1 Introduction

According to instrumental temperature records, the Himalaya of Nepal have warmed more rapidly than global mean temperatures, with the greatest rate of warming occurring at high elevations (Shrestha et al. 1999, 2012; IPCC 2007; Shrestha and Aryal 2011). However, meteorological stations in Nepal are few, particularly in mountain regions, and are sparsely distributed through the country (IPCC 2007; Shrestha et al. 1999, 2012; Cook et al. 2010), making it difficult to estimate the rate and geographic extent of recent warming. Among the widely distributed weather stations throughout Nepal, precipitation records are available only after the year 1960 and temperature after 1970 (Shrestha et al. 1999, 2000). Thus to further extend the existing climate records of the region, tree-ring based climate reconstruction is needed. Tree-ring records have been used to extend climate records in the Indian Himalayan region (Bhattacharyya et al. 1988; Borgaonkar et al. 1994, 1996; Yadav et al. 1997, 1999; Singh and Yadav 2005; Singh et al. 2009; Yadav 2010), as well as the Tibetan Plateau (Yu et al. 2006; Huang and Zhang 2007; Gou et al. 2007; Liang et al. 2008; Li et al. 2012), but Nepal remains a ‘frontier’ region for dendroclimatological studies.

Rudolf Zuber collected the first tree-ring specimens from Nepal in 1979 and 1980, and their subsequent analysis by Bhattacharyya et al. (1992) indicated that Cedrus deodara, Pinus roxburghii, P. wallichiana, Abies spectabilis and Abies pindrow exhibited clear annual bands and, in some cases, common growth patterns. To date, the most extensive survey of climate information preserved within Nepalese trees was conducted by Cook et al. (2003), who established a 46-site network of tree-ring width records spanning most of the country and used those data to reconstruct average early-season (February–June) temperatures back to the mid-sixteenth century. Ring-width, density, and stable isotope records from Abies spectabilis have been used to hindcast temperature and moisture conditions in the far-north western corner of Nepal (Sano et al. 2005, 2011). But most other dendroclimatic studies have focused on identifying the climate signals recorded by tree-ring records (Bhuju et al. 2010; Chhetri and Thapa 2010; Gaire et al. 2011; Dawadi et al. 2013; Liang et al. 2014), rather than producing palaeoclimatic reconstructions.

Dendroclimatic studies of Picea smithiana (Wall.) Boiss (West Himalayan Spruce) growing in Pakistan and northern India has established the sensitivity of the species to early-spring temperatures (Borgaonkar et al. 1994; Yadav et al. 1997; Ahmed and Naqvi 2005; Khan et al. 2008; Ahmed et al. 2009, 2011; Zafar et al. 2010; Borgaonkar et al. 2011; 2012; Ram 2012). However, this species has not previously been the target of palaeoclimatic studies within far western Nepal. In this study, we report the development of a new tree-ring width record from P. smithiana in the far-western Nepalese Himalaya. We compare this new record against climate data and produce a new reconstruction of spring temperatures back to the mid-seventeenth century.

2 Materials and methods

2.1 Sample collection, laboratory methods, and analytical methods

During September 2012, we collected 161 tree-ring samples from 80 P. smithiana trees growing within the Lokhada region on the northern half of Khaptad National Park (Fig. 1). Established in 1984, Khaptad National Park has an elevation range of 1,400–3,300 m asl (Bhuju et al. 2007). The sampling site at Lokhada (29°42′N latitude and 81°15′E longitude, 2,700 m asl) had a slope of approximately 30° and was located in the temperate forest zone. Besides P. smithiana, the dominant tree species at the site included Abies pindrow, Juglans regia and Rhododendron spp.

Fig. 1
figure 1

Map showing the location of the Lokhada P. smithiana tree-ring sample collection, nearby meteorological stations, and grid points from the Climate Research Unit’s TS 3.21 compilation. The red online marks the boundaries of Khaptad National Park

Full size image

After samples were transferred to the Tree-ring Laboratory of Nepal Academy of Science and Technology, Nepal, they were air dried, mounted, and sanded until ring boundaries were clearly visible under a stereo-zoom microscope. Samples were cross-dated using the skeleton plot technique (Stokes and Smiley 1968) and tree-ring widths were measured using a linear encoder. The program COFECHA (Holmes 1983) was used to confirm dating and identify potential measurement errors. Ring-width measurements for individual trees were standardized using negative exponential curves, and pre-whitened by fitting an autoregressive (AR) model to each series (Cook 1985; Cook and Kairiukstis 1990). The detrended, pre-whitened series were combined to create a mean-value function (‘chronology’) describing overall tree growth at the site using the program ARSTAN (Cook and Holmes 1986). To assess the quality of the common signal expressed across the entire stand of P. smithiana, we computed the between-tree correlation, the expressed population signal, and running-window versions of both statistics (Wigley et al. 1984).

