Vegetation burning is a common land management practice in Africa, where fire is used for hunting, livestock husbandry, pest control, food gathering, cropland fertilization, and wildfire prevention. Given such strong anthropogenic control of fire, we tested the hypotheses that fire activity displays weekly cycles, and that the week day with the fewest fires depends on regionally predominant religious affiliation. We also analyzed the effect of land use (anthrome) on weekly fire cycle significance. Fire density (fire counts.km-2) observed per week day in each region was modeled using a negative binomial regression model, with fire counts as response variable, region area as offset and a structured random effect to account for spatial dependence. Anthrome (settled, cropland, natural, rangeland), religion (Christian, Muslim, mixed) week day, and their 2-way and 3-way interactions were used as independent variables. Models were also built separately for each anthrome, relating regional fire density with week day and religious affiliation. Analysis revealed a significant interaction between religion and week day, i.e. regions with different religious affiliation (Christian, Muslim) display distinct weekly cycles of burning. However, the religion vs. week day interaction only is significant for croplands, i.e. fire activity in African croplands is significantly lower on Sunday in Christian regions and on Friday in Muslim regions. Magnitude of fire activity does not differ significantly among week days in rangelands and in natural areas, where fire use is under less strict control than in croplands. These findings can contribute towards improved specification of ignition patterns in regional/global vegetation fire models, and may lead to more accurate meteorological and chemical weather forecasting.
Citation: Pereira JMC, Oom D, Pereira P, Turkman AA, Turkman KF (2015) Religious Affiliation Modulates Weekly Cycles of Cropland Burning in Sub-Saharan Africa. PLoS ONE 10(9): e0139189. https://doi.org/10.1371/journal.pone.0139189
Editor: Philip Anglewicz, Tulane University School of Public Health, UNITED STATES
Received: October 14, 2014; Accepted: September 10, 2015; Published: September 29, 2015
Copyright: © 2015 Pereira et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: Anthromes data are publicly available from http://ecotope.org/anthromes/data/. The screened MODIS active fires data are publicly available at http://www.isa.utl.pt/cef/pub/modisactivefires-screend Global Administrative Areas data are publicly available at http://www.gadm.org/. Data from the 2010 World Religion database were kindly provided by the Gordon-Conwell Theological Seminary, Boston. Data source: Todd M. Johnson and Brian J. Grim, eds. World Religion Database (Leiden/Boston: Brill, accessed November, 2012). Researchers interested in obtaining the WRD data should contact email@example.com
Funding: Funding for this research was provided by Fundação para a Ciência e a Tecnologia (http://www.fct.pt/) research projects PEst-OE/AGR/UI0239/2014 (JMCP and DO), PEst-OE/MAT/UI0006/2014 (KFT, AT and PP), PTDC/MAT/118335/2010 (KFT, AT, PP and JMCP) and doctoral grant SFRH/BD/4752/2008 (DO). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Vegetation burning displays spatial and temporal patterns at various scales [1,2,3] affected by the seasonal occurrence of dry periods and thunderstorms, and by sporadic heat waves and droughts [4,5]. Land management practices, which often rely on fire, also influence those patterns, timing its use according to crop calendars and ecoclimatic constraints . African vegetation fires are mostly anthropogenic, used for hunting, livestock husbandry, pest control, facilitating food gathering, managing agricultural plot fertility, and for reducing the risk of large, uncontrolled fires . Although the annual and daily dynamics of African biomass burning have been analyzed , the occurrence of intermediate scale temporal patterns is undocumented. Biomass burning, a large source of greenhouse gases and aerosols, constitutes a major factor controlling the inter-annual variability of atmospheric composition , and has been analysed at time scales ranging from a million years , to millennia , centuries , years , and days . Patterns found in these studies were interpreted as a function both of climate dynamics and human activity. Regular cycles in vegetation burning are found at annual scale, mostly in seasonally dry areas, and at the daily scale as a consequence of underlying cycles in