The semiarid region of Brazil (Figure 1) has a tropical climate that is characterized by high temperatures (mean annual temperature of 27 °C) and annual precipitation amounts less than 800 mm, which is irregularly distributed over long periods of drought (Silva et al., 2017). In almost all regions, the rainy season occurs in the summer, between December and April, except in the west, where rains can start in November (Silva et al., 2017; Queiroz et al., 2017). In the Brazilian semiarid region, the caatinga is the main type of vegetation, where more than 4900 species have been recorded (Filardi et al., 2022; Moro et al., 2014; 2016; Queiroz et al., 2017). Fabaceae, Poaceae, Euphorbiaceae, Bromeliaceae, and Cactaceae dominate in terms of the number of species and landscape formation (Moro et al., 2014; 2016). This vegetation is characterized by a spiny seasonal forest surrounded by an herbaceous stratum that can vary in density according to climate and soil conservation (Loiola et al., 2012).

To test the hypotheses, we selected four annual grass species as biological models: Leptochloa anisopoda (B.L. Rob.) P.M.Peterson, Panicum trichoides Swart, Paspalum fimbriatum Kunth, and Paspalum scutatum Nees ex Trin. (Figure 2). These species were selected because 1) they are annual species; 2) they are widespread in the Brazilian semiarid (Figure 2); 3) their collections have a significant amount of samples; and 4) their collections start in the 19th century and fit the objective of testing variation through time. Leptochloa anisopoda and P fimbriatum are present in other semiarid regions of South America, but records show that their occurrence is predominant in the Brazilian semiarid (Maciel et al., 2009; Maciel and Valls, 2020). P. trichoides is widely distributed in North, Central, and South America (Andrade et al., 2007). P. scutatum is an endemic species from the Brazilian semiarid (Maciel et al., 2009).

We collected the data from 590 samples (L. anisopoda = 142, P. trichoides = 163, P. fimbriatum = 136, and P. scutatum = 149) in the Brazilian herbaria (e.g., IPA, PEUFR, and UFP) and from the digital collections of Specieslink, Reflora, and the Smithsonian National Museum of Natural History. The samples included the timeframe between 1859 and 2022. We recorded the geographic coordinates of each sample when available and for specimens without geographic coordinates, we used the collection site description to assign the geographic coordinate of the municipality centroid. Samples with very imprecise location descriptions were removed from the database for morphological and phenological analyses.

For morphological measurements, we selected specimens with complete representation of the root systems, leaves, and inflorescence. Exsiccatae with signs of grazing, trampling, and leaf herbivory were not measured, even when morphologically complete. We then measured plant heights (measured from the base of the culm to the tip of the inflorescence); leaf lengths (as only the measure of the leaf blade); inflorescence lengths (measured from the first raceme to the tip of the last raceme); and spikelet lengths (from the base of the spikelet at the apex). All measurements were recorded manually for herbarium samples and with ImageJ software (v1.53e; National Institutes of Health, USA; https://imagej.nih.gov/ij/), as described by Schneider et al. (2012), for digital exsiccatae.

We also obtained the date of collection of exsiccatae to calculate the day of the year for each sample (Day 1 = January 1 in the year of collection). Samples with incomplete information on the date of collection were removed from the analyses. Grasses are often collected during flowering and fruiting periods, especially annual grasses that have their lifecycle restricted to the short rainy season in the Brazilian semiarid region, which lasts between 30 and 90 days (Andrade et al., 2007; Ferreira et al., 2009; Maciel et al., 2009; Oliveira et al., 2013). Therefore, we considered only one phenophase for analyses, which is named the reproductive period and represents the flowering and fruiting window of the four species.