2.2 Climate data

Climate data are comparatively sparse in far-western Nepal, and there are no meteorological stations located within the boundaries of Khaptad National Park. The station nearest to the tree-ring sampling site is 25 km south is Khaptad having limited meteorological records. In general, the climate of this region is temperate, with cool, wet summers and cold, dry winters (Hindu Kush Himalayan (HKH) Conservation Portal, http://hkhconservationportal.icimod.org). Another meteorological record is located in Jumla, roughly 92 km east of the study site and also has short records. But the nearest long meteorological record is available at Mukteshwar, roughly 140 km west of the study site (Fig. 1). This record spans more than 117 years (AD 1897–2013), and, because of its length, has been used regularly as a reference for other dendroclimatic studies in the Nepal Himalayas (Sano et al. 2005, 2010, 2011; Thapa et al. 2013). Missing data within this record (4.5 % in temperature and 5.2 % in precipitation) was estimated using the ‘mtsdi’ package (Junger and Ponce de 2012) in the R programming environment (R Development Core Team 2014). The general climatic scenario depicted by climate records from meteorological stations, Jumla and Mukteshwar appeared had very similar pattern of monthly and seasonal distribution of temperature and precipitation (Fig. 2). The general overview of the regional climate of the area is derived using an average of four grid points of global CRU TS 3.21 gridded climate datasets (Harris et al. 2014) for the period of AD 1901–2012. The details of the entire climate data studied here are given in Table 1. Another significant climate data set is the Gridded Berkeley Earth Surface Temperature Anomaly (http://berkeleyearth.org/data/) records based on 1° × 1° latitude and longitude grid point is available for AD 1750–2013. Comparison of the temperature records of the Mukteshwar, CRU TS 3.21 and Berkeley Earth Surface Temperature Anomaly for the common time period of AD 1901–2012 showed similar trends (supplementary material, Figure 1). But the variability of Berkeley Earth Surface Temperature Anomaly was lower in comparison to both Mukteshwar and CRU TS 3.21 temperature records.

Fig. 2
figure 2

Climographs for a Jumla, Nepal, and b Mukteshwar, India. Data were obtained from Global Historical Climate Network (GHCN) and Department of Hydrology and Meteorology of Nepal. The time span and other details of the climate data are given in Table 1

Full size image
Table 1 Details of the climate (temperature and precipitation) records
Full size table

2.3 Assessing tree–climate relations

We assessed the relations between the growth of P. smithiana and local climate by comparing the residual ring-width chronology from Lokhada against the Mukteshwar climate record using the seasonal correlation procedure (SEASCORR) developed by Meko et al. (2011). This approach calculates Pearson correlations between tree-ring width and monthly mean temperature and total precipitation over a 14-month window (beginning in the September prior to growth and ending in October of the growth year). Because previous research has shown that the radial growth of P. smithiana is sensitive to early-spring temperatures, we chose temperature as the primary variable and precipitation as the secondary variable. To identify potential relations between tree growth and climate over long intervals, the monthly analysis was repeated using climate variables aggregated over 3-, 6-, and 12-months. Significance levels calculated by SEASCORR are produced by a Monte Carlo procedure that creates synthetic copies of the original tree-ring record using exact simulation (Percival and Constantine 2006). The temporal stability of the correlation results were evaluated by splitting the Mukteshwar climate record into two halves and repeating the analysis. We also compared the Lokhada ring-width chronology against gridded climate variables to evaluate the spatial extent of the relations between P. smithiana growth and climate using KNMI Climate Explorer (Trouet and Oldenborgh 2013).

