1 Introduction

Climate change and variability is recognized as a major threat for the environment and sustainable development (Lal et al. 2012). It is evident that a change in the climate, depending upon location, may cause disastrous consequences on the socioeconomic survival of millions of people (Bhutiyani et al. 2007). Therefore, studies on climate change, hydro-climatic extremes, and potential impacts on various sectors have gained momentum in recent years (Chen et al. 2007). In the last few decades, hydro-climatic variations became more prominent and were studied widely at global, regional, and local scales. These extreme climatic events like heat waves, floods, and drought induced by the hydro-climatic variability are expected to exacerbate in the future potentially due to climate change thus posing a major challenge to various sectors such as agriculture, biodiversity, and related ecosystem services that support livelihoods (Shrestha et al. 2017). The socioeconomic impacts of those changes are significant in all countries; however, low- and middle-income countries are especially vulnerable (IPCC 2013). Such countries experience higher fatalities even when exposed to hazards of similar magnitude and further with the 1 °C additional warming, risks associated with such types of extreme events increases progressively (IPCC 2013).

A warmer world is projected to bring more precipitation across the world as well as in Nepal; most models project a wetter and warmer future (2040–2059) mostly in the range of 2–3 °C, depending on the location and scenarios considered (Agrawala et al. 2003). Baidya et al. (2008) observed the general increasing trend in the temperature and precipitation extremes all over Nepal indicating more weather-related extreme events like flood and landslides in the future. Similarly, Manandhar et al. (2012) revealed a warming trend in the Kali Gandaki River Basin at higher altitudes with variable trends in precipitation indices. Bastakoti et al. (2016) also showed an increasing trend of climatic extreme in the recent past. Shrestha and Nepal (2016) observed changes in temperature and rainfall patterns at Makwanpur district of Nepal. Furthermore, Shrestha et al. (2017) found an increasing trend of extreme climatic events in Koshi river basin though long-term trend was not observed in rainfall pattern. In Western Nepal too, warmer and wetter future is projected in Chamelia watershed of Mahakali river basin (Pandey et al. 2019). The warming trends observed over the past several decades exacerbate the hydrological cycle and hydrological systems in many ways. Some of them include change in precipitation patterns, widespread melting of snow and ice; increase in atmospheric water vapor; increase in evaporation; and changes in soil moisture and runoff causing natural variability on inter-annual to decadal time-scales (Bates et al. 2008). Therefore, increased climate variability, could have influence on extreme climatic events like floods and droughts, both in frequency and intensity, affecting Nepal in various ways (Agrawala et al. 2003; Chaulagain 2006; Society of Hydrologists and Meteorologist 2012). Thus, understanding the historical as well as projected future trends in hydro-climatic variables, especially amount and significance of the trends in the extremes, are useful for informed climate-resilient development planning and decision-making.

There are various statistical methods and tools available for evaluating climatic trend as described in literatures (Helsel and Hirsch 2002; Khon et al. 2007; Some’e et al. 2012; Duhan and Pandey 2013). The knowledge on amount of the long-term trends, change points (if any), their position in the time series, and statistical significance of the trends are very important as they allow the interpretation of its possible causes (Moraes et al. 1998). There are many parametric and nonparametric methods suitable for detection and attribution of trends and breaks in hydro-climatic series. Nonparametric tests are widely used as they will work with independent data and can accommodate outliers. One of the widely used nonparametric tests for detecting a trend in hydro-climatic time series is the Mann–Kendall (Mann 1945; Kendall 1975; Shrestha et al. 1999; Nepal 2016; Khatiwada et al. 2016). RClimDex (Zhang and Yang 2004), a R-based tool, is also available in public domain for calculating trends in climatic variables (i.e., temperature and precipitation) and their statistical significance. RClimDex has been used by many studies over the years. They include but not limited to Manton et al. (2001), Kiktev et al. (2003), Alexander et al. (2006), Tank et al. (2006), Baidya et al. (2008), Islam (2009), Donat et al. (2013), and Shrestha et al. (2017). The tool can calculate 27 indices related to temperature and precipitation as defined by Expert Team on Climate Change Detection and Indices (ETCCDI) (WMO 2009).

In addition to determining the historical climatic trend, projecting future climatic extreme plays a vital role for the climate impact studies. Future climate of an area is generally projected using General Circulation Models (GCMs) or Regional Circulation Models (RCMs). However, RCM has been widely used for the climate impact studies due to its higher resolution and better capturing of regional conditions. Many recent studies have used RCMs in climate projection and impact studies (Kulkarni et al. 2013; Fang et al. 2015; Devkota and Gyawali 2015; Khadka et al. 2016; Magar et al. 2016; Rajbhandari et al. 2017; Bhattarai et al. 2018; Pandey et al. 2019). RCM projections can further be downscaled using approaches such as linear scaling (Teutschbein and Seibert 2012), quantile mapping (Gudmundsson et al. 2012), local intensity scaling (Fang et al. 2015), power transformation (Fang et al. 2015), variance scaling (Teutschbein and Seibert 2012; Fang et al. 2015), and delta change (Ruiter 2012) to make RCM projections usable for a watershed level.

Similarly, a large number of studies (Wang et al. 2012; IPCC 2013; Panda et al. 2013; Kundzewicz et al. 2015; Asadieh et al. 2016; Dery et al. 2016) have examined potential trends in observed streamflow during the twentieth century, at scales ranging from catchment to global. Some studies have detected significant trends in selected indicators of flow and demonstrated statistically significant links with trends in temperature or precipitation (Bates et al. 2008). Trends in various indicators of streamflow, one of the important hydrological components that can be altered by both climatic and nonclimatic factors, can be analyzed by various statistical approaches. Indicators of Hydrological Alteration (IHA) (Richter et al. 1996; Ritcher et al. 1997) is a tool that has a capability to analyze 33 indices related to hydrological extremes (Kiesling 2003; Bharati et al. 2016). The hydro-climatic time series may also have abrupt changes in addition to gradual changes (or trend). In such cases, one needs to calculate trends separately before and after such abrupt changes. Statistical tests such as Pettitt’s Change Point (Pettitt 1979) and Mann-Kendall (Mann 1945; Kendall 1975) are widely used (Liu et al. 2012; Xia et al. 2014; Mallakpour and Villarini 2016) to detect a change point and statistical significance in hydro-climatic time series.

