1 Introduction

As an outcome of past demographic transition and industrial revolution, most European countries have been experiencing a forest transition since the nineteenth century, consisting of a strong regain of forest area (Mather 1992; Meyfroidt and Lambin 2011). While present forests’ state has been accurately documented since 1990 with implementation of Forest Europe reporting (Forest Europe 2015) and the increasing implementation of statistical national forest inventory (NFI) programs across Europe (Tomppo et al. 2010), forest dynamics over more ancient time periods remains poorly covered (Mather et al. 1999; Meyfroidt and Lambin 2011). This has prevented from elaborating a clear view on the nature and dynamic of forest changes after the forest transition, including appraisal of areal, structural and compositional forest changes. Such long-term perspective is however of major concern to forest status and ecosystem conservation appraisal (Willis and Birks 2006; Foster et al. 2003).

Across European countries, forests of metropolitan France are those currently experiencing the strongest areal and growing stock changes (Forest Europe 2015). According to the French NFI, started in 1961, French forests have been increasing by almost 17% in area over 30 years (Denardou et al. 2017). This increase results from both land abandonment (Chakir and Madignier 2006; Cavailhès and Normandin 1993) and forest extension ensuing from natural colonization, and active afforestation programs based on conifer species (afforestation of the Landes de Gascogne in Southwestern France by the mid-nineteenth century, David 1993, Buffault 1942; Restoration of Mountain Terrain (RTM) by the end nineteenth century, Saillet 1991; or afforestation supported by the National Forest Fund (FFN) after WWII, Betolaud 1972), with unquantified consequences on forest area and composition. Also, spontaneous forest maturation may have strongly modified the prevalence of forest management types (e.g., coppice or high forest abundance) together with the growing stock.

With this respect, ancient forest data are fairly available in France as compared with other countries (Mather et al. 1999). The first comprehensive survey of French forests is provided by Daubrée statistics of 1908 (Daubrée (Ministère de l’agriculture, FRANCE) 1912; Brenac 1984). Preliminary analysis of these data revealed an increase by + 60% in the forested area over one century (Denardou et al. 2017). These ancient data were not yet based on survey sampling, and methods actually used remain unspecified. Also, no definition of forests based on dimensional and vegetation cover criteria (international forest definition of FAO 2000) was used at this remote period of time, making these figures uncertain as regards how forested land was distinguished from other types of vegetation and land uses.

In this respect, agricultural statistics provide valuable information for both the test of data consistency by cross-comparison approaches, and analyses of forest-area changes. Between 1862 and 1929, five decennial agricultural statistics have been produced by public authorities about every 10–30 years (1862, 1882, 1892, 1902, and 1929; Oberlé 2000). All of them include a section dedicated to forests. In addition to the Daubrée statistics of 1908, agricultural statistics from 1892 to 1929 were considered in the present study, with a view to cross-compare ancient statistics, and cover a period distant by around one century to present time for analyses of long-term forest changes.

The three statistical surveys describe forest area according to three categorical variables of interest: (1) the department administrative units, corresponding to the 90 NUTS-3 units of the nomenclature of the European Commission (European Parliament and Council of the European Union 2003) with an area of 3000 to 10,000 km2 per unit; (2) forest management types, where at least coppices, high forests or coppice-with-standards forests are distinguished. Of note, forest management types primarily reflected objectives of forest planning rather than physical forest structure; (3) the legal category of forest ownership. In addition, Daubrée statistics provides areal statistics for major forest tree species of the country, making it possible to infer long-term compositional changes in French forests by comparison with present NFI data that are available yearly on the whole French territory since 2005.

The first objective of this paper was to improve our knowledges about past forest dynamics based on ancient statistics. The underlying goals were: (1) to digitize the three ancient statistics, only found in paper or image format to date, in order to make them available for further analysis; (2) to analyze and cross-compare ancient statistics to evaluate their consistency; (3) to analyze forest dynamics over more than one century, according to forest management types, ownership categories, and forest tree species, where modern NFI were also used. All comparisons between statistics required harmonization of data variables and resolution.