2.4 Palaeoclimate reconstruction

The results of our analysis comparing P. smithiana ring-width chronology against local and regional climate provided a basis to reconstruct past climates in the far-western Nepalese Himalaya over the last few centuries. We used a simple linear regression model to relate tree-ring width against seasonal climate, and tested the time-stability of that model by dividing the calibration period into two equal sub-periods (Snee 1997; Meko and Graybill 1995; Touchan et al. 2011). We also tested the model through the predicted error sum-of-squares (PRESS) method (Weisberg 1985; Fritts et al. 1990; Meko 1997; Touchan et al. 2011), which is similar to the ‘leave-one-out’ cross-validation approach (Michaelsen 1987). We evaluated the model’s ability to reproduce observed changes in climate using statistical measures, including the correlation coefficient (r), the reduction of error (RE) and the coefficient of efficiency (CE). Once the model was judged to be effective and stable, we applied it to estimate past changes in climate based on tree-ring width, and truncated the reconstruction at the point where the tree-ring chronology becomes unduly influenced by non-climatic noise (Wigley et al. 1984). We also verified the significant climatic variable to be reconstructed against gridded climate variables for the same season to evaluate its regional strength through spatial correlation. The spatial correlation was carried out in KNMI Climate Explorer (Trouet and Oldenborgh 2013).

3 Results and discussion

3.1 The Lokhada P. smithiana ring-width chronology

The final set of dated ring-width measurements from Lokhada included 101 cores collected from 52 trees. Overall, the quality of the common signal exhibited by trees at this site is quite good (Table 2). The mean between-tree correlation for the residual chronology is 0.261. Average radial growth was roughly 1.2 mm year, and there were relatively few (0.1 %) locally absent rings (Schulman 1941; St. George et al. 2013). These measurements were detrended and combined to produce a 422-year long ring-width record for Lokhada, Nepal spanning AD 1591–2012 (Fig. 3). The expressed population signal (EPS) exceeded the arbitrary cut-off of 0.85 often used in dendroclimatology (Wigley et al. 1984) and, based on the EPS criterion, we judged the Lokhada chronology to be reliable back to AD 1640, although its between-tree agreement is somewhat lower during a brief period centered around AD 1680. The characteristics of the Lokhada chronology are generally comparable to results obtained by other studies of the same species (Borgaonkar et al. 1994; Yadav et al. 1997; Ahmed and Naqvi 2005; Ahmed et al. 2009, 2011, 2012; Khan et al. 2008; Zafar et al. 2010; Borgaonkar et al. 2011; Ram 2012).

Table 2 Chronology statistics of P. smithiana tree-ring chronology
Full size table
Fig. 3
figure 3

Tree-ring chronology of P. smithiana extending from AD 1591 to 2012 along with sample depth and EPS threshold cut off. The thick line represents the 10 years spline-smoothing curve

Full size image

3.2 Tree–climate relations

The seasonal correlation analysis showed that radial growth of P. smithiana is most strongly associated with spring temperatures (Fig. 4). Over the full period (1899–2012), the Lokhada ring-width chronology is negatively correlated with mean March to May temperatures (r = −0.56), with similar results obtained from both the early (1899–1955) and late (1956–2012) sub periods (r = −0.51 and −0.62, respectively). The Lokhada ring-width record when compared against gridded climate data shows that tree growth at this location is significantly correlated with mean March–May temperatures across much of the western Himalaya, north central and northeastern India (supplementary material, Figure 2). The inverse relationship between tree growth and spring temperatures might be due to increased evapo-transpiration and inadequate soil moisture at the onset of radial growth (Pant et al. 2000; Yadav and Singh 2002; Cook et al. 2003; Sano et al. 2005; Liang et al. 2008; Dawadi et al. 2013). Tree-ring records from several other species in the Nepalese Himalaya have shown similar negative responses to early-season temperatures (Cook et al. 2003; Sano et al. 2005; Chhetri and Thapa 2010; Gaire et al. 2011; Thapa et al. 2013). Outside of Nepal, studies in the western Himalayan region of India (Borgaonkar et al. 1994, 1996; Pant et al. 2000; Yadav and Singh 2002; Yadav et el. 2004; Singh and Yadav 2005) and Pakistan’s Karakoram range (Ahmed et al. 2012) also reported that the growth of conifer species is inhibited by anomalous spring warmth. Although Borgaonkar et al. (2011) showed that P. smithiana at high-altitude sites near Kinnor, India exhibited a strong positive relationship with mean annual and winter (December–February) temperatures, that behavior is not present in the Lokhada chronology.

