More stories

  • in

    Migrant birds and mammals live faster than residents

    1.
    Alerstam, T., Hedenström, A. & Åkesson, S. Long-distance migration: evolution and determinants. Oikos 103, 247–260 (2003).
    Article  Google Scholar 
    2.
    Hahn, S., Bauer, S. & Liechti, F. The natural link between Europe and Africa–2.1 billion birds on migration. Oikos 118, 624–626 (2009).
    Article  Google Scholar 

    3.
    Gill, R. E. Jr et al. Extreme endurance flights by landbirds crossing the Pacific Ocean: ecological corridor rather than barrier? Proc. R. Soc. B Biol. Sci. 276, 447–457 (2008).
    Article  Google Scholar 

    4.
    Kempenaers, B. & Valcu, M. Breeding site sampling across the Arctic by individual males of a polygynous shorebird. Nature 541, 528 (2017).
    ADS  CAS  Article  Google Scholar 

    5.
    Dingle, H. & Drake, V. A. What is migration? Bioscience 57, 113–121 (2007).
    Article  Google Scholar 

    6.
    Faaborg, J. et al. Recent advances in understanding migration systems of New World land birds. Ecol. Monogr. 80, 3–48 (2010).
    Article  Google Scholar 

    7.
    Berthold, P. Bird migration: a general survey. (Oxford University Press on Demand, 2001).

    8.
    Dingle, H. The biology of life on the move. (New York, NY: Oxford University Press, 2014).

    9.
    Rappole, J. H. The Avian Migrant: The Biology of Bird Migration. (Columbia University Press, 2013).

    10.
    Pulido, F. The genetics and evolution of avian migration. BioScience 57, 165–174 (2007).
    Article  Google Scholar 

    11.
    Berthold, P., Gwinner, E. & Sonnenschein, E. Avian Migration. (Springer Science & Business Media, 2013).

    12.
    Bearhop, S. et al. Assortative mating as a mechanism for rapid evolution of a migratory divide. Science 310, 502–504 (2005).
    ADS  CAS  Article  Google Scholar 

    13.
    Sutherland, W. J. Evidence for flexibility and constraint in migration systems. J. Avian Biol. 29, 441–446 (1998).
    Article  Google Scholar 

    14.
    Piersma, T. & van Gils, J. A.. The Flexible Phenotype: A Body-Centred Integration of Ecology, Physiology, and Behaviour. (Oxford University Press, 2011).

    15.
    Healy, K., Ezard, T. H. G., Jones, O. R., Salguero-Gómez, R. & Buckley, Y. M. Animal life history is shaped by the pace of life and the distribution of age-specific mortality and reproduction. Nat. Ecol. Evol. 3, 1217–1224 (2019).
    Article  Google Scholar 

    16.
    Stearns, S. C. The evolution of life histories. (Oxford University Press, London, 1992).

    17.
    Roff, D. Evolution Of Life Histories: Theory and Analysis. (Springer Science & Business Media, 1993).

    18.
    Boyle, W. A. & Conway, C. J. Why migrate? A test of the evolutionary precursor hypothesis. Am. Nat. 169, 344–359 (2007).
    Article  Google Scholar 

    19.
    Winger, B. M., Auteri, G. G., Pegan, T. M. & Weeks, B. C. A long winter for the Red Queen: rethinking the evolution of seasonal migration. Biol. Rev. 94, 737–752 (2019).

    20.
    Levey, D. J. & Stiles, F. G. Evolutionary precursors of long-distance migration: resource availability and movement patterns in neotropical landbirds. Am. Nat. 140, 447–476 (1992).
    Article  Google Scholar 

    21.
    Kokko, H. & Lundberg, P. Dispersal, migration, and offspring retention in saturated habitats. Am. Nat. 157, 188–202 (2001).
    CAS  Article  Google Scholar 

    22.
    Altizer, S., Bartel, R. & Han, B. A. Animal migration and infectious disease risk. Science 331, 296–302 (2011).
    ADS  CAS  Article  Google Scholar 

    23.
    Sillett, T. S. & Holmes, R. T. Variation in survivorship of a migratory songbird throughout its annual cycle. J. Anim. Ecol. 71, 296–308 (2002).
    Article  Google Scholar 

    24.
    Klaassen, R. H. G. et al. When and where does mortality occur in migratory birds? Direct evidence from long-term satellite tracking of raptors. J. Anim. Ecol. 83, 176–184 (2016).

