Introduction

The persistence of large portions of the world’s biodiversity in general and of primates in particular is threatened directly by anthropogenic pressures and indirectly by climate change effects (Barlow et al., 2018; Estrada et al., 2017). While direct effects, such as habitat conversion or hunting are obvious, climate change effects can take several forms that are not understood in their complexity and interactions. Apart from gradual changes, such as increasing mean temperature and increasing or decreasing rainfall, extreme events or an increase in the variability of environmental conditions can have more profound effects than gradual changes (Maxwell et al., 2019; Vasseur et al., 2014; Zhang et al., 2019).

Madagascar might provide an excellent place to study effects of environmental variation on adaptations and responses of various biota, including primates. The evolution of lemurs on Madagascar is hypothesized to have been shaped by poorly predictable ambient conditions and extreme weather events (Dewar & Richard, 2007; Wright, 1999). Within Madagascar, the dry southwest stands out with respect to floristic endemism, unpredictable weather conditions and desiccation over the past few hundred years (Aronson et al., 2018; Dewar & Wallis, 1999; Fenn, 2003; Waeber et al., 2015). Subfossil deposits, originating a few hundred or thousand years ago, document a rich fauna in the region with many large birds and mammals that went extinct. Many of the mammals were lemurs, the nonhuman primates of Madagascar (Goodman & Jungers, 2014; Rosenberger et al., 2015). The negative effects of climatic changes on lemurs are ongoing and are illustrated by changing population characteristics in a variety of lemur species in relation to forest changes with increasing aridity and increasing frequencies of extreme weather events (Abel et al., 2023; Birkinshaw & Randrianjanahary, 2007; Blanco et al., 2015; Brown & Yoder, 2015; Dinsmore et al., 2021; Dunham et al., 2011; Ganzhorn, 1995; Gould et al., 1999; Hending et al., 2022; Jolly et al., 2002; Kasola et al., 2020; Lawler et al., 2009; Morelli et al., 2020; Ozgul et al., 2023; Rakotondrafara et al., 2018; Richard et al., 2000; Stalenberg et al., 2018; Zhang et al., 2019). While gradual changes and extreme events in changing frequencies are important, a neglected, yet possibly crucial component of inter-annual variation is represented by the difficulty of predicting the month(s) with lowest food supply (Dewar & Wallis, 1999; Dewar & Richard, 2007). Several species of mammals have the option of embryonic diapause to adapt reproduction to favorable environmental conditions (Renfree & Shaw, 2000). Lemurs are not known to have this option and, thus, interannual variation in the months of the lean season would compromise the sequence of mating, pregnancy, and lactation. This climate scenario does not apply to all of Madagascar (Crowley et al., 2017), but its probability of occurrence increases with increasing aridity (Dewar & Wallis, 1999; Richard, 2022). Thus, the spiny forest region of southwestern Madagascar might offer opportunities to investigate coping strategies of lemurs at their ecological limits and subject to high variability in the production of food.

Today, several mammal species have a patchy distribution within the seemingly continuous dry and spiny forest in and around Tsimanampetsotse National Park on a regional scale as well as at specific sites (Bohr et al., 2011; Goodman et al., 2008; LaFleur et al., 2016; Marquard et al., 2011; Ralison, 2008; Steffens et al., 2017). This suggests that forests or forest parts where the species are absent do not provide the necessary resources for the persistence of species.

Lepilemur petteri is member of a species-rich genus with several species that are Critically Endangered with very few individuals left in forest remnants persisting by chance (Dinsmore et al., 2016; Rasoamazava et al., 2022; Schwitzer et al., 2014; Tinsman et al., 2022; Wilmet et al., 2019).

The genus Lepilemur has undergone substantial revision over the past few years. According to the present classification, most studies published as studies on L. leucopus actually concern L. petteri and not L. leucopus (Lei et al., 2017). Until now, the only studies of L. leucopus are inventories between Andohahela and east of the Mandrare River (Feistner & Schmid, 1999; Jaonasy & Birkinshaw, 2021; Rakotondranary et al., 2021). Lepilemur petteri is still widespread, but it is one of the lemur species with a patchy distribution within the spiny forest. It is mainly folivorous and occurs in densities of more than 1 individual per hectare at some sites but is absent a few hundred meters away (Hajanantenaina, 2018; Ralison, 2008; S. J. Rakotondranary and Y. R. Ratovonamana, personal observation). Their occurrence and absence seem to follow environmental gradients, but this pattern is not consistent. Since extremes and the variability of ambient conditions and extreme events affect species more than gradual changes (Maxwell et al., 2019; Vasseur et al., 2014; Zhang et al., 2019), the patchy distribution also might be due to spatial and temporal variation in ambient conditions on a small scale. This pattern of distribution makes for an ideal natural experiment for measuring the effects of fine-scale habitat variables on L. petteri’s presence in the landscape. When species have a continuous distribution, unlike L. petteri, studies risk confounding presence-absences, driven by large-scale determinants of species’ occurrence, such as climate or elevation, with changes driven by fine-scale habitat variables. Studies of species with a larger extent of occurrence with recurrent presences and absences, such as L. petteri, might allow insights into limiting factors and subsequent conservation strategies that could be extrapolated to other species where similar studies are no longer possible. The results might not only be useful for conservation activities but also might contribute to the reconstruction of evolutionary scenarios.

