Validating a continental European charcoal calibration dataset

Abstract

Large-scale training sets enabling quantitative reconstructions of past fire parameters are needed to better assess potential effects of increased fire hazard under global warming conditions. The aim of this article is to validate recently developed continental regression equations for the reconstruction of fire number, intensity and size. These transfer functions were built by linking satellite data and charcoal collected in annually sampled sediment traps. We apply these European regression equations to four annually layered lakes located on a North–South gradient in Europe. Down-core annual microscopic charcoal (MIC) and macroscopic charcoal (MAC) influx values were compared with satellite-derived time series of fire number, fire intensity and area burned. Results show that the match between predicted and observed values improves when the overall mean and median of sampled years (12 and 9 years) are considered. Especially, the comparisons of median values show a very good agreement between charcoal-inferred and satellite-observed fire-regime parameters. MIC-based predictions underestimate the variability of the observed fire parameters and MAC-based predictions overestimate it. Our results imply that median values of the fire parameters can be reconstructed well by using MIC and MAC, while it is more difficult to infer the variability of fire-regime parameters. However, when MIC- and MAC-based predictions are pooled together, the fit between observed and predicted values increases for both medians and variability. This finding suggests that MIC and MAC are complementary proxies, thus best sedimentary fire reconstructions may be achieved when they are used together. We conclude that sediment traps can be used for the construction of continental-scale training sets and that their results can be applied to Holocene sedimentary charcoal sequences.

Keywords

calibration in space charcoal fire history quantitative reconstruction sediment trap transfer function validation

Introduction

Quantitative environmental reconstructions play an important role in the fields of Palaeolimnology and Palaeoecology and have enabled the scientific community to gain new insights on topics such as past changes in climate, baseline conditions and the effects of human impact on aquatic systems (e.g. Bradshaw et al., 2005; Samartin et al., 2017; Simpson et al., 2005). Quantitative reconstruction approaches rely on either ‘calibration in time’ (e.g. Amann et al., 2014; von Gunten et al., 2009) or ‘calibration in space’ (Blois et al., 2013) approaches. In the case of discrete events such as fire and other disturbances that can vary strongly from event to event and year to year, calibration studies need to capture the full variability of these events and to ensure high chronological accuracy and precision of the training set surface samples. Temporal accuracy and precision is additionally needed to calculate reliable influx values (particles or area cm⁻² yr⁻¹), which allow comparisons across sites and periods (Davis, 1969; Welten, 1944). Influx values are crucial to link the sediment proxy (e.g. charcoal) to the environmental variable of interest (e.g. observed fire-regime parameters). Similarly, for the down-core application of transfer functions, particular accuracy and precision is needed in regard to the chronology to allow gaining reliable influx values.

A recent calibration in space study covered most biomes of Europe to allow encompassing the fire variability range observed today at the continental scale (Adolf et al., 2018). Covering such a large environmental variability is needed for reconstructing past fire-regime changes under strongly variable environmental conditions, for example, of the Quaternary. In Adolf et al. (2018), regression equations linked charcoal influx (microscopic charcoal (MIC), particles >10 µm in length; macroscopic charcoal (MAC), particles >100 µm in length) from annual sediment traps to important satellite-derived fire-regime parameters (i.e. fire number (fires yr⁻¹), fire intensity (W m⁻²) and area burned (km²)). Floating sediment traps (type: EAWAG 130, see Appendix 1, available online; following Bloesch and Burns, 1980) distributed across lakes in all main biomes of Europe were collected yearly over a period of three years. The annual recovery of the sediments enabled the calculation of reliable charcoal influx values, which were linked to satellite-derived fire data from the same time period.

The aim of this study is to provide validation datasets to check the proposed continental equations from Adolf et al. (2018) and their application to sedimentary time series. Validation of quantitative reconstructions is an important step of calibration studies (Telford and Birks, 2011). We chose sites with annually layered (i.e. varved) sediments because they provide highest chronological accuracy and precision and thus reliable MIC and MAC influx data. By using varved sediments, we also aim at assessing charcoal deposition processes after burning, which may cause lags between fire occurrence and charcoal abundance in the sediments (Tinner et al., 1998; Whitlock and Millspaugh, 1996). In our approach, we use MIC and MAC to reconstruct satellite-derived fire parameters, namely, fire number (number of fires yr⁻¹), fire radiative power (FRP; W m⁻²) and burned area (km²). It has been argued that one proxy may not be sufficient to reconstruct different environmental parameters (Telford and Birks 2011). However, choosing such a procedure is useful if the target variables are (1) expected to be highly correlated and (2) contain unique information (Telford and Birks 2011). Here, we evaluate sedimentary reconstructions by continental-scale remote sensing information which is fully independent from the proxy and transfer function approach. We assume this test to be rigorous, given that the remote sensing information has a very high quality in regard to spatial and temporal resolution and precision for the past 12 years.

Methods

Study sites

We selected four annually laminated lakes from among the lakes studied by Adolf et al. (2018) along a North-to-South gradient in Europe (Figure 1, red stars): Sarsjön in northern Sweden (64.039°N, 19.602°E), Jezioro Gościąż in central Poland (52.583°N, 19.339°E), Lago Piccolo di Avigliana in northern Italy (45.053°N, 7.393°E) and Lago Grande di Monticchio (40.932°N, 15.604°E) in southern Italy. During the years 2001–2013, the mean annual fire number per 1000 km² and mean area burned per fire in the municipalities/provinces of the study sites were 1.8 fires and 0.45 ha for Sarsjön (Vasterbottens Lan, Sweden), 9.8 fires and 0.23 ha for Jezioro Gościąż (Kujawsko-Pomorskie, central Poland), 12.6 fires and 9.1 ha for Lago Piccolo di Avigliana (province of Turin, Italy) and 27.3 fires and 9 ha for Lago Grande di Monticchio (province of Potenza, Italy; extracted from the European Forest Fire Information System (EFFIS), San-Miguel-Ayanz et al., 2012). The lakes are situated at altitudes between 76 and 608 m a.s.l. and they are rather small-sized (8 ha for Sarsjön, 42 ha for Jezioro Gościąż, 61 ha for Lago Piccolo di Avigliana and 30 ha for Lago Grande di Monticchio).

