1
- -
FIRE EFFECTS ON NITROGEN POOLS AND DYNAIMCS IN TERRESTRIAL ECOSYSTEMS: A META-ANALYSIS
Shiqiang Wan*, Dafeng Hui and Yiqi Luo
Department of Botany and Microbiology, University of Oklahoma,
Norman, OK 73019-0245, U. S. A.
* Corresponding author
Department of Botany and Microbiology
University of Oklahoma
770 Van Vleet Oval
Norman, OK 73019-0245
Tel: (405) 325-5003
FAX: (405) 325-7619
Email:
Abstract. A comprehensive and quantitative evaluation of fire effects on nitrogen (N) pools and dynamics in terrestrial ecosystems is critical for fire management. This study utilized a meta-analysis approach to synthesize 40185 datasets from 87 studies published from 1955 to 1999 describing fire effects on N pools and dynamics in terrestrial ecosystems. We examined six N response variables related to fire: fuel N amount (FNA) and concentration (FNC), soil N amount (SNA) and concentration (SNC), soil ammonium (NH4+) and nitrate (NO3) contents. When all comparisons (fire treatment vs. control) were considered together, fire significantly reduced FNA (58%) and increased soil NH4+ (94%) and NO3 (152%) contents and had no significant influences on FNC (–6%), SNA (–3%), or SNC (–3%). The responses of N to fire varied with different independent variables. The response of FNA to fire was significantly influenced by vegetation type, fuel type, fuel consumption amount and percentage. Reduction in FNA was linearly correlated with fuel consumption percentage (r2 = 0.978). FNC response to fire was only affected by fuel type. None of the seven independent variables had any effect on SNA. Sampling soil depth profoundly influenced the responses of SNC, NH4+, and NO3 to fire. The responses of both NH4+ and NO3 to fire were significantly affected by fire type and time after fire but with different temporal patterns. Soil NH4+ content increased by about two-fold right after fire, then gradually declined to pre-fire level after one year. Fire-induced increase in soil NO3 content was small (24%) right after fire, reached a maximum of three-fold of pre-fire content one year after fire, and then declined. Interactions among fire, N, and vegetation, and mechanisms and practices for replenishing and restoration of N after fire should be considered in fire management to maintain the balance of N cycling and the stability of primary productivity and structure in terrestrial ecosystems.
Key Words: Biomass, fire, forests, fuel, grasslands, meta-analysis, nitrogen, prescribed burning, response ratio, shrubland, slash burning, soil nitrogen availability, wildfire.
INTRODUCTION
Prior to 1930, fire was generally considered as a destructive and undesirable force that occurred with varying frequency in terrestrial ecosystems (Clements 1916, Fowells and Stephenson 1933). That point of view resulted in suppression of natural fire for almost half a century (Kozlowski and Ahlgren 1974). However, since the 1970s, critical scientific evaluations have indicated the potential usefulness of fire in terrestrial ecosystem managements (Kozlowski and Ahlgren 1974, Raison 1979), e.g., controlling fuel (combustible material) levels and avoiding catastrophic wildfire (Wagle and Eakle 1979), restoring forest ecosystems (Kaye et al. 1999, Vose et al. 1999), maintaining species composition and richness in grasslands (Collins and Wallace 1990, Collins et al. 1995, Pendergrass et al. 1999), and improving water yields in catchment areas (Schindler et al. 1980, Bosch et al. 1984).
As a powerful and instantaneous modifier of the environment, fire potentially has profound, long-term influence on nutrient cycles in terrestrial ecosystems. Intensive attentions have been paid to nitrogen (N) pools and dynamics associated with fire because N often limits primary productivity in natural terrestrial ecosystems (Christensen 1977, Woodmansee and Wallach 1981, Maars et al 1983, Fenn et al. 1998) and because N is easily lost during fuel combustion (Grier 1975, DeBano et al. 1979, Raison et al. 1985a, Gillon and Rapp 1989) since N volatizes at 200 ºC (White 1973), which can be easily exceeded in fire if fine fuel exceeds 3370 kg ha–1 (Stinson and Wright 1969).
