Minimum temperature and precipitation determine fish 1 richness pattern in China ’ s nature reserves 2

Abstract. Understanding the drivers of geographic variation in species richness is one of the fundamental goals in ecology and biogeography. Fish is the key element in freshwater ecosystem and the focus of fishery production and biological conservation. Chinese freshwater fish fauna is rich and largely endemic due to variable geography and climate. By compiling the published data on fish richness for 86 nature reserves, and taking environmental predictors into consideration, we aimed to test latitudinal and longitudinal gradients in fish richness and the relative roles of energy availability, physiological tolerance, climatic seasonality and habitat heterogeneity hypotheses in explaining geographic fish richness pattern. Fish richness in China's nature reserves decreases with latitude and showed a hump-shaped relationship with longitude. Latitudinal fish richness is mainly shaped by mean temperature of the coldest month. Mean elevation and associated changes in temperature lead to longitudinal fish richness gradient. Among the four hypotheses tested, physiological tolerance hypothesis performs best and accounts for 55.4 % of the spatial variance in fish richness. Minimum temperature and precipitation are the primary determinants of fish species richness. Habitat heterogeneity is not negligible since adding river density to physiological tolerance model can explain additional 2 % variance in fish richness. Our results can provide useful information for regional fish production and conservation.


Introduction
Species diversity of most taxa at different spatial scales increases from polar towards the tropics (Heino, 2010;Kinlock et al., 2017).An important task in macroecology and biogeography is to explain the geographic pattern of biodiversity along latitudinal gradient (Brown, 2014).Biological responses to the abiotic variables, especially those relate to solar radiation, are regarded as one of the key determinants of latitudinal variation in species richness (Brown, 2014).Many hypotheses have been proposed to account for the large-scale spatial patterns of species richness , among which, energy availability (Clarke & Gaston, 2006;Kreft & Jetz, 2007), physiological tolerance (Wang et al., 2011;Griffiths et al., 2014), climatic seasonality (Tello & Stevens, 2010;Dalby et al., 2014) and habitat heterogeneity (Stein and Kreft, 2015) have been frequently used.
According to energy availability hypothesis, species richness increases with energy, which can be grouped into kinetic energy and potential energy from the perspectives of physics (Evans et al., 2005;Evans et al., 2008).Physiological tolerance hypothesis states that species richness is primarily explained by extreme climate (especially low temperature), because of the limited ability for species to tolerate freezing and drought stress (Currie et al., 2004;Griffiths et al., 2014).The minimum of seasonal temperature and precipitation were usually used to represent freezing and drought stress respectively (Wang et al., 2011;Griffiths et al., 2014).Climatic seasonality hypothesis emphasizes the intra-annual variability in climate variables (Tello & Stevens, 2010).Less fluctuations create more spatial or temporal niche opportunities, thus enhance regional species diversity by promoting co-existence (Chesson & Huntly, 1997;Tello & Stevens, 2010).However, whether seasonality has important effect on the geographic species richness patterns is still extensively controversial (Chen et al., 2014;Dalby et al., 2014).Habitat heterogeneity hypothesis proposes that highly heterogeneous habitats can promote species diversity through the following three major mechanisms.First, environmentally heterogeneous areas increase the available niche space and thus allow more species to coexist (Tews et al., 2004;Stein and Kreft, 2015).Second, environmentally heterogeneous areas are more likely to provide shelters and refuges under adverse environmental conditions .This should subsequently promote species persistence and regional biodiversity in periods of climate change, (Fjeldså et al., 2012;Stein and Kreft, 2015).Third, the probability of speciation from isolation or adaptation to diverse environmental conditions should increase with higher habitat heterogeneity, thus enlarge species pool and then regional species richness (Hughes & Eastwood, 2006;Stein and Kreft, 2015).
During the past decade, more efforts have been made to study species diversity patterns in terrestrial ecosystems than in aquatic ecosystems (Heino, 2011;Beck et al., 2012).Although freshwater covers only 1% surface of the Earth, aquatic animals account for 12% of the total animals in the world (Johnson et al., 2001).In recent years, aquatic biodiversity is declining sharply due to human disturbances (Vörösmarty et al., 2010), and freshwater fishery yield has been recently proved to be highly correlated with fish species richness (Brooks et al., 2016).Therefore, it is important to understand the environmental determinants of the current fish diversity patterns for future production and conservation (Olden et al., 2010;Knouft and Ficklin, 2017).
Most of the previous studies on fish species richness focused on lakes and river basins, rather than protected areas.This may prevent the utilization of the results in conservation planning at fine scales.
Chinese freshwater fish fauna is rich and largely endemic due to variable geography and climate (Xing et al., 2016).Recent fish inventory in China's nature reserves by ichthyologists produced reliable and ample data on regional species composition.In this study, by compiling the published literatures for 86 Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.nature reserves, and taking environmental predictors into consideration, we assessed the influences of environmental variables on geographic fish richness pattern.The main aims were to address these two scientific questions: i) is there any obvious geographic pattern of fish species richness along latitudinal or longitudinal gradient?ii) which hypothesis (energy availability, physiological tolerance, climat ic seasonality and habitat heterogeneity) plays the dominant role in determining geographic fish richness pattern?
2 Materials and methods