We evaluated associations between L. petteri and to quantitative and qualitative vegetation and food characteristics at the regional and the local scale. On the regional scale, we compared the structure and floristics of forests with or without L. petteri. We complemented the measurements on the ground by and compared them with a remote sensing component. For this, we used the Enhanced Vegetation Index (EVI) from the MODIS Vegetation Index Products as a proxy for ecosystem productivity and its variability over time (Garroutte et al., 2016). At the local scale, we then determined the home range of individuals and compared the structure of vegetation plots located inside or outside these home ranges. If vegetation characteristics that are essential for L. petteri were the same at both scales, we might be able to generalize the suitability of forests for L. petteri (and possibly other lemur species).

Our objectives of the study were (i) to compare the utility of different methods to assess forest quality for L. petteri, and (ii) to identify vegetation characteristics that allow the lemur species to persist in an environment at the limit of its ecological niche. Our specific questions were:

  1. 1.

    Do forests with or without L. petteri differ in structural or floristic vegetation components on a regional and on a local scale?

  2. 2.

    Do means and their annual variation of spatially integrated productivity-related vegetation characteristics as described by remote sensing on a 30 x 30 m scale (EVI) differ between forests with or without L. petteri?

  3. 3.

    Can the EVI be linked to the vegetation characteristics measured on the ground in the regional approach?

  4. 4.

    Can the results be interpreted as strategies of L. petteri to cope with unpredictable environmental variation and if so, what can be learned for conservation, habitat restoration or reforestation projects under increasing environmental unpredictability as part of the ongoing climatic changes.

Methods

Study sites

We conducted the study in southwestern Madagascar in the western part of Tsimanampetsotse National Park and the adjacent plain towards the sea (−24.64°S, 43.95°E; Fig. 1). The park is located ca. 85 km south of the city of Toliara. The climate exhibits a dry season between April and October and a rainy season between November and March, with irregular rainfall of ca. 300–350 mm/year and an annual mean temperature of 24 °C (Goodman et al., 2018; Ratovonamana et al., 2011). Rainfall is supplemented by dew, but its effect on soil moisture and water intake by plants and animals is poorly understood (Hanisch et al., 2015).

Fig. 1
figure 1

Locations for vegetation descriptions (June 2007 – November 2019 and tracking of L. petteri (one site at Andranovao camp and three sites at Lavavolo; tracking between September 2018 and November 2019) in southwestern Madagascar. Black stars represent areas that can consist of up to four distinct “sites” as detailed in Table I.

Full size image
Table I Characteristics of sites in and around Tsimanampetsotse National Park, southwestern Madagascar with geographical coordinates for 30 x 30 m2 plots and data availability for the different aspects of the study; + indicates that L. petteri was present or that we completed radio-tracking or vegetation descriptions at the site.
Full size table

The presence or absence of different lemur species along the western border of Tsimanampetsotse National Park had been determined with systematic transect walks at three sites by Ralison (2008). We supplemented his surveys by systematic lemur inventories at one of his sites plus another 10 sites between July and September 2017. Inventories consisted of walking transects of 2 km length with approximately 1 km/h with two people equipped with headlamps between 19.30 hr and 22.30 hr. We walked each transect in two nights. We spotted nocturnal lemurs through their eyeshine. We confirmed the absence of L. petteri by information from local people living close to the sites. We did not formalize the exchange of information with local residents. For the regional surveys, we were living with the people, informed them about our intentions and integrated them in the surveys. Where we did not find L. petteri, and the local people had not seen the species in their forests either, we recorded L. petteri as being absent at that site. The sites covered the full length of Tsimanampetsotse National Park from north of “Andranovao camp” to south of “Tongaenoro” (Fig. 1; Hajanantenaina, 2018; S. J. Rakotondranary, personal observation; Y. R. Ratovonamana, personal observation). We verified and confirmed the results at 10 of these sites again by walking 2-km transects at night between March and May 2019 (Apel, 2020).

We radio-tracked L. petteri from September 2018 to November 2019 (Nevermann et al., 2023), followed by occasional checking of the presence and reproduction of L. petteri close to Andranovao camp until March 2022. “Andranovao camp” represents the region where we have installed our research and training camp with permanent staff and a variety of projects since 2006 (Rakotondranary et al., 2010; Ralambomanantsoa et al., 2023; Ratovonamana et al., 2011). We did not catch or track females with babies and did not catch animals in February (late lactation when it was difficult to notice before capture whether a female had an offspring or not). We also do not have tracking data for the months of June and August (2 months of the dry season from April to October), because we did not have the capacity for captures and radio-tracking during these 2 months.

The vegetation of the area is part of the spiny forest ecosystem (Moat & Smith, 2007). Net primary production shows very high spatial and temporal variation between years (Fig. 2a; see below for a description of remote sensing data). In particular, the months with highest or lowest productivity shift erratically between years (Fig. 2b, months with lowest productivity not shown; Dewar & Wallis, 1999). Net primary production can vary by an order of magnitude between years, ranging from less than 150 g to more than 1000 g carbon/m2/year, resulting in substantial deviations from the long-term mean (Fig. 3; Ralambomanantsoa et al., 2023).