We collected data on the monthly mean precipitation; mean, maximum, and minimum monthly temperatures, starting in 1960, from 94 conventional and automatic meteorological stations in the Brazilian semiarid region from the database of the Instituto Nacional de Meteorologia (BDMEP-INMET). We used the collection date and location of the exsiccatae as parameters to select the meteorological stations to extract climate data. We also used the altitudes of each sample to select the meteorological stations to reduce the altitudinal effects of climate variations in the data. Importantly, the limitation of meteorological data in the Brazilian semiarid region (available only from 1960 onward) does not interfere with the climate-specimen correlation analysis, as this analysis covers a significant period of 62 years (1960–2022). Accordingly, the inclusion of exsiccate data from as early as 1859 was essential to evaluate the plant morphological responses over the observed years, using the variable “year” as an explanatory factor in the linear regression model.

In order to verify whether the main effects of climate change are occurring alongside 1960-2020 in the Brazilian semiarid, we applied Mann-Kendall and Tau tests to search for signals of significant temporal trend of mean annual temperature and mean annual precipitation during the 62 years. Tau tests were also utilized to find the correlation between the two climatic variables and geographic variables.

Continuously increasing global temperatures since 1850 are the main evidence of climate change (Masson-Delmotte et al., 2022). Therefore, a temporal analysis of morphological and phenological variation has the potential to reveal a correlation between climate change and the structural and reproductive behavior of plants. We used the entire period of sample collection (1859–2020) for performing linear models of the year of collection to explore the potential morphological (plant heights, leaf lengths, leaf widths, inflorescence lengths, and spikelet lengths) and phenological (day of year) shifts in Brazilian semiarid grasses through time. For each model, we calculated the effect size of the year of collection and estimated the confidence interval of the regression coefficient after fitting the model with 5,000 bootstraps.

In order to evaluate whether spatial or climate variables explain the morphological and phenological variation better than the year of collection, we built multiple regression models for each morphological variable and for the day of the year. For these multiple regression analyses, we used the 1960-2020 time frame because of the availability of climate data. However, we tested several spatial and temporal assumptions of the dataset before running the multiple regression models.

To identify potential collinearity among the predictor variables, we performed a Pearson correlation analysis and used a correlation index >0.5 as threshold to select the most independent variables. Next, we applied Moran’s I spatial autocorrelation with 10-class distance to inspect the spatial distribution of each predictor variable. We also evaluated the interannual correlation of longitude, latitude, monthly precipitation, and temperature (monthly mean, minimum and maximum) applying partial autocorrelation analysis with a maximum number of lags limited to ten. The stationarity of the time series for monthly mean precipitation, mean, maximum, and minimum monthly temperatures was verified with the Augmented Dickey-Fuller test. All these preparatory tests were performed independently for each species.

We included, latitude, monthly precipitation, monthly temperature and collection year as predictor variables to build the full models of morphological analyses and longitude, latitude, mean annual precipitation, and temperature for phenological analyses. We performed a stepwise backward-forward model selection to remove non-informative variables from the final model. Next, we assessed the possibility of multicollinearity between the variables of the final model by variance-inflation factors. Then, we calculated the effect size of each variable in the final model. To estimate the confidence interval of the regression coefficient, we fitted the selected model with 5,000 bootstraps. In order to assess the impact of spatial autocorrelation on the model, we calculated Moran’s I of the model residuals with 100 distance classes.

Expansion of agricultural lands replaced a significant part of the natural cover of Brazilian semiarid since 1985. This change in land use may mask the historical effect of climate change on the morphology and phenology of species. Therefore, we divided the morphological and phenological data of the species into two groups to represent the period when the caatinga still preserved more than 70% of its original vegetation (pre-1985) versus the period when the caatinga was reduced to less than 59% (post-1985). To assess statistically significant changes in the morphological data, we performed Kruskal–Wallis tests for each of the morphological variables. For changes in phenology, we assessed the synchronicity of phenology between the two periods. Since our data only include the day of the year on which each fertile sample was collected, we used this date as the reproductive peak for three different period length scenarios: 1) 30-day scenario; 2) 60-day scenario; 3) 90-day scenario. We then calculated the Augspurger phenological synchronicity index for each scenario. Significant differences in synchronicity between pre and post-1985 were assessed with Kruskal–Wallis tests.

