Keywords

1 Introduction

Freshwater species play an important role in the respective ecosystem in association with the surrounding environment. Fishes are an important source of animal protein in the world and are widely recognized as a good source for maintaining body health (Ganaie and Sharma 2021). The quantification of these relationships portrays the core of predictive geographical modeling in ecology to solve the crucial problem of how various environmental factors are controlling the distribution of species and communities (Guisan and Zimmermann 2000). The diversity of freshwater species is declining swiftly worldwide, and the population of freshwater wildlife has declined by 83% since 1970 (WWF 2018; Acreman et al. 2019). The huge concentration of population and urban centers around the freshwater ecosystems has threatened the freshwater species diversity including habitat loss and degradation, water pollution, overfishing, etc. (Arthington et al. 2016).

Species distribution models (SDMs) connect the species’ occurrences with the prevailing environmental conditions and extrapolate the suitable habitat of a species (Domisch et al. 2015a). In other words, it estimates the distribution of a species within the geographic space and time with the help of environmental information as well as occurrence records (Mahato et al. 2022). Generally, SDMs can be characterized as a methodology that is based on core ecological and biogeographical principles about the association of species distributions with the physical environment (Elith and Franklin 2017). SDMs are frequently named correlative or statistical models, habitat models, or ecological niche models and are broadly separated into two categories, namely, correlative and process-based or mechanistic models (Srivastava et al. 2019). The outcomes of SDMs are also applied as input data for other investigations, especially in the field of conservation planning, impact assessment, and land use planning to identify suitable locations for habitat restoration and reintroduction of species (Elith and Franklin 2017).

Tor tor, commonly known as mahseer or Tor Barb, is the type of species of the genus Tor (Pinder et al. 2019). It was first described from the Mahananda, a tributary of the Ganges flowing through Northeast Bengal, India, by Hamilton (1822). It is a well-known game and food fish which inhabits the streams and rivers of the mountainous area and fast-flowing rivers of the plains and mostly prefers clear, fast-flowing waters having a stony, pebbly, or rocky surface (Shrestha 1997). The species has been described to reach 150 cm Total Length (TL) (Mishra 1959) and gain a maximum weight of 68 kg (Talwar and Jhingran 1991). T. tor is a native species of south Himalayan rivers extending from Pakistan to Myanmar and also toward the rivers of South India (Rayamajhi et al. 2018). The southernmost native distribution of T. tor was believed to be the Narmada River in Madhya Pradesh (Desai 2003). However, it was also discovered from the Godavari and Krishna River basins (Lal et al. 2013), which raises the question of whether it is native to tropical Peninsular India or expanded through artificial propagation (Pinder et al. 2019). T. tor has been earlier assessed as “Near Threatened” in the IUCN Red List due to swiftly decreasing populations, but recently it has been reassessed as “Data Deficient” (Rayamajhi et al. 2018). The present work is an attempt to predict the suitable habitats of T. tor in the Indian river systems by combining species occurrences, environmental variables, and geospatial technology.

2 Materials and Methods

2.1 Study Area

T. tor is native to the rivers of the Himalayas, Central India, Northeast India, and northern Peninsular India. Hence, the study area includes the rivers of the whole country. India is the seventh largest country in the world and the third largest in Asia. It is located between 8° 4′ N and 37° 6′ N latitudes to 68° 7′ E and 97° 25′ E longitudes. It shares an international border with Afghanistan and Pakistan in the northwest; China, Bhutan, and Nepal in the north; and Myanmar and Bangladesh in the east. The southern part of the country is also bounded by the Bay of Bengal to the east and the Arabian Sea to the west (Fig. 17.1). The biogeographic regions of India have been also considered to comprehend the habitat suitability of T. tor (Fig. 17.2).

Fig. 17.1
A map of India exhibits the study area region, The international boundary, state boundary, and the coastal line are marked.

Map of the study area

Full size image
Fig. 17.2
A color-coded map of India marks the biogeographic regions and includes trans-Himalayan region, himalayan zone, Indian desert zone, semi-arid region, Western Ghats, Deccan Plateau, Gangetic Plain, North East region, Coastal region, and Andaman and Nicobar Islands.

Biogeographic regions of India

Full size image

2.2 Occurrence Records

The occurrence records of T. tor have been collected from both primary and secondary sources. The majority of the occurrence records were downloaded from the Global Biodiversity Information Facility (GBIF) using the dismo package of R (version 4.0.3) following (Hijmans et al. 2011). The primary data was collected from different rivers of Arunachal Pradesh using a handheld Global Positioning System (GPS). All the occurrence records have been filtered to remove records with no coordinate information. Initially, a total of 301 occurrence records were generated. We used a snapping tolerance of 3 km to move the occurrence points to the closest freshwater pixel, and the points falling beyond this range have been removed. Further, the spThin package in R was used to remove the biases from the records. The algorithm retained only one record within a 10-km spatial grid and returned a dataset of 95 occurrence records for the final model execution.

