Simulating The Potential Suitable Habitats Of Four Relict Plant Species In China Under Climate Change Based On Artificial Neural Network

Species suitability habitat assessment under climate change is of great significance for effective management and protection of species.The development of species distribution models(SDMs)provides more possibilities for habitat suitability modeling,while also brings more difficulties in model evaluation and more uncertainties.The ensemble model has been proved that can aggregate the advantages of single model and improve the modeling accuracy.However,what is combined is the results of model operation,not the model itself,so it is impossible to directly extrapolate to future/past for ensemble model consists of single models from different modeling platforms.It is necessary to run a single model repeatedly for the traditional ensemble model,increasing the time cost and the accumulation of single model uncertainty.Artificial neural networks have been proven to have the ability to learn nonlinear features of large-scale data sets,which can be used to reconstruct the mapping relationships between known variables.Four relict species of T.chinensis,D.involucrate,M.glyptostroboides,and P.amabilis were selected as the research objects,and 10 bioclimatic variables participated in modelling.Firstly,construct ensemble model of maximum entropy(MaxEnt)and random forest(RF)to obtain the species habitat suitability index(BTi)of the current period.Then,two neural networks,BP neural network(BPNN)and long and short-term memory(LSTM)network,were established to reproduce the functional relationship between current environmental variables and BTi.Finally,the network with good simulation performance was used to simulate the potential suitable habitats of four species in China for the past three periods(last interglacial period(LIG),last glacial maximum(LGM)and middle Holocene(MH)),and two climate scenarios(SSP2-4.5 and SSP5-8.5)of two periods(2050s and 2070s)in the future,and the distribution of four relict species in China under climate change was analyzed,the glacial refuges were intoxicated,and the protected areas was determined.At the same time,for comparison,we also forecast past/future climate scenarios based on MaxEnt model using current species occurrence data.There are 4 main conclusions:(1)The simulation performance of BPNN was much better than that of LSTM.Therefore,BPNN was used to simulate the potential habitats of four species in the past/future,and the predictions were better than those based on the ensemble model.(2)The suitable habitats of the four species tended to mitigate to higher latitudes or altitudes under climate change.The total suitable habitats of T.chinensis increased and tended to northern migration.The suitable habitats of D.involucrata in China only distribute around the Sichuan Basin with a small area.And continued to decline from the LGM period,the survival of D.involucrata would face a great threat.The suitable habitats of M.glyptostroboides in China would be generally expanding and move northward across latitudes in future.The suitable habitats of P.styracifolium would downward from the current period to future,significantly disappeared in southern China and begin to look for new suitable habitats in northern China.(3)The key environmental variables affecting the growth and survival of T.chinensis were Bio6(the coldest month minimum temperature),Bio 17(the driest season precipitation),and Bio9(the driest season average temperature);D.involucrata were Bio 14(the driest month precipitation),Bio18(the hottest season precipitation),and Bio6;M.glyptostroboides were Bio14,Bio6,and Bio18;and P.styracifolium were Bio14,Bio6,and Bio 17.The habitats characteristics of the four species are warm and humid,but the suitable temperature and precipitation range are different.The lack of precipitation during the LIG period restricted the distribution of species in most areas of China.The temperature decreased and it was extremely cold during the LGM period,but the four species expanded owing to their better tolerance and suitability to low temperature.From MH to 2050s,the temperature continued to rise sharply,T.chinensis and M.glyptostroboides migrated northward,D.involucrata and P.styracifolium reduced to high altitude based on the current habitats.(4)The glacial refuges of the four species locate in southern China.However,only a small part of the habitats located in the mountain areas whose suitability kept stable or increased.The priority protected areas of the four species would locate in the south of China,while the key protected areas would locate in the northern newly emerging suitable habitats.Due to the climate warming,the habitats in the south would lose its advantages and the habitat suitability would decline,so priority should be given to the protection;while the advantage in the north would increase,priority should be given to the protection,and protected areas should be planned in advance to assist species migration.Combined with the artificial neural networks,the functional relationship between known variables can be reconstructed,separating ensemble model from the single models of different modeling platforms,so that ensemble model can be directly extrapolated without running single models many times.This method saved time and costs,and greatly reduced the accumulation of uncertainty in single models.It improved the accuracy of species habitat suitability assessment and is of great significance for effective conservation and management of species.