meteorological variables and land management practices. Weekly cycles were detected in aerosols and cloud droplet number concentration in several studies, and models confirm that the difference in emissions between week days and weekends generates this cycle. Since there is no natural forcing with a seven-day period, weekly cycles provide evidence of an anthropogenic fingerprint . A weekly cycle of aerosol optical depth (AOD), observed in the United States (U.S.) and in Central Europe, is stronger in urban sites than in rural sites. A reversed AOD weekly cycle occurs in the Middle East and India, with lower AODs on Thursday and Friday, the weekend for Middle Eastern cultures. Therefore, cultural practices and the industrial activity schedule of this region are consistent with the observed AOD weekend effect . Chemical weather forecasting models use meteorological and emissions information as input to simulate transport, chemistry, and deposition processes to predict short-term variability of air quality. Good performance of a chemical weather forecast model requires accurate representation of the temporal distribution of natural and anthropogenic emissions. However, some atmospheric chemistry modelling studies have assumed that, contrary to emissions from industry and transportation, biomass burning does not display weekly cycles [14,16]. Considering that vegetation burning in Africa is mostly anthropogenic and used as a land management tool [4,7,17], we tested the hypotheses that there is a day of the week with lower fire activity, and that this day depends on the regionally predominant religious affiliation: Sunday on predominantly Christian areas, and Friday on predominantly Muslim areas. We also tested the effect of regionally dominant land use on significance of the weekly fire activity minima, which is expected to depend on land management intensity. Our analysis is restricted to the two major religions in sub-Saharan Africa, which account for 62.9% (Christians) and 30.2% (Muslims) of the total population, and are the only ones that prescribe a weekly day of rest [18, 19]. African traditional religions are followed by 3.3% of sub-Saharan Africans, while Hindus and Buddhists jointly represent less than 0.3% of the population, 0.2% have other religious affiliations, including Judaism, and 3.2% are not religiously affiliated . Muslims are concentrated in northern hemisphere sub-Saharan Africa, in a belt extending from the southern part of Sudan, in the east, to Senegal and Sierra Leone in the west, while the southern hemisphere is very predominantly Christian, with the exception of the east coast from northern Mozambique to Somalia. Additional information on the geography of religion in Africa, as classified for the purposes of our analysis, is provided in Materials and Methods.
Materials and Methods
Religious affiliation, land use, and fire activity spatial data
The geographical basis for our analysis is the Global Administrative Areas (GADM, version 2, Jan 2012, http://www.gadm.org/) first level divisions for Africa, which contains a total of 554 units and is the spatial support for the religious affiliation data (Fig 1A) of the 2010 World Religion Database . Land use / land cover data (Fig 1B) are from the Anthropogenic Biomes of the World (Anthromes) dataset , at 5' (ca. 9 km) spatial resolution and dated from the year 2000. Anthrome data incorporate information on population density, land cover and land use and map global patterns of sustained direct human interaction with ecosystems . Given the strongly anthropogenic character of vegetation burning in Africa they were considered particularly suited for our analysis and preferred over biophysical variables such as vegetation type or land cover. The alternative of using annual land cover data (e.g. the MODIS MCD12Q1 product) would suffer from the interannual classification inconsistencies reported by . The relative stability of spatial patterns depicted by the Anthromes dataset justifies its use to analyze active fire counts from the period 2003–2011.
a) Classification of predominant religious affiliation; b) Classification of predominant anthrome.
Daily fire activity was quantified with data from the NASA MODIS MCD14ML Collection 5 active fire product  for the period 2003–2011 (TERRA and AQUA sensors) at 1km spatial resolution, after screening for false alarms and non-vegetation fires according to the procedures detailed in . Fig 2 shows regional fire count densities per week day (Fig 3A–3G) and total density.