2.3 Selection of the Best Environmental Variables

Initially, a set of 51 environmental parameters (1-km spatial resolution) have been downloaded from the EarthEnv project (www.earthenv.org) using R (Table 17.1). The sources of the data include WorldClim, Consensus Land Cover, HydroSHEDS, and World Soil Information (ISRIC) as explained by Domisch et al. (2015b). Various statistical techniques face multicollinearity problem due to the presence of highly correlated information among the variables (Miller 2010), which negatively affects the model performance and poses difficulties in interpreting the relative performance of variables in the model predictions (Dormann et al. 2013; Manzoor et al. 2018). The complex features created by MaxEnt are often highly correlated. So, it is recommended to minimize such highly correlated environmental variables (Merow et al. 2013). A threshold of |r| > 0.7 correlation coefficients between predictor variables is a proper indicator of collinearity (Dormann et al. 2013; Manzoor et al. 2018; Sony et al. 2018; Farrell et al. 2019; Feng et al. 2019). Similarly, the variance inflation factor (VIF) must be computed to check the effects of the multicollinearity of the variables. It is suspiciously high when VIF for an independent variable is greater than 5 or 10 (Tsagris and Pandis 2021). Initially, 17 (Table 17.2) out of 51 variables were selected based on the correlation coefficients of |r| < 0.7 and VIF less than 5 using usdm package in R (Naimi, 2016). Out of these, ten variables with more than 1% contribution were selected (Table 17.3). The final model was performed with three replications to predict the suitable habitat of T. tor in the Indian river systems.

Table 17.1 Environmental data used for species distribution modeling of Tor tor
Full size table
Table 17.2 Variance inflation factor and contribution of selected variables
Full size table
Table 17.3 Contribution of selected variables
Full size table

2.4 Model Setup and Evaluation

MaxEnt (version 3.4.4) was used to predict the habitat suitability of T. tor. MaxEnt is a machine learning algorithm that performs best with high prediction accuracy when used in small sample sizes and presence-only datasets (Ji et al. 2020; Li et al. 2022). It possesses several advantages over other methods such as good performance with incomplete datasets, short running time, simple operation, needs small sample size, and high simulation precision (Li et al. 2020). The settings of the model include logistic output, cross-validate replicate run type, response curves, jackknife measures of variable importance, ten percentile training presence, and rest kept as default settings (Mahato et al. 2022). The results ranging between 0 and 1 have been classified into five classes of suitability, viz., <0.1 unsuitable; 0.1–0.3, slightly suitable; 0.3–0.5, moderately suitable; 0.5–0.7, suitable; and >0.7, highly suitable. The stepwise methodology followed in this study is shown in Fig. 17.3. The threshold-independent, area under the curve (AUC) of the receiver operating characteristic (ROC) curve and true skill statistic (TSS) were used to evaluate the overall performance of the model.

Fig. 17.3
An 8-tier flowchart of the methodology starts with data collection and branches out into occurrences of Tor tor and environmental variables. After a series of steps, both lead to Maxent modelling, model validation and evaluation, and habitat suitability of Tor tor.

Flowchart of the methodology

Full size image

3 Results

3.1 Model Performance

The model results show the distribution of T. tor is mostly influenced by Hydroclim_14 with 31.2% followed by Lc_avg_01 (26.9%) and Lc_avg_03 (14.7). Elevation, Lc_avg_02, and Hydroclim_03 were other important predictors with over 5% contributions. The rest of the variables such as Lc_avg_04, Soil_avg_07, and slope also slightly influenced the distribution of T. tor (Table 17.2). The performance of the model was reasonably consistent and good with a mean AUC of 0.860 and a TSS value of 0.711.

3.2 Habitat Suitability Modeling of Tor tor

The final results of habitat suitability were considered in terms of the total river length (km). Out of the total river length of 502,255.03 km, the model showed 9214.15 km (1.83%) as highly suitable and 17,155.93 km (3.42%) as suitable followed by 37,352.21 km (7.44%) as moderately suitable, and 159,173.13 km (31.69%) as slightly suitable. More than half of the study area, i.e., 279,359.61 km (55.62%), falls under the unsuitable category (Table 17.4 and Fig. 17.4). The results were also separately calculated for each biogeographic region of India. Out of the ten biogeographic regions, the highly suitable habitat has been predicted in only five regions. Among these regions, Northeast India showed the highest percentage (9.44%), followed by the Himalayan region (8.16%), the Gangetic plain (5.76%), the semi-arid region (0.57%), and the Deccan Plateau (0.15%) (Table 17.4 and Fig. 17.4). The model predicted Andaman and Nicobar Islands (100%), Trans-Himalayan region (99.35%), coastal region (92.00%), and Indian Desert region (85.61%) as unsuitable for T. tor.

