Biogeographic Distribution of Cedrela spp. Genus in Peru Using MaxEnt Modeling: A Conservation and Restoration Approach

The increasing demand for tropical timber from natural forests has reduced the population sizes of native species such as Cedrela spp. because of their high economic value. To prevent the decline of population sizes of the species, all Cedrela species have been incorporated into Appendix II of the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES). The study presents information about the modeled distribution of the genus Cedrela in Peru that aims to identify potential habitat distribution of the genus, its availability in areas protected by national service of protected areas, and highlighted some areas because of their conservation relevance and the potential need for restoration. We modeled the distribution of the genus Cedrela in Peru using 947 occurrence records that included 10 species (C. odorata, C. montana, C. fissilis, C. longipetiolulata, C. angustifolia, C. nebulosa, C. kuelapensis, C. saltensis, C. weberbaueri, and C. molinensis). We aim to identify areas environmentally suitable for the occurrence of Cedrela that are legally protected by the National Service of Protected Areas (PAs) and those that are ideal for research and restoration projects. We used various environmental variables (19 bioclimatic variables, 3 topographic factors, 9 edaphic factors, solar radiation, and relative humidity) and the maximum entropy model (MaxEnt) to predict the probability of occurrence. We observed that 6.7% (86,916.2 km2) of Peru presents a high distribution probability of occurrence of Cedrela, distributed in 17 departments, with 4.4% (10,171.03 km2) of the area protected by PAs mainly under the category of protection forests. Another 11.65% (21,345.16 km2) of distribution covers areas highly prone to degradation, distributed mainly in the departments Ucayali, Loreto, and Madre de Dios, and needs immediate attention for its protection and restoration. We believe that the study will contribute significantly to conserve Cedrela and other endangered species, as well as to promote the sustainable use and management of timber species as a whole.


Introduction
Forest covers have been reduced drastically in the Peruvian Amazon region over the last few decades as a result of agricultural expansion and livestock activities, deforestation, mining, and urban expansion [1,2]. In Peru, 2,433,314 ha of Amazonian forests have been lost during 2001-2019 [3]. Although the tropical Amazon forest covers about 60% of Peru [4], it has now been highly fragmented because of the forest harvesting activities. The need for more agricultural land also promoted heavy migratory agricultural practices, [5]

Dataset and Methodological Design
The methodological framework used in the present study has been described graphically in Figure 2. From the cartographic standardization through the rescaling in the raster calculator of Qgis 3.16, 33 variables at a spatial resolution of 250 m were derived as input for use in modeling with MaxEnt. The bioclimatic information under current conditions (average 1970-2000) with a spatial resolution of 30 s (~1 km) was obtained from Woldclim version 2.1 (https://www.worldclim.org/data/worldclim21.html; accessed on 5 January 2021) [37]. Topographic factors such as elevation, slope, slope, and ground orientation were obtained from the 90 m spatial resolution DEM generated by the Shuttle Radar Topography Mission (SRTM) [38], United States Geological Survey (USGS) (http://srtm.usgs.gov; accessed on 28 December 2020). The edaphic variables were collected from SoilGrids 0.5.3 (http://soilgrids.org; accessed on 15 January 2021) with a spatial resolution of 250 m.
Likewise, the geographic occurrence data of 10 target species of the genus Cedrela to be used in the MaxEnt model were obtained from GBIF's Global Biodiversity Information Service (https://www.gbif.org/; accessed on 1 February 2021) through "Species Explorer" plug-in of QGIS software. The registration information of CITES species was obtained from the Ministry of the Environment of Peru (https://geoservidor.minam.gob.pe/recursos/ intercambio-de-datos/; accessed on 18 February 2021). Finally, to identify the locations of Cedrela habitats within the protected areas, and the areas prone to degradation but having high suitability for genus Cedrela habitat, the modeled potential distribution result was overlaid with the degraded areas identified by the Ministry of Environment (MINAM) and the spaces conserved by the National Service of Natural Areas Protected by the Peruvian State (SERNANP). These degraded areas were identified by the ministry mainly based on deforestation, soil erosion, forest fires, mining, illegal logging, etc.