2 Material and methods

2.1 Data description

2.1.1 Data overview

Daubrée statistics

Daubrée statistics (names after its director) as the first comprehensive forest statistics in France was published in two volumes in 1912. Data were collected by the administration of “Eaux et Forêts” (“Water and Forests”, Corvol 2013) between 1904 and 1910 (Brenac 1984). Nevertheless, all data refer to 1908 (Brenac 1984; Oberlé 2000). Protocols for historical data acquisition remain fragmental. Forest data were collected from the archives of “Eaux et Forêts”, and from forest owners and associated organization. Forest areal data were provided by public local authorities and ensued from the French land registry (Corvol 2013; IGN 2013). The statistics provides forest information also at a resolution below the NUTS-3 level, i.e. by “canton” administrative units (LAU-1 of the EC, > 3000 units in France; European Parliament and Council of the European Union 2013). While the latter resolution remains out of scope of the present study, it was, e. g., used in Dupouey and Hervé (2013) for spatial inference of forest changes.

One synoptic table of Daubrée statistics is reproduced in Annex 1.

Decennial agricultural statistics of 1892 and 1929

Decennial agricultural statistics (Ministère de l’Agriculture - Direction de l’Agriculture 1892; 1929) were intended to provide a comprehensive survey of the agricultural land of the national territory and its associated productions (Hitier 1899; Fromont 1938). A special section is dedicated to forests. We considered data from this section, only.

Decennial agricultural statistics relied on questionnaire surveys of municipalities (Oberlé 2000). Survey results were aggregated by the French administration prior to their publication (Musset 1938).

Agricultural statistics of 1892 (published in 1897 and acquired in 1892, Hitier 1899) and 1929 (published in 1936 and acquired between 1928 and 1933, Musset 1938) also provide information on forest area and harvested volume at the departmental scale.

Forest tables of both decennial agricultural statistics are reproduced in Annex 1.

Complementary data from the French national forest inventory

NFI data were used to assess changes in both forest structure and composition over a one-century period.

Inventory data rely on an annual and spatially systematic sampling of forests (Robert et al. 2010; Vidal et al. 2016) and comply with the international FAO definition of forest (FAO 2000). With aggregation of successive years of inventory increasing data precision (Wernsdörfer et al. 2012) a 5-year aggregation of 2006-to-2010 inventories (median year 2008) distant by 100 years from the Daubrée statistics was computed.

2.1.2 Standard categorical variables

Daubrée statistics

Forest area was split by ownership category and submission to the “Forest regime”, associated to legally-binding management rules defined by the forestry code of 1827 (Némoz-Rajot 1998, Annex 2) and isolating: State-owned forests, municipal forests, and forests of other territorial public bodies subjected to forest regime; municipal forests non subjected to the forest regime and private forests (never submitted to the forest regime). In addition to the three forest management types identified in all three statistics (see Introduction), two management types were specific of Daubrée statistics: (i) the transient regime of coppice conversion enforced by the forest law of 1827, and aiming at shifting to high forests for timber production (Degron 1998), and (ii) unproductive forest areas. The latter corresponded to marshes, bogs, rocks, and reforestation perimeters of forest protection programs. “Unproductive” hence needs to be understood in the literal meaning of not subjected to tree felling, as these areas are systematically encompassed for estimation of the afforestation rating, yet systematically associated with zero harvested volume.

Daubrée statistics included statistics on forest owners and annual harvested volume. All state variables available are summarized in Tables 1 and 2.

Table 1 Description of variables found in ancient forest statistics and national forest inventory
Full size table
Table 2 Summary of data found in ancient forest statistics and national forest inventory. We provide minimum and maximum values (in square brackets) at NUTS-3 unit scale. Also, we provide totals for the whole of France
Full size table

Decennial agricultural statistics of 1892 and 1929

Forest area was classified according to forest management type, ownership category, and forest regime submission (Tables 1 and 2). Both statistics split the area of high forests according to botanical classes (broadleaf- or conifer-dominated forests). Whereas no information on the method used to achieve this distinction (evaluation of class predominance), the significance of compositional mixtures in French forests (Morneau et al. 2008) makes these figures uncertain.

Like in the Daubrée statistics, specific forest management types were identified. 1892 statistics includes category decree afforestation perimeters and unproductive coastal dunes solely found in forests subjected to the forest regime. This complies with the temporary or permanent acquisition by the State, of lands dedicated to landscape protection in coastal and mountain areas, as ordinated by successive RTM laws (Saillet 1991; Sonnier 1991). 1929 statistics includes a forest management type related to recently afforested area.

In addition, annual harvested volume is included in both decennial agricultural statistics (Table 2).