Table 17.4 Habitat suitable for Tor tor in the Indian river system
Full size table
Fig. 17.4
A color-coded India map of Indian drainage system with a fish at the top right labeled as Tor tor. The legend indicates habitat suitability and includes unsuitable, slightly suitable, moderately suitable, suitable, and highly suitable. Most of the regions reflect slightly suitable.

Habitat suitability mapping of Indian drainage system

Full size image

A close examination of the model outputs reveals the rivers of Northeast India as suitable for the targeted species (Fig. 17.4). The north and south bank tributaries of the mighty river Brahmaputra appear to be the best home of T. tor. The northern tributaries of the Brahmaputra such as Kameng, Dikrong, Subansiri, Siang, etc. have been predicted as suitable habitat for the species. Besides, a portion of the Barak River system also falls under suitable habitat. In the central and northern parts of the country, Ganga and its tributaries fall under the suitable category. Narmada, Tapi, and Godavari River systems have been also found suitable for T. tor. Some of the reservoirs, namely, Indra Sagar Reservoir of Madhya Pradesh, Gobind Sagar, and Maharana Pratap Sagar of Himachal Pradesh, also fall under suitable habitat for the species (Table 17.5).

Table 17.5 Habitat suitability of Tor tor in different biogeographic regions (in %)
Full size table

4 Discussion

The distribution and occurrences of fish communities are highly influenced by the surrounding environment. The habitat of fish is largely influenced by various factors such as the meandering of streams, the gradient of river banks, riparian vegetation, and stream flow dynamics (Gebrekiros 2016). Based on the quality of water and the associated environmental factors especially rainfall and temperature, the species can travel to large rivers for breeding and feeding (Kaushik and Bordoloi 2016). The model finds a positive association of T. tor with precipitation of the driest month, land cover classes (evergreen/deciduous needleleaf trees, deciduous broadleaf trees, and evergreen broadleaf trees), upstream minimum elevation, and isothermality.

T. tor is widely distributed in Bangladesh, Bhutan, India, Myanmar, Nepal, and Pakistan. In India, the species is distributed over Arunachal Pradesh, Assam, Bihar, Haryana, Manipur, Meghalaya, Nagaland, Punjab, Sikkim, Uttar Pradesh, and Uttaranchal (Khajuria and Langer 2016). The model predicted the drainage system of the Himalayan foothill zone, Central India, Northeast India, and southern India (Narmada, Godavari, and Tapi) as the suitable habitats of T. tor. The results are in agreement with the earlier studies (Lal et al. 2013; Pinder et al. 2019). The results also show the highly suitable habitats over the hilly and mountainous rivers, which may be attributed to better growth in the rivers with a rocky surface (Brraich and Saini 2019) and clear and swiftly flowing water along with optimum water temperature and pH level (Patil and Saxena 2021).

The inland drainage systems provide fewer chances of migration to better environmental conditions and are frequently confined within the landlocked water bodies (Roy et al. 2021). Human populations are always settled along water bodies including rivers; hence, the rivers are severely affected by human influences (Zeng et al. 2022). The fish habitats have been altered or degraded resulting in the declining trend of fish assemblage, extinction of species, and replacement by other species (Aadland 1993; Gebrekiros 2016). In general, anthropogenic influences alter the natural environment through changing land use/land cover, modifying flow regimes, constructing river dams, and polluting soil and water as well as climate change which indirectly influences the extinction of various native species and/or introduction of non-native species (Su et al. 2021). Most of the cities and large urban centers in India are located near the drainage network. In general, mahseer is decreasing very hastily in their population and size in Central India because of many reasons including overexploitation, water pollution, habitat destruction, domestic effluent, and use of insecticides and pesticides (Nautiyal and Dwivedi 2020), particularly the exploitation rate of T. tor (Dwivedi and Nautiyal 2012). T. tor has been reported to possess high nutrition and economic values (Dey et al. 2015). Therefore, there is an urgent need of evolving sound conservation strategies for the species in the country before it’s too late. The results of the present study could help formulate plans and policies for the conservation and management of T. tor.

5 Conclusion

The study predicted the habitat suitability of T. tor in the Indian river systems through the MaxEnt model. The suitable habitats have been mostly predicted in the North Eastern and the Himalayan region. Although the species is widely distributed in the country, the model results show limited occupancy of the species in certain concentrated pockets. Keeping in view the reported declining population of the species due to various anthropogenic activities, there is a need of exploring the potential habitats of the species. The model predictions, particularly the moderately suitable areas, should be explored and confirmed about the occurrence of the species. Moreover, the predicted suitable and highly suitable areas need to be prioritized for the conservation and restoration of the species in the future. However, our study is constrained by the vastness of the study area. Therefore, we recommend further research at biogeographic/regional levels based on the results of the present study.