Religious affiliation and land use classification
The screened MODIS active fires dataset contains a total of 22,749,687 observations. One hundred and sixty one GADM regions with less than 0.1 km-2 active fire counts accumulated over the nine-year study period were excluded from analysis. An additional nineteen regions where the sum of Christian and Muslim believers falls below 50% of the total population were also removed from analysis, since a weekly day of rest is specific of the Abrahamic monotheisms [18, 19], while in such regions most of the population follows African traditional religions. Two more regions were excluded due to lack of data on religious affiliation, resulting in a total of 372 spatial units analyzed. Wherever Christians (Muslims) represent over 75% of the sum of Christian + Muslim believers, the region was considered Christian (Muslim). When neither religion accounted for at least 75% of the population, the region was labelled mixed Christian-Muslim (Fig 1A). Classification of each region in terms of dominant land use relied on the highest level Anthromes legend  with aggregation of the Dense Settlement and Villages classes into a single class (Settled) and the Forest class and Wildland class into a single Natural class. Cropland and Rangeland classes were not altered. The percent area in each region covered by each of the four new classes was calculated and wherever a single class represented at least 50% of the area, the region was labelled as belonging to that class. If no single class exceeded the 50% area threshold, the region was considered as having Mixed land use (Fig 1B). Religious affiliation and Anthrome classification thresholds were selected with the aim of obtaining clear dominance of a class in each region, while ensuring that the respective mixed classes do not cover an unduly large fraction of the study area. Evidently, this was more easily achieved for the three-class religious affiliation variable than for the five class Anthrome variable. Table 1 shows the number of regions by religion and by anthrome, based on these classification rules. Fifty-eight percent of regions (216/372) were classified as Christian and 19% (72/372) as Muslim. Cropland is the anthrome accounting for the highest percentage of regions (164/372, 44%), while the Natural anthrome accounts for the fewest (32/372, 9%).
According to our thresholds for defining predominant religious affiliation, Muslim regions are mostly located in West Africa, in southeastern Chad and in the southern part of Sudan. Most of southern hemisphere Africa is Christian, with the exception of Tanzania and northern Mozambique. Predominantly Christian regions are also found in the southern regions of Ghana, Togo, Benin, Nigeria, and Cameroon, the western part of Central African Republic, South Sudan, and western Ethiopia. Mixed Christian/Muslim regions are located in West Africa, throughout most of Cameroon, in southwestern Chad, the eastern part of Central African Republic and part of Ethiopia, and also in most of Tanzania and northern Mozambique. African traditional religions are dominant in a few regions of West Africa, and in western Madagascar (Fig 1A). Aggregated Anthromes data show that croplands predominate in West Africa, Ethiopia, Uganda, Rwanda, and Burundi, and in parts of the Democratic Republic of Congo, Zimbabwe, and Tanzania. Natural areas are primarily located in central Africa, from the Central African Republic and Cameroon, down to eastern Angola and western Zambia. Rangelands occupy arid areas in the Sahel, and in southern Angola, Namibia, Botswana, and South Africa. Settled regions are concentrated in West Africa, and in the Great Lakes region. Finally, regions of Mixed land use, according to our definition, predominate in southern hemisphere Africa, namely in parts of Angola, Tanzania, Zambia, Mozambique, Madagascar, and South Africa (Fig 1B). Classification of regional fire counts by week day minimum reveals that most of Africa displays a weekly minimum of fire activity on Sunday, especially evident in the southern half of Nigeria, large parts of Cameroon, the Central African Republic, western Ethiopia, the Democratic Republic of Congo, Angola, Zambia, and Zimbabwe. Friday weekly fire minima are primarily found in West Africa, in the southern part of Sudan, and in western Madagascar (Fig 2).
