Recent increase in European forest harvests as based on area estimates (Ceccherini et al. 2020a) not confirmed in the French case

A recent paper by Ceccherini et al.(2020a) reported an abrupt increase of 30% in the French harvested forest area in 2016–2018 compared to 2004–2015. A re-analysis of their data rather led us to conclude that, when accounting for the singular effect of storm Klaus, the rate of change in harvested area depended on the change year used to separate the two periods to compare. Moreover, the comparison with data on harvested volumes from different sources brought contrasted results depending on the source. Therefore, it cannot be concluded that wood harvest increased in France in 2016–2018 compared to 2004–2015. The discrepancy between Ceccherini et al.’s data and other data on harvested volumes points out the difficulty of reconciling different approaches to estimate wood harvest at a country level.

used instead of other, more emitting materials), adaptation to climate change, and the development of a forest-based circular, inclusive, and sustainable bioeconomy call for new priorities to be assigned to forests in the future. Thus, the balance between the different dimensions of forests and forest management is questioned and publicly debated. Shifting to these new priorities in the most consensual way requires precise assessment and monitoring of the different forest attributes, including wood harvest, i.e., the quantity of wood extracted from forests by human activities, whether expressed as a volume or as an area (i.e., the area where logging completely removed tree-cover canopy). This harvest is in fact not only an objective as such of forest management but also a means of ensuring one or more of the other objectives.
Based on estimates of harvested forest areas, a recent paper by Ceccherini et al. (2020a) stated that harvest had increased by 43% over Europe in 2016-2018 compared to 2004-2015, with the mean harvested patch size having increased by 34% between the same two periods. For France, the estimated increases were 30% and 44% for harvested area and patch size, respectively. However, such an unprecedented increase in wood harvest in Europe remains undocumented in national forest statistics (Niedzwiedz et al. 2018). Our primary objective here is to understand how such an abrupt change in wood harvests as claimed by Ceccherini et al. (2020a) may have gone unnoticed. Assuming that harvest intensity (i.e., the volume of harvested wood per unit of harvested forest area) and logging loss rate did not change across the 2004-2018 period but for random fluctuations, such an abrupt change in harvested forest area should go along with a marked change in harvested wood volumes. Our investigation thus is focused on the co-variation between harvested forest area as reported by Ceccherini et al. and harvested wood volumes as reported by different sources. We focused on the French case, even if the methodological limitations identified in this case may also apply to the results for the other European countries.
Harvested forest area in France here refers to the time series from 2004 to 2018 estimated by Ceccherini et al. using the Global Forest Change product by Hansen et al. (2010) after filtering out areas affected by forest fires but retaining those affected by windstorms. There are different statistics on harvested wood volumes in France for the same period. One source is the FAOSTAT database (FAO 2020) that was also used by Ceccherini et al. in their study. It indicates the total roundwood production defined as the amount of wood removed from the forest and other wooded land or other sites in m 3 under bark. It cumulates the marketed harvest and the self-consumption of wood from productive forests, other forests, and wooded lands. A second source is the Agreste database of the French Ministry of Agriculture and Food (MAA 2020). It gives an estimate of the harvested wood volumes based on a survey of forest logging companies and sawmills, and thus only includes the harvest for the inner market excluding exports (and excluding self-consumption) in all kinds of forests. A third source is the French National Forest Inventory (NFI 2020) that provides annual estimates of wood felled in productive forests (including harvest and logging losses) based on an inventory of trees felled in semipermanent sample plots re-measured after a 5-year period. These different data sources thus differ by the kinds of forests they cover, the kinds of wood products (marketed, selfconsumed, or total), and the kinds of estimation techniques (field inventory, survey of forest companies, survey on energy consumption, etc.) (Fig. 1). Ceccherini et al. (2020a) reported that the harvested area in France represented "fairly well the amount of harvests from final cut and salvage logging" as given by the harvested volumes from FAOSTAT, even if "part of the silvicultural treatments" (including thinnings) and "of the overall harvest" may have been missed in the harvested area "owing to the complex structure and heterogeneity of the management systems applied in France." Nevertheless, our re-analysis of their data (Ceccherini et al. 2020b) provided a different view. The value of Pearson's correlation Fig. 1 Simplified description of the harvested volumes included in the Agreste, NFI and FAOSTAT data. The decomposition of the total harvest is actually more complex than reported in this chart because the NFI makes a distinction between productive forests and other forests; the NFI reports the volume of stems while FAOSTAT additionally includes the volume of large branches, etc coefficient of 0.33 that Ceccherini et al. reported was due to a single leverage point corresponding to year 2010 when wood harvest was exceptionally high in France following the windstorm Klaus of 2009 in the Landes de Gascogne (south-western France) (Fig. 2). Windfall wood volume then amounted to 43 million m 3 , a level comparable to the average annual harvested volume as estimated by the NFI. Removing year 2010 (which brings Pearson's coefficient down to 0.04, p value = 0.89) or using Spearman's coefficient as a more robust alternative to Pearson's coefficient with respect to leverage points (coefficient = 0.13, p value = 0.66) showed that