Geographical Records of Forest Species
The geographic coordinates of the 10 species of the genus Cedrela (Table 1) were obtained using the GBIF and Species Explorer plug-ins in QGIS 3.16 software. It was also complemented with the records of the presence of Cedrela, obtained from the Ministry of Environment of Peru. The CITES species information was collected from the systematization of forest inventories, review of national herbaria which is available in its geoservidor (https://geoservidor.minam.gob.pe/recursos/intercambio-de-datos/; accessed on 18 February 2021), and information related to the species of the genus Cedrela was separated. The data were then re-sampled at a spatial resolution of 250 m [39] by visually excluding those samples that were falling within lagoons, rivers, and roads, or urban areas. Finally, 947 resulting data were exported into CSV to be used for modeling in MaxEnt (https://biodiversityinformatics.amnh.org/open_source/maxent/; accessed on 10 November 2020).

Geographical Records of Forest Species
The geographic coordinates of the 10 species of the genus Cedrela (Table 1) were obtained using the GBIF and Species Explorer plug-ins in QGIS 3.16 software. It was also complemented with the records of the presence of Cedrela, obtained from the Ministry of Environment of Peru. The CITES species information was collected from the systematization of forest inventories, review of national herbaria which is available in its geoservidor (https://geoservidor.minam.gob.pe/recursos/intercambio-de-datos/; accessed on 18 February 2021), and information related to the species of the genus Cedrela was separated. The data were then re-sampled at a spatial resolution of 250 m [39] by visually excluding those samples that were falling within lagoons, rivers, and roads, or urban areas. Finally, 947 resulting data were exported into CSV to be used for modeling in MaxEnt (https://biodiversityinformatics.amnh.org/open_source/maxent/; accessed on 10 November 2020).

Bioclimatic, Physiographic, and Soil Variables
The spatial distribution of species within a geographic area dependson the interaction with several environmental factors that contribute to their development and coexistence [40]. Considering this, 33 variables were selected (Table 2) to carry out the modeling. These variables include 19 bioclimatic and solar radiation obtained from WorldClim 2.1 (https://www.worldclim.org/; accessed on 5 January 2021) [37]; 3 topographic derived from digital elevation model (DEM) obtained from the United States Geological Survey (USGS) web portal (http://srtm.usgs.gov; accessed on 28 December 2020); the relative humidity obtained from the Climate Research Unit (CRU) [41] (www.cru.uea.ac.uk; accessed on 1 May 2021); and 9 soil properties collected from SoilGrids 0.5.3 (http://soilgrids.org; accessed on 15 January 2021) [42]. All variables were rescaled into a spatial resolution of 250 m to overcome the issues such as collinearity between variables, which causes overfitting problems, increases uncertainty, and decreases the statistical power of the model [43]. Therefore, using the function "remove collinearity" from the package "virtual species" [44] in R 3.6, the variables were grouped (clustering) according to the Pearson correlation coefficient, and only variables having Pearson's r ≥ 0.7 were considered. This threshold is an acceptable measure to minimize the multicollinearity of fitted models [43]. To select an important variable for each cluster, a preliminary MaxEnt model was run (the configuration is explained in Section 3.2.) using all the variables. The variable with the best performance in the Jackknife test [25] was selected (i.e., the smallest difference in regularized training gains obtained from a model generated with all criteria except that of interest and a model generated only with the criterion of interest [21] (Table 2).

Execution of the Model
The biogeographic distribution model for the 10 species of the genus Cedrela was performed using a maximum entropy algorithm [31] which estimates the probability of potential distribution of each species from the presence data (location) using the open-source software MaxEnt ver. 3.4.1 (https://biodiversityinformatics.amnh.org/open_source/maxent/; access on 10 November 2020). For the validation of this model, 75% of the randomly selected presence data were used for training purposes, and 25% were used for validation [31]. The algorithm was run using 100 repetitions in 5000 iterations with different random partitions (Bootstrap method), and other configurations (i.e., extrapolation, graph drawing, etc.) were kept as default [45].

