Study area S. robustais distributed in Purbachal and neighboring areas of Bangladesh. Purbachal covers an area of 2489 ha that includes a large terrace area of Madhupur tracts developed in the Pleistocene Era in the central part of Bangladesh (Zaman, 2016). The annual mean air temperature of Purbachal is 28C, and the annual precipitation is 2400 mm (Shapla et al., 2015).The terrace comprises low, gentle-edged hills and ridges separated by shallow valleys and depressions that flood extensively during the rainy season. The lithological landforms comprise Madhupur clay deposits from the Pleistocene Epoch and more recent alluvial deposits (BBS, 2013), which are associated with valleys and depressions characterized by silty clay, silt, finesand, and graye light gray and dark gray soil. The Madhupur clay deposit comprises silty clay with fine sand, redereddish-brown, and yellowish-brown soil found mostly in the hills. Oxidized soil with the accumulation of nodules is a soil characteristic of the Madhupur clay deposit.The hilly areas of Purbachal include scattered homesteads (i.e., settlement and residential areas) and homestead vegetation (including trees, shrubs, and herbs on and around the settlement). At the bottom of the valleys and depressions, onecrop is cultivated annually (Shapla et al., 2015). Although the crop lands are developed, Purbachal is a sanctuary of natural ecosystems supporting ecologically important species and habitats (Mamun, 2007).2.2. Data sampling The localities of S. robustain the Purbachal were collected in 2016 and 2017 by field investigations. We recorded 280localities using a random-sampling method that included all inhabitants and isolated patches of S. robusta forests using the Global Positioning System (Garmin 64; Garmin Corporation, Taipei, Taiwan). S. robusta patches differed in size, ranging from3 ha to 29 ha.S. robusta localities prior to the urban growth were extracted from IKONOS satellite imagery at 04:35 (GMT) on May 1, 2001 and 4:44 (GMT) on February 16, 2002 and from World View-2 (WV2) at 04:41 (GMT) on December 9, 2015 (Digital Globe; Apollo Mapping, Longmont, CO, USA). These GPS and remote-sensing data were integrated via ArcMap (v.10.2; ESRI Redlands, CA, USA) for data processing. Analyses were performed after checking the quality of and pre-processing the data to remove noise and unify geo-references (Dewan and Yamaguchi, 2009). We collected 12 environmental variables pertaining to climate, physical, and soil conditions. The climatic data (i.e., precipitation and air temperature), maintaining equivalent climatic trends to the previous 20 years, were provided by the Bangladesh Meteorological Department (BMD, 2017). Elevation data with a spatial resolution of 9.99 m were obtained from Rajdhani Unnayan Kartripakkha (RAJUK) (RAJUK, 2016). Data (anthropogenic impacts) regarding the distance from road and settlement with 500 m altitude from the ground, were digitized and saved as Kmlfiles using Google Earth Pro software. The ArcMap software (v.10.2; ESRI) was used to prepare the Kml data into raster grids. Data concerning the pH and organic matter(OM), phosphorus (P), and potassium (K) contents in soil were obtained from the Soil Resources Development Institute (SRDI, 2015). Organic carbon (OC), calcium (Ca), and nitrogen (N) in soil were derived from the Bangladesh Country Almanac (BCA, 2009), the national database of Bangladesh. These data were converted into ASCII format in raster grids with the same geographic boundary according to the World Geodetic System 1984 longitude-latitude projection (resolution: 9.999.99 m).We used the bilinear method to obtain environmental rasters the same cell size as those from ArcMap software.Two scenarios (RCP4.5 and RCP8.5) were used to predict the potential distribution, changes in distribution, and the long-term distribution possibilities of S. robusta. Among GCMs, we used 19 bioclimatic variables from the Australian Community Climate and Earth System Simulator (ACCESS1-0) and Community Climate System Model 4.0 (CCSM4) for 2070 from the Coupled Model Inter comparison Project (CMIP5) Phase 5 developed by the IPCC at 30 arc seconds and ~1 km2 resolution (http://worldclim.org/)(Yang et al., 2017). The bioclimatic variables denote annual trends (e.g., mean annual temperature,annual precipitation), seasonality (e.g., annual range in temperature and precipitation), and extreme or limiting environ-mental factors (e.g., temperature of the coldest and warmest month, and precipitation of the wet and dry quarters). The projected global temperature of RCP4.5 ranged from 1.1C to 2.6C, and for RCP8.5, the range was from 2.6C to 4.8C(2061e2080; IPCC AR5 WG1, 2013). In case of regional climatic modeling (RCM), 14 climatic variables were used (eight seasonal and six annual variables) for both models of ACCESS1-0 and CCSM4 for the future period of the 2080s (2070e2100)(IPCC, 2014;Wang et al., 2017). 2.3. Analyses of species distribution We used green-red vegetation index (GRVI) to evaluate the density of S. robusta (Shishir and Tsuyuzaki, 2018; Xue and Su,2017) based on the precision of GRVI for detecting large-canopy tree phenology and coverage. Sensitive to the canopy surface of forests, GRVI is an effective threshold for phenology detection and is able to distinguish dense vegetation in areas of high greenness from other types of ground covers (Motohka et al., 2010; Nagai et al., 2012). Canopy density and forest degradation due to urban growth was assessed using green and red bands by GRVI and was calculated, as follows:GRVI greenredÞ=green þred Þ;range:1 to 1(1) where green and red represent band reflectance, respectively. GRVI ranges from 0.20 to 0.24 showed the most plausible distribution of S. robusta orests. The presence of S. robusta forests in previous (2001) and current (2015) stages was examined using GRVI with ArcGIS to investigate the accuracy of distributions predicted by Maxent modeling. The green bands ranged from 506 nm to 595 nm (IKONOS) and 510 nme580 nm (WV2), and the red bands ranged from 632 nm to 698 nm (IKONOS)and 630 nme690 nm (WV2), with the resolutions at 0.8 m for IKONOS and 0.5 m for WV2. Therefore, data quality andquantity from the two satellites were similar.2.4. ClimateAP and Maxent modeling Climate AP software (ClimateAP v2.20) which extracts and down scales gridded climate data with the spatial resolution of0.250.25 arc min (44 km) in Purbachal region with an increased spatial accuracy was used in RCM (Wang et al., 2017;Hijmans et al., 2005). The down scaling is attained through a combination of bilinear interpolation and dynamic local elevational adjustment. The 280 sample locations were manipulated in Excel to generate the inputfiles such as location, region,elevation, coordinates, for Climate AP in a same order and saved as“comma delimited textile”. Climate variables were appended following the input of the created csvfiles into Climate AP. The surface maps were generated using the climate variables in Maxent and spatial analysis were performed in ArcMap.We used a Maxent model (v.3.4.1) to predict current and future suitable geographic distributions of S. robusta species (Phillips, 2017). Maxent is developed by deterministic algorithms that converge to the optimal (maximum entropy) probability distribution (Phillips et al., 2006). It uses only present data of species distribution to model interactions between species occurrences and environmental variables (Marcer et al., 2013;Elith et al., 2011). Maxent builds composite, nonlinear response curves by selecting various feature classes, bounds model complication, and defends against over fitting by regularization(Phillips and Dudik, 2008).Maxent prediction is deri.