PCR), regression on correspondence analysis components (CA) and partial least squares regression (PLS) to infer N-NH4+ mean concentrations from assemblages of 10 aquatic bryophyte species in calcareous running freshwaters in four hydrographic networks. The combination of all the data into a fifth data set provided a N–NH4+ spectrum covering the whole range of the species. The investigated methods exhibited very contrasting performances. As a function of the data sets, the highest r2cv were obtained by STEPWISE, PLS and PCR. However, only the regression on CA components, PCR and PLS gave positive r2cv values whatever the data set. It was thus impossible to select a priori the method producing the lowest prediction error for all the investigated data sets. Among the methods producing the lowest prediction errors, PLS, PCR and regression on CA components provided acceptable predictions for all the investigated data sets. These techniques, although not producing the best predictions in all the cases, could therefore be seen as the best compromise between prediction accuracy and constancy of the prediction quality.

Efficiency of inventory plot patterns in quantitative analysis of vegetation: a case study of tropical woodland and dense forest in Benin

The main issue in forest inventory is the reliability of data collected, which depends on the shape and size of inventoried plots. There is also a need for harmonisation of inventoried plot patterns in West Africa. This study focused on the impact of plot patterns on the quantitative analysis of two vegetation types of West Africa based on case studies from Benin. Twenty and fifteen plots of 1 ha each were demarcated in dense forest and woodland, respectively. Each 1 ha plot was divided into 100 quadrats of 100 m2 each and diameter at breast height (dbh) of trees was recorded in each quadrat. The required time to measuring trees diameter in each 1 ha plot was also recorded to compute the mean inventory effort. From the 100 quadrats in each 1 ha plot, 14 subplots of different shapes and sizes were considered by grouping together adjacent quadrats. The basal area of each subplot was computed and the relationship between estimation bias of the basal area and the size of subplots was modeled using Smiths Law (Smith 1938). The mean absolute error of the shape parameter c of Weibull distribution was computed for each of the subplot shape, size and direction. The direction and shape of subplots did not influence significantly (P > 0.05) the precision of the quantitative analysis of vegetation. However, square subplots were suitable in practice. On the contrary, plot size was significantly (P < 0.05) and inversely correlated to estimation efficiency. The optimal plot size for quantitative analysis of vegetation was 1 800 and 2 000 m2 with an inventory effort of 0.51 and 0.85 man-days per subplot in woodland and dense forest, respectively. It is concluded that use of standard sample sizes will help to harmonise a forestry database and to carry out comparisons at regional level.

Empirical comparison of error rate-estimators in logistic discriminant analysis

A simulation study related to the comparison of 15 estimators of the actual-error rate in two-group logistic discriminant analysis is achieved in 528 cases taking into account some characteristics of the groups such as the common distribution of the variables, the overlap of the populations, the number of variables, etc. For each combination of these parameters, the relative errors of the estimators with regard to the theoretical actual-error rate (empirically determined) are computed and help to compare the performance of the estimators. The results show an overall best performance of the estimators e632 and eOS. On the contrary, the estimators eA, epp and eppcv record the lowest performance in terms of the mean relative error and the mean relative bias. Moreover, the performance of most of the parametric estimators depends on the normality and heteroscedasticity degrees of the populations.