The fire density observed in each region (Fig 3A–3G) was modeled using a negative binomial model  with fire counts as response variable, region area as offset and a structured random effect, in the form of a ICAR model , to account for spatial dependence. Anthrome, religion and week day, together with their two-way and three-way interactions were used as independent variables. A Bayesian methodology was used to fit these spatial generalized linear model using R-INLA ) and the respective package (www.r-inla.org). The best model was chosen using the deviance information criterion (DIC) . For each region i = 1, …, 372 characterized by anthrome A = 1, …, 5 and religion R = 1,2,3, the full model assumes that the total number of fires per year in each week day W = 1, …, 7 (YiW) are conditionally independent observations following the full negative binomial model, (Eq 1) with shape parameter θ and mean μiW such that: (Eq 2) where α is a vector of parameters accounting for the intercept, (α0) the effect of the three main factors anthrome (α1A), religion (α2R), and week day (α3W), together with the corresponding two-way (α1RW, α2RA, α3WA) and three-way (αRWA) interactions. The area of the region enters in the linear predictor as an offset, ln(areai). The term zi is the structured spatial random effect for region i, defined by the ICAR model.
To prevent confounding due to association of the anthrome variable both with fire density and religion, the hypothesis of a significant interaction between religion and week day was tested for each of the negative binomial regression models built separately for the different anthromes, taking spatial dependence into account. Again under a Bayesian framework, DIC values of the models were compared with and without the interaction term, and the model with the lowest DIC was selected.
We analyzed specific pairwise contrasts of interest, namely between Sunday and every other week day in Christian regions, and between Friday and every other week day in Muslim regions. The following hypotheses were tested: where 1 = Monday, 2 = Tuesday, 3 = Wednesday, 4 = Thursday, 5 = Friday, 6 = Saturday, and 7 = Sunday. All testing was performed under a Bayesian methodology, by computing, for each of the respective contrasts, the posterior contour probability of zero. The contour probability equals one minus the content of the highest posterior density (HPD) interval just covering zero , and is the probability that the posterior density function of the contrast is less than or equal to the posterior density function of the contrast at zero. This probability can be interpreted as a p-value in checking the posterior support of specific values of interest, an approach motivated by well-known analogies between Bayesian and classical inference concepts . To account for multiple comparisons, adjusted values for these probabilities were calculated using the Benjamini-Hochberg method . In addition 95% credible intervals were calculated and displayed.
Statistical models relating regional annual fire counts and regional area with week day of occurrence (nine annual replicates, 2003–2011), religious affiliation, and anthrome, and taking spatial dependence into account, were fit to the data (Table 2). The best model selected using DIC contained the two-way interactions between week day and land use, and between religion and land use, and the spatial random effect in the form of a ICAR model, revealing the existence of a significant interaction between religion and week day, i.e. regions with different religious affiliation (Christian, Muslim, and mixed) display distinct weekly cycles of vegetation burning (Fig 4A). The anthrome variable is strongly associated with religious affiliation and with fire density, acting as a confounding factor in the model. To address this problem, statistical models relating fire density in each region with week day of occurrence, religious affiliation, and the corresponding interaction term were built separately for each anthrome, taking into account spatial dependence. The best model in each anthrome was chosen using DIC. Only for the croplands anthrome (which accounts for 31.2% of the total number of active fires used in the study) did the best model contain the interaction between weekday and religion (Table 3 and Fig 4B).
a) All anthromes; b) Croplands. Solid line: Christian regions; Large dashed line: Muslim regions; Fine dashed line: mixed Christian-Muslim regions.
The best model (lowest DIC, bold) includes all single variables, the two-way interactions between religion and week day, between religion and anthrome, and the spatial term.
To test the hypotheses that the weekly fire counts minima occur on Sunday (Friday) in Christian (Muslim) regions, we analyzed the pairwise contrasts Sunday versus every other week day in Christian regions, and Friday versus every other week day in Muslim regions. We also tested contrasts between fire counts on Sunday in Christian areas versus Friday in Muslim areas; on Sunday in Christian areas versus on Sunday in mixed religion areas; on Friday in Muslim areas versus on Friday in mixed areas (Table 4). The contrast analysis showed that, as hypothesized, there are significantly fewer fires on Sunday than on any other week day in Christian regions and on Friday than on any other week day in Muslim regions.