    25.
    Lindström, Å. Finch flock size and risk of hawk predation at a migratory stopover site. Auk Ornithol. Adv. 106, 225–232 (1989).
    Google Scholar 

    26.
    Conklin, J. R., Senner, N. R., Battley, P. F. & Piersma, T. Extreme migration and the individual quality spectrum. J. Avian Biol. 48, 19–36 (2017).
    Article  Google Scholar 

    27.
    Böhning-Gaese, K., Halbe, B., Lemoine, N. & Oberrath, R. Factors influencing the clutch size, number of broods and annual fecundity of North American and European land birds. Evol. Ecol. Res. 2, 823–839 (2000).
    Google Scholar 

    28.
    Jetz, W., Sekercioglu, C. H. & Böhning-Gaese, K. The worldwide variation in avian clutch size across species and space. PLOS Biol. 6, e303 (2008).
    Article  CAS  Google Scholar 

    29.
    Ricklefs, R. E. & Wikelski, M. The physiology/life-history nexus. Trends Ecol. Evol. 17, 462–468 (2002).
    Article  Google Scholar 

    30.
    Peters, P. H. Ecological Implication of Body Size. (Cambridge Studies in Ecology). (Cambridge University Press, cambridge, 1983).

    31.
    Schmidt-Nielsen, K. & Knut, S.-N. Scaling: Why is Animal Size So Important? (Cambridge University Press, 1984).

    32.
    Hedenström, A. Scaling migration speed in animals that run, swim and fly. J. Zool. 259, 155–160 (2003).
    Article  Google Scholar 

    33.
    Hedenström Anders. Adaptations to migration in birds: behavioural strategies, morphology and scaling effects. Philos. Trans. R. Soc. B Biol. Sci. 363, 287–299 (2008).
    Article  Google Scholar 

    34.
    Hein, A. M., Hou, C. & Gillooly, J. F. Energetic and biomechanical constraints on animal migration distance. Ecol. Lett. 15, 104–110 (2012).
    Article  Google Scholar 

    35.
    Teitelbaum, C. S. et al. How far to go? Determinants of migration distance in land mammals. Ecol. Lett. 18, 545–552 (2015).
    Article  Google Scholar 

    36.
    Watanabe, Y. Y. Flight mode affects allometry of migration range in birds. Ecol. Lett. 19, 907–914 (2016).
    Article  Google Scholar 

    37.
    Newton, I. The migration ecology of birds. (Academic Press: Oxford, 2008).

    38.
    Speakman, J. R. & Król, E. Maximal heat dissipation capacity and hyperthermia risk: neglected key factors in the ecology of endotherms. J. Anim. Ecol. 79, 726–746 (2010).
    PubMed  PubMed Central  Google Scholar 

    39.
    Alexander, R. M. C. N. When is migration worthwhile for animals that walk, swim or fly? J. Avian Biol. 29, 387–394 (1998).
    Article  Google Scholar 

    40.
    Klaassen, M. Metabolic constraints on long-distance migration in birds. J. Exp. Biol. 199, 57–64 (1996).
    CAS  PubMed  PubMed Central  Google Scholar 

    41.
    Klaassen, M. & Lindström, Å. Departure fuel loads in time-minimizing migating birds can be explained by the energy costs of being heavy. J. Theor. Biol. 183, 29–34 (1996).
    Article  Google Scholar 

    42.
    Lindström, Å. Fuel deposition rates in migrating birds: causes, constraints and consequences. in Avian Migration (eds Berthold, P., Gwinner, E. & Sonnenschein, E.) 307–320 (Springer, 2003).

    43.
    Newton, I. Weather-related mass-mortality events in migrants. Ibis 149, 453–467 (2007).
    Article  Google Scholar 

    44.
    Gylfe, Å., Bergström, S., Lundstróm, J. & Olsen, B. Reactivation of Borrelia infection in birds. Nature 403, 724 (2000).
    ADS  CAS  Article  Google Scholar 

    45.
    Walter, H. Eleonora’s Falcon: Adaptations to Prey and Habitat in a Social Raptor. (University of Chicago Press, 1979).