French national forest inventory

We used the breakdown of the forest area according to ownership category and the forest regime submission (State-owned forests, municipal and other public forests subjected to forest regime, private and other public forests non-subjected to forest regime merged in the NFI process).

Forest area at a NUTS-3 resolution was further broken down by stand vertical structure (i.e., coppice, coppice-with-standards, and high forest). The definitions of these structures are provided in Annex 3.

2.1.3 Tree species composition

This data was available in Daubrée statistics and current forest inventory data. In the first one, forest area of a given NUTS-3 unit was split by forest tree species at a 10% resolution, in two broad ownership categories: (i) public forests subjected to the forest regime, (ii) forests not subjected to the forest regime (predominantly private forests). Tree species listed in Daubrée statistics varied across NUTS-3 units and were partially described by genus or common names, requesting technical harmonization (Annex 4).

In NFI data, plots (and associated estimated area) are described by a dominant species of the canopy layer of the census strata (trees with diameter ≥ 7.5 cm). 3.3% of total forest area has a non-identified dominant species (open forests).

2.1.4 Statistics for Bas-Rhin, Haut-Rhin, and Moselle units

Territories of three NUTS-3 units (Bas-Rhin, Haut-Rhin and Moselle) were annexed by Germany between 1870 and 1918, and their forests were not described in Daubrée statistics and in decennial agricultural statistics of 1892. Owing to the existence of the German statistics of 1907 (Ministerium für Elsaß-Lothringen 1909), we supplemented the statistics to cover the whole territory in Daubrée statistics.

German statistics provided information on the legal category of ownership (State-owned forests, municipal forests, and private forests). However, forest areas were not split by management type. Thus, German data were used in relation with Daubrée statistics solely for information on the total forest area and on forest area split by ownership category.

Bas-Rhin, Haut-Rhin, and Moselle NUTS-3 units in the 1892 statistics could not be supplemented by German data.

2.2 Data entry and error correction

Ancient statistics were available in paper textbook format and image format. Areal data were given for each NUTS-3 unit (table rows) and intersections of forest management type and ownership categories (table columns). Marginal totals were also provided. Digitization of the three statistics was the first step of the present work (Audinot et al. 2020). Data from Daubrée statistics were digitized only at a NUTS-3 unit. Initial typing errors of different types were identified in ancient statistics by comparing row/column totals, and were corrected. They are summarized in Annex 5.

2.3 Harmonization of statistics

For the purpose of cross-comparison of forest area across the three ancient statistics, categorical variables were harmonized including tree species composition, forest management type, and ownership categories (Annex 4).

As an overview, forest areas by forest management type were found in equal proportions among the ancient statistics (Fig. 1), except for the other forests’ areas found smaller, mainly in private forests (0% in 1892, 4.3% in 1908, and 2.1% in 1929, Table 3) in line with their heterogeneity. By contrast, high forests that formed a restricted category in ancient statistics are currently predominant (53%) in the total forest area (Fig. 1). Forest ownership categories were distributed in roughly the same proportions between the three ancient statistics (68% of the forest area was private in 1892, 66% in 1908, and 67% in 1929; Fig. 1).

Fig. 1
figure 1

French forest area split by forest management (a), ownership (b), and species composition (c) categories according to three ancient statistics (1892, 1908, and 1929) and national forest inventory data (2008), Bas-Rhin, Haut-Rhin, and Moselle NUTS-3 units were excluded as these were not documented in the Daubrée statistics (German annexation), Corsican NUTS-3, and those surrounding Paris were considered as a single unit. Data about tree species composition do not exist in 1892 and 1929 statistics

Full size image
Table 3 Summary statistics for the French forest area according to the four data sources under study.
Full size table

Moreover, NUTS-3 units’ boundaries were unstable between ancient statistics. The standardization of NUTS-3 units’ boundaries and the analysis of their potential effects of the forest area changes are provided in Annex 6.

2.4 Data analysis

2.4.1 Principle

Forest area differences between successive statistics, including NFI data, were computed, with NUTS-3 units as the reference population. Forest area differences were analyzed at NUTS-3 level, as a whole, or conditionally to forest management type and ownership categories. Paired t tests were used to analyze forest area differences. Maps at NUTS-3 level were produced to detect spatial structures of forest changes.

