Introduction

Biodiversity decline is a pressing global environmental issue (Fischer et al. 2023), where human-induced landscape changes play a pivotal role (Thomas et al. 2004; Newbold et al. 2015). Birds, as widely distributed and easily observable species, serve as vital indicators of biodiversity status (Strohbach et al. 2013). However, bird populations are dwindling due to extensive habitat fragmentation and loss (Haddad et al. 2015; Püttker et al. 2020). Since 1990, 57 bird species have become extinct (Ceballos et al. 2015), and 13% are currently at risk of extinction on a global scale (Tilman et al. 2017), impacting the provision of ecosystem services and human well-being (Cox et al. 2018; Jahani et al. 2021). Therefore, understanding the impact of landscape patterns on bird diversity is crucial for developing effective conservation strategies and achieving SDG15 (Life on Land).

Landscape pattern encompasses the spatial distribution, composition and configuration of various land use types (Krummel et al. 1987; Turner and Gardner 2015; Liu et al. 2020). It is influenced by multiple factors, with human activities playing a pivotal role in shaping land types that impact landscape patterns (Turner 2005). Anthropogenic and natural factors have collectively driven landscape fragmentation, leading to a shift from landscape homogeneity to complexity (Turner 1990; Ryser et al. 2021). Research findings indicate a significant 95.8% decrease in the average patch area of global land during the early twenty-first century (Jacobson et al. 2019), accompanied by a 15.5% increase in fragmented forest patches (Rocchini et al. 2010). The dynamics of landscape change involve modifications in patch area, edge length, shape, and isolation, which contribute to the emergence of diverse landscape pattern characteristics (Haddad et al. 2015). The aspects of composition, complexity, and heterogeneity provide valuable insights into the fundamental structure and dynamics of landscapes (Duarte et al. 2018; Jin et al. 2023). Composition refers to the proportional distribution of land use patches (McGarigal and Marks 1995). Complexity captures the morphological and structural features of patches, such as higher fractal dimension indicating more complex patch shapes (Feng and Liu 2015). Heterogeneity manifests as the diversity of patch types within the landscape (Rocchini et al. 2010).

Bird diversity is closely linked to landscape patterns, with species richness serving as a valuable indicator for quantifying biodiversity. Research findings indicate that species richness plays a pivotal role in driving ecosystem functionality, emphasizing its significance beyond species turnover (Albrecht et al. 2021). Given its clear and intuitive portrayal of species diversity, this metric is widely utilized on a global scale (Santini et al. 2017; Stevenson et al. 2024). Previous research has indicated a positive correlation between forest cover and bird richness (Bonfim et al. 2021). Additionally, studies have found that bird richness is positively associated with the proportion of native forest but negatively correlated with the proportion of mixed forest (Lisón et al. 2022). The habitat heterogeneity hypothesis suggests that greater habitat heterogeneity supports higher species richness (MacArthur et al. 1966; Stein et al. 2014; Albrecht et al. 2021). Furthermore, it has been demonstrated that heterogeneity in agricultural landscapes contributes to increased bird richness (Anderle et al. 2023). In urban environments, green space heterogeneity positively influences the richness of urban avoider species (Chiron et al. 2024). However, the impact of landscape complexity on birds is comparatively weaker than that of composition and heterogeneity (Bonfim et al. 2021; Semper-Pascual et al. 2021; Dong et al. 2023).

Despite numerous published studies, inconsistent findings persist regarding the relationship between landscape patterns and bird richness, especially concerning landscape complexity (Henden et al. 2013; Schindler et al. 2013; Bonfim et al. 2021; Molina-Marin et al. 2022), and the potential sources contributing to these variations remain unclear. This inconsistency presents challenges for providing cohesive guidance in landscape planning and design. Therefore, conducting a comprehensive and systematic literature review is essential to comprehend the diverse results and their sources. This will contribute to a better understanding of how landscape patterns influence bird diversity, allowing landscape practitioners to develop effective strategies for bird conservation.