Fig. 2
figure 2

(a) Variability of annual primary production from 2009 to 2020 in the region in and around Tsimanampetsotse National Park, southwestern Madagascar between Beheloke (north) and Tongaenoro (south) as shown in Fig. 1. (b) Month with highest productivity in different years. Productivity is illustrated by EVI integral (EVIi), a surrogate of the annual ecosystem primary productivity. The eastern part of the region is dominated by grassland and agriculture. Readers of the printed journal are referred to the online article for a color version of this figure.

Full size image
Fig. 3
figure 3

(a) Mean of annual EVI integrals (proxy of primary production) from 2009 to 2020 in and around Tsimanampetsotse National Park, southwestern Madagascar between Beheloke (north) and Tongaenoro (south) as shown in Fig. 1. (b) Deviation of annual EVIi from the 12-year mean. Readers of the printed journal are referred to the online article for a color version of this figure.

Full size image

Quantifying interannual fluctuations of the month with highest or lowest primary productivity

We used circular statistics to analyze the strength of seasonality (as a proxy for predictability) in primary production (Batschelet, 1981). This approach is easier to visualize and to quantify than the more elaborate approach of Colwell (1974). For this, we assigned the months with the highest or lowest primary production for the 12-year period from 2009 to 2020 in 30° intervals to a circle with 360°. January is equivalent to 30°, February to 60°, etc. until December that is equivalent to 360° = 0° (Fig. 4). The month with the highest (or lowest) primary production is characterized by a vector with the associated angle and a vector length of 1. The vectors of the 12 months with the highest (or lowest) primary production for each year of the 12-year period are added, resulting in a “resulting vector (R)” with a given length and a certain angle (α) that can be retransformed to the corresponding month. Dividing the resulting vector by the number of months (12) results in a mean vector (rvector) with the same angle as the resulting vector and a mean length between rvector = 1 (if the highest production falls always into the same month) and rvector = 0 (if the months with the highest production would be distributed uniformly around the circle). Conventionally, the length of the mean vector is labelled by “r” (Batschelet, 1981). To avoid confusion with Pearson and Spearman correlation coefficients, we labelled the length of the mean vector as “rvector” and kept “r” for Pearson and rs for Spearman correlation coefficient.

Fig. 4
figure 4

Visualization and quantitative analyses of the predictability of seasonality between 2009 and 2020 using the site Vohombositse as an example. Each symbol at the periphery of the circle represents (a) the month with maximum EVI (primary production), or (b) the month with the lowest EVI per year for the 12-year period from 2009 to 2020. Months are assigned to the annual circle in 30° intervals (a). The circle has a radius of 1. The length of the mean vector (rvector) is proportional to the predictability that the month with the highest or lowest EVI occurs at the time of year indicated by the direction of the mean vector; (a) months with highest EVI: rvector = 0.763, α = 102°, p < 0.001, N = 12 (reflecting highest productivity at the end of March with high predictability as the p-value indicates highly significant directionality according to the Rayleigh test); (b) months with lowest EVI: rvector = 0.140, α = 287°, p = 0.797, N = 12 (reflecting lowest productivity at the beginning of October, but with very low predictability as the length of the mean vector does not indicate any significant directionality).

Full size image

The length of the vector can be assigned levels of significance, indicating whether the distribution deviates from random or indicates a predominating (significant) direction (indicated by the p-value according to the Rayleigh test; Batschelet, 1981). A long mean vector indicates that the months with the highest productivity are the same or deviate little in different years, indicating high predictability of high primary production (Fig. 4a). A short mean vector reflects that the time of highest productivity falls into different months in different years, indicating low predictability of plant productivity between years. We applied the same approach using the months with the lowest plant productivity (Fig. 4b).

Enhanced vegetation index

We quantified the annual variability in primary production, using the EVI time series from 2009 to 2020. The EVI includes additional wavelengths (blue band) to make EVI more robust than the Normalized Difference Vegetation Index (NDVI), used widely since this information had become available for biologists. Apart from applications for dense forests, EVI seems to provide more reliable information than NDVI in areas with high soil exposure, as it is characteristic for the spiny forest ecosystem (Feeley et al., 2005; Garroutte et al., 2016; Rafanoharana et al., 2023). To minimize the influence of atmospheric signal and view-angle effects, we obtained 16-day EVI composites at a nominal spatial resolution of 250 m (product MOD13Q1, Version 6.1 (Didan, 2021)). After downloading and pre-processing all scenes, we removed pixels potentially affected by clouds based on the pixel reliability layer. We then filled in removed pixels using linear temporal interpolation between dates. We performed the EVI-based analysis with the R statistical environment (R Core Team 2024) using the packages MODIStsp (Busetto & Ranghetti, 2016), terra (Hijmans 2023) and tidyverse (Wickham et al., 2019).

We calculated two indices from the EVI data: the annual EVI integral (EVIi) as a surrogate of the annual ecosystem primary productivity, and the months of maximal or minimal EVI as a measure of interannual fluctuations in the months of maximal or minimal vegetation productivity. We used the indices to characterize different forest of the region (see previous section). Additionally, we compared the remote-sensing based measures of primary production with the ground-based vegetation descriptions of the 30 x 30 m2 plots. The analyses on the levels of the 30 x 30 m2 plots requires higher resolution than on the landscape level and even finer resolution when breaking the analyses down to monthly EVI measures. This was not a problem for the calculation of annual means but was not possible for monthly measures of two of the sites (Emisampy and Vohibataza). This reduced the number of sites to 15 for the analyses of monthly primary production.