Direct N losses during fuel combustion are usually in the forms of gasification, volatilization and ash convection (Christensen 1994). Related to types of vegetation, fire, and fuel (Binkley et al. 1992a), the amount of fuel N losses during burning can vary from 8.5 kg ha–1 in grasslands (Medina 1982) to 907 kg ha–1 in coniferous forest (Grier 1975) to 1604 kg ha–1 in Brazilian tropical forest (Kauffman et al. 1995). It is commonly accepted that the loss of fuel N through fuel combustion is significantly correlated with fuel consumption and/or fire intensity (DeBano and Conrad 1978, Raison et al. 1985a, Schoch and Binkley 1986, Feller 1988, Binkley and Christensen 1992, O’Connell and McCaw 1997, Belillas and Feller 1998).
Reports about fire impacts on total soil N contents are highly variable due to differences in vegetation (Dyrness et al. 1989), topography (Turner et al. 1997, Vose et al. 1999), fire regime (including frequency, intensity, and season) (Covington and Sackett 1984, Blair 1997), and sampling methods (Monleon et al. 1997). In contrast, most studies suggest a consistent pattern that fire can increase the availability of soil ammonium (NH4+) and nitrate (NO3) (Christensen 1973, Kovacic et al. 1986, Rapp 1990, Covington et al. 1991, Baldwin and Morse 1994, Kaye and Hart 1998). Covington and Sackett (1984) attribute these increases to the leaching of N from forest floor into the soil. But others suggest that the increases result from pyrolysis of organic material and increased mineralization after fire (Grove et al. 1986, Knoepp and Swank 1993, Baldwin and Morse 1994, Kaye and Hart 1998, Lynds and Baldwin 1998).
As the common and most limiting nutrient, N plays an extremely important role in post-fire recovery of primary productivity and vegetation in terrestrial ecosystems and need to be given special attention in fire management. An integration of the highly variable results about fire effects on N across various terrestrial ecosystems is essential to policy-making in fire management to preserve the stability and sustainability of terrestrial ecosystems. Such syntheses have been made via narrative reviews (Ahlgren and Ahlgren 1960, Kozlowski and Ahlgren 1974, Wells et al. 1979, Raison 1979, Woodmansee and Wallach 1981, Johnson et al. 1998). However, the conclusion regarding fire effects on terrestrial ecosystem N, particularly on total soil N content, remains controversy, partly because changes in N (e.g., total amount, forms, availability) are not always expressed on the same unit and/or sampled from the same soil depth (DeBano et al. 1998). To resolve the controversy, a comprehensive evaluation of fire effects on N with quantitative methods becomes necessary.
By integrating experimental results from different vegetation types, fire types, fuel types, time after fire, fuel consumption amounts and percentages, and sampling soil depths, this study uses a meta-analysis approach, which provides unbiased estimates of fire treatment effects across multiple studies, to address the following questions that are essential to policy makers and land managers. First, to what extent will various terrestrial ecosystem N pools be affected by fire? Second, how will different types of vegetation, fire, fuel, and sampling soil depth affect N responses to fire? Finally, what is the temporal pattern of N response after fire?
MATERIAL AND METHODS
Approach
Meta-analysis is a quantitative method to compare and synthesize the results of multiple studies. In a meta-analysis, studies are first gathered to address a common question or hypothesis, then the results of these studies are combined into an estimate of the magnitude of the effect (commonly called "effect size", which essentially means the magnitude of the effect of manipulation) across studies, after that, the significance and the homogeneity of the effect size are calculated, finally, the studies can be grouped according to various independent variables and the effect sizes between these groups of studies can be statistically compared. The response ratio (the ratio of mean of some measured variable between experimental and control groups) is commonly used as a measure of effect size because it quantifies the proportionate change that results from an experimental manipulation (Curtis and Wang 1998, Hedges et al. 1999). Compared to traditional narrative review, meta-analysis has the advantages of objectivity and better control of Type II errors (Failure to reject null hypotheses that are false) (Arnqvist and Wooster 1995) and has the potential to resolute of longstanding scientific debates (Gurevitch et al. 1992). A mixed model, which assumes that the studies within a group share a common mean effect and that there is random variation among studies in a group in addition to sampling variation, was used in our meta-analysis.
Extracting data from published results
Our literature survey was intended to be inclusive. We extracted data from publications in the literature on six N response variables: fuel N amount (FNA), fuel N concentration (FNC), soil N amount (SNA), soil N concentration (SNC), soil ammonium content (NH4+), and soil nitrate content (NO3). The seven independent variables were vegetation type, fire type, fuel type, time after fire (TAF), fuel consumption amount (FCA), fuel consumption percentage (FCP), and sampling soil depth. Vegetation types included broad-leaved forest (BF), coniferous forest (CF), grassland (GL), and shrubland (SB). Fire types were prescribed burning (Pres–B), slash burning (SL–B), and wildfire (WF). Fuel types were aboveground biomass (AGB), forest floor (FF), forest floor plus understory (FF+US), and slash plus forest floor (SL+FF). From 87 studies published between 1955 and 1999 (Database References), we examined 57, 48, 40, 62, 184, and 185 comparisons (control vs. fire treatment) for the analyses of FNA, FNC, SNA, SNC, NH4+, and NO3, respectively (see Table1).