Fig. 4
figure 4

SEASCORR results summary of seasonal climatic signals in tree-ring data of P. smithiana. (Top) correlation of tree-ring variable with monthly, 3-month total, 6-month total and 12-month mean temperature for ending months from September preceding the growth year through October of the growth year. (Bottom) partial correlations between tree-ring chronology with monthly total precipitation. Monte-Carlo-derived significance of correlation or partial correlation is shown in colors (Meko et al. 2011) for levels 0.01 and 0.05

Full size image

3.3 Spring temperature reconstruction

Based on the seasonal correlation results, we identified mean spring (March–May) temperatures as the most promising candidate for palaeoclimate reconstruction. We constructed a linear regression model with spring temperature as the predictand and the Lohkada ring-width chronology as the sole predictor. Over the full calibration period, the regression model explained 32 % of the total variance in spring temperatures, and had positive values for both RE and CE, indicating that the model has skill in reproducing the observed variability in temperature (Table 3). Our cross-validation tests, which split the calibration into two equal sub-periods and the PRESS procedure, indicate the model was stable over the period of comparison. The observed and estimated spring temperatures when compared over the period of calibration indicated that the linear regression model is skillful for reconstructing spring temperature variability (Fig. 5). But comparing smoothed versions of the observed and estimated temperatures indicated that the model is also applicable in recovering low-frequency variations (supplementary material, Figure 3). The low frequency variability in estimated temperature generally follows pattern similar to the low frequency variability of observed data but lack resemblance in trends in some periods. This might be due to the low temperature variance captured in the reconstruction. The observed and estimated spring temperature when compared against average gridded spring temperature data showed significant positive spatial correlation across much of the western Himalaya, north central and northeastern India (Fig. 6). This showed that the reconstruction produced roughly the same spatial correlation pattern with regional temperatures as the observed record.

Table 3 Calibration verification statistics for March–May temperature reconstructions
Full size table
Fig. 5
figure 5

Observed mean spring (March–May) temperatures at Mukteshwar, India and temperature estimated from the Lokhada P. smithiana tree-ring width chronology. The time period for both the data is AD 1897–2012

Full size image
Fig. 6
figure 6

Spatial correlations between a observed and b estimated temperature with gridded average March–May CRU TS 3.21 temperature records. Spatial correlation was carried out for the common period of AD 1901–2012

Full size image

3.4 Analysis and comparison of reconstructed temperature

We applied the regression model to reconstruct spring (March–May) temperatures back to AD 1640 (Fig. 7). We chose not to extend the reconstruction over the entire length of the Lokhada record because the earlier portion is comprised of fewer tree-ring specimens and has a weaker common signal. Comparing the reconstructed temperatures to the mean of the calibration period (AD 1897–2012) identified several cool and warm periods. If we consider lower-frequency variations, the reconstruction exhibits cooling trends during its earlier portion and warming in the recent few decades (Fig. 7). The longest cool period extended from the latter part of the eighteenth century to the late nineteenth century. Working in the western Himalayan region of India, Borgaonkar et al. (2011) identified an unprecedented enhancement in growth during the last few decades in the tree-ring chronologies based on the same taxa which is absent in our record. This may be due to the different standardization approach applied on the tree-ring data based on local site characteristics.

Fig. 7
figure 7

The reconstructed March–May temperature data filtered with a smoothing spline with a 50 % frequency cut off of 10 years