    46.
    Somveille, M., Rodrigues, A. S. L. & Manica, A. Why do birds migrate? A macroecological perspective. Glob. Ecol. Biogeogr. 24, 664–674 (2015).
    Article  Google Scholar 

    47.
    Dalby, L., McGill, B. J., Fox, A. D. & Svenning, J.-C. Seasonality drives global-scale diversity patterns in waterfowl (Anseriformes) via temporal niche exploitation. Glob. Ecol. Biogeogr. 23, 550–562 (2014).
    Article  Google Scholar 

    48.
    Able, K. P. & Belthoff, J. R. Rapid ‘evolution’ of migratory behaviour in the introduced house finch of eastern North America. Proc. R. Soc. Lond. B Biol. Sci. 265, 2063–2071 (1998).
    Article  Google Scholar 

    49.
    Pérez-Tris, J. & Tellería, J. L. Migratory and sedentary blackcaps in sympatric non-breeding grounds: implications for the evolution of avian migration. J. Anim. Ecol. 71, 211–224 (2002).
    Article  Google Scholar 

    50.
    Chapman, B. B., Brönmark, C., Nilsson, J.-Å. & Hansson, L.-A. The ecology and evolution of partial migration. Oikos 120, 1764–1775 (2011).
    Article  Google Scholar 

    51.
    Fogarty, M. J., Sissenwine, M. P. & Cohen, E. B. Recruitment variability and the dynamics of exploited marine populations. Trends Ecol. Evol. 6, 241–246 (1991).
    CAS  Article  Google Scholar 

    52.
    Forcada, J., Trathan, P. N. & Murphy, E. J. Life history buffering in Antarctic mammals and birds against changing patterns of climate and environmental variation. Glob. Change Biol. 14, 2473–2488 (2008).
    Google Scholar 

    53.
    Winger, B. M. & Pegan, T. M. The evolution of seasonal migration and the slow-fast continuum of life history in birds. bioRxiv 2020.06.27.175539 (2020), https://doi.org/10.1101/2020.06.27.175539.

    54.
    Martin, T. E. Nest predation and nest sites. BioScience 43, 523–532 (1993).
    Article  Google Scholar 

    55.
    Hurlbert, A. H. & Haskell, J. P. The effect of energy and seasonality on avian species richness and community composition. Am. Nat. 161, 83–97 (2003).
    Article  Google Scholar 

    56.
    Buckley, L. B., Hurlbert, A. H. & Jetz, W. Broad-scale ecological implications of ectothermy and endothermy in changing environments. Glob. Ecol. Biogeogr. 21, 873–885 (2012).
    Article  Google Scholar 

    57.
    Wilcove, D. S. & Wikelski, M. Going, going, gone: is animal migration disappearing. PLoS Biol. 6, e188 (2008).
    Article  CAS  Google Scholar 

    58.
    van Gils, J. A. et al. Body shrinkage due to Arctic warming reduces red knot fitness in tropical wintering range. Science 352, 819–821 (2016).
    ADS  Article  CAS  Google Scholar 

    59.
    Wikelski, M. & Tertitski, G. Living sentinels for climate change effects. Science 352, 775–776 (2016).
    ADS  CAS  Article  Google Scholar 

    60.
    Myhrvold, N. P. et al. An amniote life-history database to perform comparative analyses with birds, mammals, and reptiles. Ecology 96, 3109–3109 (2015).
    Article  Google Scholar 

    61.
    Eyres, A., Böhning-Gaese, K. & Fritz, S. A. Quantification of climatic niches in birds: adding the temporal dimension. J. Avian Biol. 48, 1517–1531 (2017).
    Article  Google Scholar 

    62.
    Gnanadesikan, G. E., Pearse, W. D. & Shaw, A. K. Evolution of mammalian migrations for refuge, breeding, and food. Ecol. Evol. 7, 5891–5900 (2017).
    Article  Google Scholar 

    63.
    Jetz, W., Thomas, G. H., Joy, J. B., Hartmann, K. & Mooers, A. O. The global diversity of birds in space and time. Nature 491, 444–448 (2012).
    ADS  CAS  Article  Google Scholar 

    64.
    Revell, L. J. phytools: an R package for phylogenetic comparative biology (and other things). Methods Ecol. Evol. 217–223 (2014), https://doi.org/10.1111/j.2041-210X.2011.00169.x@10.1111/(ISSN)2041-210X.TOPMETHODS.