Figure 1.

Sites of study (red stars) and sites of calibration dataset from Adolf et al. (2018) (white). Modified map from Digital Map of European Ecological Regions (European Topic Centre on Biological Diversity (ETC/BD), 2009).

Coring, chronology and sampling

During the months of April and May 2013, we recovered undisturbed surface sediments from the deepest part of the four study lakes by using a freeze-corer device (Wright, 1991). With the information from previously published articles (Sarsjön: Snowball et al., 1999; Jezioro Gościąż: Ralska-Jasiewiczowa et al., 1998; Lago Piccolo di Avigliana: Finsinger and Tinner, 2006; Finsinger et al., 2011 and Lago Grande di Monticchio: Allen et al., 2002; Zolitschka and Negendank, 1996), varves were characterized and sampled by scraping or cutting/sawing off the layers while they were still frozen. We chose an annual sampling resolution to enable reliable charcoal influx calculations and counted the annual layers of the cores five times to enable mean and standard deviation calculations. Two standard deviations were used as error estimations of the varve chronologies. Constant rate of supply (CRS) depth-age models based on ²¹⁰Pb- and ¹³⁷Cs-isotope dating (Department of Chemistry and Biochemistry at the University of Bern, Switzerland) following the calculations described in Appleby (2001) were used to check the varve chronologies for Lago Piccolo di Avigliana and Lago Grande di Monticchio. As the varves were not clearly visible in the sequence of Jezioro Gościąż, samples were taken in 1 and 0.75 cm intervals, based on sedimentation rate estimates from Adolf et al. (2018). The age of each sample was then estimated by using the CRS depth-age model. In the case of Sarsjön, we assessed the varve chronology by comparing distinct wide and light layers in the core to historical records of nearby floods in Northern Sweden (Myndigheten för samhällsskydd och beredskap (MSB), 2012). These layers were assumed to be erosion-related sediment depositions caused by floods.

MIC samples with volumes between 0.1 and 3.75 cm³ (mean sample volume: 0.92 cm³) were prepared according to standard pollen preparation procedures (Moore et al., 1991), adding a known number of Lycopodium clavatum spores to enable concentration calculations (Stockmarr, 1971). MIC was counted on pollen slides following Finsinger and Tinner (2005). MAC samples with volumes between 1 and 2 cm³ (with the exception Jezioro Gościąż, where samples had to be amalgamated resulting in volumes between 1.75 and 4 cm³) were sieved through a 100 µm mesh-size sieve (Colombaroli et al., 2014), and particles were analysed under a stereo microscope according to the method described in Whitlock and Anderson (2003). For Sarsjön, not enough sediment was available for MAC analysis because varves were too thin (<1–2 mm in thickness, refer to Appendix 2, available online, for sample details).

We standardized MIC and MAC concentrations to 1 cm³. Subsequently, MIC and MAC concentration were multiplied with the varve thickness (corresponding to the annual sedimentation rate in cm yr⁻¹) to obtain influx values. We corrected for an apparent higher varve thickness resulting from the uncompressed (i.e. high water content) sediments at the surface of the frozen core by averaging the varve thicknesses of the five previous yearly layers and by using this estimate to calculate influx for the first layer.

Satellite data

Moderate Resolution Imaging Spectroradiometer (MODIS)-derived satellite products were used to estimate fire number, FRP (in W m⁻²) and total burned area (km²) per year since the year 2001. The Global Monthly Fire Location product (MCD14ML, collection 5.1; Giglio et al., 2003; Justice et al., 2006; Kaufman et al., 2003) was implemented to retrieve fire number and FRP data, while the shapefile version of the MODIS burned area product (MCD45monthly, collection 5.1; provided by the University of Maryland; Roy et al., 2002, 2005, 2008) was used to build the burned area database. The satellite products were further processed as described in Adolf et al. (2018) to remove false fires and to reduce biases due to temporal and spatial autocorrelations (Giglio, 2013). This was done by using an urban area filter from GlobCover 2009 (Arino et al., 2012) and by removing all fire pixels of which the centres were within 100 m of each other and which were detected within eight days (Adolf et al., 2018; Oliveras et al., 2014). We extracted fire number, FRP and burned area data from buffers with increasing radii (in km) from the lakes’ deepest part to identify local (0–1 km), extra-local (1–5 km) and regional (5–100 km) fires through their associated, remotely sensed fire parameters. The radii increased from 1 to 25 km in 1 km steps, from 30 to 200 km in 10 km steps, and also included the radii of 75 and 125 km.