Meta-analysis assumed the independence of elements being synthesized. Violation of this assumption (e.g., including multiple results from a single study) might alter the structure of the data, inflate samples and significance levels for statistical tests, and increase the probability of Type–I errors (Rejecting the null hypothesis when it is true) (Wolf 1986, Vander Werf 1992). Some researchers considered lack of independence to be a serious problem for meta-analysis, and thus they advocated the inclusion of only one result per study (Vander Werf 1992, Tonhasca and Byrne 1994, Koricheva et al. 1998). However, the loss of information caused by omission of multiple results in a single study might become a more serious problem than that caused by their nonindependence of those results (Hedges and Olkin 1985, Gurevitch et al. 1992). Many researchers therefore included more than one result from a single study in their meta-analysis (Gurevitch et al. 1992, Poulin 1994, Wooster 1994, Arnqvist and Wooster 1995, Curtis 1996, Curtis and Wang 1998). In the analysis of FNA, FNC, SNA, and SNC, when more than one vegetation types or stands were burned in a single fire or in several fires, the results were considered to be independent and are included. When there were data with different sampling dates from a single stand or vegetation type, we only took the result with the earliest sampling date. Soil available N (NH4+ and NO3) had an obviously seasonal and annual pattern after fire and was more dynamic than total soil N (Singh et al. 1991, Singh 1994, Covington et al. 1991). Two steps were used in analyses of soil NH4+ and NO3. First, all the results from different sampling dates in a single study (k = 184 and 185 for NH4+ and NO3, respectively) were taken together to address the temporal dynamics of NH4+ and NO3 after fire, whereas vegetation type, fuel type, fire type and soil sampling depth were not considered due to the large replications of these independent variables. Second, data groups from the peaks of fire responses of NH4+ (with TAF = 0 month, right after fire, k = 29) and NO3 (TAF = 7~12 month, k = 36) were selected and analyzed to address the effect of independent variables on the responses of soil NH4+ and NO3 contents to fire.
Means, standard deviations, and sample sizes were needed for the experimental and control groups in order to conduct the meta-analysis. We only included studies in which means, standard deviations, and sample sizes for the experimental and control groups could be derived or inferred from information in the articles. Where means and standard deviations of each treatment were reported, these data were used directly. Where data (means and some measures of variance) were presented in the form of graphs, figures were enlarged and manually digitized. If raw data of experimental and control groups were given, means and standard deviations were calculated. When means and standard error (SE) of each treatment were reported, as in most studies, the standard deviation (sd) was calculated as:
(1),
where n was the sample size. If data were given with a mean and a confidence interval (CI), the standard deviation was calculated as:
(2),
where CIu and CIl were the upper and lower limits of CI, and up was the significant level and equaled 1.96 when = 0.05 and 1.645 when = 0.10. When pre-burn and post-burn fuel N concentrations in different fuel components (such as forest floor, understory, and slash with different diameters) were given, means and standard deviations representing whole-system FNC were calculated. When there were several SNA and SNC data from different soil layers in a single study, only the value of the surface soil layer was used in our analysis. The units with which measurements were reported were not important since the calculated response ratios were dimensionless (Curtis 1996).
Response ratio
The meta-analysis was conducted on the natural logarithm of the response ratio using the statistical software MetaWin (Rosenberg et al. 1997). The response ratio, , was the ratio of mean in an experimental group () to that of the control group (). The response ratio was converted to the metric of the natural log as:
(3).
If and were normally distributed and both were greater than zero, lnRR was approximately normally distributed (Curtis and Wang 1998) with a mean equaled to the true log response ratio and variance (v) approximately equaled to
(4),
where ne and nc were the sample sizes in the experimental and control groups, respectively; seand sc were the standard deviations of comparisons in the experimental and control groups, respectively.