    65.
    Fritz, S. A., Bininda-Emonds, O. R. P. & Purvis, A. Geographical variation in predictors of mammalian extinction risk: big is bad, but only in the tropics. Ecol. Lett. 12, 538–549 (2009).
    Article  Google Scholar 

    66.
    Healy, K. et al. Ecology and mode-of-life explain lifespan variation in birds and mammals. Proc. R. Soc. B Biol. Sci. 281, 20140298 (2014).
    Article  Google Scholar 

    67.
    Revell, L. J. Size-correction and principal components for interspecific comparative studies. Evolution 63, 3258–3268 (2009).
    Article  Google Scholar 

    68.
    Pagel, M. Inferring the historical patterns of biological evolution. Nature 401, 877 (1999).
    ADS  CAS  Article  Google Scholar 

    69.
    Hadfield, J. D. MCMC methods for multi-response generalized linear mixed models: the MCMCglmm R package. J. Stat. Softw. 33 (2010).

    70.
    Hadfield, J. D. & Nakagawa, S. General quantitative genetic methods for comparative biology: phylogenies, taxonomies and multi-trait models for continuous and categorical characters. J. Evol. Biol. 23, 494–508 (2010).
    CAS  Article  Google Scholar  More

  • in

    Seasonal weather and climate prediction over area burned in grasslands of northeast China

    Consistent with best practices for homoscedastic stationary series using small data sets, we used the ADF, PP, KPSS, and DF-GLS tests to confirm model validation (Table 1). Maximum temperature and wind speed were stationary I(0) at a 1% level of significance while other variables were integrated to order one I(1) indicating stationarity at first difference. According to the DF-GLS test, the mean minimum monthly temperature was also stationary I(0). The absence of integration to order two I(2) for any of the variables indicated the suitability of the ARDL approach.
    Table 1 Unit root test results of climatic variables for Augmented Dickey–Fuller (ADF), Phillips–Perron (PP), Kwiatkowski, Phillips, Schmidt, Shin (KPSS), and modified Dickey-Fuller generalized least squares (DF-GLS).
    Full size table