Previous research based on remotely sensed data revealed different aspects of the anthropogenic nature of African vegetation burning. The very low interannual variability and strong annual periodicity of African fire activity, combined with the presence of a diurnal cycle were considered as evidence of markedly anthropogenic fire, caused by large-scale burning for land management purposes . Anticipation of the fire season relatively to the dry season for periods of over one month throughout most of Africa was shown by , and interpreted as due to preventative break-up of fuel continuity at the landscape scale to avoid large, damaging fires, later in the season . Statistical significance of the weekly cycle of fire activity in croplands, as opposed to the other anthromes, is understandable, considering that cropland fires typically are closely managed, small and low intensity. Less strictly controlled burning in natural areas and rangelands means that fires may be more intense , larger, and last longer than in agricultural areas [7,17]. Therefore, rangeland and forest fires observed by satellite on Friday (Sunday) in Muslim (Christian) regions are more likely to have been set in previous work days than the smaller, carefully controlled burns in fragmented agricultural landscapes, potentially reducing the difference between fire counts on holidays and week days. A study of preventative landscape burning in Mali  found that cropland fires were very well contained, with only a small number of farmers interviewed reporting that fires lit to prepare fields for agriculture had burned into the lands adjacent to their plots. Also in Mali, the occurrence of few, large fires in the northern pastoral areas of the Sahel, is opposed to a pattern of many small fires in agricultural south . The same study refers smaller burns in croplands than in slash-and-burn farming in forest areas of Madagascar, while in southern Mozambique fires set to clear agricultural fields are smaller than those used to manage pastures .
Identification of a previously undetected weekly cycle in cropland burning further highlights the extent to which African fire is anthropogenic, and provides evidence that religion, although an essentially symbolic cultural system, has material impacts on the land surface that can be observed from space. Detailed geographical analysis of weekly cycles of fire activity will further elucidate the global geography and magnitude of anthropogenic vegetation burning, and may inform specification of the ignition component of fire modules incorporated in dynamic global vegetation models. Our detection of weekly cycles in vegetation burning contradicts the assumption of [14, 16] and lends support to the use of operational chemical weather forecasting models that rely on daily fire remote sensing data to estimate biomass burning emissions , which have important climatic effects [38,39] and impacts on public health . Since biomass burning aerosols are strong absorbers of sunlight, they may also alter the thermodynamic structure of the atmosphere and contribute towards observed large scale weekly cycles in meteorological variables .
Regional religious affiliation data for Africa were kindly provided by the Gordon-Conwell Theological Seminary, Boston. Funding for this research was provided by Fundação para a Ciência e a Tecnologia (http://www.fct.pt/) research projects PEst-OE/AGR/UI0239/2014 (JMCP and DO), PEst-OE/MAT/UI0006/2014 (KFT, AT and PP), PTDC/MAT/118335/2010 (KFT, AT, PP and JMCP) and doctoral grant SFRH/BD/4752/2008 (DO).
Conceived and designed the experiments: JMCP AAT KFT. Performed the experiments: AAT PP DO KFT. Analyzed the data: DO PP AAT. Contributed reagents/materials/analysis tools: JMCP DO. Wrote the paper: JMCP AAT KFT.
- 1. Dwyer E, Pereira JMC, DaCamara CC, Grégoire J-M. Characterization of the spatio-temporal patterns of global fire activity using satellite imagery for the period April 1992 to March 1993. J Biogeogr, 1999; 27 (1), 57–69.
- 2. Balzter H, Gerard FF, George CT, Rowland CS, Jupp TE, McCallum I, et al. Impact of the Arctic Oscillation pattern on interannual forest fire variability in Central Siberia. Geophys Res Lett, 2005; 32 (14) L14709,
- 3. Archibald S, Roy DP, van Wilgen BW, Scholes RJ. What limits fire? An examination of drivers of burnt area in Southern Africa. Glob Chang Biol, 2009; 15: 613–630,
- 4. Bond WJ. Fire. In: Cowling RM, Richardson DM, and Pierce SM, editors. Vegetation of Southern Africa. Cambridge UK: Cambridge University Press; 1997, pp. 421–446.