Meta-analysis has been proven to be a scientific and quantitative statistical method (Muruga et al. 2024), providing an effective approach for extensively understanding the relationship between landscape patterns and bird richness on a broad scale. It has been applied to examine the impacts of anthropogenic disturbances on bird functional diversity (Matuoka et al. 2020), the effects of roads on bird richness (Kroeger et al. 2022), as well as the influence of urbanization on bird abundance (Lakatos et al. 2022). Additionally, meta-analysis has revealed a positive association between landscape complexity and the richness and abundance of arthropods (Marja et al. 2022). However, there is a notable dearth of quantitative synthesis regarding the correlation between bird richness and landscape patterns. Consequently, this study aims to bridge this gap by undertaking a systematic evaluation and meta-analysis. We addressed the following questions: (1) How does landscape composition, complexity, and heterogeneity influence bird richness? (2) What are the sources of variation in the outcomes? (3) What threshold values for crucial indicators can be identified to inform landscape planning and design?

Materials and methods

Literature search

We conducted a comprehensive systematic search in the Web of Science literature database using the search strategy (“landscape pattern OR complex* OR fragment* OR heterogeneity OR configuration OR composition OR landscape metric or landscape index OR landscape indices”) AND (“bird* or avian*”), AND (“diversity or species richness or biodiversity”). Initially, three sets of search terms were established to encompass landscape patterns, birds, and species richness. These sets were searched using the OR operator. Subsequently, the results from the three searches were combined using the AND operator to perform a comprehensive search (Table S1 in the Supplemental Material contains the detailed search expressions). No limitations were imposed on the publication dates of the retrieved articles to ensure a comprehensive coverage of relevant literature. The same search strategy was applied in the Scopus database using the TITLE-ABS-KEY field. The search results were refined to include articles written in English and categorized as “Article”. This process yielded a total of 18,300 articles.

After removing duplicates, 11,462 articles underwent screening. Non-original research papers were excluded, and further screening based on titles and abstracts resulted in 1155 articles. Full-text articles were then screened using the following exclusion criteria (Fig. 1): (1) Absence of landscape pattern metrics, as this would not provide the necessary data relevant to our research objectives; (2) Lack of bird richness data, impeding the quantitative analysis of the relationship with landscape patterns; (3) Studies that were unable to provide the necessary effect sizes were excluded from the meta-analysis, regardless of their relevance. (4) Lastly, studies not based on real research case sites were also excluded, as simulated results would not contribute to our research objectives. Ultimately, 101 articles were included in the review.

Fig. 1
figure 1

Flow diagram of the literature selection process

Full size image

Data extraction

We extracted various information from the literature, including author names, publication year, geographic location, bird data sources, bird richness, sample area, sample size, landscape metrics, and correlation coefficients. Data were recorded in Excel spreadsheets for subsequent analysis. Table 1 presents the landscape metrics mentioned in the collected literature, along with their corresponding descriptions. The literature data extraction was independently performed by two researchers, who also assessed the quality of the papers. Heterogeneity among the studies will be evaluated and addressed through heterogeneity evaluation and sensitivity analysis.

Table 1 Landscape metrics used in publications
Full size table

In constructing our literature database, the calculation of landscape metrics involved two approaches: field data collection and the utilization of land use data in Fragstats software. Some metrics, calculated for individual land use types (e.g., “class” metrics in Fragstats), may vary in their effects on birds due to different land use attributes. For example, the effect of the number of forest patches on bird cannot be equated to the impact of the overall number of landscape patches. Thus, we classified the landscape metrics into two categories: hybrid landscape metrics and single landscape metrics. This categorization facilitates a nuanced analysis of the impact of each metric on bird richness.

To elucidate the heterogeneity of research results, we incorporated external explanatory variables and collected data from WorldClim (https://worldclim.org/) for the year 2020. This included global annual total precipitation, average maximum temperature, and average minimum temperature, all with a spatial resolution of 21 km2. We also acquired 2020 global net primary productivity (NPP) data from MODIS with 500 m resolution. The vector data for the Ecoregion map were obtained from Olson et al. (2001). All the aforementioned data were stored in ArcGIS 10.8.