All analyses were performed in the Statistical Environment R using the packages letsR (Vilela and Villalobos, 2015), aTSA (Qiu, 2024), boot (Canty and Ripley, 2025), MASS (Venables and Ripley, 2002), car (Fox and Weisberg, 2019), effectsize (Ben-Shachar et al., 2020) and AugspurgerIndex (Benkendorf, 2023). Datasets and R-codes used to produce our results are available in the Supplementary Material.

Temperatures increased significantly in the semiarid region of Brazil between 1960 and 2021 (Figure 3). The mean temperature increased by 1.89 °C (tau = 0.57, z = 6.56, P < 0.001), the minimum temperature was 1.54 °C higher (tau = 0.60, z = 6.86, P < 0.001), and the maximum temperature increased by 2.05 °C (tau = 0.56, z = 6.36, P < 0.001). Mean annual precipitation presented a trend to reduction, but without significance (tau = −0.11, z = −1.22, P = 0.2244042).

Moran’s I never crossed the threshold of 0.4 revealing low levels of spatial autocorrelation in all variables of all species (Supplementary Figures S1, S2). We also found that precipitation and temperature data were stationary in all species. Partial autocorrelation function reported no significant values beyond the lag one for precipitation and temperature in all species (Supplementary Figures S3, S4). To reduce collinearity among the independent variables, we removed maximum and minimum temperature from the multiple linear regressions analyses of morphological variation for all species (Supplementary Tables S1, S4). We retain those variables considered effectively independent: year, latitude, longitude, mean monthly precipitation, and mean monthly temperature. We compiled all preliminary results in the Supplementary Material.

Our results indicate negative responses of the morphological variables throughout 1859 and 2020 (Figure 4). P. scutatum presented reductions in plant heights (9.2 cm, CI = 17.4–1.0 cm, P < 0.05, R ² = 0.03), leaf lengths (2.77 cm, CI: 4.7–0.9 cm, P < 0.005, R ² = 0.05) and spikelet lengths (0.015 cm, 0.030–0.001 cm, P < 0.05, R ² = 0.02). We recorded reductions in the heights (33.5 cm, CI: 52.0–15.1 cm, P < 0.0005, R² = 0.09), leaf lengths (4.2 cm, CI: 8.2–0.07 cm, P < 0.05, R² = 0.02) and inflorescence lengths (3.1 cm, CI: 5.95–0.26, P < 0.05, R² = 0.03) of P. fimbriatum. In P. trichoides, we also observed reductions in plant sizes (15.3 cm, CI: 26.8–3.8 cm, P < 0.001, R² = 0.04) and inflorescence sizes (4.1 cm, CI: 6.7–1.5 cm, P < 0.005, R² = 0.05). L. anisopoda was the only species for which no responses were observed.

Seven models explaining the morphological variation of all species were identified. The variable year consistently emerged in six models and was the most significant factor in four models, highlighting that temporal changes play a central role in the morphological variations (Table 1). We found a correlation among geographic and climatic drivers and variations in the morphological traits of the analyzed species. L atitude correlated positively with inflorescence length in P. trichoides and P. fimbriatum and with spikelet length in L. anisopoda, while longitude associated with increases in leaf length in P. scutatum (Table 1). Among the climatic variables, temperature had a positive effect on leaf length in P. fimbriatum and P. scutatum and on spikelet length in L. anisopoda (Table 1). Mean precipitation positively influenced leaf length in P. fimbriatum, but negatively affected leaf length in P. scutatum and inflorescence length in P. trichoides (Table 1). Plant height was the only trait explained solely by year in the multiple regressions, indicating changes over time (Table 1).

We found no significant changes for all morphological traits of evaluated species between pre-1985 and post-1985 periods (Supplementary Table S2). Thus, our data are not able to track any morphological changes related with land-use changes in Brazilian semiarid.

None of the species analyzed showed a significant change in the day of the year of flowering and fruiting throughout 1859-2020 (Supplementary Table S3). When applying multiple regression to the period of 1960–2020, we observed that reproductive time was influenced differently in each species, revealing more complex patterns of phenological responses to environmental changes (Table 2).