- 5. Wooster MJ, Perry GLW, Zoumas A. Fire, drought and El Niño relationships on Borneo (Southeast Asia) in the pre-MODIS era (1980–2000). Biogeosciences, 2012; 9, 317–340,
- 6. Le Page Y, Oom D, Silva JMN, Jönsson P, Pereira JMC. Seasonality of vegetation fires as modified by human action: observing the deviation from eco-climatic fire regimes. Glob Ecol Biogeogr, 2010; 19 (4), 575–588.
- 7. Kull C, Laris P. Fire ecology and fire politics in Mali and Madagascar. In: Cochrane MA, editor. Tropical Fire Ecology: Climate Change, Land Use and Ecosystem Dynamics. Heidelberg: Springer-Praxis; 2009, pp. 171–226.
- 8. Roberts G, Wooster MJ, Lagoudakis E. Annual and diurnal African biomass burning temporal dynamics. Biogeosciences, 2009; 6, 849–866.
- 9. Schultz MG, Heil A, Hoelzemann JJ, Spessa A, Thonicke K, Goldammer JG, et al. Global wildland fire emissions from 1960 to 2000. Global Biogeochem Cycles, 2008; 22 (2) GB2002,
- 10. Bird MI, Cali JA. A million-year record of fire in sub-Saharan Africa. Nature, 1998; 394: 767–769
- 11. Marlon JR, Bartlein PJ, Carcaillet C, Gavin DG, Harrison SP, Higuera PE, et al. Climate and human influences on global biomass burning over the past two millennia. Nat Geosci, 2008; 1, 697–702
- 12. Lamarque J-F, Bond TC, Eyring V, Granier C, Heil A, Klimont Z, et al. Historical (1850–2000) gridded anthropogenic and biomass burning emissions of reactive gases and aerosols: methodology and application. Atmos Chem Phys, 2010; 10, 7017–7039,
- 13. Le Page Y, Pereira JMC, Trigo R, da Camara C, Oom D, Mota B. Global fire activity patterns (1996–2006) and climatic influence: an analysis using the World Fire Atlas. Atmos Chem Phys, 2008; 8, 1911–1924.
- 14. Quaas JO, Boucher O, Jones A, Weedon GP, Kieser J, Joos H. Exploiting the weekly cycle as observed over Europe to analyse aerosol indirect effects in two climate models. Atmos Chem Phys, 2009; 9, 8493–8501.
- 15. Xia X, Eck TF, Holben BN, Goloub P, Chen H. Analysis of the weekly cycle of aerosol optical depth using AERONET and MODIS data. J Geophys Res Atmos, 2008; 113 D14, 27,
- 16. Beirle S, Platt U, Wenig M, Wagner T. Weekly cycle of NO2 by GOME measurements: a signature of anthropogenic sources. Atmos Chem Phys, 2003; 3: 2225–2232.
- 17. Archibald S, Scholes RJ, Roy DP, Roberts G, Boschetti L. Southern African fire regimes as revealed by remote sensing. Int J Wildland Fire, 19: 861–878.
- 18. Goitein SD (1959) The origin and nature of the Muslim Friday worship. The Muslim World, 2010; 49 (3), 183–195.
- 19. Ringwald CD. A Day Apart: How Jews, Christians, and Muslims Find Faith, Freedom and Joy on the Sabbath. New York: Oxford University Press; 2007.
- 20. Hackett C, Grim B, Stonawski M, Skirbekk V, Potančoková M. The Global Religious Landscape. Pew Research Center’s Forum on Religion & Public Life. Washington D.C., 2012.