There are many studies focused on various aspects of climate change in Nepal, ranging from climate change impact assessments (e.g., Pandey et al. 2019, 2020) to flood risk assessments in climate-change context (e.g., Devkota and Bhattarai 2015; Devkota and Maraseni 2018). A summary of selected studies related to climate change in Nepal are provided in Appendix Table 6. However, there are limited studies focusing on climate projection using RCMs and the most recent representative concentration pathways (RCP) scenarios in Central Nepal in general, and Marshyangdi watershed. Furthermore, studies focusing on both historical and future climatic extremes as well as hydrological extremes are almost nonexistent. Marshyangdi is a snow-fed/Himalayan catchment having high potential for water infrastructure development, and it hosts a good number of hydropower projects. The watershed has a potential to generate at least 3251.8 MW of electricity (Jha 2010). Currently, three hydropower projects, namely, Marshyangdi (69 MW), Middle Marshyangdi (70 MW), and Upper Marshyangdi-A (50MW) are in operation and six more have got license. Therefore, understanding hydro-climatic extremes in that watershed is important for informed-adaptation planning. Thus, we aim to unpack hydro-climatic extremes in a Marshyangdi, located in Central Nepal which feds to Narayani river basin (Fig. 1). The objectives of this study are as follows: (i) to characterize historical and projected trends in climatic extremes and (ii) to characterize hydrological extremes in the watershed.

Fig. 1
figure 1

Location and topographical details of the Marshyangdi watershed in Nepal

Full size image

2 Materials and methods

Overall methodological framework is shown in Fig. 2. It consists of preparation of historical time series of climatic data (temperature and precipitation) at selected stations, projection of future climate, selection of suitable set of indices for climatic extremes, evaluation of trends in those indices, evaluation of hydrological indices related to extremes, and finally direction and magnitude of hydro-climatic trends with its significance obtained. All these aspects are elaborated in the following sub-sections.

Fig. 2
figure 2

Methodological framework. RCP, representative concentration pathways; RCMs, regional climate models; NF, near future; MF, mid-future; T, temperature; IHA, indicators of hydrological alteration; Met, meteorological

Full size image

2.1 Study area

Marshyangdi watershed is a sub-basin of one of the major river systems, the Gandaki River Basin, in Central Nepal. It is located between 27° 50′ 42″ and 28° 54′ 11″ N latitudes and 83° 47′ 24″ and 84° 48′ 04″ E longitudes (Fig. 1), covering an area of 4148 km2. The elevation of this watershed varies between 274 and 8042 m above the mean sea level (masl). The major portion of the watershed lies above 45% slope and are covered by snow and glacier, i.e., most of the area lies between 4000 and 6000 masl.

Climate in the watershed varies from Tropical Savannah in the lower belt to Polar frost type in the higher altitudes (Karki et al. 2016). The mean slope of this basin is 29.42°. Average annual maximum and minimum temperatures are 26°°C (June) and − 6°°C (January), respectively (Sharma 2017). Population in the four districts covered by the study watershed is 0.77 million (CBS 2019). Major land use/cover pattern in the watershed is the Grassland, followed by Barren land and Agricultural land, respectively (Sharma 2017). This region is a part of the major Annapurna Trekking route from Besisahar (in Lamjung district) where the local economy also depends upon it.

The Marshyangdi River is perennial in nature and has a typical dendritic drainage system which begins at the confluence of two mountain rivers, the Khangsar and Jharsang, northwest of the Annapurna massif at an altitude of 3600 masl. Then it flows eastward through Manang district and southward through the Lamjung district covering other districts Gorkha and Tanahu. Finally, it joins the Trishuli river system at Mugling as one of the major tributaries of the Saptagandaki River system. Major tributaries of the Marshyangdi River includes Khudi, Dordi, Chepe, and Daraudi.

2.2 Historical trend analysis

Historical time series of daily observed temperature, both maximum and minimum, and precipitation at 12 climatic stations and observed river discharge data at 2 stations were collected for the period of 1970–2018 from the Department of Hydrology and Meteorology, Nepal. However, data at only 11 stations (Appendix Table 7), including one hydrological station, were selected for further use after an exploratory data analysis. Then suitable data length was selected (Appendix Table 7) considering the missing values calculated for each month per year for variables like maximum temperature (Tmax), minimum temperature (Tmin), precipitation, as well as for discharge.

Data quality control (QC) was carried out using RClimDex statistical tool, with the purpose of identifying errors in data processing, such as errors in manual keying (Alexander et al. 2006). Months with missing values of more than 10 days were considered month with missing data and coded accordingly while preparing data for RClimDex. We defined outliers in daily maximum and minimum temperatures as the values beyond the range of three standard deviations (SD) of the mean (i.e., mean ± 3*SD)) (Zhang and Yang 2004; Vincent et al. 2005). Similarly, 25 and 0 °C were defined as upper and lower thresholds of daily maximum temperature and 25 mm as the threshold of daily precipitation. A set of indices used in this analysis are based on the 27 indices related to daily temperature and precipitation developed by an Expert Team on Climate Change Detection and Indices (ETCCDI) (WMO 2009). RClimDex (1.0) was used to calculate trends in the climatic indices on an annual basis at various stations using daily precipitation and temperature data of varying length, as presented in Appendix Table 7. Out of 27 extreme indices, 23 indices (13 related to temperature;10 related to precipitation) selected for analyzing climatic extremes in this study are presented in Table 1. The trends in terms of magnitude, direction, and statistical significance were estimated using the methods described by Zhang and Yang (2004).