Unlike simple linear regression, in the multiple regressions the variable year exhibited a significant trend of accelerated reproductive period in L. anisopoda (1.30 days per year), P. fimbriatum (1.24 days per year), and P. scutatum (1.78 days per year). All multiple regressions retrieved the mean annual temperature as an informative variable, showing positive and significant relationship with phenological changes and indicating that higher temperatures tend to delay the reproductive time of grasses (Table 2). Latitude exhibited a negative relationship with reproductive time, suggesting that specimens located further north tend to flower and fructify earlier, as observed in L. anisopoda and P. trichoides (Table 2). Longitude showed a positive relationship with DOY in L. anisopoda and P. scutatum (specimens further east tend to flower later) and a negative relationship in P. trichoides (Table 2). Finally, mean annual precipitation did not show a significant relationship with the reproductive time of the analyzed species (Table 2).

Despite the four species showing a trend to reducing phenological synchronicity between pre-1985 and post-1985 (Figure 5), significant reductions in all reproductive scenarios were found for L. anisopoda, P. fimbriatum and P. scutatum. In L. anisopoda 30-day scenarios presented higher reduction in Augspurger Index (0.67–0.26, χ² = 72.889, df = 46, P = 0.007), followed by 60-day (0.77–0.41, χ² = 82.747, df = 58, P = 0.018) and 90-day (0.85–0.51, χ² = 92.596, df = 65, P = 0.014) scenarios. For P. scutatum, the same pattern was found (30-day: 0.51–0.27, χ² = 91.602, df = 52, P = 0.006; 60-day: 0.62–0.43, χ² = 92.993, df = 63, P = 0.008; 90-day: 0.69–0.56, χ² = 106.44, df = 71, P = 0.004). For P. fimbriatum 60-day scenario had a higher change (0.77–0.41, χ² = 82.747, df = 58, P = 0.018) than 90-day (0.85–0.51, χ² = 92.596, df = 65, P = 0.014) and 30-day (0.53–0.37, χ² = 89.853, df = 45, P < 0.001). Panicum trichoides show reduction in phenological synchronicity only for 90-day scenarios (0.77–0.43, χ² = 103.55, df = 78, P = 0.028).

In this study, we demonstrated that grass species of the Brazilian semiarid region have been undergoing reductions in the sizes of their vegetative and reproductive organs since the 19th century and, especially in the last 62 years, when the annual mean temperature has increased by 1.89 °C in the region. Thus, we detected morphological shifts in semiarid grasses at the same time climate change is impacting the ecosystems. Only L. anisopoda exhibited no significant morphological shifts over time in our study. However, we also detected regional variations in the morphological attributes of the species from the Brazilian semiarid region, in which local variations in precipitation and temperature explained the shifts in leaf, inflorescence and spikelet sizes. These regional variations confirm the ability of grasses to invest in strategies to improve inflorescence sizes and spikelet sizes in places with low precipitation and high temperatures.

In addition to the variable year, we found that temperature was the second climatic variable that better explained the morphological shifts in the analyzed grass species. Reducing size of plant organs can be connected to physiological responses to temperature increases caused by climate change (Gardner et al., 2011). Consequently, we rejected the role of precipitation as expected in our second hypothesis. This finding highlights the importance of temperature reported in other studies (Guerin et al., 2012; Leger, 2013). There is evidence that plants and animals are decreasing in size as a consequence of climate change over time, and this negative relationship can be explained by increased temperatures (Barber et al., 2000; Sheridan and Bickford, 2011). In an experimental study with the grass, Austrodanthonia caespitosa (Gaudich.) H.P. Linder, a reduction in seedlings in response to an increase in CO² and temperature was observed (Hovenden et al., 2008). However, the possible influences of temperature increases caused by climate change on the morphological shifts of plant organs in grasses had never been explored before our study.