Leads and lags and application of regression equations

We compared charcoal influx to the fire products for corresponding 12 years (April/May 2001–April/May 2013) by using an approach similar to Adolf et al. (2018), involving the calculation of correlation coefficients. To assess whether leads and lags are present when comparing charcoal to the remotely sensed fire parameters, we calculated cross-correlations (Bahrenberg et al., 1992) for all radii at ±3 lags (in years) according to the prerequisites for time-series analyses (≤n/4; Box et al., 2008). Highest correlations between the fire products (x = fire parameters) and charcoal (y = charcoal influx) at positive lags were interpreted as delayed charcoal deposition after fires. Highest correlations at negative lags (i.e. charcoal increases preceding the fire) were interpreted as indicative of chronological problems and thus not considered. Indeed, it is impossible that charcoal particles are incorporated and sealed in already existing varves if contamination issues during coring and/or sampling are excluded. However, as varves were thin and the limit between varves was not always sharp and uniform, we expect errors in the sampling of exact years, which may cause high correlations at negative lags. We used the following regression equations from Adolf et al. (2018) to calculate predicted values for the different fire products:

l o g_{10} (FN + 1) = 0.659 * l o g_{10} (tMICi + 1) - 1.6577

(1)

l o g_{10} (FN + 1) = 2.1802 * \log_{10} (tMACi + 1) + 0.207

(2)

\log_{10} (FRP + 1) = 0.9349 * \log_{10} (tMICi + 1) - 1.5111

(3)

\log_{10} (FRP + 1) = 3.2172 * \log_{10} (tMACi + 1) + 1.1002

(4)

\log_{10} (tBA + 1) = 2.6429 * \log_{10} (tMACi + 1) + 0.712

(5)

where FN stands for fire number, tMICi for total MIC influx and tMACi for total MAC influx. The equations listed refer to the Eqs 1, 3, 4, 6 and 9, respectively, of the article by Adolf et al. (2018). They are based on a European calibration dataset and include charcoal data from 37 lakes. Because Eqs 1 and 3 have negative intercept values, they are only valid for tMICi values ≥326.7 and ≥40.3, where the target variables become positive. We do not present the equation linking MIC influx with burned area because this relationship was not found significant in Adolf et al. (2018). While Eqs 1–4 are expected to reconstruct fire number and FRP within a radius of 40 km from lakes’ deepest points, Eq. (5) is expected to reconstruct burned area within a radius of up to 180 km. By using the source areas on which each regression equation is based, we extracted the observed values from the satellite data for the period 2001–2013. We calculated Pearson’s product–moment correlation coefficient (r) and the coefficient of determination (R², adjusted) for the linear relationship between observed and predicted values by using MIC and MAC influx from the four study sites. The regression equations were built on the basis of logarithmically transformed data (log₁₀(x + 1)), to correct for non-normality of residuals (Juggins and Birks, 2012). Therefore, we plotted the untransformed data on a logarithmic scale to show the original units of the fire products. In a final step, we displayed the distributions of the predicted values per MIC and MAC and observed fire parameters (taking all sites together) by using kernel density estimation (Parzen, 1962; Rosenblatt, 1956) to estimate the shape of the data distributions. We then compared the predicted and observed distributions per MIC/MAC and per satellite-derived fire parameter by indicating the median and the median absolute deviation. In a last step, we pooled MIC and MAC predicted values and compared their distribution to the distribution of the observed data for fire number and fire intensity. It was not possible to apply this last step to burned areas because the MIC regression developed for this parameter is not recommended for reconstructions (Adolf et al., 2018). All calculations and dataset processing were done in R software (R Core Team, 2016), by using the packages stats, ggplot2, readxl, plyr, dplyr and reshape (R Core Team, 2016; Wickham, 2007, 2009, 2011; Wickham and Bryan, 2017; Wickham et al., 2017).

Results

Chronology and lag analysis

Varves were clearest for Sarsjön and Lago Piccolo di Avigliana, while the annual layers of Lago Grande di Monticchio were less regular and defined and the laminations of Jezioro Gościąż could not be identified with sufficient confidence. The Lago Piccolo di Avigliana and Lago Grande di Monticchio varve chronologies show good agreement with the ²¹⁰Pb-based CRS depth-age models (Figure 2, Appendix 3, available online). The ¹³⁷Cs curve further supports the chosen CRS values (Appendix 3, available online). The Jezioro Gościąż CRS depth-age model revealed that the samples taken at regular intervals contain mostly only half a year. We therefore amalgamated samples to match the annual resolution of the satellite data, so that the time series for Jezioro Gościąż was reduced to 9 instead of 12 years. We then calculated influx values by using the sedimentation rates derived from the Jezioro Gościąż CRS depth-age model. The Lago Grande di Monticchio and Jezioro Gościąż cores were additionally too short, so that the depth where the equilibrium between supported and unsupported ²¹⁰Pb is reached could not be measured. We therefore calculated the depth of equilibrium based on the rate of decline of unsupported ²¹⁰Pb concentration from the available measurements per sediment core, assuming equilibrium at unsupported ²¹⁰Pb concentrations of 1 Bq kg⁻¹ (as measured in Morlock et al., 2017 and seen in Lago Piccolo di Avigliana, the best dated sequence of this study). This calculated equilibrium depth was in turn used for the age estimations using the CRS model (Appleby, 2001). The Sarsjön varve chronology is in good agreement with thicker silt layers and nearby flood events registered in the Swedish flood database of Västerbottens Län (MSB, 2012; Figure 2). However, both varve and CRS depth-age models have errors of more than 1.5 years (Appendix 3, available online). Cross correlation analysis revealed the presence of either no lags or predominant lags at +1 to +2 (Figure 3, Appendix 4, available online). As each lag represents a year, this indicates that on average, charcoal is deposited within the same year of the fire (lag 0, see Figure 3) or up to 1 or 2 years after a fire (lag +1 and +2). For Sarsjön, the predominant lag was of +2. Significant positive correlations at negative lags were found for all lakes and were mainly significant at lags −1 and −2, probably indicating sampling limitations at such fine scales. However, lags for each site are not consistent and varied when different satellite-derived fire parameters were taken into consideration. A tentative correlative comparison unrelated to evaluating the transfer functions reveals no clear spatial sources for MIC and MAC, respectively (Figure 3, Appendices 5a and b, available online). Nevertheless, correlation coefficients increase and stabilize at or after 40 km radius (Appendices 5a and b, available online) for most sites, irrespective whether MIC or MAC is considered. This suggests overall regional source areas for both proxies, in agreement with the findings of the calibrations study of Adolf et al. (2018).

Figure 2.