Effect size calculation

Species richness is a readily comprehensible and uncontroversial indicator for describing biodiversity (Fjellstad et al. 2001). In this study, we employed the Pearson’s correlation coefficient (Pearson’s r) between bird richness and landscape metrics as the effect size for meta-analysis. Pearson’s r values were obtained through the following approaches: (1) Direct reporting of Pearson’s r values in the original studies (n = 17; 16.83%); (2) Conversion of Spearman’s correlation coefficients provided in the articles to Pearson’s r (n = 5; 4.95%) (Yang and Konrath 2023); (3) Calculation of Pearson’s r using raw data from the articles (n = 54; 53.47%); (4) Extraction of numerical values for bird richness and landscape metrics from graphical data using GetData Graph Digitizer 2.26 software for Pearson’s r computation (n = 25; 24.75%) (Cauvy-Fraunié and Dangles 2019; Liu et al. 2023). To ensure data normality prior to meta-analysis, Fisher’s z transformation was applied to Pearson’s r (Koricheva et al. 2013). The formula for Fisher’s z transformation is as follows (Borenstein et al. 2009):

$$z=\frac{1}{2}\text{ln}(\frac{1+r}{1-r})$$
(1)

where \(z\) represents the Pearson’s r that has been transformed using Fisher’s z. The variance of \(z\) can be calculated as follows:

$${V}_{z}=\frac{1}{n-3}$$
(2)

where \({V}_{z}\) denotes the variance of \(z\), and \(n\) represents the sample size.

Statistical analysis

We conducted a meta-analysis on the Fisher’s z transformed Pearson’s r, including publications with at least 5 counts (Wu et al. 2023; Li et al. 2024). For publications with fewer counts, we provided narrative description. The random effects model assumes that the estimated effects in the studies originate from different population effects, and the heterogeneity among studies includes not only sampling errors but also genuine effect differences (Borenstein et al. 2009). The data collection for this study encompasses studies with varied landscape characteristics and experimental designs, rendering the random effects model more appropriate. Employing this model, we obtained a 95% confidence interval (CI) and visualized the results with a forest plot. A non-zero 95% CI indicated a statistically significant combined effect size of landscape pattern on bird richness, while a CI including zero suggested a significant difference in the results (Song et al. 2019). To facilitate interpretation, we converted Fisher’s z values back to Pearson’s r.

When heterogeneity exists in the meta-analysis results, meta-regression can be employed to explore the sources of heterogeneity (Hedges et al. 2010). By introducing covariates that may impact the variation in study outcomes and assessing their associations with the effect sizes, meta-regression provides valuable insights. Therefore, we performed meta-regression with seven moderators: sample area, latitude and longitude of the study area, publication year, NPP, annual total precipitation, average maximum temperature, and average minimum temperature. We also assessed publication bias, which can affect the credibility of the meta-analysis, by employing a funnel plot and conducting Egger test. Funnel plot asymmetry and a p-value less than 0.05 in the Egger test indicated the presence of publication bias (Egger et al. 1997). In the presence of publication bias, we employed the Trim-and-Fill method to assess its impact on the results. If the effect size estimate remains unchanged after accounting for potentially missing studies, it suggests that publication bias has minimal influence on the meta-analysis results.

The robustness of the meta-analysis results was assessed using sensitivity analysis. We used the “one study removed” approach, sequentially excluding one study at a time while combining the effect sizes of the remaining studies. Changes in the combined effect size and 95% CI were compared to determine consistency with the original meta-analysis. Consistent results would indicate the reliability of the meta-analysis. The aforementioned analyses utilized the “metafor” package in RStudio3.1 (Viechtbauer 2010).

To test thresholds for guiding landscape planning and design, we obtained the bioclimatic zones associated with each individual study and conducted both linear and nonlinear regression analyses. This comprehensive analysis encompassed a range of regression models, including linear, polynomial, exponential, power, and piecewise models. By exploring the fitting performance of different functional forms, we identified the most suitable model for characterizing the relationship between forest cover and bird richness. The model performance was evaluated using the Akaike Information Criterion (AIC) to determine the most suitable fit. These analyses were carried out in Rstudio 3.1, utilizing the “ggtrendline” package for linear, polynomial, exponential, and power regressions (Mei et al. 2022), the “segmented” package for segmented regression (Muggeo 2008), and the “cluster” package for k-means clustering (Mächler et al. 2012).