Full size image

Three other tree-ring based temperature reconstruction for Nepal are available viz., February–June (Cook et al. 2003), October–February (Cook et al. 2003) and March–September (Sano et al. 2005). We compared the temperature reconstruction from present study with temperature reconstruction of February–June (Cook et al. 2003). The February–June temperature reconstruction of Cook et al. (2003) is based on the dense tree-ring network covering Nepal. Our reconstruction does not perfectly matches with Cook et al. (2003) February–June temperature reconstruction, which could be due to the difference in season and methodology involved in chronology development. However, the multi-decadal scales of warming and cooling trends are shared and we observed appearance of warming trends has comparable coherency (Fig. 8). Cook et al. (2003) recorded sharp cold period centered in 1810s, which they considered as cooling trend due to the eruption of Tambora in 1815–1816, and another unknown eruption in 1809 (Dai et al. 1991; Cole-Dai et al. 2009). However, we did not noticed such sharp cold period during the 1810s but AD 1815–1823 cooling trends are observed in low frequency smoothing of 30 and 50 years (supplementary material, Figure 4). The March-September temperature reconstruction based on A. spectabilis tree-ring data from western Nepal Himalaya (Sano et al. 2005) is not compared directly with the present temperature reconstruction because there is absence of cool and warm periods description in their record. Further Sano et al. (2005) also compared their reconstruction with February–June temperature reconstruction (Cook et al. 2003). Sano et al. (2005) also reported that their temperature reconstruction does not perfectly matches with temperature reconstruction of Cook et al. (2003) but shared multi-decadal to century-scale warming and cooling trends.

Fig. 8
figure 8

Reconstructed temperature anomalies filtered using smoothing spline with a 50 % frequency cut off of 10 years: a spring (March–May) temperature reconstruction for far western Nepal Himalaya (present study), b February–June temperature reconstruction for Nepal (Cook et al. 2003), c May–June maximum temperature for north central China (Chen and Yuan 2014), d April–September for Tibet (Briffa et al. 2001), e April–September for Central Asia (Briffa et al. 2001), f seasonal March–May temperature for Europe (Xoplaki et al. 2005), g summer (June–July) temperature for Yamalia, northwest Siberia (Briffa et al. 2013), h growing-season, April–September temperature for northern Hemisphere (Briffa et al. 1998) and i annual temperature for extra-tropical northern Hemisphere (Wilson et al. 2007)

Full size image

The spring temperature reconstruction from present study is also compared with other regional temperature reconstructions (Briffa et al. 1998, 2001, 2013; Xoplaki et al. 2005; Wilson et al. 2007; Chen and Yuan 2014) along with records from Nepal (Cook et al. 2003) (Fig. 8). For comparison all the reconstructed temperature anomalies were filtered using smoothing spline with a 50 % frequency cut-off of 10 years. There are similar cooling trends observed during the early eighteenth century temperature reconstruction records from Nepal (Cook et al. 2003), North central China (Chen and Yuan 2014) and Tibetan Plateau (Briffa et al. 2001) (Fig. 8a–d). Increasing trends in the spring temperature reconstruction of present study during recent few decades are also observed in the temperature reconstructions from North central China (Chen and Yuan 2014), Europe (Xoplaki et al. 2005), Siberia (Briffa et al. 2013) and extratropical northern Hemisphere (Wilson et al. 2007) (Fig. 8c, f, g, i). From the Mid eighteenth century towards recent there is a close resemblance between the spring (March–May) temperatures of Europe (Xoplaki et al. 2005) and the present study. Both these spring temperature reconstructions do not show cooling trends for few years after Tambora eruption that is witnessed in all other records considered. Thus it was observed that the present study has closest relationship with records of spring temperature from Europe (Xoplaki et al. 2005). However, there were comparable instances in the given period of all the records (Fig. 8) where there are similar trends as obtained in present study. At the end of 1940s i.e., mid twentieth century the records including the present study showed a peculiar rise in temperature, even including records from cold and arid regions such as North central China, Central Asia and Siberia (Fig. 8c, e, g). However, temperature in Central Asia and northern Hemisphere revert back towards cooling trends in the late twentieth century. Thus looking at all the records we observed that our present study is following patterns of temperature variations observed in surrounding regions. Probably all these records have slight variation amongst each other and with present study due to regional climate factors, species and parameters used for tree-ring based temperature reconstruction.