Age-depth models for the four study sites. Green, hollow symbols refer to the samples taken according to varve counts; red, filled symbols refer to the ²¹⁰Pb-based dates according to the constant rate of supply depth-age model. Blue, filled symbols for Sarsjön refer to wide and light layers in the Sarsjön sediment sequence.

Figure 3.

Leads and lags in charcoal deposition if compared with satellite data, unrelated to the transfer functions developed in Adolf et al. (2018). Cross-correlation results showing the three highest correlation coefficients r among satellite-derived fire parameters and microscopic (MIC) and macroscopic (MAC) charcoal influx for variable spatial scales; ordered according to site (rows) and satellite-derived fire parameter (columns). Km values refer to the radius at which the highest correlation was found. Dashed lines indicate the 95% confidence intervals.

Comparison of satellite products

The satellites did not register any local fires around the four sites from 2001 to 2013. However, extra-local fires were detected at Lago Piccolo di Avigliana and Lago Grande di Monticchio, where the closest fire pixels were situated at 3–4 km distance. The closest burned area pixels were 11 km away from Lago Piccolo di Avigliana, indicating a mismatch between the fires detected by the active fire product and those by the burned area product. This mismatch is less evident for Lago Grande di Monticchio, where both active fire and burned area products indicate the presence of fires within 3–4 km. At Sarsjön and Jezioro Gościąż, MODIS sensors only detected regional fires (at a distance >5 km radius) during the study period. We found the closest active fire pixel at 16 and 22 km, and the closest burned area pixel at only 60 and 21 km for Sarsjön and Jezioro Gościąż, respectively. We attribute detection differences to land-cover-dependent (i.e. forested versus non-forested) fire-detection differences among both MODIS products algorithms (Roy et al., 2008), as was already observed in the study by Adolf et al. (2018).

Validation of regression models

We applied the above-mentioned regression equations to the MIC and MAC records and compared predicted versus observed fire number, FRP and burned area (Figure 4, Appendix 6, available online). The regression models of Adolf et al. (2018) were created for log₁₀(x + 1)-transformed data; therefore, we show the R² for both the transformed and untransformed data. The relationships between predicted and observed values for all variables taken into account are significant (p < 0.05) reaching r values of 0.37–0.77 and explaining 12–59% (R²) of the variability in the data. Interestingly, when the predicted and observed means (9 or 12 years) for all sites were correlated against each other, relationships improved with r values ranging from 0.53 to 0.98 for the log-transformed data, suggesting that high inter-sample variability may have reduced the correlation coefficients. Yet, the sample size for correlations of means is extremely small (N = 4 for MIC and N = 3 for MAC) and correlations with the untransformed data were strongly biased by leverage effects of single data points. The unrealistic correlation of r = 1 (Figure 4) between untransformed observed and predicted burned area is thus primarily an artefact of insufficient numbers. Despite such difficulties, when single lakes are considered, the overall means of predicted vs observed fire parameters fit best for Lago Piccolo di Avigliana and Sarsjön (Figure 4, means consistently closer to 1:1 line), the two lakes with clear varves.

Figure 4.

Observed versus predicted values for microscopic charcoal (MIC, left column) and macroscopic charcoal (MAC, right column) influx. Top r and R² values represent the correlation coefficient and the coefficient of determination calculated on untransformed values with r and R² values just below referring to those calculated on log₁₀(x + 1)-transformed data. Bigger, salmon-coloured symbols represent the mean values for each site. The salmon-coloured r values in the lower part of the graphs show the correlation coefficients for the mean values, the first line referring to untransformed values and the last line to log₁₀(x + 1)-transformed values.

It is evident that predicted fire parameter values from MIC are overestimated at the lower end of the gradient and underestimated at the higher end of the gradient, meaning that the MIC-based regressions are not able to estimate the full (observed) fire variability. In contrast, MAC-inferred fire parameters overestimate values in the higher end of the gradient, meaning that the MAC-based regressions overestimate maximum fire activity. In general, burned areas are overestimated by MAC. While the means are to some extent influenced by the outliers, the median of the observed fire parameters can be accurately predicted by either using MIC or MAC. The results indicate that there is more uncertainty associated with the reconstruction of the variability of the observed fire parameters, than when the median values are reconstructed (Figure 5, Table 1). However, when the predicted values from MIC and MAC are pooled together, the correspondence among the variability of predicted and observed fire parameters improves (Figure 6), suggesting that MIC and MAC reconstructions are complementary.

Figure 5.

Comparison among predicted (dot-dashed lines) and observed (continuous lines) data distributions (kernel density estimations, Parzen (1962) and Rosenblatt (1956)), ordered by fire parameters (rows) and charcoal type (columns). Medians are indicated by full circles and the median absolute deviations by dashed lines.

Table 1.

Comparison among overall means and medians between predicted and observed values for untransformed and transformed data.

	FN (all sites)		FRP (all sites)		BA (all sites)
	Pred	Obs	Pred	Obs	Pred	Obs
Untransformed
Mean MIC	7.6 ± 5.6	43.8 ± 68.7	162.3 ± 152.4	1,397.6 ± 2,379.3	–	–
Mean MAC	426.1 ± 822.7	59.0 ± 74.9	91,211 ± 212,844	1,844 ± 2,643	5,930 ± 12,683	321.0 ± 505.0
Median MIC	5.4 ± 2.0	11 ± 10.9	95.3 ± 39.0	356.4 ± 341.0	–	–
Median MAC	45.5 ± 44.9	24 ± 22	1798.7 ± 1787.1	436.3 ± 420.9	302.6 ± 298.5	19.9 ± 19.1
Log₁₀(x + 1)-transformed
Mean MIC	0.9 ± 0.3	1.1 ± 0.8	2.1 ± 0.4	2.3 ± 1.2	–	–
Mean MAC	1.6 ± 1.1	1.4 ± 0.7	3.2 ± 1.6	2.9 ± 0.7	2.4 ± 1.3	1.5 ± 1.1
Median MIC	0.8 ± 0.1	1.1 ± 0.6	2.0 ± 0.2	2.6 ± 0.7	–	–
Median MAC	1.7 ± 1.01	1.4 ± 0.6	3.3 ± 1.5	2.7 ± 0.6	2.5 ± 1.2	1.3 ± 1.1

Pred: predicted; Obs: observed; FN: fire number; FRP: fire radiative power (W m⁻²); BA: burned area (km²); MIC: microscopic charcoal; MAC: macroscopic charcoal.