The meta-analysis calculated a weighted log response ratio (lnRR++) from individual
ln RRij (i = 1, 2, …, m; j = 1, 2, …, ki) by giving greater weight to studies whose estimates had greater precision (lower v) so that the precision of the combined estimate and the power of the tests would increase (Hedges and Olkin 1985, Gurevitch and Hedges 1999). The weighted mean log response ratio (ln RR++) was calculated by
(5)
with the standard error as:
(6),
where wij was the weighting factor and was estimated by
(7).
The 95% confidence interval for log response ratio was
(8).
The corresponding confidence limits for the response ratio were obtained by computing their respective antilogs. If the 95% CI of a response variable overlapped with zero, the response ratio was not significantly changed. If the 95% CIs of two groups overlapped, the response ratios of the two groups were not significantly different from each other.
Homogeneity test
One important purpose of meta-analysis was to determine whether independent groups are homogeneous with respect to response ratio (i.e., that observed differences in ln RRij among studies were due to sampling error) and whether there were significant differences in mean responses between these groups (Hedges and Olkin 1985). The total homogeneity (QT) could be partitioned into within-group homogeneity (the variation among comparisons within groups, QW) and between-group homogeneity (the variation in mean response ratio between groups, QB)as:
QB + QW = QT (9),
which was analogous to the practice of partitioning variation in an ANOVA. The total homogeneity (QT) was calculated as
(10)
with degrees of freedom (df) and comparisons partitioned into m groups, each group including ki comparisons. The between-group homogeneity (QB) was calculated as
(11)
with m-1 df. The within-group homogeneity was calculated as:
(12)
with df.
The Q statistic followed a 2 distribution, allowing a significance test of the null hypothesis. The greater the value of Q, the greater the homogeneity in response ratios among comparisons. If QB was larger than a critical value, an independent variable was said to have a significant influence on the response ratio. If comparisons within a group did not share a common response ratio (i.e. there was significant within-group homogeneity), then groups could be further subdivided and the process was repeated. In principle, groups could be subdivided until Qw was no longer significant (Gurevitch et al. 1992).
In our meta-analysis, in order to explore ecological causes of fire effects, the total data sets in each of the six response variables were divided according to the seven independent variables, which were types of vegetation, fire, fuel, time after fire (TAF), fuel consumption amount (FCA), fuel consumption percentage (FCP), and sampling soil depth (Table 2 and 3). Those statistically not significant interactions between the response variables and independent variables were not further examined in this study except the interactions of SNA and SNC with TAF. The latter interactions were examined to show the time course of fire effects on SNA and SNC.
RESULTS
Fire caused a significant reduction in FNA by 58% (Figure 1). The FNA response to fire was strongly influenced by the vegetation type (p < 0.01) (Table 3). The reduction in FNA in response to fire was 71% for the broad-leaved forests, 48% for the coniferous forests, 60% for the grasslands, and 71% for the shrublands (Figure 2a). A paired comparison indicated no overlapping of 95% confidence intervals between the broad-leaved and coniferous forests, suggesting that the FNA responses of the two terrestrial ecosystems to fire were significantly different. No significant difference was found among other vegetation types.
Fire-induced reductions in FNA also varied with fuel types, being 73% for the aboveground biomass (AGB), 54% for the forest floor (FF), 64% for FF plus understory biomass (US), and 49% for slash fuel (SL) plus FF (Figure 2b). The paired comparison indicated that fire effects on FNA were significantly different only between AGB and SL+FF.
Both FCA and FCP had significant influences on the response of FNA to fire (Table 2). But it did not appear to have a linear relationship between the fire-induced reduction in FNA and FCA across all studies. When FCA increased from 10 to 80 t ha–1, the reduction in FNA increased. When FCA was greater than 80 t ha–1, the deviation of the response of FNA to fire (Figure 3a) resulted from incomplete burning of coarse fuel and piled slash. Further partitioning of this data set according to vegetation types showed a similar variability (Figure 4a, 4b, and 4c). However, there was a significant linear correlation between fire-induced reduction in FNA (y) and FCP (x) as y = 0.926x 2.750 with the determinant coefficient R2 = 0.978) (Figure 3b). When FCP increased from 12% to 96%, fire-reduction in FNA increased from 12% to 97%. When the reduction in FNA was partitioned into different vegetation types, the relationships were still significant for the broad-leaved forests (y = 1.114x + 9.517, R2 = 0.965) (Figure 4d) and the coniferous forests (y = 1.044x + 4.219, R2 = 0.970) (Figure 4e) but not for the grasslands (Figure 4f) due to smaller sample sizes.