In our study, time, as a proxy for climate change, explained the reductions in the plant heights of three annual species (e.g., P. scutatum, P. fimbriatum and P. trichoides); the leaf lengths of two other species (e.g., P. scutatum and P. fimbriatum); the inflorescence sizes of two other species (e.g., P. fimbriatum and P. trichoides); and the spikelet size of one species (e.g., P. scutatum). Leger (2013) observed divergent responses over time in the morphological changes of annual plants; While some species decreased in size (Collinsia parviflora Lindl., Cryptantha pterocarya (Torr.) Greene, Eriastrum signatum D. Gowen, Gilia inconspicua (Sm.) Sweet and Microsteris gracilis (Hook.). Greene), other (Mentzelia albicaulis (Hook.) Torr. and A. Gray) increased in height (Leger, 2013).

Since 1985, a drastic replacement of natural cover by agricultural lands in Brazilian semiarid has been documented (Beuchle et al., 2015; Silva et al., 2017; Silva and Barbosa, 2017; Franca Rocha et al., 2024). Our assessment of the effect of agricultural land expansion over the morphology of species has not found any significant differences between the pre-1985 and post-1985 periods. This contradicts the general assumption that size reductions can also be an outcome of other effects of climate change, such as soil degradation and grazing (Guerin et al., 2012; Leger, 2013; Tavares et al., 2019). For example, Fahnestock and Detling (2009) analyzed the effects of long-term grazing (>100 years) and reported that grasses presented shorter vegetative shoot heights and smaller leaf areas in grazed areas than in ungrazed areas. Therefore, climate change provides a more plausible explanation for the observed morphological trends in semiarid Brazil, given its extensive impact throughout the entire region.

Several authors have discussed the causality of these morphological shifts over time as plasticity responses to environmental changes or genetic transformations (Schlichting, 1986; Leger, 2013). Short-term studies have shown that, compared with taller plants, small plants can adapt better to drought conditions due to their low transpiration levels (Hendrix and Trapp, 1992; Casper, 1996; Angadi and Entz, 2002). Authors agree with the persistent difficulties in determining the main mechanism of these morphological alterations because of the lack of knowledge of specific genes that affect the size of organisms combined with the idea that plasticity cannot be sustained for a long period (Tilman et al., 1998; Gardner et al., 2011). However, new concepts related to the contribution of phenotypic plasticity to macroevolution are increasingly common in the literature, such as the role of genetic accommodation in determining the plasticity of the development of a population resulting in greater frequencies of genes that encode adaptive features for the next generations (West-Eberhard, 2003; West-Eberhard, 2005; Simon, 2018).

Our results provide new evidence of geographical patterns of morphological variation in the semiarid region of Brazil on the basis of significant relationships between geographical and climatic variables and morphological shifts in the analyzed species. This pattern was also reported by Guerin et al. (2012). Environmental stress factors that occur in isolated or short-term events promote plastic responses that differ from those resulting from continuous long-term changes (Ehrenreich and Pfennig, 2016). These changes may be related to the concept of phenotypic plasticity, which can be immediate, temporary and reversible (Simon, 2018). This strategy is also associated with the C4 photosynthetic pathway of the selected plants, which allows them to adapt to dry and hot habitats (Braga et al., 2021). This evidence provides clues concerning evolution and diversification drivers in the semiarid region of Brazil once they likely act as environmental filters of plasticity or adaptive responses.

Our results demonstrate that the phenological variations of grasses in the Brazilian semiarid region are more associated with spatial and thermal gradients than with temporal changes. No evidence of advancing or delaying reproductive period of grasses in the Brazilian semiarid region throughout 163 years was found. Only in combination with other variables, year of collection yielded significant effect for some species, but in the opposite direction to what would be expected if climate change were the main driver of phenological shifts. In our results, acceleration of reproductive period is associated with time whereas the realized increase in temperature caused by climate change is often associated with delayed phenology in other plant groups (Nicotra et al., 2010; Munson and Long, 2017).