2.4.2 Cross-comparison of ancient statistics

We analyzed forest area differences over two sets of successive statistics: the ancient periods 1892–1908 and 1908–1929. The following consistency criteria were considered:

  • absence of erratic variations between forest area differences at successive periods (1892–1908 and 1908–1929 periods);

  • absence of bias in the variables of interest documented in the three ancient statistics (forest management type and ownership category);

  • accordance of forest area differences (direction and magnitude) with forest policies at play at both periods (1892–1908 and 1908–1929 periods), and with forest transition features.

2.4.3 Long-term changes in forest area

Forest area differences by NUTS-3 units, forest management types, and ownership categories were also analyzed over two long periods: 1892–1929 (ca. 40 years) and 1908–2008 (ca. 100 years). Over the 1908–2008 period, we analyzed forest area differences at two levels of tree-species composition including botanical classes and tree species.

A synoptic view of periods studied, corresponding statistics, and NUTS-3 populations considered is summarized in Table 4.

Table 4 Statistics and population of NUTS-3 units used by study periods, The transition from 89 to 86 department administrative units (NUTS-3) is due to the annexation of the Alsace-Moselle region. Gironde was excluded from the whole paired t tests but it was kept in all other calculations (Annex 4)
Full size table

3 Results

3.1 Cross-comparison of ancient statistics (1892–1908 and 1908–1929 periods)

3.1.1 Cross-comparison of forest statistics at NUTS-3 units level

Positive forest area differences were found in many NUTS-3 units in 1908–1929 period (70%, Fig. 2), less in 1892–1908 period (49%, Fig. 2). However, the average difference in NUTS-3 forest area was positive and significant at both periods (p < 0.05, Table 5). Strong standard-deviations of forest area differences indicated large heterogeneity across NUTS-3 units (Table 5). Yet, differences lower than − 2000 ha of forest at both periods were only observed for two NUTS-3 units (Pyrénées-Atlantiques and Gers).

Fig. 2
figure 2

Forest area differences across NUTS-3 units of the territory over the 1892–1908 and 1908–1929 periods of the statistical cross–comparison, See Fig. 1 for details on NUTS-3 data

Full size image
Table 5 Summary of forest area differences across NUTS-3 units over the four periods under study, Forest area differences calculated for the 1892–1908, 1908–1929, 1892–1929, and 1908–2008 periods according department administrative units (NUTS-3) population. Significance assessed from paired t tests
Full size table

An obvious spatial structure in positive/negative differences was identified (Fig. 2). Over the 1892–1908 period, most NUTS-3 units showing positive differences were situated in Eastern France, essentially formed of public forests, with negative differences located in Western and central France, where forests remain largely private. Over the 1908–1929 period, positive differences extended to northern France, with differences found to reverse in Central and Southern France. Differences in Western France turned null or positive (Fig. 2). Also, some southern NUTS-3 units showed negative differences after positive differences in the previous period (for example forest area in Ariège with + 5000 ha/− 60,000 ha over 1892–1908/1908–1929 periods, respectively). Such shifts suggested a bias in Daubrée statistics relative to other decennial agricultural statistics, corresponding to underestimation of private forests.

3.1.2 Cross-comparison of forest statistics by forest management type

The analysis of forest area differences by forest management types showed significant and inverse differences between coppice forests (± over 1892–1908/1908–1929 periods; p < 0.001) and coppice-with-standards forests (± over the same periods; p < 0.001, Table 5). Hence, 87% of the NUTS-3 units had negative differences over the 1892–1908 period in coppice forests, and conversely 71% of these units had positive differences over the “1908–1929” period Fig. 3). In coppice-with-standards forests, 76% of the NUTS-3 units had positive differences over the 1892–1908 period, while 74% of these had negative differences over the 1908–1929 period, Fig. 3). Daubrée statistics thus systematically over-estimated the area of coppice-with-standards forests at the expense of those of coppice forests, as compared to other decennials agricultural statistics.