- 21. Johnson TM, Grim BJ, eds. World Religion Database. Boston University Institute on Religion, Culture and World Affairs (CURA), Gordon-Conwell Theological Seminary. Brill Publishers, Leiden/Boston, 2012.
- 22. Ellis EC, Goldewijk KK, Siebert S, Lightman D, Ramankutty N. Anthropogenic transformation of the biomes, 1700 to 2000. Glob Ecol Biogeogr, 2010; 19 (5), 589–606.
- 23. Ellis EC, Ramankutty N. Putting people in the map: anthropogenic biomes of the world. Front Ecol Environ, 2008; 6: 439–447
- 24. Friedl MA, Sulla-Menashe D, Tan B, Schneider A, Ramankutty N, Sibley A, et al. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sensing of Environment, 2010; 114 (1): 168–182.
- 25. Justice C, Giglio L, Korontzi S, Owens J, Morisette JT, Roy D, et al. The MODIS fire products. Remote Sens Environ, 2002; 83 (1–2), 244–262.
- 26. Oom D, Pereira JMC. Exploratory spatial data analysis of global MODIS active fire data. Int J Appl Earth Obs, 2013; 21, 326–340.
- 27. Hilbe JM. Negative Binomial Regression, 2nd ed. Cambridge, UK: Cambridge,: University Press, 2011.
- 28. Banerjee S, Carlin BP, Gelfand AE. Hierarchical Modeling and Analysis for Spatial Data, 2nd. ed. Boca Raton, FL: CRC Press, 2014.
- 29. Rue H, Martino S, Chopin N. Approximate Bayesian inference for latent Gaussian models using integrated nested Laplace approximations (with discussion). J R Stat Soc B, 2009; 71(2):319–392,.
- 30. Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde A. Bayesian measures of model complexity and fit. J R Stat Soc B, 2002; 64, 4: 583–639.
- 31. Held L. Simultaneous posterior probability statements from Monte Carlo output. J. Comput Graph Statist, 2004;13, 20–35.
- 32. Box GEP, Tiao GC. Bayesian Inference in Statistical Analysis, Reading, MA: Addison-Wiley; 1973.
- 33. Benjamini Y, Hochberg Y. Controling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol, 1995; 57: 289–300.
- 34. Giglio L, Csiszar I, Justice CO. Global distribution and seasonality of active fires as observed with the Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) sensors. J Geophys Res Biogeosci, 2006; 111 (G2) G02016,
- 35. Laris P. Burning the seasonal mosaic: Preventative burning strategies in the wooded savanna of southern Mali. Hum Ecol, 2002; 30 (2), 155–186.
- 36. Shafer LJ. Indigenous fire use to manage savanna landscapes in southern Mozambique. Fire Ecol, 2010; 6 (2), 43–59.
- 37. Kaiser JW, Heil A, Andreae MO, Benedetti A, Chubarova N, Jones L, et al. Biomass burning emissions estimated with a global fire assimilation system based on observed fire radiative power. Biogeosciences, 2012; 9, 527–554,
- 38. Crutzen PJ, Andreae MO. Biomass burning in the tropics: impact on atmospheric chemistry and biogeochemical cycles. Science, 1990; 250 (4988), 1669–1678 pmid:17734705
- 39. Penner JE, Dickinson RE, O'Neill CA. Effects of aerosol from biomass burning on the global radiation budget. Science, 1992; 256 (5062), 1432–1434 pmid:17791612
- 40. Johnston FH, Henderson SB, Chen Y, Randerson JT, Marlier M, DeFries RS, et al. Estimated global mortality attributable to smoke from landscape fires. Environ Health Perspect, 2012; 120 (5), 695–701. pmid:22456494
- 41. Sanchez-Lorenzo A, Laux P, Franssen H-J, Calbó J, Vogl S, Georgoulias AK, et al. Assessing large-scale weekly cycles in meteorological variables: a review. Atmos Chem Phys, 2012; 12, 5755–5771