Identification of Potential Areas for Restoration and Conservation
Subsequently, the areas of high distribution potential were overlapped with the Protected Natural Areas (PNA) information obtained from GeoServer (https://geo.sernPNA. gob.pe/visorsernPNA/; access on 18 February 2021) of the National Service of Natural Areas, which is protected by the State (SERNPNA) to promote conservation of the genus Cedrela, currently considered as endangered and overexploited in Peru.
Similarly, the raster layer (30 m resolution) of degraded areas as identified by the Ministry of the Environment of Peru (MINAM) in 2019 was also obtained from its geoservidor, (https://geoservidor.minam.gob.pe/recursos/intercambio-de-datos/; access on 18 February 2021) and overlapped with the potential distribution of Cedrela. Finally, the distribution surfaces of the 10 species within the PAs and degraded areas were quantified. This way, the analysis had made it possible to identify the protected areas that conserve the genus Cedrela and those degraded spaces that could be restored with one or more of the species under study.

Model Performance and the Importance of Environmental Variables
Model performance evaluation aims to estimate the accuracy of machine learningbased prediction models and ensures confidence in the results obtained. The performance of this model obtained an area under the curve (AUC) value of 0.866 (Figure 3a), which is considered good (0.8 < AUC < 0.9). The response curves (Figure 3b-n) reflect the dependence of predicted suitability, both on the selected variable and on dependencies induced by correlations between the selected variable and other variables. Overall, 83% of the potential distribution of Cedrela was found to be driven mainly by four environmental variables, i.e., bio19 (precipitation of coldest quarter), soc (organic carbon), dem (elevation above mean sea level), and cec (cation exchange capacity) ( Table 3). On the other hand, silt (slime content), bdod (bulk density of the fine earth fraction), and nitrogen were the three environmental variables that contributed the least. Figure 3o shows the results of jackknife test of variable importance. The environmental variable that reported the highest gain when used in isolation was bio19, which therefore appeared to have the most useful information by itself. The environmental variable that decreased the gain the most on its omission was dem, which therefore appeared to have the most information that was not present in the other variables. Likewise, the Jackknife test (Figure 3o) identified that the variables bio 19 (coldest quarter precipitation), bio 12 (annual precipitation), soil pH, and elevation (DEM) contributed highly to the biogeographic distribution model of the Cedrela species.

Potential Distribution of the Genus Cedrela
The areas of a high probability of occurrence of genus Cedrela under present climatic and environmental conditions were identified mainly across the lowland Amazonia, covering 86,916.2 km 2 (6.7%) area of the study region. This potential habitat distribution covers about 17 regions of the Peruvian territory ( Figure 4)

High-Priority Areas for Research, Conservation, and Restoration
The study identified that 4.4% (10,171.03 km 2 ) of the areas of the high-occurrence probability of genus Cedrela was distributed in the designated Peruvian conservation areas (Figure 5a), out of which the PNA cover of 35.5% (8995.64 km 2 ) was distributed among the reserved zones (85.18 km 2 ), national sanctuary (130.76 km 2 ), historic sanctuary (20.18 km 2 ), wildlife refuge (0.13 km 2 ), national reserve (2323.46 km 2 ), communal reserve (1023.63 km 2 ), national park (5000.93) and protection forest (411.37 km 2 ). The distribution also included conservation areas administered by regional governments (1020.71 km 2 ), and by individuals or institutions at a private level through private conservation areas, whose high-occurrence potential covered a total of 154.68 km 2 (Table 5) area of the study region.

Potential Distribution of Genus Cedrela
Our study is the first attempt that makes use of SDMs as a probabilistic decision-making tool [49], which allows the prediction and identification of geographic spaces of the genus Cedrela [50] through maximum entropy modeling technique [51]. The proposed model can be applied at regional [24,25] to national scale [52][53][54] that will sig- After compiling the potential habitat distribution results with the information on degraded areas, a high-distribution potentiality of genus Cedrela was observed over 20,857.0 km 2 areas of the central and western parts of Peru (accounts for 11.4% of the study area) that are highly prone to degradation (Table 6). In other words, with proper conservation and management practices in these areas, 11.4% of degraded Peruvian Amazon can be potentially restored.