Selection of study sites

Extensive biodiversity surveys had revealed gradual changes in community composition and patchy species distributions in the Tsimanampetsotse region that seemed to follow north-south and east west gradients of humidity (Feldt & Schlecht, 2016; Goodman et al., 2008; Goodman & Raselimanana, 2008; Nopper et al., 2017; Ralison, 2008; Ratovonamana et al., 2013). We used the patchy distribution patterns to compare sites with and without L. petteri. While we could determine the presence of L. petteri with certainty, absence of the species is less certain. For example, we had never observed L. petteri at Andranovao Rive during 15 years of field studies and extensive studies at night prior to this study (Rakotondranary et al., 2010; Bohr et al. 2011; Marquard et al., 2011; Nopper et al., 2017; Abel et al., 2023). Obviously, our standard study sites there had not included the parts of the forest suitable for L. petteri. Thus, “absence” could mean “lower density” than at sites where the species had been recorded as being present. The conclusions would not be affected by this classification. At the three sites of “Andranovao camp,” we consider L. petteri to be present at “Andranovao Rive” but absent at “Ambolely Nord” and “Degoly” (Table I).

Vegetation descriptions

We described vegetation characteristics on the regional scale for forests in general and on the local scale on the level of home ranges. We assessed regional differences in vegetation structure and tree species composition by establishing 17 plots of 30 x 30 m2 over a distance of 100 km along the western edge of Tsimanampetsotse National Park between Beheloke in the north and Tongaenoro in the south (Fig. 1, Table I; supplemented from Ratovonamana, 2016; Ratovonamana et al., 2011). We placed the 30 x 30 m2 plots at sites that seemed representative for the forest in the region. We based the decision of a forest being “representative” on our intimate knowledge of the region, supported by staff of the Association Analasoa, composed of people from the region. We installed the 30 x 30 m2 plots at three sites around Andranovao camp between June 2007 and March 2009 (Ratovonamana et al., 2011). In the other forests, we described the vegetation of the 30 x 30 m2 plots between July 2017 and November 2019. We described the 10 x 10 m2 vegetation plots at the sites of radio-tracking between September 2018 and November 2019.

We identified the MODIS pixels that matched the location of the 30 x 30 m2 plots and extracted the EVIi for each year between 2009 and 2020. We used these 12 EVIi values to calculate the EVIi mean and 95% coefficient of variation per 30 x 30 m2 plot during the 12-year period. We used the latter as a proxy for the inter-annual variability in productivity.

We added 10 x 10 m2 plots of vegetation descriptions related to the home ranges of L. petteri. These did not overlap with the descriptions of the 30 x 30 m2 plots. Once we could define the home ranges of the various individuals through radio-tracking, we established 10 x 10 m2 plots spaced 20-m apart inside and outside their home ranges, and we measured and identified trees as in the 30 x 30 m2 plots (Fig. S1). We established these 10 x 10 m2 plots at Ambory, Ankola Ebale, Andafiabo, and Andranovao camp. We took care that the plots outside the home ranges did not fall into the possible home range of neighboring animals that had not been radio-tracked. Our assignments of plots to be “inside” or “outside” the animals’ home ranges entail some uncertainty and could also be considered as plots of “low” versus “high” utilization frequency. In total, we completed 83 plots inside home ranges and 122 plots outside the home ranges.

At all plots, we identified and measured the diameter at breast height (DBH) of all trees ≥5 cm DBH, and estimated their height (Ratovonamana, 2016). The small DBH of only 5 cm was chosen because trees in the spiny forest are much smaller than trees in most other forest habitats inhabited by arboreal primates. Therefore, the tree size categories used elsewhere (DBH ≥ 10 cm or DBH ≥ 30 cm) would not describe this type of forest appropriately (Moonlight et al., 2021). For each plot (30 x 30 m2 as well as 10 x 10 m2), we calculated the mean height and mean DBH of all trees ≥5 cm DBH. We assigned the trees to three size categories: ≥5 cm DBH (all trees); 5–9.9 cm DBH (small trees); and ≥10 cm DBH (large trees). For each size category we counted the number of trees, the number of tree species, the number of food trees, and the number of food tree species. This resulted in 14 structural and floristic variables. In addition, we analyzed plots for the occurrence of specific tree species. For the analyses of occurrences of distinct tree species within the 30 x 30 m2 plots, we used all tree species that occurred with more than a total of 17 individuals. This number corresponds to the number of 30 x 30 m2 plots and would have allowed for the presence of one individual of the species in question in case of a uniform distribution. For the occurrence of distinct tree species in the 10 x 10 m2 plots, we used all tree species that were recorded with at least 205 individuals in the 205 total 10-x 10-m2 plots. This number corresponds to the number of 10 x 10 m2 plots and would have allowed for the presence of one individual of the species in question in case of a uniform distribution. We did not include herbaceous vegetation as L. petteri is not known to come to the ground for feeding. Herbaceous vegetation is mostly absent during the long dry season (Ratovonamana et al., 2013).