The mean for predicted and observed values for MIC include data from all four sites, while those for MAC only include three sites (all except Sarsjön).

Figure 6.

Pooled microscopic (MIC) and macroscopic (MAC) charcoal distribution curves (kernel density estimations, Parzen (1962) and Rosenblatt (1956)) for predicted (dot-dashed lines) and observed (continuous lines) values of fire number and fire radiative power (FRP) (W m⁻²).

Discussion

Signal of sediment charcoal in response to fire parameters

Several studies provide evidence for the existence of a lagged charcoal signal after fire of up to 5 years (Millspaugh and Whitlock, 1995; Tinner et al., 1998; Whitlock and Millspaugh, 1996). This lagged signal in the sediments can primarily be explained by long-term floating, by secondary transport of charcoal from the hydrological catchment due to streams or erosion and/or by secondary re-deposition within the lake (Clark et al., 1996; Whitlock and Millspaugh, 1996). Secondary transport of charcoal through erosion is especially important if a fire has burned within the watershed of a lake (Higuera et al., 2007; Whitlock and Millspaugh, 1996). In this study, we examined lags 0–3 years after a fire, finding mainly evidence for charcoal deposition within the same year (no lag) or up to 2 years later (lag 2), roughly agreeing with previous estimates of charcoal lags (Millspaugh and Whitlock, 1995; Tinner et al., 1998; Whitlock and Millspaugh, 1996). However, different lags prevail in each of the study lakes and there are different lags within sites when different satellite-derived fire parameters are considered (Figure 3). This is explained by chronological uncertainties and the prevailing differences of the study lakes in respect to size, steepness of surrounding slopes, number of inlets and outlets as well as type of vegetation cover – all factors which affect the amount of sedimentary charcoal delivered to the lake bottom (Whitlock and Larsen, 2001). In addition, differences may be related to the two remote-sensing fire products and their distinct algorithms used to register fire number, FRP and burned area. For instance, the MODIS active fire products, which detect fire number and FRP, can only register fires that burn at the time of satellite overpass (Giglio, 2013). While the MODIS burned area product is not limited by satellite overfly time, it has more difficulties in detecting burned areas in, for example, densely forested regions (Roy et al., 2008). However, there are also significant correlations with negative lags, hinting towards chronological uncertainties and sampling issues. In lakes which do not show annual layers, sediment mixing due to bioturbation of, for example, sediment-dwelling organisms may lead to homogenization of the ca. 10–20 years of deposited sediments (Whitlock and Larsen, 2001). On the basis of the varves, we can exclude such blurring effects at least at the three visibly annually laminated sites (Sarsjön, Lago Piccolo di Avigliana and Lago Grande di Monticchio).

After applying the transfer functions to the charcoal influx data, all correlations between charcoal influx–predicted and satellite-observed fire parameters were significant, the best relationships explaining up to 59% of variability in the data (Figure 4). The relationships between observed and predicted values are different for MIC and MAC. For instance, MAC is better than MIC in predicting low values. Indeed, MIC influx was never zero in the continental-scale calibration study (Adolf et al., 2018), even where no fires were registered at the largest radii around the lakes. In addition, MIC has also been found on remote places such as Svalbard, far away from any large vegetation fire sources (Hicks and Isaksson, 2006). This suggests that a certain amount of MIC comes from large source areas (e.g. sub-continental), potentially affecting regional fire activity reconstructions. In addition, even though largest MIC influx values in the samples vary considerably, ranging from 21,667 to 39,590 particles cm⁻² yr⁻¹, the corresponding predicted fire parameter values change only moderately (e.g. predicted fire number from 19 to 23, respectively), resulting in low variability of predicted values. This is explained by the regression slopes which are less steep for MIC than for MAC (see Eqs 1–4).

Overestimation of predicted values at the high end of the gradient by MAC mainly results from Lago Grande di Monticchio. The majority of MAC particles in Lago Grande di Monticchio samples were identified as belonging to the ‘grass’ morphotype (data not shown, but see Adolf et al. (2018), Colombaroli et al. (2014), Jensen et al. (2007) and Umbanhowar and McGrath (1998) for details on morphological classification). These are long (width-to-length ratio larger than 4) and thin particles that are more likely to break during transport into the sediment or sample preparations, potentially artificially increasing charcoal count values. However, for MIC, there is evidence against significant charcoal-breakage even during rough pollen preparation procedures (Tinner and Hu, 2003) which may explain why this problem does not occur for MIC. In Jezioro Gościąż, consistently overestimated predicted fire parameter values from MIC and MAC are observed. This can be explained by sedimentological studies of Jezioro Gościąż, which found that sediment focusing occurs in the deep part of the lake as a result of sediment from shallow parts being redistributed into the deeper parts of the lake (Mieszczankin, 1997; Mieszczankin and Noryśkiewicz, 2000; Rozanski et al., 2010). Combining MIC and MAC reconstructions as described in this study is a tentative approach that is based on the observation that at the continental scale, MIC and MAC come from similar source areas (Adolf et al., 2018). Applying them together may not be necessary, since both approaches yield reliable median estimates of important fire parameters.