Table 1 Definitions of extreme climatic indices used in this study (source: Zhang and Yang 2004)
Full size table

2.3 Future climate extremes analysis

Future climate projection is based on outputs of the Coupled Model Inter-comparison Project-Phase 5 (CMIP5), a collaborative climate-modelling process coordinated by the World Climate Research Programme (WCRP) using different climate forcing’s. Future climate projection was carried out at four meteorological stations, namely, Thakmarpha (index:604), Khudi Bazzar (index:802), Gorkha (index:809), and Chame (index:816) as these stations have long-term time series of both temperature and precipitation data. Missing values in the daily time series were filled with long-term average daily values for all the variables. For example, value for day 1 (i.e., 1 January) was calculated as an average of 31 values of 1st January (i.e., from 1983 to 2013) and that for day 365 (i.e., 31 December) was calculated as an average of 31 values of 31st December (i.e., from 1983 to 2013). Then three different Regional Climate Models (RCMs), namely, ACCESS-1, CNRM-CM5, and MPI-ESM-LR of 0.5 × 0.5° horizontal resolution were downscaled from the South Asia CORDEX data portal (http://cccr.tropmet.res.in/home/index.jsp) and then divided into two periods namely, near-future (2014–2033) and mid-future (2034–2053) to project future scenarios. Considering the focus of this study on river health in connection to water infrastructure development, this study considered future period up to the mid-century only. These RCMs were selected based on literature review (Appendix Table 6). Generally, RCM outputs are only available for RCP4.5 and RCP8.5 and occasionally for RCP2.6. Hence, in this study, RCP4.5 is selected as a medium-stabilizing scenario, stabilization without overshoot pathway leading to 4.5 W/m2 (~ 650 ppm CO2) at stabilization after 2100 and RCP8.5 as a very high emission scenario, which refers to rising radiative forcing pathways leading to 8.5 W/m2 (~ 1370 ppm CO2) by 2100. In order to remove the systematic bias in the downscaled data, quantile mapping bias correction technique was applied to all the raw daily temperature and precipitation time series prior to the calculation of the extreme climatic indices by using RClimDex as mentioned in Sect. 2.2. Future climatic extreme indices were analyzed based on ensemble time series data.

2.4 Hydrological extreme analysis

IHA as described in Mathews and Ritcher (2007) were used for evaluating hydrological extremes in the Marshyangdi watershed. IHA uses a nonparametric range of variability approach (RVA) (Richter et al. 1997) to characterize alterations in inter- and intra-annual variation in river flow. RVA is based upon comprehensive statistical characterization of the temporal variability in hydrologic regime quantifying the degree of alteration of 33 ecologically relevant hydrological parameters (Appendix Table 10) that describe crucial relationships between flow and ecological functions.

RVA analysis places the category boundaries of 17 percentiles from the median yielding an automatic delineation of three categories of equal size, as follows: the lowest category contains all values less than or equal to the 33rd percentile (low alteration); the middle category contains all values falling in the range of the 34th to 67th percentiles (moderate alteration); and the highest category contains all values greater than the 67th percentile (high alteration) (Richter et al. 1998). A positive hydrological alteration value means that the frequency of values in the category has increased from the pre- to the post-impact period (with a maximum value of infinity), while a negative value means that the frequency of values has decreased (with a minimum value of − 1). Each IHA is calculated in terms of median value, deviation degree, and degree of hydrological alteration (Appendix 1) between two periods to assess impacts of intervention on alterations. The pre- and post-impact periods were determined as per Pettitt’s (1979) test on the annual average data to identify any abrupt change points in the streamflow time series in the Marshyangdi watershed.

2.5 Identification of change point

The approach after Pettitt 1979) was applied to detect a single abrupt change point in climatic as well as hydrological data (Pohlert 2018) to provide input in the IHA tool. The Pettit’s test is a nonparametric test, which is useful for evaluating the occurrence of abrupt changes in climatic records (Sneyers 1990; Tarhule and Wool 1998; Smadi and Zghoul 2006; Gao et al. 2011). It tests the H0: The variables follow one or more distributions that have the same location parameter (no change), against the alternative: a change point exists. The Pettitt’s test is one of the most commonly used tests for change point detection because of its sensitivity to breaks in the middle of any time series (Wijngarrd et al. 2003). This test is based on the Mann-Whitney two-sample test (rank-based test) and allows the detection of a single shift at an unknown point in time because of the lack of distributional assumptions (Javari 2016).

3 Results and discussion

3.1 Historical trends in climatic extremes

3.1.1 Temperature-based indices

Decadal trends of the 13 temperature-based extreme climatic indices at four meteorological stations are shown in Table 2. Statistically significant trend values at 5% (p < 0.05) level of significance are marked with an asterisk in Table 2. The results clearly indicate varying amount, direction, and significance (statistically) of the trends across the stations. Furthermore, such variations are clearly visible also among the indices under the same group (i.e., fixed-threshold index, absolute extreme index, percentile-based index, and duration-based index).