Animal capture and Ethical Statement

We caught animals by hand from their day shelters and anesthetized them with 0.4 ml of Ketanest (100 mg/ml; Nevermann et al., 2023). While anesthetized, we fitted the animals with radio transmitters (TW‐3 button‐cell tags, Biotrack, UK). We fastened the transmitters around the neck of the animals using a coated brass loop that also functioned as antenna. We kept the animals until they had fully recovered and released them at their capture site. The same individuals later reused sleeping trees where they had been captured. We removed radio‐collars after the end of the study by capturing the animals again.

We confirm that the research adhered to the legal requirements of Madagascar and that this research adhered to the American Society of Primatologists (ASP) Principles for the Ethical Treatment of Non-Human Primates. The research followed standard protocols for animal handling, capture, and radio-tracking, and was approved by the Ethics commission of the Institute of Cell and Systems Biology of Animals of the Universität Hamburg. Fieldwork was authorized by the Ministère de L’Environnement et du Developpement Durable: N° 187/18/MEEF/SG/DGF/DSAP/SCB.Re du 02/08/18.

Radiotelemetry

We tracked the animals sequentially by triangulation to assess their spatial distribution for two half-nights (18:00–23:00 hr) per animal and season, taking 1 fix/hr for each individual. We aimed for 12 fixes for each animal per season. We could not relocate some animals in subsequent seasons, thus providing only a home range proxy for one season. Transmitters were moved between animals once 12 fixes per season and two seasons had been completed. Because animals were not habituated, positions were mostly based on triangulation to avoid chasing the animals. We verified the accuracy of the triangulation during the day. We plotted home ranges based on minimum convex polygons using QGIS (Apel, 2020; Straßburg, 2021; Nevermann et al., 2023). While 12 fixes per season or 24 fixes per year seem too few to represent an animal’s home range, Lepilemur spp. have very small and stable home ranges (Droscher & Kappeler, 2013; Zinner et al., 2003). A much more extensive tracking study estimated home ranges of L. petteri at Berenty to be around 0.33 ha (Droscher & Kappeler, 2013). In our study, the individual home ranges overlapped extensively between seasons with median home range sizes of 0.31 ha and 0.47 ha during the dry and wet season, respectively. Also, home range size did not increase significantly after 20 fixes any more (Apel, 2020; Nevermann et al., 2023).

Competing interest

JUG is on the editorial board of the International Journal of Primatology. The authors declare that they have no conflict of interest.

Statistical analyses

30 x 30 m2 plots: The numbers of several of the single tree species deviated from normality. For consistency, we used Mann-Whitney U tests for the comparison of vegetation measures between sites with and without L. petteri for all vegetation variables measured on the ground. We used t-tests to compare EVIi measures between 30 x 30 m2 plots with and without L. petteri. We calculated Pearson and Spearman correlations to explore relationships between vegetation measures on the ground and the measures of EVIi.

10 x 10 m2 plots: To test for differences between plots located inside or outside the home ranges of L. petteri, we used multiple analyses of variance (MANOVA) with “home range” and “site” as group variable. In order to evaluate the importance of the different variables that showed significant differences between 10 x 10 m2 plots located inside or outside the home range of L. petteri, we used these variables in a binary logistic regression. For this analysis, we removed variables with a correlation coefficient ≥0.5 between the vegetation variables (Table S6).

We ran circular statistical tests with PAST (Hammer et al., 2001) and all other tests with SPSS. P-values are two-tailed with a level of significance set at 0.05.

Results

30 x 30 m2 plots

The 30 x 30 m2 plots covering 17 forest sites contained 2810 trees with a DBH ≥ 5 cm belonging to 123 different species. The nine sites with L. petteri had more large trees with a DBH ≥ 10 cm, the trees had a larger DBH, and the sites contained more different food species consisting of trees with a DBH ≥ 10 cm than sites where L. petteri was absent (Mann-Whitney U tests: p < 0.05; Table S2).

Of the 39 tree species that occurred with more than 17 individuals in the regional data set, Alluaudia procera and Euphorbia plagiantha occurred with higher abundances at sites with than without L. petteri. Alluaudia procera is an important food species found with 276 individuals (Mann-Whitney U test for differences between sites with and without L. petteri: p = 0.007; Table S2). Euphorbia plagiantha (N = 89) has not been seen to be eaten by L. petteri but also occurred only in forests with L. petteri (p = 0.039; Table S2).

The EVIi averaged for the 12 years of records for each region did not differ between the 30 x 30 m2 plots with or without L. petteri (F- and t-test: F = 0.049, p = 0.827; t = 0.211, df = 15, p = 0.836; Fig. 5a). The coefficients of variation of the EVIs varied more but were smaller for the sites with L. petteri than for the sites without L. petteri, indicating that sites with L. petteri showed less variation in annual primary production than sites where L. petteri was absent. However, the mean difference was not significant (F = 4.874, p = 0.043; t = 2.084, df = 13, p = 0.058; Fig. 5b).

Fig. 5
figure 5

Primary productivity (EVI integral) and predictability of production from 2009 to 2020 at sites with or without L. petteri in and around Tsimanampetsotse National Park, southwestern Madagascar. Values are means and 95% confidence intervals of (a) mean annual plant productivity, (b) coefficients of variation of the mean annual production, (c) predictability of the month with the highest primary production, and (d) predictability of the month with the lowest primary production; predictability characterized by the length of the mean vector rvector as illustrated in Fig. 4; **p < 0.01 based on t-tests (Table S2).