Fig. 3
figure 3

Comparison of successive 1892–1908 and 1908–1929 periods forest area differences at a NUTS-3 level across forest management and ownership categories, See Fig. 1 for details on NUTS-3 data, FAD, forest area differences. a FAD in public forests. b: FAD in private forests. c FAD in coppice forests, d FAD in coppice-with-standards forests, e FAD in high forests, f FAD for other forests (see Annex 4 of the manuscript)

Full size image

The large standard-deviations of forest area differences found in other forests’ areas despite their small areas (Fig. 3 and Table 3) pointed out to the heterogeneous definition of this forest management type (Table 5). For example, other forests’ areas in Gironde and Landes showed large differences over the one or two periods (+ 78,000 ha over the 1892–1908 period and − 73,000 ha over the 1908–1929 period in Gironde; + 1400 ha over the 1892–1908 period and + 58,000 ha over the 1908–1929 period in Landes) at a time of active afforestation, bound to explain these strong differences.

In contrast, means of forest area differences were much more constant in high forests (Table 5) and both positive and statistically significant (p < 0.01 for 1892–1908; p < 0.05 for 1908–1929, Table 5). Positive differences represented 56% of NUTS-3 units over the 1892–1908 period and 76% units over the 1908–1929 period (Fig. 3).

3.1.3 Cross-comparison of forest statistics by ownership category

Over 1892–1908 and 1908–1929 periods, a substantial fraction of NUTS-3 units showed positive forest area differences in public forests (64% of NUTS-3 units over the 1892–1908 period, 69% of NUTS-3 units over the 1908–1929 period, Fig. 3), however, statistically significant only in the first period (p < 0.01, Table 5). The strong increase in public forests area over the 1892–1908 period was in agreement with the spatial structure identified in the Eastern half of France (Fig. 2) where there is a large share of public forest.

Fifty-five percent of NUTS-3 units in private forests also showed a negative forest area differences over the 1892–1908 period. Among them, 68% then had positive differences over the period 1908–1929. Means of forest area differences means between these NUTS-3 units showed inverted differences over the two periods (− 5200 ha/NUTS-3 units over the 1892–1908 period and + 4900 ha/NUTS-3 units over the 1908–1929 period), but these differences were statistically significant only for the 1908–1929 period (p < 0.01, Table 5). These results confirmed the underestimation of private forests areas in the Daubrée statistics, as inferred in 3.1.1.

Area of private forest in Gironde and Landes (southwestern France) last showed tremendous differences at a period of active afforestation (+ 110,000 ha in Gironde over the 1892–1908 period, + 110,000 ha in Landes over the 1908–1929 period).

3.2 Long-term changes in forest area (1892–1929 and 1908–2008 periods)

3.2.1 Historical differences in whole French forest area

Significantly positive forest area differences in total forest area were identified at both periods (p < 0.001 in 1892–1929 period and p < 10−10 in 1908–2008 period, Table 5), suggesting a continuous increase in forest area since 1892. The magnitude of the total forest area difference at 1892–1929 period was + 700,000 ha, this corresponding to a relative increase by + 7% and to an annual rate of change by + 0.19% over the 1892–1929 period. However, large standard-deviations of forest area differences indicated that not all NUTS-3 units were subjected to positive differences over this period (30% of NUTS-3 units showed negative differences). Also, the total forest area differences in 1908–2008 period was around + 5,800,000 ha corresponding to a relative difference by + 36% and to an annual rate of change by + 0.45%, thus foot-printing an acceleration in forest areal increases between 1892 and 1929 and 1908–2008 periods (p < 10−7).

3.2.2 Historical differences in forest area by forest management type

Forest area differences at both periods showed a decline in coppice forests on almost the whole territory, together with a large increase in high-forests over the 1908–2008 period, indicating a massive change in the structure of these forests over the period. In 1892, the area of coppice forests was of 3,500,000 ha. About one century later (2008), coppice forests have reached 1,700,000 ha, as documented by NFI data. Assuming that ancient coppice forests have fully evolved into high forests, and since the total increase in forest area is 5,800,000 ha, this suggested that an additional + 4,100,000 ha of high forests had another origin than coppice conversion since 1908. High forests, therefore, have formed the majority of afforested or returning forests since 1908 (Fig. 1 and Fig. 4).

Fig. 4
figure 4

Changes in forest area classified by forest management type across NUTS-3 units of the territory over the 1892–1929 and 1908–2008 periods, See Fig. 1 for details on NUTS-3 data

Full size image

Forest area differences in coppice-with-standard forests showed a North-South division over the 1908–2008 period (Fig. 4), with a marked decrease in Northern and Northeastern France (− 1,200,000 ha), where the area of high forests has increased (+ 2,500,000 ha), and a conversely marked increase in South France and Brittany (+ 1,600,000 ha) where coppice forests area has decreased (− 360,000 ha). Maps of forest area differences therefore suggested similar maturation in forest structure (coppice–coppice-with-standards–high forest) across the whole territory, though with a much longer apparent delay in southern France.