Table 2 Historical decadal trends in the climate extreme indices in the Marshyangdi watershed
Full size table

Among the threshold-based indices, “ice days (ID)” index does not have any trend (trend = 0) at three stations and insignificant negative trend at one station, i.e., at Chame (index:816); therefore, it was discarded for further analysis. However, “summer days (SU25)” index has shown significant positive trends at Khudi (index:802) and Gorkha (index:809) stations, insignificant positive trend at Thakmarpha (index:604), and insignificant negative trend at Chame (index:816) (Table 2). Among the absolute extreme indices, for example, “tropical nights (TR20)” index has significant positive trend at Gorkha (index:809) stations and insignificant positive trend at Khudi (index:802) whereas no trends were observed (trend = 0) at the other two stations. Similarly, index TXn has positive trends across all the stations with varying amount and level of significance and other three indices (TNx, TNn, TXx) have mixed trends in terms of magnitude, direction, and level of significance (Table 2). In case of three percentile-based indices, number of warm days (TX90p) has positive trends at all the stations, number of cool nights (TN10p) has positive trend only at two stations, while the number of warm nights (TN90p) has positive trend at only one station Khudi (index:802) and number of cool days (TX10p) has no positive trends (i.e., significant negative trends) at all the stations (Table 2). It again reflects heterogeneity in percentile-based climate extreme indices derived from temperature time series. Finally, among the duration-based indices, DTR has statistically significant positive trends at three stations, WSDI also has positive trends at those three stations but are statistically insignificant, and CSDI has positive trends only at two stations (Table 2). Such rising trends in the warm temperature indices and decreasing trends in the cool temperature indices are reported in other recent studies as well (e.g., Karki et al. 2020; Poudel et al. 2020).

Though the temperature extreme indices have certain magnitude of trends over the years, the actual index value varies from year to year as indicated in Appendix Fig. 14 for TX90p as an example. For the period of 1984–2013, TX90p at Khudi Bazzar (index:802) station has an average value of 9.3 °C, trend of + 0.36°C/year, and the index value varies from 0.81 to 25.8 °C. This indicates the significance of understanding these variabilities in addition to average annual value and long-term trend while applying the results for informed decision-making.

3.1.2 Precipitation-based indices

Trends in 10 precipitation-based climate extreme indices are shown in Table 2. They are also grouped under the following four categories, namely, fixed threshold indies (2), absolute extreme indices (3), percentile-based indices (2), and duration-based indices (3). Trends in the indices are evaluated at 5% level of significance (i.e., p < 0.05) at 10 stations distributed across the Marshyangdi watershed. Precipitation-based indices also show variation in trend amounts, directions, and statistical significance across the 10 stations with no distinct spatial trends.

Both the indices under threshold-based category, namely, R10 (number of heavy precipitation days) and R20 (number of very heavy precipitation days), show insignificant decreasing trends at six out of 10 stations. Both indices, in most of the cases, show the same direction and significance in the trends, though the magnitude of trends are different (Table 2). Similar results are reported in other studies as well (e.g., Lamichhane et al. 2020). Among the duration-based indices, “consecutive dry days (CDD)” shows insignificant positive trends at seven out of 10 stations whereas at two stations Kunchha (index: 807) and Chame (index: 816), trends are significantly positive. “Simple daily intensity index (SDII)” show insignificant positive trends at five out of 10 stations, and “consecutive wet days (CWD)” show insignificant positive trends only at three out of 10 stations (Table 2). All the trends in remaining stations are negative but statistically insignificant. In case of three absolute extreme indices, RX1day and PRCPTOT have negative indices with varying magnitude at seven out of 10 stations. Some of those indices are also statistically significant, for example, PRCPTOT at three stations (i.e., Larke Samdo, Bandipur, and Manang Bhot) and RX1day at Khudi (Table 2). The RX5day index on the other hand has negative trends only at five out of 10 stations, with only one of them (i.e., Khudi) being statistically significant. There magnitude of indices varies widely across the stations for all three indices. Finally, for two percentile-based indices (i.e., R95p and R99p), R99p trends are negative (insignificant) at all eight stations and that of R95p are negative, two of them being statistically significant, at five out of 10 stations (Table 2).

The increasing trend of CDD index is also observed over the southern and northern slopes of Central Himalayas and across the Narayani river basin (Sigdel and Ma 2016; Lamichhane et al. 2020). Similarly, Karki et al. (2017) observed the similar trend of CDD across the country warning that such increase in the dry period can impact negatively in agricultural activities and hydropower generation, thus affecting economic aspects of the livelihood. Increasing trends in the climatic indices have also been reported at many stations of the Koshi basin of Nepal (Shrestha et al. 2017). Such climatic extremes may have implications in public health as well as it may cause respiratory-related health problems in Nepal (Karki et al.2017).

Like temperature-based extremes, precipitation-based extreme indices also have inter-annual variability as shown in Appendix Fig. 14 for Rx5day as an example. For the period of 1984–2013, Rx5day at Khudi Bazzar station (index: 802) has an average value of 299.6 mm, trend of − 3.78 mm/year, and the index value varies from 244 to 414 mm with a coefficient of variation of 43.6 mm. Understanding such variabilities are helpful to use the results cautiously.

3.2 Projected future trends in climatic extremes

Projected future trends in climatic extremes are based on an ensemble of three RCMs (CNRM, ACCESS, and MPI) under RCP4.5 and RCP8.5 scenarios for both NF and MF periods and presented in Table 3 (please refer to Appendix Table 6 for the detailed characteristics of the RCMs). The RCM outputs were bias corrected for the historical period (1983–2013) and projected for the future periods. The performance of the RCM outputs for the historical periods was of acceptable quality after bias corrections.

Table 3 Projected trends in future climatic extreme indices based on an ensemble of three RCM outputs
Full size table

3.2.1 Projected climatic extreme trends in the Marshyangdi watershed

Temperature-based indices

Across the stations, the trends in extreme climatic indices were evaluated for an ensemble time series generated based on the three RCMS (i.e., CNRM, ACCESS, and MPI). Result shows a gradual increase in the extreme temperature indices at some stations from baseline to MF, while some indices are observed to be decreasing gradually at all the stations and some indices shows mixed trends (Table 3). For example, summer days (SU25) index shows a gradual increase (insignificant) at the Chame (index: 816) station from baseline (− 0.03 days/year) to MF (0.11 days/year) under RCP4.5 and RCP8.5 scenarios, but it does not show any trend during NF under RCP8.5. Similarly, TXx at Chame shows increasing (insignificant) trend from baseline (− 0.07 °C/year) to NF and MF for both RCPs with a similar trend of 0.05 °C/year, but during MF under RCP4.5 shows a slight decreasing trend of 0.03 °C/year. TNx shows no change from baseline (− 0.01 °C/year) to NF under RCP4.5; however, it increases (significant) to NF and MF for both RCP4.5 and RCP8.5 with a trend amount of 0.07 °C/year at Thakmarpha (index:604). TNx at Chame (index:816) shows increasing (significant) trend from baseline (− 0.42 °C/year) to MF (0.09 °C/year) under RCP4.5 and RCP8.5; however, the trend was insignificant during NF under both RCPs.