The reconstruction did not show the significant pattern of cooling during the Little Ice Age but there were few cold episodes recorded. The millennium-long reconstruction of mean summer (May–August) temperature extending back to AD 940 derived from tree-ring width data of Himalayan Pencil Juniper (Juniperus polycarpos C. Koch) from the monsoon-shadow zone in the western Himalaya, India revealed medieval warm period and little ice age cooling and again warming since the nineteenth century (Yadav et al. 2011). The cool episodes noted in their reconstruction around AD 1537–1622, 1731–1780, 1817–1846 represent the LIA was widely observed in many regions (PAGES 2K Consortium 2013), however, there exist regional to local temporal offsets and differences in the magnitude of cooling (Cook et al. 2003, 2013; Yadav et al. 2004, 2011; PAGES 2K Consortium 2013). Some tree-ring based summer temperature reconstructions from Asian mountain regions like Tibet and central Asia (Briffa et al. 2001), Nepal (Cook et al. 2003) and India (Yadav et al. 2004) document cooling during the last decades of the twentieth century. However, other high-resolution climate proxies from high-latitudes of the northern and southern Hemispheres indicated warming in the twentieth century (Mann and Jones 2003; Yadav et al. 2011; Cook et al. 2013; PAGES Consortium 2013).

3.5 Regional tele-connections

The average March–May observed and estimated temperature was compared with average March–May temperature of All-India and homogenous regions, viz., Western Himalaya, Northwest, North Central, Northeast, West Coast, East Coast and Interior Peninsula temperature records (http://tropmet.res.in) for the common time period of AD 1901–2007. Except West coast region, all other six region including all India showed significant correlation at p < 0.05 with reconstructed temperature series (Table 4). This observed positive correlation shows the large-scale regional coherency pattern and also the validation of the reconstructed data. However, the observed average March–May temperature showed higher correlation with all these regions wise Indian temperature as compare to the estimated temperature.

Table 4 Correlation of observed and estimated March–May temperature with different climatic zone of India (www.tropmet.res.in). The correlation analysis was carried out for the time period of AD 1901–2007
Full size table

Applying wavelet analysis (Torrence and Compo 1998) to the Lokhada reconstructions shows that most of its variance is concentrated at higher frequencies (Fig. 9). The higher frequency bands of less than 7 years observed may be linked with ENSO activity. The periodicity of this high frequency variability was dominant during mid seventeenth century, early eighteenth, nineteenth and twentieth century and late period of twentieth century. Based on this result we performed correlation analysis of both observed and estimated average March–May temperature with Sea Surface Temperature (SST) index of different region of the equatorial Pacific viz., NINO1.2, NINO3, NINO3.4 and NINO4 (Kaplan et al. 1998) and extended multivariate ENSO Index (MEI) (Wolter and Timlin 2011). Both observed and estimated March–May temperatures had significant negative correlations with SSTs index and MEI for monsoon and post monsoons seasons (supplementary material, Table 1). These negative relationships suggest that warm (cool) spring season over the western Nepal Himalaya region are associated with cool (warm) SSTs in the following seasons. Similar observation between spring temperature reconstruction and SST are also noted in Yadav et al. (1997). We also carried out spatial correlation of observed and estimated temperature with Sea Surface Temperature (Kaplan et al. 1998) and observed similar negative correlation in the equatorial belt of the Pacific Ocean (Fig. 10) for monsoon months. However the relationship is stronger in case of observed data as compare to the estimated. But this tele-connection showed temperature reconstruction based on the high-resolution proxy from the high altitude region of the western Nepal Himalaya has linkages with climatic variability in a global scale.

Fig. 9
figure 9

Wavelet power spectrum of reconstructed temperature data for AD 1640–2012. The power has been scaled by the global wavelet spectrum. The cross-hatched region is the cone of influence, where zero padding has reduced the variance. Black contour is the 10 % significance level, using a red-noise background spectrum. The global wavelet power spectrum (black line) and 10 % significance level global wavelet spectrum (dashed line) shown at right

Full size image
Fig. 10
figure 10

Spatial correlations between a observed and b estimated temperature with Kaplan Sea Surface Temperature in global scale. The spatial correlation was carried out for monsoon months (June–September) covering a time span from AD 1897 to 2012

Full size image

4 Conclusions

The study presents the longest dendroclimatic temperature reconstruction for the Far-western Nepal Himalayan region of Nepal. This new reconstruction provides a basis for studying past temperature variability in this region where very short period of the instrumental data is available. The long term climate management planning and climate modeling work of the region shall be possible with the available long-term records of temperature. The relationship of Nepal Himalayan climate with ENSO indicated climate records based on high-resolution proxy such as tree-ring from this region help to understand climate change linkages and coherency pattern in global terms. In the future, combination additional tree-ring records from P. smithiana with tree-ring data from other species should help study the temporal and spatial structure of temperature variability in Nepal.