    Published studies on fire-climate relationship23,32 generally do not resolve the non-stationarity issue that often arises in time series data that measures climate variables during recent years. Particularly, for explaining the variability of grassland area burned, ignoring the non-stationarity of the series leads to bad specified models that invalidate some of the results and generate high prediction error. In addition, modeling the series as if they were stationary may produce spurious relations that, although perceived as statistically significant, are not real and do not have a correct interpretation33,34. These potential problems generated by non-stationarity provide possible explanations for the large errors we observed in predictions and estimated relationships between area burned and climate. Cointegration is therefore a solution to model the complex dynamics of multivariate ecological time series when some of them are non-stationary. However, even under cointegration, all model parameters, including the lags of independent variables and the error correction term must be correctly specified. Moreover, the error term should not be serially dependent, in order to prevent spurious relationships arising from temporal correlations33,35. Implementation of statistical tests for stationarity of time series, known as unit root tests, is then a fundamental tool for the ecologist to analyze longitudinal data.
    The ADF is among the most used and powerful unit root tests under constant variance and if the lag p is correctly selected. Alternatively, PP unit root tests differ mainly in how they address serial correlation and heteroskedasticity in the errors. ADF tests use a parametric autoregression to approximate the autoregressive moving averages structure of the errors in the test regression, whereas the PP tests ignore any serial correlation in the test regression. Unlike ADF tests, PP tests are robust to general forms of heteroskedasticity and do not require a specified lag length for the test regression23. The ADF and PP tests are asymptotically equivalent but may differ substantially in finite samples due to differences in correcting serial correlation in the test regression. In contrast, DF-GLS is a sensitive test that may increase the power of the previous, especially near stationarity and for small samples. Another recommendation, especially for short time series, is the simultaneous use of the KPSS test.
    The lag selection criteria of log-likelihood36 (LR), the final prediction error37 (FPE), Akaike information criterion38 (AIC), Schwarz Bayesian criterion39 (SBC), and Hannan-Quinn information criterion40 (HQ) were used to select the optimal lag on all variables prior to testing for cointegration (see Supplementary Information Table S1). All these criteria are based on the minus maximized value of the respective likelihood function including a penalization term that increases with the number of estimated parameters to favor a more parsimonious model. Balancing out these two terms creates a metric that can be used as a model selection criterion. Both SBC and HQ are harsher in the selection than AIC, tending to choose a more parsimonious model. Therefore, we preferentially select the lag(1) as the optimal lag for subsequent model estimation in order to keep the model more compact and interpretable. The final prediction error, Schwarz Bayesian criterion, and Hannan-Quinn information criterion were used to confirm lag(1) as the optimal lag for subsequent model estimation.
    The ARDL bounds testing approach is advantageous over traditional cointegration methods because it can examine a mixed order of integrations. We used the F-statistic to test for the null hypothesis of no-level relationship41 (SI Table S2). The bounds test for cointegration that the F-statistic was above both lower [1(0)] and upper [1(1)] bounds critical values, thus, all the models presented reject the null hypothesis of no cointegration. To validate the cointegration test, we further examined the null hypothesis following42 in the light of response surface regressions that report critical p-value approximations43. The p-values strongly rejected the null hypothesis at a 1% significance level, confirming a level relationship in the proposed models.
    We implemented a graphical representation of the ARDL model simulation to provide an easily understandable representation of an instantaneous change (i.e., a shock) in one independent variable on the response in posterior observations (Fig. 3). This contrasts with the more elaborate interpretation of estimated coefficients, given their separation in long and short-run effects. For example, the dynamic plot shows that a positive shock (one standard deviation increase) at time t = 10 in wind speed produces a rapid positive short-run increase in burned grassland area.
    Figure 3

    Plots of the dynamic simulated ARDL model showing the effect of: (a,b) positive and negative one standard deviation change in predicted wind speed on burned grasslands (c,d) positive and negative one standard deviation change in predicted maximum temperature on burned grasslands (e,f) positive and negative one standard deviation change in predicted carbon emission on burned grasslands (g,h) positive and negative one standard deviation change in predicted sunlight on burned grasslands. Dots represent average predicted value while dark blue to light blue lines denote 75, 90 and 95% confidence intervals. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

    Full size image

    At t = 15 this effect vanishes and the response value stabilizes thereafter (Fig. 3a). The confidence intervals (blue lines) indicate that the long-run change in burned grassland area is statistically significant using 75% confidence intervals, rather than significant with 90% confidence intervals. In contrast, the resulting plot of negative shock wind speed at time t = 10 decreased the mean response over several periods, resulting in a statistically significant equilibrium shift over four periods (Fig. 3b).
    To validate the model, we used diagnostic tests to examine the residual independence (Table 2). The coefficient of determination (R-squared) resulted in a goodness of fit of ~ 78.3%. The diagnostic tests showed that functional misspecification error, first-order serial correlation, heteroskedasticity, and non-normality assumptions are adequately controlled in the estimated model. To corroborate the independence of the residuals, we further employed the recursive cumulative sum (CUSUM) plots to examine the stability of the data series over the sampled period (Fig. 4).
    Table 2 Short-run and long-run coefficients of dynamic ARDL.
    Full size table

    Figure 4

    Recursive CUSUM plot of (a) maximum temperature, (b) minimum temperature, (c) sunlight, (d) wind speed, (e) vapor pressure deficit, (f) relative humidity, (g) precipitation, (h) carbon emission with 95% confidence bands around the null. For all variables, residuals fall between the limits, validating the ARDL bounds testing model and showing stability of the variances of coefficients over time.