At the Khudi Bazzar (index:802) station, TXn shows gradual increasing (insignificant) trend from baseline (0.06 °C/year) to increasing (significant) trend in MF under RCP4.5 and RCP8.5 (0.11 °C/year). However, some percentile-based extreme temperature indices show an increasing trend at more than one stations for the future scenarios. For example, TN90p increases from baseline at stations Thakmarpha (insignificant), Khudi (insignificant), as well as Chame (significant) (0.1 days/year, 0.15 days/year, − 0.48 days/year) to MF with significant trends at the three stations for both RCPs (0.56 days/year, 0.61 days/year, 0.62 days/year). At the Gorkha (index:809) station, though it does not show any change in trend in the baseline, it increases significantly from 0.2 days/year in NF4.5 to 0.5 days/year from NF to MF under both RCPs except during MF under RCP8.5 where it decreases slightly to 0.07 days/year. Other climatic extremes like “Warm Spell Duration Indicator (WSDI)”shows increasing (insignificant) trend from baseline (0.17 days/year) to MF under both RCP scenarios (0.52 days/year) at the Khudi bazzar station, but the trend is significant in NF for both RCPs (Table 3).

Precipitation-based indices

Precipitation-based climate extreme indices do not show gradual increasing trend at any stations from baseline to future periods (NF and MF) under both RCP scenarios (4.5 and 8.5) (Table 3). For example, Rx1day at the Khudi Bazzar (index:802) is projected to increase (insignificant) from baseline (− 2.47 mm/year) to NF and MF (0.1 mm/days) under both RCPs, but the magnitude of trend value was less in comparison with NF for both RCPs. Index like “simple daily intensity index (SDII)” and number of very heavy precipitation days (R20) do not show any gradual increases from baseline to MF under both RCPs (mixed trend) at all the stations (Table 3). Similarly, annual total wet-day precipitation (PRCPTOT) at the Chame station shows increasing (insignificant) trend from baseline (− 0.20 mm/year) to MF (3.18 mm/year) under both RCPs whereas during NF under RCP8.5, it shows negative (insignificant) trend (− 1.81 mm/year).

Some extreme precipitation indices show increasing trend only from baseline to NF and then decreases during MF. For example, RX5day increases (insignificant) from baseline (0.82 mm/year) to NF (1.38 mm/year) and then decreases (insignificant) during MF under RCP4.5 (− 1.22 mm/year) and RCP8.5 (− 0.27 mm/year) at the Thakmarpha (index:604) station (Table 3). Similarly, the number of heavy precipitation days (R10) at the same station shows an increasing (insignificant) trend from baseline (− 0.20 mm/year) to a significant increasing trend in NF (0.29 mm/year) for RCP4.5, but it decreases (insignificant) in MF (− 0.15 mm/year) under RCP8.5.

Furthermore, some stations show decreasing trends in extreme precipitation indices from baseline to NF and then to MF under both RCP scenarios. For example, consecutive dry days (CDD) as well as CWD decrease (insignificant) at the Chame station from baseline (2.45 days/year, 0.91 days/year) to NF and MF (0.24 days/year, − 0.30 days/year) under RCP4.5 and RCP8.5. Simple daily intensity index (SDII) also shows decreasing trend from the baseline (0.08 days/year) to MF (0.02 days/year), but it does not show any trend during MF under RCP4.5.

3.2.2 Projected changes in future climatic extreme indices w.r.t. baseline

Projected changes in temperature-based extremes

Changes in the model-based extreme climatic indices at the four meteorological stations Thakmarpha (index:604), Khudi (index:802), Gorkha (index:809), and Chame (index:816) of the Marshyangdi basin for NF (2014–2033) and MF (2034–2053) against the historical period (1983–2013) for RCP4.5 and RCP8.5 scenarios are shown in Appendix Table 8 and Table 4, respectively. Temperature-based extreme indices for both scenarios are projected to increase more in MF compared with NF, with higher magnitude of changes being projected in RCP8.5 than in RCP4.5. For example, TNx increases at all the stations gradually from NF to MF but at the Chame station, its magnitude was the highest at NF (15.9%) as well as at MF (23.6%) for RCP4.5. In contrast, TXx decreases at all the stations except during MF at the Chame and Khudi, respectively.

Table 4 Projected changes in future climatic extreme indices (based on ensemble time series) for RCP8.5 scenarios across four stations in the Marshyangdi watershed
Full size table

Likewise, for RCP4.5 (Appendix Table 8), summer days (SU25) gradually increases at two stations only, Khudi and Gorkha. In case of RCP8.5 (Table 4), it increases for both NF and MF at Khudi, only during NF at Gorkha and only during MF at Chame, with the highest average values (360.4%). Furthermore, index TXx at Chame is the highest for RCP4.5 (1.9%) and RCP8.5 (4.4%) in MF but decreases at the rest of the three stations for both RCPs except at Khudi.

In overall, trends in hot nights (TN90p, significant), warm days (TX90p; insignificant at 1 station), Max Tmin (TNx, significant), hot days (TXx; insignificant), and WSDI are projected to increase from baseline to future under both RCP scenarios at all the stations. On the contrary, cool nights (TN10p; significant), cool days (TX10p; significant), and CSDI (insignificant) are projected to decrease from baseline to future under both RCP scenarios. The diurnal temperature range (DTR) is also projected to decrease (insignificant) in future periods for both RCP scenarios.