Full size image

The predictability of the month with the highest plant productivity during the 12-year period (length of the mean vector rvector) did not differ between sites with or without L. petteri (F = 0.895, p = 0.361; t = −1.875, df = 15, p = 0.083; Fig. 5c). The predictability of the month with the lowest plant productivity during the 12-year period was significantly lower at sites where L. petteri was absent than where the species was present (F = 3.340, p = 0.091; t = 3.102, df = 13, p = 0.008; Fig. 5d).

The EVIi means derived from MODIS for the years 2009 to 2020 were correlated significantly with the height of trees (Pearson correlation: r = 0.594, n = 17, p = 0.012; Table S3). The coefficient of variation of the EVIi was negatively correlated with the number of small tree species (r = −0.512, n = 17, p = 0.036) and negatively with the number of Euphorbia plagiantha (Spearman correlation: rs = −0.616, n = 17, p = 0.008). Annual EVIi means and their coefficients of variation were not correlated significantly with any other of the vegetation measurements on the ground (Table S3).

The predictability of the month with the highest or lowest plant productivity was represented by the length of the mean vector, derived from the location of the months with the highest productivity during the 12 years covered by our MODIS analyses. For the months with the highest plant productivity, the length of the mean vector was correlated significantly and negatively only with the number of large food tree species (r = −0.570, n = 15, p = 0.027; Table S3). For the months with the lowest plant productivity, the length of the mean vector was correlated significantly but positively with the total number of Alluaudia procera (r = 0.650, n = 15, p = 0.009), with the number of large trees (r = 0.527, n = 15, p = 0.044), and with all food trees (r = 0.592, n = 15, p = 0.020), reflected mainly by the number of large food trees (r = 0.567, n = 15, p = 0.027), but not by the number of small food trees (r = 0.304, n = 15, p = 0.271). The length of the mean vector for the distribution of months with the lowest plant productivity was correlated most significantly with the number pf Alluaudia procera (rs = 0.738, n = 15, p = 0.002; Table S3).

10 x 10 m2 plots in and outside home ranges

We had installed the 10 x 10 m2 plots in and outside home ranges of radio-tracked L. petteri in three of the 17 forest sites plus at one site (Ambory), where we did not have a 30 x 30 m2 plot (Table I). In total, we tracked 13 individuals (Table S4). The 205 total number of 10  x 10 m2 plots contained 4508 trees with a DBH ≥ 5 cm belonging to 86 species; 22 of the trees could not be identified. The structure and floristic composition of the four sites differed in almost all structural and floristic variables. Within the four sites, 10 x 10 m2 plots located inside the home ranges of L. petteri (N = 83) contained significantly higher and bigger trees, more large trees, more food trees, and more food tree species than plots outside the home ranges (N = 122; Fig. 6; Table S5). The differences in food trees and food tree species are mainly due to large but not to small trees. This is indicated by the more significant effect for large than for small food trees and tree species between plots located inside versus outside the home ranges of L. petteri (Table S5). Tree height showed the most significant difference between plots inside and outside the home ranges of L. petteri (MANOVA: F = 44.64, p < 0.001; Table S5). For other variables, pairwise comparisons were not significant for all four sites, but the combined sites were (Figs. 6b-h). At Ankora Ebale, many of the vegetation variables did not show significant differences between plots located inside versus outside the home ranges. The measures in question reached high values already in plots outside the home ranges (e.g., number of food trees, food tree species, Alluaudia procera; Fig. 6), possibly making the choice unnecessary.

Fig. 6
figure 6

Vegetation characteristics of 10 x 10 m2 plots with significant differences between plots located outside (white circles; N = 122) or inside (black dots; N = 83) home ranges of L. petteri in and around Tsimanampetsotse National Park, southwestern Madagascar; p-values according to MANOVA with “region” and “inside or outside the home range of L. petteri” as group variables. Black stars indicate significant differences for the site in question (p < 0.05) according to t-tests; values are means and standard errors; statistical details for the MANOVA in Table S5.

Full size image

We recorded only four tree species with more than 205 individuals in the local dataset. These were Alluaudia procera (N = 897), Gyrocarpus americanus (N = 820), Euphorbia plagiantha (N = 277), and Commiphora orbicularis (N = 234). Of these four species, only A. procera occurred in significantly higher numbers inside the home ranges than outside (MANOVA: F = 8.891, p = 0.003; Table S5). Commiphora orbicularis (also a food species) consistently occurred in higher densities inside than outside the home ranges, but the differences were not significant according to the MANOVA (F = 0.139, p = 0.239; Table S5). The binary regression identified the height of trees, the number of large food tree species, and the number of A. procera as the most important components for the placement of home ranges by L. petteri (Table II).

Table II Binary regression with vegetation characteristics from 10 x 10 m2 plots located within or outside the home ranges of L. petteri in and around Tsimanampetsotse National Park, southwestern Madagascar in 2018/2019
Full size table

Comparison of scales

Lepilemur petteri used sites and had their home ranges in areas with higher tree densities (number of trees) than found in forests where the species was absent. Because this variable was correlated with all other vegetation measures, we did not consider it in the binary regression. Of all other variables, the number of large food tree species (DBH ≥ 10 cm) and the number of Alluaudia procera were the only variables that, in all comparisons, showed higher values in parts of the forests that were occupied by L. petteri than forests or parts of the forests that were not (Table III; Table S2 for the 30 x 30 m2 plots; Table S5 for the 10 x 10 m2 plots).