Results

Narrative synthesis of study characteristics

Our literature database comprises 101 publications sourced from 56 different journals. Research on the correlation between landscape patterns and birds has increased significantly over the past three decades (Fig. 2). The earliest publication in our database dates back to 1993, with 77.2% of the literature published after 2013. The years 2018 and 2022 emerge as pivotal, each hosting the highest number of included publications (n = 11). These publications span across 51 countries globally, with a predominant focus on temperate broadleaf and mixed forest biomes (Fig. 3). Geographically, the distribution of publications is primarily concentrated in Europe (n = 49), followed by Asia (n = 24), North America (n = 23), South America (n = 20), and a smaller number of studies conducted in Africa (n = 9) and Oceania (n = 9). China leads in the number of publications (n = 14), followed by the United States (n = 11), Australia (n = 9), and Brazil (n = 9), indicating a concentration of research efforts in these countries.

Fig. 2
figure 2

The number of reviewed studies from 1993 to 2023

Full size image
Fig. 3
figure 3

The geographical distribution of reviewed studies (where a single document may encompass multiple countries), with the Ecoregion map sourced from Olson et al. (2001)

Full size image

A significant majority of articles (81.19%) utilized structured surveys to collect bird data (Fig. 4a). Some studies relied on data extracted from previous literature (n = 8) or utilized bird observation data from official institutions or organizations (n = 8). Within our literature dataset, two publications in 2018 made use of citizen science data. The use of semi-structured citizen science data, gathered by volunteers, has emerged as a valuable source for bird research. This data type offers wide observational coverage and substantial volume, effectively addressing data gaps (Planillo et al. 2021; Squires et al. 2021). Additionally, certain studies combined field surveys with supplementary bird observation records obtained from historical literature.

Fig. 4
figure 4

The frequency of a bird data and b research methods used in reviewed papers

Full size image

The association between bird richness and landscape patterns has been investigated using various models, with regression models being the most commonly utilized. This includes Multiple Linear Regression (n = 17), Generalized Linear Model (n = 16), and Generalized Linear Mixed Model (n = 14) (Fig. 4b). The Poisson distribution is frequently used to characterize count data distribution (Mohd-Azlan and Lawes 2011; Bain et al. 2020). Bird observation data is often collected in the form of counts, aligning with the assumptions of the Poisson-based Generalized Linear Model. Pearson correlation analysis was used in 13 instances to examine the relationship between landscape metrics and bird richness. In recent years, machine learning algorithms have gained prominence as a valuable tool. For example, Steel et al. (2017) employed an enhanced regression tree model, a self-learning binary algorithm that combines regression tree algorithms and boosting methods, to capture the nonlinear relationship between bird richness and predictive factors. Ferenc et al. (2016) utilized a random forest model to evaluate the ranking of landscape factors’ importance.

Effects of landscape composition, complexity, and heterogeneity on bird richness

Landscape composition

He meta-analysis included 31 studies examining the impact of Forest% on bird richness. Among these studies, 54.84% reported positive effects, 12.90% reported negative effects. And the remaining 32.26% of the studies reported neutral responses (Fig. 5a). The random effects model revealed that Forest% had the highest combined effect size (r = 0.32; 95% CI 0.18, 0.48) (Fig. 5a). Following Forest%, Shrubland% (r = 0.29; 95% CI 0.20, 0.39) and Water% (r = 0.21; 95% CI 0.08, 0.35) exhibited the second and third highest promoting effects on bird richness, respectively (Fig. 5b). 85.71% and 40.00% of the studies reported positive results for Shrubland% and Water%, respectively (Fig. 5a). In contrast, increasing Grassland% (r = − 0.17; 95% CI − 0.31, − 0.02), Agricultural% (r = − 0.35; 95% CI − 0.64, − 0.11), and Urban% (r = − 0.19; 95% CI − 0.34, − 0.04) would negatively impact species richness (Fig. 5b).

Fig. 5
figure 5

a The effect size and 95% CI for each study examining the impact of landscape composition on bird richness. Each square represents the Pearson’s r after Fisher z transformation, while the vertical line depicts the distribution of the 95% CI. An orange vertical line indicates a significant positive effect, green represents a significant negative effect, and gray signifies a neutral response. b Summary effect sizes for bird richness responses to landscape composition. Asterisk within the parentheses (*) indicates the conversion of effect sizes to Pearson r. Significance code: < 0.001 *** < 0.01 ** < 0.05*