3.2.3 Historical differences in forest area by ownership category

Increase in private forests area was found substantial (corresponding to an annual rate of change + 0.27%/+ 0.56% over 1892–1929/1908–2008 periods, respectively) and concerned 97% of NUTS-3 units over the 1908–2008 period. In 1908, the area of private forests was of 6,800,000 ha. About one century later (2008), private forests area has almost doubled, reaching 12,000,000 ha. In private forests, mean of forest area differences was very strongly significant over the 1908–2008 period (p < 10−10, Table 5). As a consequence, whereas two-thirds of French forests were private in ancient statistics, three-fourths of the forest area is nowadays private (Fig. 1).

Public forests also revealed significant differences over the 1908–2008 period (+ 700,000 ha, p < 10−8, annual rate of change by + 0.17%, Table 5). The slight increase over 1892–1929 period was not statistically significant (p = 0.21, annual rate of change by + 0.39%, Table 5).

3.3 Long-term changes in forest composition

Broadleaf- and conifer-dominated forests showed significant positive differences over the 1908–2008 period (Fig. 5) with extension of broadleaved forests twice as intense in magnitude as in conifer forests (+ 3,600,000 ha, p < 10−10 and + 1,600,000 ha, p < 10−7, respectively), yet favored in afforestation programs. Consistently, various broadleaves also progressed three times faster than various conifers (Table 6 in Annex 4). These relative changes were strikingly similar (+ 34/+38% in broadleaved/conifer forests), making the proportion of these forests unchanged over one century (Fig. 1).

Fig. 5
figure 5

Changes in broadleaf- and conifer-dominated forest area over the 1908–2008 period across NUTS-3 units of the territory, See Fig. 1 for details on NUTS-3 data

Full size image

Tree species (without various broadleaves and various conifers) showed a significant positive differences over 100 years (+ 3,900,000 ha or 67% of total extension, p < 10−10), indicating that species identified in Daubrée statistics had been playing a role in forest extension.

Significant positive areal changes were detected in all tree species/species groups but hornbeam (Carpinus betulus L.) and cork oak (Quercus suber L.) showing negative and non-significant negative changes (Fig. 6 and Table 6 in Annex 4), two species that were important contributors to coppice production systems in France.

White oaks showed sharp significant positive areal changes (Fig. 6 and Table 6 in Annex 4) in a large part of territory (81% of NUTS-3 units). White oaks privileged in both coppice-with-standards and high forests took advantage of changes in the structure of the French forest.

Norway spruce (Picea abies Karst.) and pine species used in afforestation programs showed areal increases by + 40 to + 70% larger than those of, e.g., Silver fir (Abies alba Mill.), a coniferous tree species regenerated naturally (Fig. 6 and Table 6 in Annex 4). Also, European larch (Larix decidua Mill.) privileged in landscape protection programs of the nineteenth century still showed progression after Daubrée statistics (Fig. 6 and Table 6 in Annex 4).

In summary, the strong absolute progression of broadleaved-dominated forests was emphasized, with afforested coniferous tree species and coppice broadleaved species found at extreme magnitudes.

4 Discussion

4.1 Statistical value of ancient statistics

This study is based on historical sources of forest information: Daubrée statistics (1908) completed by a German forest statistic (1907) for Bas-Rhin, Haut-Rhin, and Moselle NUTS-3 units and decennial agricultural statistics of 1892 and 1929. Their supplementations with modern NFI data offered an extended overview on broad long-term forest changes, and in this sense, also increase the value of modern data.

The lack of a clear forest definition and of a specific description of methods used at that time to produce ancient statistics questions their consistency. Also, the absence of foundations in survey sampling at that time period (Frayer and Furnival 1999) and of subsequent error estimation makes these statistics of lower statistical value than modern forest inventory. However, over a long time period, changes in the forest can turn strong enough for these statistics’ uncertainties to appear much lower in magnitude than the main shifts detected (Table 5). This suggests their practical use to identify broad patterns and the footprints of past forest policies and natural processes.