Potential Distribution of Genus Cedrela
Our study is the first attempt that makes use of SDMs as a probabilistic decisionmaking tool [49], which allows the prediction and identification of geographic spaces of the genus Cedrela [50] through maximum entropy modeling technique [51]. The proposed model can be applied at regional [24,25] to national scale [52][53][54] that will significantly contribute to the decision-making system for the Peruvian Amazon authorities. Our model is evident with higher accuracies represented by the strong AUC values of 0.866. Among different topographic and bioclimatic variables, altitude emerges as the most significant variable, which proved to be a determining factor in distribution ranges [24,25]. Species such as C. montana and C. lilloi, are mostly located and distributed at the higher altitudes. However, the distribution of species depends on the biogeographic conditions and also has a strong influence on historical or evolutionary constraints along with biogeographical, physiological, and ecological factors [55]. In this study, we observed that the 10 species of the genus Cedrela covered 17 departments related to the National Forestry and Wildlife Service, as of 2021 [16], and evaluated the location of botanical collections and inventories of the species [13,53,54]. Overall, the modeled distribution of genus Cedrela will also help to understand the historic evaluation of genus Cedrela species under a spatiotemporal framework. Therefore, we believe that our modeling framework will help in the future in order to establish forest management strategies.

Conservation and Restoration of Genus Cedrela
Peru is one of the most megadiverse countries in the world and is enriched with the biogeographic distribution of various species that requires the implementation of adequate strategies for species conservation [56,57]. Among the 10 species of genus Cedrela, C. odorata is currently one of the important timber species, threatened by deforestation and unsustainable logging [58]. However, the PAs that harbor C. odorata, together with the other species of the Cedrela genus (10,171.03 km 2 ), will allow the implementation of mechanisms to maintain its population and genetic diversity, given that the PAs constitute territorial protection reserves [59,60]. Similarly, the degraded areas are the result of anthropogenic pressure and forest fires in 2019 in Peru, occupying an area of 183,288.15 km 2 [35]. Among the degraded region, 11.4% of the area is currently having a high probability of recovery through the plantation of species of high economic value such as the Cedrela genus. The Cedrela genus needs to be protected from selective logging and overexploitation over time [9,12,18,61]. This is possible through the implementation of sustainable forest management strategies [62], strengthening forest monitoring and surveillance actions [48,63], and forming strategic alliances for conservation to protect these vulnerable species [56].
This study modeled the potential distribution of the genus Cedrela in Peru under current climatic conditions and identified which part of this potential distribution is protected by conservation areas or coincides with degraded areas. However, future studies could evaluate the distribution in future conditions of climate change, similar to Rojas et al. [25], who studied five timber forest species in Amazonas (northeastern Peru). However, it should be noted that species distribution models in climate change scenarios should be interpreted with caution since they may overestimate the decline or increase, by not considering the qualities of the species to adapt in situ to new conditions or persist outside the conditions in which they have been observed [64,65]. Despite the above, the relatively stable distribution sites (same current and future potential) of species would be of interest and essential to ensure the success of any conservation or restoration initiative.

Conclusions
The current biogeographic distribution of the 10 genus Cedrela (C. odorata, C. montana, C. fissilis, C. longipetiolulata, C. angustifolia, C. nebulosa, C. kuelapensis, C. saltensis, C. weberbaueri and C. molinensis) using MaxEnt, covers around 6.7% of Peru, found in 17 departments under Ucayali, Loreto, and Madre de Dios that are more likely to be located in the Amazon region. Likewise, the Natural Protected Areas categorized by the Peruvian State play a fundamental role in allowing the conservation of 4.4% of the potentially high-distribution regions of its territory. Such regions have high potentiality forthe genus Cedrela plantations, and such plantations could possibly be protected through appropriate conservation strategies.
Our research has also allowed us to quantify that 11.4% of the degraded areas identified in Peru as of 2019, can possibly be recovered through the plantation of one or more species of Cedrela genus. Therefore, our study has a strong potentiality to serve as a tool for identifying geographic spaces of genus Cedrela under a spatiotemporal framework in order to conserve or recover its local populations in areas degraded by anthropic or natural factors.