Full size image

Landscape complexity

The meta-analysis incorporated six landscape metrics to assess landscape complexity. When not distinguishing between landscape types, the 95% CI of all metrics’ effect sizes encompassed 0, indicating heterogeneous results. Specifically, ENN (r = 0.17; 95% CI − 0.23, 0.57), FRAC (r = 0.04; 95% CI − 0.31, 0.39), PD (r = 0.14; 95% CI − 0.24, 0.52), and SHAPE (r = 0.09; 95% CI − 0.15, 0.34) exhibited a weak positive trend, but these associations were not statistically significant. Similarly, ED (r = − 0.02; 95% CI − 0.18, 0.14) and NP (r = − 0.01; 95% CI −0.18, 0.17) showed non-significant negative relationships (Fig. 6b). However, when considering landscape types separately, the summary effect size of PD in natural landscapes was statistically significant (r = 0.42; 95% CI 0.17, 0.70), while the other metrics continued to display heterogeneity.

Fig. 6
figure 6

a The effect size and 95% CI for each study examining the impact of landscape complexity on bird richness. b Summary effect sizes for bird richness responses to landscape complexity. The annotations carry the same meaning as in Fig. 5

Full size image

For landscape complexity metrics with limited sample sizes, we provide the distribution of Pearson’s r numerically (Fig. 7). Concerning DIVISION, three studies found a negative influence on species richness, while one study found a weak positive correlation. Regarding ECON, conflicting conclusions were presented in two articles. In terms of IJI, both studies indicated a weak correlation with bird richness. Only one source in the literature reported a positive correlation between the LPI and bird richness. For the PROX and TE, two studies each reported positive outcomes.

Fig. 7
figure 7

The distribution of Pearson’s r between landscape metrics and bird richness within a hybrid landscape context, considering the limited literature coverage

Full size image

Figure 8 illustrates five instances of a positive correlation between AI and bird richness, along with two instances of a negative correlation, specifically observed in water bodies and urban land. Additionally, three results indicate a positive correlation between CONTIG and bird richness, while one study reveals a negative correlation. CIRCLE and COMPLEX generally exhibit a positive correlation with bird richness. Conversely, six studies assume a negative correlation between NP and bird richness, implying that the fragmentation of landscape context may have a detrimental impact on bird richness. ED, FRAC, LPI, and SHAPE present inconsistent outcomes, even within the same land cover category, indicating noteworthy disparities.

Fig. 8
figure 8

The distribution of Pearson’s r between landscape metrics and bird richness within a single landscape context

Full size image

Landscape heterogeneity and area

Concerning landscape heterogeneity, 55.56% of the studies reported a positive influence of PR, while 22.22% reported negative or non-significant results (Fig. 9a). Similarly, SHDI (r = 0.31; 95% CI 0.20, 0.45) showed a positive correlation with bird richness, demonstrating its positive influence in both natural and agricultural landscapes. The analysis revealed a positive effect of landscape area (r = 0.53; 95% CI 0.47, 0.71), observed across natural, agricultural, and urban landscapes (Fig. 9b).

Fig. 9
figure 9

a The effect size and 95% CI for each study examining the impact of landscape heterogeneity on bird richness. b Summary effect sizes for bird richness responses to landscape heterogeneity. The annotations carry the same meaning as in Fig. 5

Full size image

Evaluation of publication bias and sensitivity analysis

The assessment of publication bias revealed that the Egger test p-values for all indicators, except Agricultural%, were greater than 0.05 (Fig. S1–S3 in the Supplemental Material). Therefore, Agricultural% was identified as having publication bias. Applying the Trim-and-Fill method and including two additional studies, the meta-analysis produced a combined result of r = − 0.29; 95% CI − 0.54, − 0.05, indicating a slight change compared to the previous value (r = − 0.35; 95% CI − 0.64, − 0.11). The correlation trend did not reverse. Sensitivity analysis demonstrated the stability of the meta-analysis results for landscape composition, landscape heterogeneity, and area metrics. However, certain metrics of landscape complexity showed a change in the trend of the combined effect size, although the 95% CI did not reach statistical significance, consistent with the original results (Tables S4–S18 in the Supplemental Material). In summary, the influence of landscape composition, heterogeneity, and area on bird richness remains stable, while the results for landscape complexity exhibit notable variability.