Also, despite the reduced availability of information technologies at that time (calculator, computer), only few errors were detected in all of the ancient statistics in comparison to the total number of digits entered (Annex 5). The rate of errors in primary unit figures however remains out of reach.

4.2 Cross-comparison of ancient statistics

4.2.1 Potential systematic biases in Daubrée statistics

Forest area differences across NUTS-3 units showed systematic positive shifts in forest area at both periods (1892–1908 and 1908–1929, Table 5). All successive statistics suggested an increase in the total forest area, in line with current knowledge of the French forest transition (Cinotti 1996; Denardou et al. 2017). Yet, many of these units experienced successive increases/decreases in area over the two periods (notably NUTS-3 units of Southern France, Fig. 2), suggesting that either Daubrée statistics is biased positively, or decennial agricultural statistics are so, negatively. Over-estimation of forest area in Daubrée statistics may have several causes: a lower attention paid to forests of lower economic interest like Mediterranean forests of southern France or for mountain forest areas, also poorly described in administrative documents (Alps, Pyrenees, and Corsica). Negative forest area differences on 1892–1929 period followed by positives differences on 1908–1929 period affected Massif Central (Fig. 2), mainly private (Dodane 2009) and as such prone to lower attention in Daubrée statistics. These regional heterogeneities are discussed in Denardou et al. (2017).

4.2.2 Interplay between political will and past forest reporting

Cross-comparisons suggested a relative bias in the forest area of coppice/coppice-with-standards in Daubrée statistics with respect to other decennial agricultural statistics (Fig. 3). Over the 1892–1908 period, coppice forests regressed by 1,200,000 ha and coppice-with-standards forests increased by 930,000 ha. Over the 1908–1929 period, however, coppice forests increased by 860,000 ha and coppice-with-standards forests decreased by 700,000 ha. This over-estimation of coppice-with-standards (including coppice conversion) pattern in Daubrée statistics may find root in the coppice conversion policy, hindered by political and economic issues, which promoted a climate of suspicion between the State and the forest administration (Degron 1998). Such context is bound to have led to over-estimations of the conversion effort for better agreement with policy objectives (Corvol 2013). Moreover, the very short study period (1892–1908-1929) with respect to the pace of coppice conversion into high forest makes this evaluation highly dependent on survey processes.

The moderate negative bias in private forests in Daubrée statistics (Fig. 3) could stem from forest owners’ distrust towards State policies. The difficult economic context of that period (Degron 1998) pushed the State to reflect on the privatization of public forests (Dabat 1920), or to interfere in the management of private forests (Corvol 2013), encouraging private owners to voluntarily underestimate the extent of their forests.

4.3 Increases in the forest area

4.3.1 Returning forests

Forest area differences over a longer time scale (1892–1929 and 1908–2008 periods) highlighted the sharp significant increase in forest area (+ 5,800,000 in whole French forest area and an average + 65,000 ha/NUTS-3 unit over the 1908–2008 period; p < 10−10, Table 5) and particularly in private forests (+ 5,100,000 in whole French forest area and + 58,000/NUTS-3 unit over the 1908–2008 period; p < 10−10, Table 5). Abandonment of agriculture is alleged to be one of the main drivers of this increase (Chakir and Madignier 2006; Cavailhès and Normandin 1993) and remains acute in rural areas (Denardou et al. 2017).

4.3.2 Afforestation programs

Major afforestation programs have taken place during 19th and 20th centuries including the landscape protection of the RTM program (1860 and 1882, Larrère et al. 1980) which led to planting of 170,000 ha by 1914 (Métailié 1999), and the much more extended FFN program (1946–1999) intended to foster the French forest industry and which reached a net 1.2 million hectares (Dodane 2009). This total afforestation (1400,000 ha) only represented 24% of total forest area increase (5,800,000 ha) over the 1908–2008 period room for undocumented secondary afforestation programs (e.g., national lottery program in 1920, Bazire and Gadant 1991), or more largely to natural expansion processes.

4.4 Structural and compositional changes in forests

4.4.1 Structural forest changes

Regarding forest management, the shift between coppice and high forest area over both long periods 1892–1929 and 1908–2008 is obvious (Fig. 4)—while at different stages depending on regions—and it allows concluding that forest conversion (1908–2008 period: − 600,000 ha in coppice, p < 0.01 and + 4,900,000 ha in high forest, p < 10−10; Table 5) over the territory (Fig. 4) has been a very long-lasting and still ongoing process. Of importance, such process is bound to also have also dramatically increased the forest growing stock (Pignard 2000; Forest Europe 2015).