    Full size image

    Simulation and analysis of ARDL dynamics
    The grassland ecosystems of Xilingol are characterized by highly dynamic and complex interactions among climate, human settlement patterns, ignition, and vegetation cover. Increased grazing constrains grassland burning leading to an “area burned deficit”. Otherwise, the system is highly responsive to high-speed seasonal winds with low relative humidity, quickly drying fuel, and increasing the potential for extensive grassland fires across the region28.
    The long-run and short-run dynamics illustrated in Table 2 show that wind speed represents the dominant variable that extensively increases grassland area burned over both the long and short run. Arithmetically, a 1% increase in mean monthly wind speed increases the burned grassland area by 6.9% (different from zero in a 99% confidence interval (99% CI)) in the long run, and 2.2% (99% CI) in the short run, while holding other variables constant. Besides the direct impact of wind speed on the grassland area burned, it is likely to affect other environmental conditions which also influence fire behavior throughout the Xilingol region. That is, the dynamics of the whole system show complex causal relations that may not be completely explicit on single equation coefficients. For example, our data suggest that wind speed increased following the burning of grasslands (see Supplementary Information Fig. S1). This is consistent with the observation of more extensive burned grasslands over the study period also associated with a dramatic increase in wind speed, an indicator of potential positive feedback not considered in previous studies23. Importantly, high wind speeds were measured in May–August accompanied by the dry land surface conditions and atmospheric aridity, contributing to the influence of burning grasslands feeding back to further promote pyrogenic conditions. Above the threshold, each incremental increase in wind speed on a monthly scale leads to larger burned grasslands across the Xilingol region, overriding other ecological properties, and grassland functions.
    Threshold maximum temperature causes uncertainty in ecological change
    The maximum temperature is also an important predictor of the increasing extent of burned grasslands. The estimated elasticity of maximum temperature in the long-run indicates that a 1% increase in maximum temperature results in a 2.4% increase in burned grassland extent (99% CI) while holding other variables constant (Table 2). The short-run estimated elasticity of maximum temperature is similar to the long-run response, as a 1% increase in maximum temperature leads to a 2.3% (99% CI) increase in burned grasslands. The results corroborate similar studies13,14,15,22.
    The effect of maximum temperature could have more uncertainty because significant relationships between it and burned grasslands occur only when the regional summer maximum temperature exceeded a threshold level, producing a nonlinear relationship22. That is, slight changes in maximum temperature are unlikely to change the ecosystem except when a threshold is crossed, above which effects can be dramatic.
    A more meaningful understanding of the effect of maximum temperature on burned grassland areas can be achieved by the use of the simulations of the ARDL model, given that all the relationships among variables are considered. At time t = 10, a small increase in maximum temperature in the short run is not statistically significant, but becomes statistically significant in the long run. Following a negative shock on the temperature at t = 10, the maximum temperature decreases, stabilizing three periods later. The equilibrium temperature shift over the long-run scenario time is statistically significant (95% CI). Discrepancies between burned grassland area and maximum temperature trends in Xilingol may be rooted in urbanization, vegetation type transitions, and/or natural variability to climate response. Additionally, the frequency of human ignitions, along with other anthropogenic disturbances, strongly supported the abrupt impact of maximum temperature overpowering the influence of other environmental variables (SI Fig. S2).
    Influence of sunlight on burned grasslands and ecological change
    The amount of sunlight exacerbates changes in burned grassland area both directly via the drying effect of radiant heat on fuel and indirectly by influencing other climatic parameters such as temperature and humidity. A 1% increase in sunlight resulted in a 1.98% and 0.24% (99% CI) increase in burned grasslands in the long and short-runs, respectively (Table 2). A negative shock in ln(sunlight) at time (t = 10), insignificantly impacted burned grassland area over the next several periods (Fig. 3). This rapid response is explained by the inclusion of sunlight with a lagged-difference as a dependent variable. Conversely, a positive shock to ln(sunlight) at time (t = 10) increases area burned, which, in turn, shifts the equilibrium and predicts the ecological change just beyond t = 10.
    Impacts of burned grasslands to carbon emissions and ecological change
