Tree crown defoliation data collected on 102 managed forest plots of the RENECOFOR programme in France were investigated to identify (i) short-term (annual) changes and medium term (1994-2009) trends, and (ii) possible correlates of such changes and trends. Methodological aspects (trees assessed, changes in methods and reporting units, observers, assessment dates) were considered. To account for the specificity of individual plots in terms of tree provenance, age, site condition and management regime, an individual plot approach was adopted. Results showed highly frequent, statistically significant and methodologically meaningful (>5% of the expected measurement error) annual defoliation changes, with pulses of increasing defoliation occurring in 1994-1997 (with a possible methodological bias), in 2002-2004 and 2008-2009. A meta-analysis of individual plot results revealed a significant overall increase, in defoliation over the examination period; when the potentially biased 1994-1996 data were excluded from the analysis, the increase in defoliation was also significant. Within this overall increasing trend, cases of stability (11-24% of the plots) or even decreasing defoliation (11-18%) were frequent. We used a Partial Least Square (PLS) regression to model defoliation on 87 plots where sufficient data was available for a standard set of predictors, including meteorology, nutrition, phenology, reported health problems, management regime and assessment methodology. The most frequent correlates of defoliation were precipitation-related variables (of the current and previous years), tree density and frequency of trees with reported health problems. Foliar nutrients, air temperature, assessment method and observers were never found to be important predictors. Within this general pattern, interactions among predictors varied on a plot basis, leading to divergent estimated effects for the same predictor. The adopted plot-based approach avoids the bias that affects traditional cross-sectional, correlative studies and makes it possible to estimate correlates of change at the scale of individual plots; it is therefore a powerful tool to identify response patterns that can be of value when considering (or re-considering) management options.