Projected changes in precipitation-based extremes

Precipitation-based extreme indices are projected to decrease in the future at most of the stations for both RCPs, but with varying magnitudes (Appendix Table 8; Table 4). For example, SDII is projected to decrease at all the stations for both the RCPs. Also, CDD are projected to decrease at all the stations for future scenarios for both RCPs and CWD is projected to increase. PRCPTOT is projected to increase gradually at most of the stations from NF to MF for RCP4.5 with the highest percentage increase during MF (16%) at Thakmarpha. However, for RCP8.5, PRCPTOT increases only during NF at the three stations (except Gorkha during MF) with the highest percentage increase during NF at Thakmarpha (15.4%). Importantly, R99p at two stations Thakmarpha and Khudi is projected to increase from NF to MF for both the RCP scenarios whereas at Chame only for MF. Likewise, RX1day is projected to increase widely (from − 1.8 to 8.4%) at Thakmarpha but decrease from 27.4% in NF to 24.8% in MF at Chame. Other intense precipitation indices like RX5day are projected to decrease at all the stations (except in NF at Thakmarpha) for both the RCP scenarios. The R10, on the other hand, is projected to increase with higher percentages in NF in comparison with MF at Khudi and Gorkha stations and decrease at other two stations with varying percentages. Furthermore, R20 for RCP4.5and RCP8.5 is projected to increase only at the Khudi station albeit with different magnitudes.

The changes in the annual occurrence of summer days (SU), annual occurrence of tropical nights (TR), number of heavy precipitation days (R10), and number of very heavy precipitation days (R20) can have profound impacts on various sectors including ecosystems, as elaborated in literatures such as Alexander et al. (2006). And increase in projected warm temperature indices (warm days and nights (WSDI)) and the corresponding decreases in cool temperature indices (cool days, cool nights (CSDI)) have been observed all over Nepal under both RCP scenarios (4.5, 8.5) for the future (MoFE 2019).

3.2.3 Spatial distribution of climatic extremes

The spatial distribution in projected changes of the extreme climatic indices for RCP4.5 and RCP8.5 from the baseline and future periods (i.e., NF and MF) at four meteorological stations are discussed hereunder. As expected, different indices show varying degree of variation across the stations in terms of magnitude and direction of trends. Some indices may have some distinct trends from upstream to downstream while some may not. This section discusses spatial variation in selected set of indices.

Trends in monthly minimum of daily maximum temperature (TXn) are increasing significantly at all the stations in baseline (expect no significant trend at a station located in the middle of the watershed) and MF (both the scenarios), increasing insignificant trends in MF under both RCP scenarios (significant at middle of the watershed), and decreasing trends when towards north under both RCPs and scenarios (Fig. 3a). The rate of increase is higher from south towards north in the watershed.

Fig. 3
figure 3

Spatial distribution in trends across the stations a TXn and b TN90p

Full size image

Consecutive dry days (CDD) show distinct increasing trends from baseline to near and mid-future (significant trends towards north for baseline and NF4.5). However, the Gorkha station does not show any trends during baseline and NF8.5 while at some stations, insignificant decreasing trends are depicted (Appendix Fig. 6). In case of CWD, the trends for baseline as well as all future periods and scenarios are also increasing insignificantly at all the stations, except the case for MF under RCP4.5 (mixed trends) and insignificantly decreasing trends at all the stations in MF8.5 scenarios as well as at Thakmarpha and Khudi for NF both scenarios (Appendix Fig. 7).

Trends in warm days (TX90p) are increasing insignificantly at three stations (except Khudi located in the southern part of the watershed) during the baseline. However, it shows significant increasing trends at all the stations in NF for both RCPs except at Khudi where the trend was significantly decreasing and at Gorkha showing no distinct trend. Similar increasing trends were also observed across all the stations for mid-future for both RCP scenarios (significant at MF8.5; Appendix Fig. 8). Similar historical (or baseline) trends were also observed by other studies in different parts of Nepal (e.g., Baidya et al. 2008; Shrestha et al. 2016a; Rajbhandari et al. 2017).

Trends in warm nights (TN90p) are significantly increasing in northern mountainous part of the watershed from the baseline to NF and MF under both scenarios. However, it shows significant decreasing trends in baseline at the Chame station while insignificant decreasing trend at Thakmarpha (Fig. 3b). Trends in cold nights (TN10p) are decreasing at all the stations for all future periods and scenarios; however, mixed for baseline, with increasing trends at the stations located in the northern mountainous part of the watershed (Appendix Fig. 9). These results are consistent with the result of Panday et al. (2015) and Rajbhandari et al. (2017) for the Eastern Himalaya.

Trends in extreme wet days (R99p) show insignificant mixed trends for baseline and NF under RCP4.5 scenarios while it shows insignificant decreasing trends at most of the stations for NF8.5 (except Gorkha) and MF in both scenarios (expect at Khudi) (Appendix Fig. 10). In case of very wet days (R95p), the trends are also increasing insignificantly at all the stations for future except at that at baseline, it shows mixed trends (Appendix Fig. 11).

3.3 Trends in hydrological extremes

3.3.1 Change point analysis

Change point analysis (CPA), a powerful statistical technique to determine abrupt changes in a time series (Chang and Byun 2012), was used to identify pre- and post-impact periods for assessing hydrological extremes. In this study, the change points were identified for annual maximum (Tmax), minimum (Tmin), and average (Tav) temperatures at four stations; annual precipitation at nine stations; and annual flow time series at one hydrological station in the Marshyangdi watershed. Results are tabulated in Appendix Table 9. The change points for various variables occurred at different years, some with statistical significance and some without. For example, change point for Tmax and Tmin at the Thakmarpha station is identified in 2002 (significant) and 2000 (insignificant). Similarly, at the Khudi station, change points for Tmax, Tmin, and precipitation are detected in 1992 (insignificant), 1986 (significant), and 1990/2004 (insignificant), respectively. In case of river discharge, CP in time series is detected in 1999 (insignificant). The CP at other stations are reported in Appendix Table 9. Following the CP detected in discharge time series as 1999, as well as significant CP for the precipitation at Manang Bhot, we have taken this year to separate pre- and post-impact periods in the IHA tool. Literatures (e.g., Khadka and Pathak 2016) also report flood causalities in the Barpak village of Gorkha district, located within the watershed, during the same year the CP was detected. However, no specific mechanism for the CP could be identified.