no correlation could be evidenced between harvested forest area detected from remote sensing and FAOSTAT harvested volume in France over 2004-2018. Similar results were obtained with the other national statistics on wood harvest. The positive correlation between harvested forest area and the Agreste data on marketed volumes were slightly higher (Pearson's coefficient of 0.44, p value = 0.09) than with the FAOSTAT data on total harvested volumes, but again, this correlation was due to the year 2010 alone acting as a leverage point and disappeared when removing this point (coefficient = 0.15, p value = 0.60) or when using Spearman's coefficient (coefficient = 0.21, p value = 0.45). The correlation between harvested forest area and the NFI data on wood removals from forests available for wood supply was even negative and not significantly different from zero (coefficient = − 0.11, p value = 0.71). It is worth noticing that correlations among national statistics on wood harvest were not higher (Pearson's correlation between FAOSTAT data and Agreste data = 0.07, p value = 0.81; correlation between Agreste data and NFI data = 0.41, p value = 0.15), or even negative (correlation between FAOSTAT data and NFI data = − 0.53, p value = 0.05).
The absence of a significant correlation between harvested forest area and harvested volumes in France from three data sources  Fig. 3). In contrast, Ceccherini et al. adopted a historical rather than a statistical approach. The statistical approach indicates that the direction of the change in harvested forest area is sensitive to the breakpoint used and that the time series is too short to conclude that a change occurred in 2015.
Hence, the re-analysis of Ceccherini et al.'s data and its comparison with alternative data on harvested wood brings no justification for comparing 2004-2015 to 2016-2018 rather than some other periods (Fig. 3). Reporting on the reconstructed estimate of the total harvested volume (FAOSTAT data) or some component of this total harvest (Agreste or NFI data) is already enough to bring differences that are greater than the temporal changes of any of these quantities over 2004-2018 (compare panels in Fig. 3, also shown by the absence of significant correlation between those data). Therefore, it cannot be concluded that wood harvest increased in France in 2016-2018 compared to 2004-2015. The main interest of Ceccherini et al.'s study rather lies in illustrating the difficulty of reconciling different approaches to estimate wood harvest at a country level (Gschwantner et al. 2019). Reconstructing total wood harvest in France relies on data from surveys and sampling techniques, and additional coefficients to convert or extrapolate these estimates (e.g., the damage rate to convert wood removals into wood harvest for the NFI data, or the self-consumption rate given by the French National Institute of Statistics and Economic Studies (INSEE) housing survey to extrapolate marketed harvested volumes to total harvested volumes) (Niedzwiedz et al. 2018). Additional difficulties may come from the tree components considered (e.g., the French NFI reports wood volumes for the stem only whereas large branches are included in harvested volumes in FAOSTAT) and cumulate with inherent uncertainties in sampling techniques (NFI 2011;Waggoner 2009).
Rather than relying on estimates of harvested volumes, Ceccherini et al. used a different approach based on a cartographic product, the Global Forest Change by Hansen et al. (2010), to estimate the area of harvested patch sizes. Several difficulties linked to this approach were highlighted by the authors, including the difficulty to delineate forests based on canopy cover or to filter out canopy gaps due to other factors than logging (viz. wildfires and storms). More importantly, Ceccherini et al. relied on the finest pixel grain of 30 m × 30 m of the GFC to delineate harvested patches while the GFC map is itself the outcome of a regression estimator procedure that restricts its capacity to bring precise estimates on small areas (Hansen et al. 2010;Wernick et al. 2010). Despite its current methodological limitations and its lack of consistency with statistics on harvested volumes, Ceccherini et al.'s study brought the idea of using high-resolution satellite images to infer harvested forest area. Additional work is still needed to turn this idea into an operational tool, especially because remote sensing is efficient at detecting clear-cuts and canopy gaps but may miss thinnings and selective logging that does not heavily affect canopy closure.
Even if data do not support the presence of an abrupt increase in wood harvest after 2015 in France, Ceccherini et al.'s study raises the question of the planned intensification of wood harvest in European countries in the context of the development of a wood-based bioeconomy. Several forest policy documents in France have assigned an objective of increased wood harvest. The National Forest and Wood Programme 2016-2026 has set an objective of an additional 12 million m 3 of wood harvest by 2026 (MAA 2017), while the National Low-Carbon Strategy has set an objective to increase wood harvest from 48 million m 3 in 2015 to 65 million m 3 in 2030 and 83 million m 3 in 2050 (MTES 2020). Such an increase in harvest has to be compared to the simultaneous increase in growing stock (Bontemps et al. 2020), the ratio of wood harvest over growing stock being remarkably stable (around 1.6%) in France over the last 20 years. Sustainability of forest management and the role and functions to be assigned to forests in the uncertain context of global changes is beyond the scope of this paper. However, as pointed out in the French Forest Research and Innovation Plan 2015-2025, current limitations in reconciling different approaches to estimate wood harvest at national level calls for additional efforts in forest monitoring systems using diversified techniques (remote sensing, field-based inventories, possibly crowdsourcing of data).
Data availability Data sharing not applicable to this article as no datasets were generated during the current study.

Compliance with ethical standards
Conflicts of interest The authors declare that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons . org/licen ses/by/4.0/.