Species data
We collected literatures (monographs, dissertations, journal articles, and investigation reports) on fish diversity in 103 nature reserves.In case of two publications in the same area, we chose the latest one.
Nature reserves without data of coordinates, topography, or complete fish species list were ruled out of our investigation.To keep consistency in habitat type, we also excluded nature reserves entirely for wetlands and lakes.To make the data comparable, we checked the methods used to collect fish specimens in each nature reserve.Finally, 86 nature reserves were retained, covering 22 provinces, autonomous regions and municipalities throughout China (Fig. 1; Table S1), with total areas of 174,589 km 2 , and average area of 2030.1 ± 1029.8 km 2 .Fish richness (FR) from the original sources was response variable in the following analyses.

Environmental data
To compare the power of the four hypotheses (energy availability, physiological tolerance, climat ic seasonality and habitat heterogeneity) in explaining regional fish richness pattern, we selected 14 Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.environmental variables.For energy availability hypothesis, we chose mean annual temperature (TEM), growing degree days (GDD), annual potential evapotranspiration (PET), annual precipitation (PRE), net primary productivity (NPP), and annual actual evapotranspiration (AET).For physiological tolerance hypothesis, we chose mean temperature of the coldest month (TEMmin), precipitation of driest month (PREmin), and moisture index (MI, i.e.AET/PET).For climatic seasonality hypothesis, we chose standard deviation of mean monthly temperature (TS) and the coefficient of variation of mean monthly precipitation (PS).For habitat heterogeneity hypothesis, we chose elevational range (ELEran) and river density (RIVden) within nature reserves.To account for sampling effect, nature reserve area (AREA ) was included in each hypothesis testing.Most variables have been frequently used to explain fish richness patterns at large scales (Oberdorff et al., 1995;Gué gan et al., 1998;Zhao et al., 2006;Knouft & Page, 2011;Griffiths et al., 2014).
We extracted mean values of the environmental variables according to longitudinal and latitudinal extent.Data for NPP and GDD were obtained from Center for Sustainability and the Global Environmen t (http://nelson.wisc.edu/sage/).Data for AET and PET were from the Global Evapotrans piration and Water Balance Data Sets (Ahn and Tateishi, 1994).TEM, PRE, TS, PS, TEMmin, and PREmin were obtained from WorldClim (http://www.worldclim.org)(Hijmans et al., 2005).AREA and ELEran (maximu m elevation-minimu m elevation) were collected from the original publications.RIVden refers to the length of the river per unit area, and was derived from land use data for 2005 (http://ngcc.sbsm.gov.cn/).Mean elevation for each nature reserve was calculated by dividing the sum of maximu m and minimum elevations.

Statistical analysis
First, we tested the normality of all data, and ln-transformed FR, AREA, ELEran, and RIVden with high skewness.Then, simple linear regression analysis was used to examine the relationships between FR and each environmental predictor (Table S2).Adjusted r 2 was used to estimate the explained variance.
Statistical significance of regressions was calculated by Dutilleul's modified t-test based on corrected degrees of freedom (Dutilleul, 1993).
For each hypothesis, we used Akaike Information Criterion (AIC) to select the best model due to the significant correlations between some environmental predictors (Table S3).To account for the influence of sampling effect, AREA was used as fixed variables in model selection.The selected environmental predictors in the best model and AREA were used to run multiple regressions.Interactive and quadratic terms were also included in model selection to assess their potential effect.After testing each hypothesis, we develop an environmental model, by repeating model selection on all the selected predictors for each hypothesis.Moran's I at different distances were used to test whether FR and residual FR of different models show evident spatial autocorrelation.All the data analyses were performed in Spatial Analysis in Macroecology software (Rangel et al., 2010).