Factors influencing bird richness in response to landscape complexity metrics

To examine the sources of result variability, a meta-regression analysis was performed on three landscape complexity metrics, leading to the exclusion of sample area, latitude, longitude, publication year, and temperature as influential factors (Fig. S4–S6 in the Supplemental Material). However, NPP and annual precipitation were found to partially account for the variation (Fig. 10). In regions characterized by higher NPP, increasing ED and NP tend to have more negative effects, while SHAPE shows a tendency towards more positive impacts. Moreover, in regions characterized by higher precipitation, there is an indication that increasing ED and NP may have a propensity to result in more negative impacts.

Fig. 10
figure 10

The results of the meta-regression analysis, with the dashed line indicating the 95% CI. The size of the circles is positively correlated with the variance of the effect sizes

Full size image

Test for forest cover thresholds

Forest cover plays a significant role in promoting biodiversity within landscape composition. Establishing thresholds for forest cover serves as a measure to restrict uncontrolled expansion of construction land. Figure 11 presents the regression results between global forest cover and species richness. Among the five models used, segmented regression demonstrates good fitting performance (the fitting results of each model can be found in Table S19 in the Supplemental Material). The results reveal variations in forest cover thresholds across different bioclimatic zones, with an average threshold value of 45.28%. Specifically, a study in the Deserts and Xeric Shrublands region indicates a threshold of 60.91%, while the Tropical Biome region includes seven studies with an average threshold of 51.25%. The Temperate Biome region incorporates four studies with an average threshold of 45.02%.

Fig. 11
figure 11

Forest cover thresholds in different bioclimatic zones

Full size image

Discussion

Association between landscape patterns and bird richness

The research findings highlight the importance of forested habitats in preserving bird richness, as they offer intricate community structures, abundant food resources, and suitable nesting habitats, while also alleviating anthropogenic disturbances (MacArthur and MacArthur 1961 and Dong et al. 2023). However, the relationship between specific bird populations and forests can vary. Frugivorous birds, for instance, often rely heavily on forest coverage (Bonfim et al. 2021), while studies have found a negative correlation between forest coverage and the diversity of habitat generalist birds (Carrara et al. 2015). The meta-analysis results further demonstrate the positive contribution of shrublands to bird richness. Moreover, several studies have demonstrated synergistic effects on bird richness in mixed landscapes of forests and shrublands (Deschênes et al. 2003; Powell and Steidl 2015; Li et al. 2019). These effects can be attributed to the complex vegetation structure and diverse plant communities, providing additional ecological niches for birds (Deschênes et al. 2003). For instance, insectivorous and omnivorous birds show a stronger preference for such habitats (Ouin et al. 2015; ); Li et al. 2019 suggest that reducing understory vegetation or deadwood may negatively impact insectivorous birds. Moreover, water bodies and their surroundings provide suitable habitats for birds due to the presence of aquatic plants, which offer nesting sites and cover for waterfowl (Dong et al. 2023; Aubrechtová et al. 2024).

The habitat heterogeneity hypothesis posits that increasing heterogeneity in habitats contributes to promoting biodiversity (Pianka 1966). In this study, the combined effect size of the landscape Shannon’s diversity index showed a significant positive correlation in both natural and agricultural landscapes. Heterogeneous landscapes are recognized for providing more ecological niches compared to homogeneous landscapes (Allouche et al. 2012; Lorenzón et al. 2016). In natural landscapes, forest habitats provide diverse habitat choices and utilization for different species, thanks to their complex spatial structures with vertical layers of varying heights and habitat sizes (Laiolo 2002). In agricultural landscapes, a range of environmental conditions, including farmlands, water bodies, irrigation systems, and natural habitats, create diverse habitat environments (Belfrage et al. 2015). Furthermore, Li et al. (2020) highlighted the potential food sources for birds in diverse agricultural landscape types. Gardens and flowerbeds, for example, with their moist soil, attract a greater abundance of insects and vegetation, benefiting both insectivorous and frugivorous birds (Li et al. 2020). Due to the strong predictive performance of habitat heterogeneity in determining bird richness, Martinez-Núñez et al. (2023) employed land cover diversity data to predict global patterns of bird species populations and functional diversity.