Table III Vegetation variables showing significant differences between sites with or without L. petteri or plots located inside or outside the home ranges of L. petteri in and around Tsimanampetsotse National Park, southwestern Madagascar in 2018/2019
Full size table

Discussion

In our study, the regional occurrence and use of home ranges by Lepilemur petteri was linked to low interannual variability in primary production. More specifically, the species was associated with higher densities of large trees with a diameter at breast height ≥ 10 cm, with higher numbers of different large food tree species, and with the occurrence of their main food tree species, Alluaudia procera (Fig. 7). Apart from providing food year-round, bundles of the spiny branches of A. procera serve as safe sleeping sites, obviously an important component for habitat suitability for Lepilemur spp. in the spiny forest, where only very few trees are large enough to provide tree holes for shelter. Alluaudia spp. have leaves year-round (Ratovonamana et al., 2011). This seems to be important for surviving prolonged dry seasons. Possibly, the absence of this species as a food source can be compensated for by a larger number of other different food tree species. This might be the case at Andranovao camp and at Soarano. In the region around Andranovao camp, A. procera does not occur. In Soarano, A. procera seems to reach its northern limit and was not found in the 30 x 30 m2 plots. Both sites have more varied large food species than any of the sites with A. procera (Table S2). Possibly, at Andranovao camp and at Soarano, the larger diversity of food tree species can replace A. procera as staple food during the dry season. This interpretation is supported by the results of the binary regression where A. procera and the number of food tree species entered the model significantly (Table II). Yet, we also found L. petteri at Ampitanake. There, we recorded only two food species (the lowest number of any site) and no A. procera, although A. procera occurs in this region. Because we did not track L. petteri at Ampitanake, we cannot resolve this discrepancy. But the pair of L. petteri at Andranovao Rive (the site without A. procera neither in the plots, nor in the region) has now been seen consistently in the same place since 2019, and they actually reproduced. According to our impression, this site might receive slightly more ground water than the surroundings, and this slight increase in humidity might make the difference because of prolonged availability of leaves and the occurrence of bigger trees, including holes for shelter.

Fig. 7
figure 7

Alluaudia procera at Lavavolo in April 2019 serving as daytime shelter and food (note the leaves between the thorns) for Lepilemur petteri; also note the occurrence of lichens on one side of the tree but not on the other, indicating dew and wind direction as an important component in the system, which has not been paid attention to sufficiently (photo by S. Nevermann).

Full size image

Apart from environmental variables not included in our approach, we must consider anthropogenic effects, such as hunting or other disturbances that blur the results. Given that we did find some trends and significant habitat effects despite these overlaying disturbances, the relationships of the animals with the various habitat components might actually be stronger than indicated here. This highlights the concept that field work need not occur only in pristine conditions to find valid results that will help conserve threatened species. Given the decreasing size of suitable habitats within an anthropogenic matrix (Rafanoharana et al., 2024), studies of this kind ought to complement studies in more pristine areas. Our findings for L. petteri are consistent with habitat use by a variety of lemur species that are positively associated with large trees, habitat productivity, and tree species diversity—all components that can be linked to food, shelter or support (Baden, 2019; Campera et al., 2020; Eppley et al., 2017a, 2017b; Ganzhorn et al., 1997; Herrera, 2017; Nevermann et al., 2023; Seiler et al., 2014; Steffens et al., 2023; Vasey et al., 2018). Certainly, the various niche dimensions do not act in isolation, but their interplay has to be considered in their hierarchical or scale dependent context (Klopfer, 1969; Owens et al., 2020).

The preference for reliable food supply also might be reflected by the tendency that Lepilemur petteri is more likely to occur in areas with low interannual variation in primary production, as indicated by the lower variability of the EVIi (Fig. 5b), and at sites where the occurrence the lean season is predictable (Fig. 5d). Because Lepilemur spp. have small home ranges that remain s over the year and also for many years, the animals ought to avoid ranges that bear the risk of very low primary production at unpredictable times (changing months) of the year (Campera et al., 2021; Charles-Dominique & Hladik, 1971; Dinsmore et al., 2016; Droscher & Kappeler, 2013; Ganzhorn, 2002; Mandl et al., 2019; Nash, 1998; Radespiel et al., 2022; Thalmann, 2001; Wilmet et al., 2019; Zinner et al., 2003). In the spiny forest, the variation of primary production as measured by remote sensing could be linked to the number of small tree species per unit area but not to large trees. Small tree species seemed of lesser importance for L. petteri in our study but had been found to be important at Beza-Mahafaly (Nash, 1998). This discrepancy between sites remains unresolved. The lack of correlation between the variation in primary production and large tree species diversity is likely the result of the patchy distribution of large trees and their small crowns that make it difficult to capture characteristics of Madagascar’s dry and spiny forests with standardized remote sensing procedures (Rafanoharana et al., 2023). This problem might have been exaggerated in our analyses of the predictability of the month with the lowest primary production, because the month with lowest primary production represents the time with least foliage in an environment that is already characterized by low vegetation cover. Therefore, the correlations between the predictability of seasonality and vegetation measures on the ground should be considered with care as our MODIS analyses might not be suitable for such a combination of very high (for MODIS standards) spatial and temporal resolution. Nevertheless, the clear difference in the predictability of the month with the lowest primary productivity between sites where L. petteri was present compared with sites where the species was absent might highlight an important constraint for the distribution of L. petteri.