Results
Fish richness (FR) in the 86 nature reserves showed evident spatial pattern at regional scale (Fig. 1; Fig. S1a).FR averaged 32, ranging from 2 to 98.The hotspots with high FR mainly centered on southeastern and southwestern China, while areas in north China and Qinghai-Tibet plateau harbored lower FR.FR decreased significantly with latitude (Fig. 2a), and showed a hump-shaped relationship with longitude (Fig. 2b).Taking latitudinal bands of 25-30 o N, and 30-35 o N as analytical units, the stronger the Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.longitudinal trend in elevation, the stronger longitudinal gradient in fish richness.After regressing fish richness against mean elevation, no relationships between longitude and residual fish richness were found (Fig. 3).
The results of simple linear regression showed that, most environmental variables explained considerable spatial variance in FR.TEMmin had the highest explanatory power (49.9%), followed by TEM (48.7%) and PREmin (46.1%) (Table S2).
Summary of the best regression model for each hypothesis showed that physiological tolerance model (AREA, TEMmin, PREmin and PREmin 2 ) had the strongest power to explain the spatial variance in FR (55.4%).The second one is energy availability model (TEM, NPP, AREA, 53.

Latitudinal and longitudinal gradients in fish richness
We found negative relationship between fish richness and latitude based on data from 86 nature reserves in China.This accords with the general law of biodiversity along latitudinal gradient (Kinlock et al., 2017) and resembles the previous research conclusions on fish richness in lakes (Barbour and Brown , Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.
1974; Amarasinghe & Welcomme, 2002), rivers (Oberdorff et al., 1995;Hof et al., 2008;Feld et al., 2016), and reservoirs (Bailly et al., 2016) at global or continental scale.An analysis on fish richness from 292 rivers at global scale showed a negative relationship between fish richness and mean latitude (Oberdoff et al., 1995).Similarly, the analysis on fish richness in 25 European biogeographical regions made the same conclusion in groundwater and lotic (running waters from springs to large streams) habitats (Hof et al., 2008).Taking a circular area with a 500-km diameter as the analytical spatial unit, Knouft and Page (2011) found negative correlations with latitude for total fish richness but not for Cottidae and Salmonidae.However, some researchers found no latitudinal trend in fish richness because of large variance in river basin area (e.g.Oberdoff et al., 1997).In this study, we used the data on fish richness from nature reserves with similar area, which may diminish the sampling effect.In fact, nature reserve area has no significant effect on fish richness and the latitudinal trend is still significant even when the effect of area has been removed.
In Eastern China, both strength and magnitude of the latitudinal gradient of fish richness are less than those of butterfly richness (Chen et al., 2014).This can be reflected by the higher explanatory power (0.443 vs 0.321) and lower standardized regression coefficient (-0.665 vs -0.574) for butterfly richness than for fish richness.This may be partly explained by their different responses to low temperature at high latitudes.Fishes from Salmonidae and Cottidae prefer cold-water systems and are less diverse at lower latitudes than at higher latitudes (Knouft and Page, 2011), but very few clades of butterflies are more diverse at high latitudes (Chen et al., 2014).
Many previous studies examined latitudinal diversity gradient, while much less explicit along the longitudinal gradient (Araújo et al., 2009;Chen et al., 2015).We found a positive relationship between fish richness and longitude in China's nature reserves.This is consistent with the finding of Ibañez et al. (2009) that across major landmasses, freshwater fish richness increases along longitudinal fluvial gradients.We argue that this longitudinal trend in fish richness is caused by t he obvious negative relationship between mean elevation and longitude.That means, the mean elevation increases from east to west, and this tendency provokes changes of temperature and precipitation.Zhao (2006) also found that elevation could account for 74.5% of the species richness in China's lakes.Likewise, fish richness in the Yangtze River basin decrease with elevation (Fu et al., 2004).In eastern coastal China, the oceanic climate provides a favorable environment for fishes (Kang et al., 2014).Western China is dominated by continental or alpine-cold climate, which prevents fish migration and survival (Fu et al., 2004;Kang et al., 2014).