Validation of calibration studies

Previous charcoal calibration studies have been conducted in different biomes, finding significant links between sedimentary charcoal abundance and either fire number, intensity or area burned on local and regional scales (e.g. Duffin et al., 2008; Kelly et al., 2013; Leys et al., 2015; MacDonald et al., 1991; Swain, 1973; Tinner et al., 1998; Whitlock and Millspaugh, 1996). However, to our knowledge, this study is the first to validate a continental-scale calibration study with independent charcoal sequences covering a continental gradient of vegetation. We observe significant relations among all predicted and observed variables that were tested, indicating that the regression equations developed in Adolf et al. (2018) may be applied to other sequences in Europe. We do, however, see important site-specific differences in regard to the strength of correlations. Nevertheless, when mean values of 12 years of samples and fire parameter data (9 years in the case of Jezioro Gościąż) are considered, relationships improve and predicted and observed values fit better (Figure 4). This possibly reflects the reduced effect of chronological errors when longer time spans are considered, as well as dampening the effect of stochastic erosion events or sediment focusing processes affecting charcoal influx values. This point is of particular importance given that the highest temporal resolutions reached in palaeoecological studies very rarely reach <10–20 years (e.g. Birks, 1997; Whitlock and Larsen, 2001) implying that the equations may perform better if applied to Holocene or Quaternary time series. However, as MODIS satellite data are only available since 2001, no longer time series could be validated and the sample size is greatly reduced by calculating correlations among means. Such validations might be aspired in future by comparing sedimentary data with historical observations (e.g. Tinner et al., 1998).

Conclusion

If applied to down-core sedimentary MIC and MAC series, recently developed regression equations (Adolf et al., 2018) are able to produce fire-regime parameter predictions (fire numbers, fire intensity and burned areas) that are significantly correlated with the observed fire-regime parameters as derived from satellite images. Relationships substantially improve when mean and median values over 12 or 9 years are considered. The observed variability in fire parameters is not well captured by the predicted MIC and MAC values. However, when the MIC and MAC predicted values are pooled together, the accuracy of the predictions in regard to both, medians and variability, improves. This result advocates for combining MIC and MAC for numeric, transfer-function-based reconstructions of important fire-regime parameters. Our validation study demonstrates that the available regression equations (Adolf et al., 2018) may be used to reconstruct fire parameters such as fire number, FRP and burned area in Europe. Assuming a robust physical link between fire parameters (e.g. frequency, intensity, size) and charcoal production, the transfer functions of Adolf et al. (2018) may be applied to other continents, as long as the reconstructed charcoal influx values remain comparable to those observed in the training set. Ultimately, our study shows that sediment-trap-based calibration studies can be applied to sediment records at continental scales. This may further encourage the use of sediment traps to improve the understanding of the link between lake-sediment proxies and present-day environmental variables for quantitative reconstructions of environmental variables.

Supplemental Material

SI_Adolf_et_al_The_Holocene – Supplemental material for Validating a continental European charcoal calibration dataset

Supplemental material, SI_Adolf_et_al_The_Holocene for Validating a continental European charcoal calibration dataset by Carole Adolf, Fabienne Doyon, Fabian Klimmek and Willy Tinner in The Holocene

Footnotes

Acknowledgements

The authors thank Christian Bigler for the introduction into freeze-coring and freeze-core sampling in Sweden. Sönke Szidat is greatly thanked for his availability regarding questions concerning the isotopic dating of sequences. Walter Finsinger and an anonymous reviewer are greatly thanked for their valuable comments which significantly improved the manuscript. The authors also thank Mariusz Gałka, Tiziana Pedrotta and Willi Tanner for assistance during field work and the Polish Environmental office, the associations of Laghi di Avigliana in Piemonte and Laghi di Monticchio for research permits.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

This research was funded by the Swiss National Science Foundation (project number: 200021_134616/1).

ORCID iD

Carole Adolf

References

Adolf

Wunderle

Colombaroli

et al . (2018) The sedimentary and remote-sensing reflection of biomass burning in Europe. Global Ecology and Biogeography 27(2): 199–212.

Allen

JRM

Watts

McGee

et al . (2002) Holocene environmental variability – The record from Lago Grande di Monticchio, Italy. Quaternary International 88: 69–80.

Amann

Mauchle

Grosjean

(2014) Quantitative high-resolution warm season rainfall recorded in varved sediments of Lake Oeschinen, northern Swiss Alps: Calibration and validation AD 1901–2008. Journal of Paleolimnology 51(3): 375–391.

Appleby

(2001) Chronostratigraphic techniques in recent sediments. In: Last

Smol

(eds) Tracking Environmental Change Using Lake Sediments. Volume 1: Basin Analysis, Coring, and Chronological Techniques. Dordrecht: Kluwer Academic Publishers, pp. 171–203.

Arino

Ramos Perez

Kalogirou

et al . (2012) Global Land Cover Map for 2009 (GlobCover 2009). © European Space Agency (ESA) & Université catholique de Louvain (UCL), PANGAEA. Available at: https://doi.pangaea.de/10.1594/PANGAEA.787668.

Bahrenberg

Giese

Nipper

(1992) Statistische Methoden in der Geographie. Band 2, Multivariate Statistik. Stuttgart: Teubner.

Birks

HJB

(1997) Reconstructing environmental impacts of fire from the Holocene sedimentary record. In: Clark

Cachier

Goldammer

et al (eds) Sediment Records of Biomass Burning and Global Change. Berlin: Springer, pp. 295–311.

Bloesch

Burns

(1980) A critical review of sedimentation trap technique. Schweizerische Zeitschrift fur Hydrologie 42: 15–55.

Blois

Williams

Fitzpatrick

et al . (2013) Space can substitute for time in predicting climate-change effects on biodiversity. Proceedings of the National Academy of Sciences of the United States of America 110(23): 9374–9379.