The results underscore landscape area as the dominant factor, aligning with the species-area relationship theory (Boecklen 1986; Lomolino 2000). This theory suggests that larger habitat areas generally harbor a greater species richness. Nevertheless, landscape fragmentation diminishes patch size and intensifies patch isolation, hindering species dispersal and gene flow, ultimately compromising species diversity (Liu et al. 2022). Landscape area is the most crucial factor influencing species richness, and efforts should be made to increase the area of ecological land (e.g., forests, shrublands, wetlands) under limited conditions. Moreover, it is important to consider that increasing heterogeneity comes with the trade-off of reduced patch size. Urban avoider species exhibit a preference for large and highly heterogeneous green spaces, but the positive impact of increased heterogeneity on species diversity is less pronounced in small habitats (Chiron et al. 2024).

Variability in landscape complexity results may related to the differences in NPP and precipitation

The results of the meta-regression showed that NPP and precipitation significantly explained the variation in bird richness related to the ED, NP, and SHAPE metrics. Drawing on existing knowledge, we offer explanations for these findings. The “Species-energy theory” supports a positive association between NPP and bird richness (Wright 1983; Leveau 2019). In regions with high NPP, increased ED and NP manifest as landscape fragmentation, coupled with reduced habitat connectivity resulting from edge effects (Vieira-Alencar et al. 2023). These factors collectively have the potential to amplify the adverse effects on bird richness. To our knowledge, most studies have focused on examining the individual effects of vegetation productivity and landscape complexity on bird diversity. However, there has been comparatively limited research that integrates these factors to illuminate their underlying mechanisms. Leveau (2019) proposes a direct influence of vegetation productivity on bird richness, with habitat heterogeneity serving as an indirect factor. The results implies that the contributions of landscape patterns and vegetation productivity to bird richness might differ, potentially diminishing the explanatory power of indirect factors. Meanwhile, the meta-regression reveals that in regions with higher precipitation, the effects of ED, NP, and SHAPE on bird richness become more negative. Previous studies have demonstrated the adverse effects of precipitation on montane bird richness (Santillán et al. 2018, 2020), and increased precipitation reduces bird nesting success (Zuckerberg et al. 2018). Consequently, we propose that variations in vegetation productivity and precipitation across different regions worldwide could lead to diverse effects of landscape complexity on bird richness.

Setting thresholds for forest cover requires attention to regional variations

The concept of a threshold suggests a transition or change from one state to another within a specific regional context (Huggett 2005). This study aimed to identify the forest cover threshold as an indicator for managing urban expansion and preserving bird diversity in landscape planning. Specifically, we examined threshold variations in bird richness and forest cover across different bioclimatic regions. Our findings demonstrate a notable trend in tropical regions, where a higher threshold exists between bird richness and forest cover. This implies that birds in tropical areas depend on a greater extent of forest cover, supporting prior research (Betts et al. 2019; Weeks et al. 2023) that highlights the heightened sensitivity of bird species to forest fragmentation in tropical environments. For instance, certain tropical insectivorous bird species exhibit reluctance or inability to cross small gaps like roads, making them vulnerable to the adverse effects of habitat disturbance (Tobias et al. 2013).

We observed that within the same bioclimatic region, there could be variations in the threshold outcomes. To explain this finding, we considered potential reasons. Firstly, the impact of forest cover on bird richness could be influenced by other factors, such as the significant effects of water bodies and shrublands, as indicated by the findings of meta-analysis. Consequently, disparities in landscape composition among regions could affect the accuracy of the relationship between forest cover and bird richness. Secondly, the establishment of thresholds relied on regression analysis using independent research data, which might have been influenced by sample variations across studies.

The establishment of thresholds aims to mitigate the negative impacts of ongoing forest loss on biodiversity (Radford et al. 2005). However, it is crucial to acknowledge that ecosystem complexity imposes limitations on threshold determination (Fahrig 2003; Fischer and Lindenmayer 2007). Habitat fragmentation results in structural and qualitative changes that often yield reciprocal effects (Ewers and Didham 2005). Therefore, thresholds frequently fail to capture the intricate relationships within ecosystems (Ewers and Didham 2007). In addition to bioclimatic region variations, previous studies have also revealed inter-species differences in thresholds (Betts et al. 2007; Macchi et al. 2019). Guénette and Villard (2005) found lower tolerance to forest logging in Setophaga fusca, Empidonax flaviventris, and Certhia americana, while Vireo solitarius, Regulus satrapa, and Seiurus aurocapilla demonstrated higher tolerance to silvicultural activities. Overall, it is imperative to address the issue of thresholds diligently in biodiversity conservation. Adequate species sampling, consideration of inter-species differences, and local landscape characteristics should be taken into account prior to establishing thresholds.