3.3.2 Changes in flow characteristics after change point

The median value, degree of deviation, and degree of alteration for the IHA parameters characterizing five groups of extreme flow regimes at Bimalnagar hydrological station in the Marshyangdi watershed are listed in Table 5 and elaborated hereunder. Percentage of deviation (P), degree of hydrological alteration (HA), and overall hydrological alteration were calculated using Eqs. (1), (2), and (3) detailed in Appendix 1. Values of hydrological alterations for each indicator are shown in Fig. 4. An overall mean hydrological alteration of all the 32 parameters is estimated as low, with a value of 27.9%, whereas alterations of 32 parameters within five groups vary widely as elaborated in following sub-sections.

Table 5 Degree of hydrological alterations of IHA parameters at Bimalnagar hydrological station in Marshyangdi watershed
Full size table
Fig. 4
figure 4

Hydrological alterations (values shown in bars) for all 33 IHAs

Full size image

Alterations in magnitude of monthly streamflow

As depicted from Table 5 and plotted in Appendix Fig. 12, the monthly median parameters increased from pre- to post-impact period, and the increases in monthly median values under high RVA category, except in January and March, have been observed indicating an insignificant increase of the frequency of observed values than the upper RVA limit. The degree of deviation (P) is negative only for the January and March out of 12 months (Table 5), suggesting that streamflow has been increased from pre- to post-impacted period. Calculation of degree of HA for monthly stream flow shows that monthly stream flows fall within the category of low alteration (D < 33%), except in the month of January where the alteration is moderate (33% < D < 67%). In a nutshell, median values among the Group-1 IHAs show low hydrological alteration (24.4%).

Alterations in annual extreme flow conditions

Analysis of median values of degree of deviation and degree of hydrological alteration for the annual extreme flow conditions (11 IHAs under Group-2 and 2 IHAs under Group-3) reveal that degree of deviation is highest (16.1%) for a 90-day maximum, followed by 30-day maximum (12.9%) (Fig. 5) whereas it decreases in 1-day, 3-day, and 90-day minimum extreme flow parameter from pre- to post-impact period. This suggests a possible increase in flood magnitude, which may have been both beneficial as well as harmful effects depending on channel morphology, types of substrate, depth, and other geomorphological characteristics (Stefanidis et al. 2016). Furthermore, increase in 1-day, 3-day, and 7-day maximum flow causes change in the floodplains due to dominant particle size of bed materials inducing ecological implications like low oxygen and prolongation of duration of stressful high temperatures (Graf 2006). HA indicates low degree of alteration (< 33%) among the indices; however, for the 3-day maximum, it shows moderate alteration (Table 5). Overall, the degree of hydrological alteration for Group-2 is 16.8%, indicating the low hydrological alteration (D < 33%).

Fig. 5
figure 5

Annual extreme flows—flow value (left) and degree of deviation (right)

Full size image

In case of timing of annual minimum extremes (i.e., indicators under Group-3), the degree of deviations (%) is negative, i.e., timing of 1-day minimum is moving backward from the 69th day to 73rd (delayed by 5 days); however, 1-day maximum is also moving forward from the 207th day to 208th day from pre- to post-impact period. Thus, lagging of Julian date of minimum streamflow indicates that annual minimum values will appear early in the year threating the riverine environment (Xue et al. 2017). The hydrological alterations for timing of 1-day minimum (Jmax) and 1-day maximum (Jul-min) are identified as moderate and low category, respectively (Table 5). The overall degree of hydrological alteration of IHAs under Group-3 is low with a value of 29.3% (33% < D < 67%). Thus, observed shift in occurrence of low flows implies earlier drying up of the downstream channel, which may have adverse consequences on the flood plain habitats, ecology, and navigability of a river (Sharma et al. 2019).

Alterations in frequency and duration of high and low flow pulses

Among the Group-4 parameters, the frequency of low (25th percentile) pulse count increases from the pre- to post-impact period; however, the and high (75th percentile) pulse counts does not show any change, while the duration of high and low pulse counts decreases from the pre- to post-impact period (Table 5). The degree of deviation (5, Appendix Fig. 13) for the low pulse counts is 25% which falls under high RVA category (Fig. 4). Degree of hydrological alteration for low pulse count is characterized as “moderate (M)” but for other three parameters under Group-4 are characterized as “low (L).” Thus, four parameters under Group-4, altogether, show low degree of hydrological alteration (16%). The increase in the low pulse counts may cause frequent dry and wet situations, thus, potentially worsening the ecological development of the Marshyangdi river floodplain. However, the low alteration in the high pulse count as well as duration may not favor the riverine ecosystem due to the limited nutrients availability for plants along the riverbank affecting the promotion of river biodiversity (Xue et al. 2017). Thus, low pulse count may induce geomorphic implications like the prolongation of a channel and bank stability, which increased frequency of depositional regimes in the channels. Concomitantly associated ecological implications include stress for plants due to the changes in frequency and magnitude of soil moisture, which causes anaerobic condition and may lack availability of floodplain for aquatic organisms (Graf 2006).