10.

Box

Jenkins

Reinsel

(2008) Time Series Analysis: Forecasting and Control. 4th edn. Hoboken, NJ: John Wiley & Sons, Inc.

11.

Bradshaw

Rasmussen

Nielsen

et al . (2005) Mid- to late-Holocene land-use change and lake development at Dallund Sø, Denmark: Trends in lake primary production as reflected by algal and macrophyte remains. The Holocene 15(8): 1130–1142.

12.

Clark

Hussey

Tristram

(1996) Estimating the mass flux of charcoal from sedimentary records: Effects of particle size, morphology, and orientation. The Holocene 6(2): 129–144.

13.

Colombaroli

Ssemmanda

Gelorini

et al . (2014) Contrasting long-term records of biomass burning in wet and dry savannas of equatorial East Africa. Global Change Biology 20(9): 2903–2914.

14.

Davis

(1969) Palynology and environmental history during the quaternary period. American Scientist 57(3): 317–332.

15.

Duffin

Gillson

Willis

(2008) Testing the sensitivity of charcoal as an indicator of fire events in savanna environments: Quantitative predictions of fire proximity, area and intensity. The Holocene 18(2): 279–291.

16.

European Topic Centre on Biological Diversity (ETC/BD) (2009) DMEER: Digital Map of European Ecological Regions – European Environment Agency. Available at: http://www.eea.europa.eu/data-and-maps/figures/dmeer-digital-map-of-european-ecological-regions.

17.

Finsinger

Tinner

(2005) Minimum count sums for charcoal-concentration estimates in pollen slides: Accuracy and potential errors. The Holocene 15: 293–297.

18.

Finsinger

Tinner

(2006) Holocene vegetation and land-use changes in response to climatic changes in the forelands of the southwestern Alps, Italy. Journal of Quaternary Science 21(3): 243–258.

19.

Finsinger

Lane

van Den Brand

et al . (2011) The lateglacial Quercus expansion in the southern European Alps: Rapid vegetation response to a late Allerød climate warming? Journal of Quaternary Science 26(7): 694–702.

20.

Giglio

(2013) MODIS Collection 5 Active Fire Product User’s Guide (version 2.5). College Park, MD: University of Maryland, Department of Geographical Sciences. Available at: http://modis-fire.umd.edu/files/MODIS_Fire_Users_Guide_2.5.pdf.

21.

Giglio

Descloitres

Justice

et al . (2003) An enhanced contextual fire detection algorithm for MODIS. Remote Sensing of Environment 87(2–3): 273–282.

22.

Hicks

Isaksson

(2006) Assessing source areas of pollutants from studies of fly ash, charcoal, and pollen from Svalbard snow and ice. Journal of Geophysical Research Atmospheres 111(2): 1–9.

23.

Higuera

Peters

Brubaker

et al . (2007) Understanding the origin and analysis of sediment-charcoal records with a simulation model. Quaternary Science Reviews 26(13–14): 1790–1809.

24.

Jensen

Lynch

Calcote

et al . (2007) Interpretation of charcoal morphotypes in sediments from Ferry Lake, Wisconsin, USA: Do different plant fuel sources produce distinctive charcoal morphotypes? The Holocene 17(7): 907–915.

25.

Juggins

Birks

HJB

(2012) Quantitative environmental reconstructions from biological data. In: Birks

Lotter

Juggins

et al (eds) Tracking Environmental Change Using Lake Sediments, Data Handling and Numerical Techniques, vol. 5. Dordrecht: Springer, pp. 431–494.

26.

Justice

Giglio

Boschetti

et al . (2006) The MODIS fire products. Remote Sensing of Environment 83: 244–262.

27.

Kaufman

Ichoku

Giglio

et al . (2003) Fire and smoke observed from the Earth Observing System MODIS instrument – Products, validation, and operational use. International Journal of Remote Sensing 24(8): 1765–1781.

28.

Kelly

Chipman

Higuera

et al . (2013) Recent burning of boreal forests exceeds fire regime limits of the past 10,000 years. Proceedings of the National Academy of Sciences of the United States of America 110(32): 13055–13060.

29.

Leys

Brewer

McConaghy

et al . (2015) Fire history reconstruction in grassland ecosystems: Amount of charcoal reflects local area burned. Environmental Research Letters 10(11): 114009.

30.

MacDonald

Larsen

CPS

Szeicz

et al . (1991) The reconstruction of boreal forest fire history from lake sediments: A comparison of charcoal, pollen, sedimentological, and geochemical indices. Quaternary Science Reviews 10(1): 53–71.

31.

Mieszczankin

(1997) A spacio-temporal pattern of pollen sedimentation in a dimictic lake with laminated sediments. Water, Air and Soil Pollution 99: 587–592.

32.

Mieszczankin

Noryśkiewicz

(2000) Processes that can disturb the chronostratigraphy of laminated sediments and pollen deposition. Journal of Paleolimnology 23(2): 129–140.

33.

Millspaugh

Whitlock

(1995) A 750-year fire history based on lake sediment records in central Yellowstone National Park, USA. The Holocene 5(3): 283–292.

34.

Moore

Weeb

Collinson

(1991) Pollen Analysis. Oxford: Blackwell Science.

35.

Morlock

Schilder

van Hardenbroek

et al . (2017) Seasonality of cladoceran and bryozoan resting stage δ¹³C values and implications for their use as palaeolimnological indicators of lacustrine carbon cycle dynamics. Journal of Paleolimnology 57(2): 141–156.

36.

Myndigheten för samhällsskydd och beredskap (MSB). (2012) Översvämningar i Sverige 1901–2010. Karlstad: MSB.

37.

