This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Several studies attempt to describe changes in the spatial patterns of forests over time, resorting to the comparison of landscape pattern indices (LPI), but new methods for quantifying landscape differences in a statistical context are necessary. In this paper, we quantified and assessed the statistical significance of the forests pattern changes, which have occurred since the end of WWII in Central Italy (Isernia). To do this; based on the proportion of forest cover (

Landscapes change in both structure and function [

The quantitative, spatial and temporal analysis of natural forest re-growth in abandoned farmlands has acquired increasing relevance due to the effects of forest expansion on many important ecosystem functions [

Among the spatial models developed in landscape ecology, the neutral models (NLMs) are those able to produce an expected pattern in the absence of specific landscape processes [

Considering the aforementioned points, the present work aims to describe forest cover dynamics in the hilly landscapes of Central Italy in the last 60 years, analyzing the spatial pattern of temperate forest patches of the area surrounding a small city in a rural European setting in detail (Isernia municipality). In particular, we focus on two questions:

How did the spatial pattern of temperate forests change over time (1954–1981–2006)?

Is the spatial pattern change of these forests significantly different over time?

To properly handle such an issue, on one hand, we used a set of landscape pattern indices to describe the forest spatial pattern dynamics and, on the other hand, we compared LPI of real and simulated landscapes to assess the statistical significance of any possible differences. In order to simulate fractal maps, we chose the midpoint displacement algorithm, because it looks very promising for modeling the natural reforestation process, which occurs in many hilly landscapes, as it is able to represent continuous autocorrelated pattern variations [

Isernia (Central Italy) was selected for our analysis (

Map showing the Isernia district (Italy) and the location of the study area.

We analyzed the spatial pattern of forests and assessed the significance of the observed differences over time, following the general framework proposed by Remmel and Csillag [

Three land cover maps, relative to the years 1954, 1981 [

Due to the fact that the midpoint displacement algorithms can generate only square maps [

Proposed framework for comparing and testing differences of forest pattern over time.

Three forest cover maps extracted from a multitemporal dataset of Isernia, Italy. Each image is 256 × 256 pixels, with a spatial resolution of 10 m. The binary classification separates oak forest (dark green) from other cover types (arable land, permanent crops, pastures, shrub and herbaceous vegetation—soft grey).

Firstly, two parameters necessary for running the NLM simulations were calculated for each date (1954, 1981 and 2006): proportion of forest cover (

To analyze the spatial pattern of forests through time, a set of landscape pattern indices on both real and simulated maps was calculated by using FRAGSTATS 4.0 [

By computing the selected LPI (NP, LFP, AWSI and COHESION) on simulated landscapes, their empirical distributions (

Scatter plots among two paired combinations of observed landscape pattern indices (LPI) values for simulated landscapes (N = 100; 1954, 1981 and 2006): (

The analysis of the 1954, 1986 and 2006 maps shows consistent changes on both the abundance and spatial distribution of forests (

The rate of forest spread, which in the first time span (1954–1981) was of 5.9%, increased to 11% in the second one (1982–2006) and is in contrast with previous studies, which pointed out an overall decrease in the rate change of secondary successions over time [

In

The observed increment in extension and the significant changes in spatial distribution of forests suggest that the analyzed area underwent an intense process of natural re-colonization, which has slowly begun after World War II and which is still in progress. The phenomenon we observed could be considered as reforestation (

Although the accurate description of the huge ecological consequences of such transformations is beyond the scope of this work, we can point out possible effects, such as an increment in true forest species [_{2} sink services [

The NLMs effectively modeled the pattern of forests over time at a specific spatial resolution and could be also very useful for exploring future scenarios, responding in this way to the urgent need to predict natural reforestation process [

Many of the insights and conclusions obtained in this study have been facilitated by the statistical framework provided by the utilization of NLMs. In particular, the application of the midpoint displacement algorithm:

allows for modeling a reliable set of maps, which adequately describe the spatial pattern of forests through time and which can be directly compared with real landscapes [

generates a set of landscape replications, which account for the most relevant information;

defines the landscape expectations, which allows the statistical comparison of patterns through time [

It is important to note that the obtained results are strongly dependent on both the specific type of landscape and the chosen spatio-temporal scale. Since, by tuning the scale of analysis, the observed patterns and the underlying processes become finer or coarser, a sound study of landscape evolution over time should include the spatio-temporal scale [

Furthermore, statistical analysis could become a standard method when comparing maps [

We gratefully acknowledge Mita Drius for the careful and critical reading of the draft, Angela Pitassi for the language review, as well as two anonymous referees for helping to improve the original version of the manuscript.