Question: Multi-temporal analysis of remotely sensed imagery has proven to be a powerful tool for assessment and monitoring of landscape diversity. Here the feasibility of assessing land-use diversity and land-use change was tested at multiple scales and over time by means of statistical linear estimators based on a probabilistic sampling design. Location: The study area (the district of Asciano, Tuscany, Italy) is characterized by erosional forms typical of Pliocene claystone (i.e. calanchi and biancane) that have been subject to the phenomenon of biancane reworking over the past 50 years, mainly owing to the expansion of intensive agriculture. Methods: Cells at two different scales (50m×50m and lOm×lOm) were classified by two operators according to a multilevel legend, using 1954 and 2000 aerial photographs. Inter-operator agreement and accuracy were tested by Cohen's K coefficient. Total land cover estimation for each class was carried out using a multistage estimator, while the variance was estimated by means of the Wolter estimator. Field-based information on plant species composition was recorded in order to test for a relationship between land use and plant community composition by ANOVA and indicator species analysis. Results: Agreement between photointerpreters and accuracy were significantly higher than those expected by chance, proving that the approach proposed is reproducible, as long as proper quality assurance methods are used. Our data show that, at the two scales considered (50 m × 50 m and lOm×lOm), crops have increased against woodlands and semi-natural areas, the latter showing the highest and significantly different mean species richness. Meanwhile, an increase in the coverage of trees and shrubs was found within the semi-natural areas, probably as a result of secondary succession occurring on typical landscape elements such as biancane. Conclusions: Inferential statistics made it possible to acquire quantitative information on the abundance of land cover classes, allowing formal multi-temporal and multi-scale analysis. Sampling design-based statistical linear estimators were found to be a powerful tool for assessing landscape trends considering both time expenditure and other costs. They make it possible to maintain the same scale of analysis over time series data and to detect both coarse- and fine-grained changes in spatial patterns.