On the other hand, we discovered that temperature explained better the variations in the phenology of Brazilian semiarid grasses, following a similar pattern reported in several other studies in which the flowering time changed with increasing temperature (Cayan et al., 2001; Cleland et al., 2007; Zhang et al., 2007; Park and Schwartz, 2015; Anderson et al., 2012; Munson and Long, 2017; Geissler et al., 2023). In addition, we found a shift in phenological synchronicity post-1985 for most of the species analyzed, when changes in land-use of Brazilian semiarid rose significantly. This dissociation between temporal and other underlying cues suggests that the temporal component does not reflect global warming, but rather local adjustments and species-specific ecological responses to geographical changes in temperature and shifts in land-use.

We identified spatial variations in phenology for some species, which reflects the spatial‒temporal complexities of the climate in the semiarid region of Brazil. These phenological responses to geographic variables may be related to the interactions of several abiotic factors that influence reproductive times, such as rainfall seasonality, temperature, humidity and soil type (Cleland et al., 2007; Parmesan and Yohe, 2003; Primack et al., 2009; Primack et al., 2004; Silva et al., 2021; Plos et al., 2024; Karthikeyan et al., 2025). The reproductive period of L. anisopoda, P. trichoides and P. scutatum occurs late eastward, accompanied by lower soil humidity, despite the milder average temperature observed most of the year (Silva et al., 2017). In contrast, the reproductive period of L. anisopoda and P. trichoides accelerated in a northern direction, which can be explained by the spatial mosaic of climatic seasonality in the Brazilian semiarid region. In the northern region (north), the concentration of rain occurs in the summer and is influenced by the Intertropical Convergence Zone (ZCIT), with a maximum temperature in November and a minimum in June, whereas in the eastern region (east), the rain concentration occurs in autumn and is influenced by the Atlantic Tropical Mass (MTA) and ZCIT, with a maximum temperature in March and a minimum in July (Silva et al., 2017).

Spatial patterns of phenological variation of the four species from Brazilian semiarid reinforce the capacity of grass to yield rapid responses for local variation of climate. In terms of the phenology‒climate relationship, the flowering period of a population is expected to be in balance with the climate in the long term, as reproductive production can be considerably lower if flowering peaks occur during unfavorable periods, such as those with extreme heat and water scarcity (Nicotra et al., 2010). Annual grasses have the ability to follow rapid rates of climate change throughout a territory and respond to the interactions of other factors that influence the phenology‒climate relationship and may present a delay in flowering in a given place and an acceleration in another, even if both localities have the same temperature (Crimmins et al., 2011; Munson and Long, 2017). We confirmed this pattern in our study of Brazilian semiarid grasses.

Functional traits explain phenological variation and changes of herbaceous plants communities in several cases (König et al., 2018; Plos et al., 2024). Species with different life cycles and photosynthetic pathways present divergent phenological responses to changes in climate, but our results reveal a slightly different pattern of that found by Munson and Long (2017). We found that increases in temperature are related to delay in the flowering/fruiting in our study for all four species, among which L. anisopoda, P. fimbriatum and P. scutatum have C4 photosynthetic pathway and P. trichoides is a C3 species (Osborne et al., 2014). Munson and Long (2017) discovered that C4 grass species delayed while C3 species advanced the flowering time in parallel to the increases in temperature.

Synchronicity of phenology of three species from Brazilian semiarid reduced at the same time agricultural lands and anthropization expanded 39% throughout last 38 years (Beuchle et al., 2015; Silva et al., 2017; Silva and Barbosa, 2017; Franca Rocha et al., 2024; MapBiomas, 2025). Changes in phenological synchronicity of L. anisopoda, P. fimbriatum and P. scutatum post-1985 endorse the possibility that spatial patterns of temperature and land use influenced more the phenology of grasses than climate change in the Brazilian semiarid. Asynchrony or alterations in phenology is often attributed to increasing annual temperatures caused by climate change (Cleland et al., 2007; Parmesan and Yohe, 2003; Primack et al., 2009; Primack et al., 2004; Karthikeyan et al., 2025). However, in some situations other spatial and temporal clues—i.e., land use shifts—offer higher explanatory potential rather than climate change (Bertin, 2008; Parker, 2021; Plos et al., 2024).