Oliveras

Anderson

Malhi

(2014) Application of remote sensing to understanding fire regimes and biomass burning emissions of the tropical Andes. Global Biogeochemical Cycles 28(4): 480–496.

38.

Parzen

(1962) On estimation of a probability density function and mode. Annals of Mathematical Statistics 33(3): 1065–1076.

39.

R Core Team (2016) R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.

40.

Ralska-Jasiewiczowa

Goslar

Madeyska

et al . (1998) Lake Gościąż, Central Poland. A Monographic Study. Part 1 (ed Ralska-Jasiewiczowa

Goslar

Madeyska

et al .). Kraków: W. Szafer Institute of Botany, Polish Academy of Sciences.

41.

Rosenblatt

(1956) Remarks on some nonparametric estimates of a density function. Annals of Mathematical Statistics 27(3): 832–837.

42.

Roy

Lewis

Justice

(2002) Burned area mapping using multi-temporal moderate spatial resolution data-a bi-directional reflectance model-based expectation approach. Remote Sensing of Environment 83(1–2): 263–286.

43.

Roy

Boschetti

Justice

et al . (2008) The collection 5 MODIS burned area product – Global evaluation by comparison with the MODIS active fire product. Remote Sensing of Environment 112(9): 3690–3707.

44.

Roy

Jin

Lewis

et al . (2005) Prototyping a global algorithm for systematic fire-affected area mapping using MODIS time series data. Remote Sensing of Environment 97(2): 137–162.

45.

Rozanski

Klisch

Wachniew

et al . (2010) Oxygen-isotope geothermometers in lacustrine sediments: New insights through combined d18O analyses of aquatic cellulose, authigenic calcite and biogenic silica in Lake Goscciaz, central Poland. Geochimica et Cosmochimica Acta 74(10): 2957–2969.

46.

Samartin

Heiri

Joos

et al . (2017) Warm Mediterranean mid-Holocene summers inferred from fossil midge assemblages. Nature Geoscience 10(3): 207–212.

47.

San-Miguel-Ayanz

Schulte

Schmuck

et al . (2012) Comprehensive Monitoring of Wildfires in Europe: The European Forest Fire Information System (EFFIS). Brussels: European Commission.

48.

Simpson

Shilland

Winterbottom

et al . (2005) Defining reference conditions for acidified waters using a modern analogue approach. Environmental Pollution 137(1): 119–133.

49.

Snowball

Sandgren

Petterson

(1999) The mineral magnetic properties of an annually laminated Holocene lake-sediment sequence in northern Sweden. The Holocene 9(3): 353–362.

50.

Stockmarr

(1971) Tablets with spores used in absolute pollen analysis. Pollen et Spores 8(4): 615–621.

51.

Swain

(1973) A history of fire and vegetation in Northwestern Minnesota as recorded in lake sediments. Quaternary Research 3: 383–396.

52.

Telford

Birks

HJB

(2011) A novel method for assessing the statistical significance of quantitative reconstructions inferred from biotic assemblages. Quaternary Science Reviews 20(9): 1272–1278.

53.

Tinner

(2003) Size parameters, size-class distribution and area-number relationship of microscopic charcoal: Relevance for fire reconstruction. The Holocene 13(4): 499–505.

54.

Tinner

Conedera

Ammann

et al . (1998) Pollen and charcoal in lake sediments compared with historically documented forest fires in southern Switzerland since AD 1920. The Holocene 8(1): 31–42.

55.

Umbanhowar

Jr McGrath

(1998) Experimental production and analysis of microscopic charcoal from wood, leaves and grasses. The Holocene 8: 341–346.

56.

von Gunten

Grosjean

Rein

et al . (2009) A quantitative high-resolution summer temperature reconstruction based on sedimentary pigments from Laguna Aculeo, central Chile, back to AD 850. The Holocene 19(6): 873–881.

57.

Welten

(1944) Pollenanalytische, stratigraphische und geochronologische Untersuchungen aus dem Faulenseemoos bei Spiez. Veröffentlichungen Geobotanisches Institut Rübel, Zürich 21: 1–201.

58.

Whitlock

Anderson

(2003) Fire history reconstructions based on sediment records from lakes and wetlands. In: Baldwin

Caldwell

Heldmaier

et al (eds) Fire and Climatic Change in Temperate Ecosystems of the Western Americas. New York: Springer Verlag, pp. 3–31.

59.

Whitlock

Larsen

(2001) Charcoal as a fire proxy. In: Smol

Birks

HJB

Last

(eds) Tracking Environmental Change Using Lake Sediments. Vol. 3: Terrestrial, Algal, and Siliceous Indicators. Dordrecht: Kluwer Academic Publishers, pp. 75–97.

60.

Whitlock

Millspaugh

(1996) Testing the assumptions of fire-history studies: An examination of modern charcoal accumulation in Yellowstone National Park, USA. The Holocene 6(1): 7–15.

61.

Wickham

(2007) Reshaping Data with the reshape Package. Journal of Statistical Software 21(12): 1–20.

62.

Wickham

(2009) ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag.

63.

Wickham

(2011) The split-apply-combine strategy for data analysis. Journal of Statistical Software 40(1): 1–29. Available at: http://www.jstatsoft.org/v40/i01/.

64.

Wickham

Bryan

(2017) readxl: Read excel files. Available at: https://cran.r-project.org/package=readxl.

65.

Wickham

Francois

Henry

et al . (2017) dplyr: A grammar of data manipulation. Available at: https://cran.r-project.org/package=dplyr.

66.

Wright

(1991) Coring tips. Journal of Paleolimnology 6(1): 37–49.

67.

Zolitschka

Negendank

JFW

(1996) Sedimentology, dating and palaeoclimatic interpretation of a 76.3 ka record from Lago Grande di Monticchio, southern Italy. Quaternary Science Reviews 15(2–3): 101–112.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.36 MB