Assuming that extreme events are more important than gradual changes for the survival of species in general (Maxwell et al., 2019; Vasseur et al., 2014; Zhang et al., 2019) and for Malagasy species in particular (Dewar & Richard, 2007; Wright, 1999), the identification of limiting habitat components and their variation in space and time are key for the understanding of the evolution of Madagascar’s biota and their conservation under the impact of climate change. The linkage between remote sensing and vegetation measures on the ground on such a small scale as used in the present study warrants further verification. In previous studies, remote sensing and ground-based vegetation descriptions in broad leaf dry deciduous forest were also both suitable to predict habitat suitability for lemurs, but the two approaches based their classification on complementary information (Mercado Malabet et al., 2020; Steffens et al., 2023).

Few primate species are tightly associated with a single food resource, such as Lepilemur petteri (Charles-Dominique & Hladik, 1971; Droscher & Kappeler, 2014). Similar specializations among primates are known from bamboo lemurs in Madagascar (Hapalemur spp.: Tan et al., 2022), Daubentonia madagascariensis and Canarium spp. (Sterling, 2003), or Nasalis larvatus in Borneo (Yeager et al., 1997). The tight association observed in some places may not be obligatory, as illustrated by the occurrence of bamboo lemurs (Hapalemur spp.) in areas without bamboo (Eppley, et al., 2017a, 2017b; Mutschler, 1999), by the dietary flexibility of D. madagascariensis (Sterling et al., 1994), or by the rather diverse diets described for the supposedly mainly folivorous Lepilemur spp. (Rasoamazava et al., 2022). For some of these specialist species as well as for L. petteri, which also occurred at sites without or with only low densities of A. procera, higher numbers of food tree species may compensate for staple food sources because many different food species have a higher probability that at least one or a few species provide food when others fail. This is in line with the phenomenon that species-diverse ecosystems can withstand disturbances better than systems with fewer species (Craven et al., 2018; Isbell et al., 2015; Liang et al., 2016). Yet, in the case of Lepilemur spp., adaptations to survive the lean dry season are not well understood. Lepilemur spp. are among the mammalian species with the lowest metabolic rates but do not show any sign of heterothermy as do other lemur species (Bethge et al., 2017; Dausmann, 2014; Schmid & Ganzhorn, 1996). During the dry lean season, other Lepilemur spp. seem to eat whatever is available and adopt a time-minimizing strategy supposedly to save energy (Campera et al., 2021; Ganzhorn, 2002; Mandl et al., 2018). This also has been seen in the sympatric Lemur catta that reduce their energy demands during times of stress (Gould et al., 2011; Pereira, Strohecker et al., 1999). Contrary to these findings and assumptions, L. petteri increases its metabolic rate during the lean season (Bethge et al., 2021). These diverging results cannot be reconciled for the time being.

The relation between the occurrence of animals and either single key tree species or forests with more diverse tree species composition is relevant for ongoing reforestation initiatives worldwide, such as the African Great Green Wall, but also for initiatives in Madagascar. For practical reasons, reforestation is often based on a handful of species (Wade et al., 2018). While the objectives of reforestation and restoration initiatives certainly differ, the conservation value of species-poor reforestation projects as buffer zones or corridors for animals are limited, but could be improved considerably by adding more species (Denryter & Fischer, 2022; Flesch, 2023). For Madagascar, and also for the southern region, there is a comprehensive body of literature on tree species of value for people and animals alike, which could be used for reforestations to the mutual benefits of people and native animals alike (Aronson et al., 2018; Deleporte et al, 1996; Eppley et al., 2022; Gérard et al., 2015; Holloway & Short, 2014; Konersmann et al., 2022; Randrianasolo et al., 1996; Račevska et al., 2022; Sagar et al., 2021). Although some local projects try to maximize the tree species diversity in reforestation (Birkinshaw et al., 2013; Donati et al., 2021; Manjaribe et al., 2013), reforestation of large surfaces is often completed with a very limited number of species because of the lack of seeds or seedlings and the need to achieve the targets set by authorities (Ministère de l'Environnement de l'Ecologie et des Forêts, 2017). Given the longevity of these tree plantations and the efforts needed to put them in place, it might be worth reconsidering the approach in projects that do not have the means to produce species diverse plantations.

Climate change and the needs of the growing human population in Madagascar will have profound impacts on the remaining forests (Hending et al., 2022; Morelli et al., 2020). Even if forests within protected areas seem to be better protected than the forest as a whole (Rafanoharana et al., 2024), they are also likely to suffer from climate change and human pressure. If the present deforestation rates continue, protected areas will represent islands in an anthropogenic, nonforested matrix. Effective conservation will then need additional buffer zones and corridors between otherwise isolated blocs of forest (Manjaribe et al., 2013). Given the long-term perspective of forest restoration and the need for restored forest habitats, reforestation projects should put more emphasis on increasing the number of tree species used for reforestation or rehabilitation.