Alterations in rate and frequency of flow conditions

The Group-5 parameters, altogether, exhibit moderate hydrological alteration of 54.4%. However, the hydrological alteration varies across the parameters; showing highest alteration in reversals (79.9%) followed by fall rate (55.7%) (Table 5). The increase in rise rate and fall rate in the post-impact period, suggests that the rate of change from high flow to low flow conditions and vice versa would be accelerated. It implies early arrival of peak streamflow in the downstream channel (Sharma et al. 2019), which corresponds well to the results of backward shifting of timing of 1-day maximum as discussed earlier. Number of reversals (number of times that flow switches from one type of period to another) of streamflow conditions which indicate change from rising water condition to falling water condition and vice versa decreases from pre- to post-impact period (Table 5), indicating low intra-annual fluctuations in water conditions of the downstream channel. Higher degree of deviation (53.2%) for the fall rate compared with deviations in number of reversals (− 3.3%) indicates high hydrological alteration for the number of reversals and moderate alteration for fall rate, respectively. This increase in rise and fall rate indicates the rise in abruptness in streamflow. The accelerated rise and fall rate, which indicates the rise in abruptness of streamflow, could trap aquatic organisms in floodplains and strand terrestrial organisms on floodplain island The Nature Conservancy (2009). This may affect the stability of plant and animal habitat (Xue et al. 2017). Based on RVA results, most of the parameters in groups 1 and 2 as well as indicators related to low pulse count (group 4) and group 5 parameters increased under high RVA category, which reflects increase in frequency of observed values than the upper RVA limit (Fig. 4).

3.3.3 Trends in extreme hydrological indices

We analyzed trends in 15 hydrological extreme indices and tabulated results (i.e., trends, average value of indices, and inter-annual variability) in Appendix Table 11. Fifteen selected hydrological extreme indices having varying direction and magnitude of trends. Some indices such as 3-, 7-, and 30-day maximum flows show increasing (statistically insignificant) annual trends of + 7.5 m3/s/year, 4.4 m3/s/year, and 5.5 m3/s/year, respectively; whereas 1-day and 90-day maximum flows have increasing (statistically significant) trends of 0.4 m3/s/year and 4.8 m3/s/year, respectively. However, 3-, 7-, 30- and 90-day minimum flows have insignificantly decreasing trend (Appendix Table 11).

Like climate extreme indices, hydrological extremes also have inter-annual variability in the index value, as shown in Appendix 2, Table 11 for 1-day maximum flow as an example. The index has an average value of 1287.21 m3/s with a trend (statistically significant) of + 0.36 m3/s/year during1987–2015; however, the index value varies from 679 to 2270 m3/s for the aforementioned period, with a coefficient of variation of 0.30. It is therefore important to note those variabilities too while using the results for informed decision-making on managing hydrological extremes.

4 Conclusions

This study assessed hydro-climatic extremes in the Marshyangdi watershed for historical as well as future periods at four climatic stations and one hydrological station. An ensemble of three regional climate models (RCMs) were used to project future climate and assess projected trends in the climatic extremes under RCP4.5 and RCP8.5 scenarios for two future periods, namely, near future and mid-future. Key conclusions specific to the study of watershed, based on analysis of the results, are listed hereunder. Though it reflects some aspects of the Himalayan watersheds, we need to study more on watersheds to generalize the conclusions for the entire Himalayan region.

  • Climatic extremes over the historical period (1983–2013) indicate hotter and drier conditions in the watershed, albeit with varying amount and significance (statistical) across the stations. Temperature-related indices such as WSDI and TX90p have increasing trends whereas TN90p and TX10p have decreasing trends. Heavy precipitation indices such as R10, R20, RX1day, and R99p show increasing trends for almost half of the 10 stations where CDD are increasing at most of the stations.

  • Future is projected to be wetter and warmer in the Marshyangdi watershed. Significant increase in trends for maximum temperature-related extremes (e.g., TN90p, TX90p, and TNx) and significant decrease for minimum-temperature-related extremes (e.g., TN10p, TX10p) from baseline to future periods under both the scenarios and future periods are considered. However, precipitation extremes such as R99, RX5day, CWD, and CDD are projected to decrease, with potential implications on water availability and its distribution across seasons.

  • Overall hydrological alterations as an indication of hydrological extremes is estimated as low (HA = 27.9%) in the Marshyangdi watershed. Increase in the median flow values especially during the period of March–August and consequent increase in the 30- and 90-day maximum values indicates the possibility of flood in the basin. On the other hand, increase in rise rate and decreases in the fall rate represent abruptness in the streamflow and inter-annual variability. Further, projected increase in climate change along with anthropogenic influences may affect the natural flow regime of the Marshyangdi watershed which may exacerbate in the future and the implications pointed out could have severe ecological consequences with the high degree of hydrological alteration

The watershed spans from the mountains in the north to the plains in the south, and the climatic extremes are analyzed at multiple stations spanning from north to south. The results therefore could be indicative of extremes at particular physiographic regions at other basins in Nepal as well. Like other snow-fed and glacierized catchments, which are highly influenced by climate change and associated melting of snow/glacier (Mingjie et al. 2013), projected increase in average annual precipitation at all the stations and higher increase in temperature in the Marshyangdi watershed and associated melting of snow/glacier may lead to an increase in water availability. As there are many hydropower projects as well as agricultural areas in the watershed, increase in water availability can lead to positive implications, if that could be harnessed properly. However, increase in average annual rainfall is associated with increase in number of heavy and very heavy precipitation days too, which means, available rainfall will be highly skewed over certain periods in a year. If excess precipitations are not stored in watersheds, hydrological extremes may increase in magnitude as well as frequency, which is clearly evident in the analysis of hydrological alteration indicators in this study. The increase in hydrological extremes means potential increase in loss and damage in the watershed. These results imply that investment in various water storage mechanisms, such as soil water storage in watershed, rainwater harvesting at household and community levels, storage in aquifers, and storage in river corridor itself is required to make productive use of excess rainfall and runoff and at the same time reduce potential losses and damages associated with hydrological extremes. Therefore, analyzing historical as well as future climatic extremes together with hydrological extremes are valuable to get a bigger picture for designing interventions for water resources development and management in a holistic way.