4.4.2 Compositional forest changes

The two main afforestation policies (section 4.3.2) mainly concerned conifer plantations. For example, the area of Norway spruce plantations in the Massif Central has increased from 360,000 to 610,000 ha over 20 years (Bazire and Gadant 1991). This massive afforestation has supported the idea, widespread in France and Europe (Moriniaux 1999) of an increase in conifer prevalence, at the expense of broadleaves. Whereas we confirmed the significant expansion of conifer-dominated forests (+ 1,600,000 ha, p < 10−7), we also showed a parallel massive expansion of broadleaf-dominated forests (+ 3,600,000 ha, p < 10−14) over the 1908–2008 period (Fig. 5). Hence, the proportion of broadleaf- and conifer-dominated forests has not changed in a century (Fig. 1), conflicting the view of conifer progression across the territory.

The significant sharp increase in white oaks areas (+ 1,500,000 ha over the 1908–2008 period, Fig. 6 and Table 6 in Annex 4) has a dynamic interpretation, as oak species are found in both coppice-with-standards forests and high forests owing to their ability to grow from sprouts (Boppe 1886), and their wood quality has led to favor them in the conversion of coppice. The non-significant small decrease of hornbeam in the 1908–2008 period (− 35,000 ha, Fig. 6 and Table 6 in Annex 4), a species predominantly found in coppice and coppice-with-standards forests due to its ability to sprout (Boppe 1886), is in accordance with conversion in favor of oaks. In addition, maintaining an understory of hornbeam benefits to the oak silviculture (Office National des Forêts 2010).

Fig. 6
figure 6

Changes in forest area classified by dominating tree species/tree species groups of the French forests over the 1908–2008 period, See Fig. 1 for details on NUTS-3 data. Significance of forest area differences was assessed by paired t tests across the NUTS-3 population. Individual tree species not identified in the tree species groups white oaks, pines, various broadleaves and various conifers, (*): p value < 0.1. *p value < 0.05. **p value < 0.01. ***p value < 0.001

Full size image

The significant increases of area in Norway spruce (+ 340,000 ha, Fig. 6 and Table 6 in Annex 4) and in European larch (+ 40,000 ha, Fig. 6 and Table 6 in Annex 4) reflect afforestation conducted by the FFN (Bazire and Gadant 1991; Bouchon 1984). Parallel significant increases in chestnut (Castanea sativa Mill., + 520,000 ha), various broadleaves (+ 1,000,000 ha), and various conifers (+ 320,000 ha) areas (Fig. 6 and Table 6 in Annex 4) can also reflect an effect of tree species diversification in afforestation programs (de Rochebouet 1987).

Note that declining cork oak areas can be explained by a decline in the demand for oak cork (Bugalho et al. 2011). Also, the mortality of cork oak has increased in the Mediterranean region, due to the increased intensity of droughts (Costa et al. 2010).

Statistics over this 1908–2008 period therefore suggested that intensive afforestation programs (RTM, FFN) did not develop at the expense of tree species diversity (Fig. 6 and Table 6 in Annex 4), but instead allowed for a tree species diversity increase at country scale, by introducing new tree species while maintaining and even increasing prevalence of the others. The illustration that of forest transition can increase tree diversity as a country scale is major.

5 Conclusion

  • Cross-comparison of the three ancient French forest statistics showed that their exploitation was effective. Specific inconsistencies between the three statistics highlighted directional biases in forest management types and ownership categories, namely in Daubrée statistics, and must be dealt with when exploiting the data.

  • These ancient statistics can support many fields of research. Examples include the impacts of the two World Wars on forests, the evolution of tree biodiversity, and spatial analyses of land-use change.

  • The study of forest-area changes over more than one a century, using the three ancient statistics and NFI data (2008) highlighted features of the French forest expansion, found mainly in private forests, and characterized a slow conversion of coppices into high forests on the whole territory. Moreover, a statistically significant increases in the area of all major tree species found in the early twentieth century, with only minor areal decreases in two species common in coppice systems (hornbeam and cork oak) together with a clear effect of conifer afforestation programs (Restoration of Mountain Terrain and National Forest Fund) were identified, highlighting the major outcome of a long-term increase in major tree diversity across the whole territory.