    The consequence of carbon emissions from burned grasslands was found to be an important relationship in the long run. Carbon emissions, whether from these fires or other sources in Xilingol may influence the extent of burning in this region by contributing to the forcing of atmospheric temperature. More directly, grassland fires resets succession, potentially increasing ecosystem flammability (SI Fig. S3). The estimated elasticity of carbon emission, in the long run, is 0.91, hence, a 1% increase in carbon emissions is related to a 0.86% increase in burned grasslands (95% CI) in the long run and 0.91% (99% CI) in the short run (Table 2).
    The graphical representation of Fig. 3 shows that a change in predicted carbon emissions on burned grasslands and ecological change. The positive shock of carbon emission on burned grasslands and ecological change was associated with a small increase in carbon emission in the short run that is statistically significant at time t = 10. Consistent with other modeled changes, carbon emissions also increase significantly over the predicted value in the long run.
    Ecological properties under threshold-governed ecosystems
    The mechanism underpinning ecological properties can vary widely with significant uncertainty where threshold relationships among different biophysical processes occur22. The model suggests significant changes in the frequency of climatic variations prior to and in response to, the increased areal extent of burned grasslands (SI Figs. S1–S3). Our results highlight the importance of varying patterns in dominant burned grassland climatic factors (e.g., SI Figs. S1–S3) may also hint at broader implications, suggesting that sudden ecosystem state changes, fire-related or otherwise, may naturally exhibit significant variability across space and time22. The results strongly suggest that area burned increases mainly due to increases in wind speeds during the summer season when characteristically warm temperatures and low humidity facilitate fuel drying and fire spread (see SI Fig. S2) and not due to concurrent changes in anthropogenic activities such as fire suppression practices, land management, and human ignitions. This is not to say that anthropogenic activities are negligible in dictating the modern extent of annually burned grasslands. Rather, increased human ignition is associated with increased burned grassland extent, further confounding a modeling effort based solely on physical ecosystem drivers.
    Our long-run analyses suggested that predicted burned grassland extent is insensitive to change in maximum temperature, vapor pressure deficit, precipitation, or relative humidity. These findings contrast with studies that predict increased fire impacts in north-eastern China in response to higher maximum temperature, decreased precipitation, greater vapor pressure deficit, and lower humidity28,44,45,46,47. We posit four potential explanations for this divergence: First, although inter-annual precipitation outside the fire season varies significantly in Xilingol grasslands, little precipitation arrives during the fire season itself. Therefore, inter-annual variation in fire season precipitation is insufficient to impact burned grassland extent. Second, grassland soil field capacity in this region is quite high. Therefore, fire season fuel moisture levels are accordingly unresponsive to precipitation variation. Third, the vast majority of fires in Xilingol originate from human activities23,28. These ignitions, and wind speeds and temperatures coincident with them, rather than fuel conditioning, may explain burned grassland extent. Finally, it is also possible that our statistical approach is insufficiently sensitive to detect a precipitation response under the conditions experienced across the 18-year time series used in this study.
    Model estimation
    Our empirical strategy for estimating the proposed models follows a series of tests required before the application of the novel dynamic simulations of ARDL27. First, the dynamic ARDL simulations require that the series are integrated below order two, I(2), meaning that the series can be integrated of order one, I(1); integrated of order zero, I(0), or a mixture of both I(1) and I(0). To test the level of integration in the series, we utilized the Augmented Dickey–Fuller (ADF)48, Phillips–Perron (PP)49, Kwiatkowski, Phillips, Schmidt, Shin (KPSS)50, and modified Dickey-Fuller generalized least squares (DF-GLS)51 unit root techniques. While all the unit root techniques are tested based on the null hypothesis of unit root, KPSS is tested based on the null hypothesis of stationarity. Testing for unit root is an important component in the model estimation to prevent a spurious relationship between the series. A simplified version of testing for unit root can be expressed in AR (1) as:

    $$y_{t} = a + rho y_{t – 1} + varepsilon_{t} ,quad t = 1 ldots T$$
    (1)

    where yt is the series to be tested for unit root, εt is the error term. εt is independent and identically distributed, independent of the initial detected value y0, and has a zero-mean if the previous observation of (mathrm{y}) is: (Eleft( {varepsilon_{t} |y_{t – 1} ldots y_{0} } right) = 0). The null hypothesis of the unit root can be tested by H0: ρ = 1 against the alternative H1: ρ  More