The importance of physiological tolerance hypothesis in shaping fish richness pattern
Among the four candidate hypotheses, physiological tolerance hypothesis performed best and accounted for 55.4% of the spatial variance in fish richness.The environment model contained all the terms in physiological tolerance model and explained additional 2% of variance.For other faunas, water perhaps plays a relatively weak role, while our results obviously showed that water availability is vital for fish live in streams and rivers of mountainous nature reserves .Most previous studies focusing on lakes would not consider water availability and thus underestimate the effect of water (e.g.Amarasinghe & Welcomme, 2002;Zhao et al., 2006).Other than focusing on river basins or lakes, we selected mountain streams or rivers with small basins as analytical units, where precipitation exerts an explicitly important impact on run-off volume.Small run-off induces lower oxygen content in water, and thus may result in fish's death due to lack of oxygen.For mountain streams, a great amount of seasonal precipitation in the driest month may increase water turbidity (Poff and Ward, 1989).Under this condition, gills of most Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.freshwater fishes would be attached with suspended particulates caused by turbidity.This directly affects respiration efficiency and increases pathogen infection risk (Robinson and Tonn, 1989).This interpreted well the quadratic relationships between fish richness and precipitation of driest month.
The simple linear regression and model selection indicated that TEMmin played a vital role in shaping geographic pattern of fish richness.This is consistent with the results from previous studies on trees (Wang et al., 2011) and butterflies (Hawkins and DeVries, 2009).Temperature plays a vital role in metabolic level of fishes (especially stenothermal fishes).The general optimal range of temperature is 18-30 o C for tropical fishes and 6-17 o C for coldwater fishes (Ficke et al., 2007).For fishes, low water temperature would decrease feeding, which in turn results in diseases and even death.
Comparing with TEM, TEMmin is more tightly correlated with latitude, and can explain more spatial variance in fish richness, therefore latitudinal gradient in fish richness is more likely driven by TEMmin.This indicated that tropical niche conservatism hypothesis could be used to explain geographic pattern of fish richness (Wiens and Donoghue, 2004).By applying indices of functional diversity and niche overlap on the fish species in French lakes, Mason et al. (2008) argued that increased temperature may have permitted increased species richness by allowing increased niche specialization.

The potential biases
Although we made great effort to compile data on fish composition and environmental variables in China's nature reserves, there are still some potential biases.First, fish inventory may not be complete due to seasonal and inter-annual changes in fish occurrences and spatial biases in selecting sampling sites within nature reserves.Second, local extinction of fishes due to climate change and human disturbance is possible and this information is not included in the dataset (Knouft and Ficklin, 2017).Third, some Biogeosciences Discuss., https://doi.org/10.5194/bg-2017-389Manuscript under review for journal Biogeosciences Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License.environmental variables at the fine scale are hard to obtain, such as total and seasonal run-off volume, total length of streams or rivers, and flow regime in each nature reserve.These variables have been thought to influence regional fish richness by previous studies (Datry et al., 2014;Daniel et al., 2016;Knouft1 and Ficklin, 2017).
1%), followed by climatic seasonality model (TS, PS, AREA, 35.1%), and to a lesser extent habitat heterogeneity model (RIVden, RIVden×ELEran, AREA, 8.3%).Environmental model (TEM, RIVden, PREmin, PREmin 2 , AREA) explained 57.4% of spatial variance of FR.Moran's I correlograms of FR were positive at relatively short distance and were negative at longer distance.Significant spatial patterns were found in the residuals of energy availability model, climat ic seasonality model, and habitat heterogeneity model, especially at short distance.Physiological tolerance model and environment model left no spatial patterns in residuals (Fig S1).

Figure 1
Figure1The geographic pattern of fish richness in 86 nature reserves of China.

Table 1 .
Summary of the multiple regression models evaluating the effects of energy availability, climat ic 386 seasonality, physiological tolerance, habitat heterogeneity, and environment on fish richness.Nature 387 reserve area (AREA) was included in each model to account for its potential effect.