Limitations and future directions

In the study, an extensive search was conducted across multiple literature databases, and a meticulous assessment of the selected publications was performed. However, it is worth noting that the availability of publications focusing on specific landscape metrics, such as ENN and PD, was limited. This limitation could potentially affect the robustness of meta-analysis results. Furthermore, the heterogeneity observed in the meta-analysis outcomes may stem from variations in the calculation methods employed for landscape pattern metrics. The utilization of field sampling data versus the application of Fragstats software for land use data calculations could introduce discrepancies in the meta-analysis findings. Moreover, our study focused solely on bird richness as an indicator, overlooking important metrics such as bird abundance and Shannon diversity at the alpha diversity level. Future research can also further explore beta diversity, capturing variations between communities. Furthermore, a more comprehensive understanding of the heterogeneity of results can be achieved by considering the distinct responses of species communities to landscape patterns through taxonomic classification.

We propose several questions for future research. Primarily, existing studies have primarily examined the relationship between landscape patterns and bird diversity at a single temporal scale. Yet, due to human activities, landscapes undergo dynamic changes. Understanding the implications of this variability on bird diversity necessitates analyzing landscape pattern changes over extended temporal scales, which should be a future research focus. In particular, it is important to examine variations in the impacts of landscape composition, complexity, and heterogeneity changes on bird diversity across different bioclimatic regions, as well as the role of these complex interactions in shaping species communities. In addition, exploring species-specific responses to landscape changes and their effects on colonization and extinction patterns can provide valuable insights.

The emergence of citizen science data and machine learning algorithms offers advantageous tools for exploring the relationship between landscape patterns and bird diversity at broader spatiotemporal scales. Considering the current methods of acquiring bird observation data, integrating diverse data sources, such as citizen science and historical literature, can supplement historical bird observations. Moreover, regarding data analysis models, the presence of multicollinearity among landscape pattern factors limits the effectiveness of traditional regression algorithms in addressing this issue. Therefore, the future application of machine learning algorithms holds greater potential. For instance, in elucidating biodiversity hotspots, large-scale citizen science data on birds and environmental variables can be collected, and key features can be extracted using machine learning algorithms to simulate spatial patterns of bird diversity.

Currently, there is a wealth of evidence on the impacts of landscape composition, heterogeneity, and area on bird richness. However, the effects of landscape complexity, especially morphology, on bird species are still debated and require further investigation. During the meta-regression process, we identified a divergence in the effects of climate and vegetation productivity, which contributed to variations in landscape complexity and species richness. This variability in outcomes can be attributed to the intricate relationship between landscape characteristics and climate change. Changes in climate can affect the timing, types, and density of vegetation growth, thereby impacting the availability of food resources (Lorel et al. 2021). Additionally, climate change can modify habitat structure and connectivity, ultimately influencing bird migration patterns, habitat selection, and population distribution (Guan et al. 2021). In the future, conducting comprehensive analyses of landscape complexity, vegetation productivity, and climate will aid in understanding their underlying mechanisms and revealing the responses of bird diversity to complex environments.

Conclusion

This study systematically reviewed the impact of landscape patterns on bird richness. The findings underscoring the significance of landscape composition, heterogeneity, and area metrics in assessing bird richness. However, the role of landscape complexity remains a subject of ongoing debate. The proportion of forests, shrublands, and water positively contribute to species richness, while grasslands, agricultural land, and urban land have adverse effects. Among these land types, forests exhibit the highest correlation with bird richness, and a recommend forest cover ranging from 15 to 40% to promote bird survival. Landscape heterogeneity and area demonstrate a positive role in bird richness. The impact of landscape complexity varies significantly. By controlling for moderators such as publication time, location, sample area, and temperature, studies propose that this variability may be attributed to precipitation and vegetation productivity in the study regions, emphasizing the combined influence of multiple environmental factors on bird communities. Furthermore, this study proposes further investigation into the long-term effects of dynamic landscape changes on birds, as well as the potential utilization of citizen science data and machine learning algorithms. The findings of our study offer compelling evidence to inform landscape planning and design decisions aimed at conserving bird communities.