There is imperative need of robust basin-scale data for climate impact studies over the topographically varying and landuse heterogenous river basins in the Indian Himalayan Region (IHR). Even finer resolution regional climate models’ (RCMs) information is elusive for these purposes. Based on available model fields and corresponding in-situ observed fields, bias correction for precipitation over Upper Ganga River Basin (UGRB) and temperature over Satluj River Basin (SRB) is demonstrated. These chosen river basins are in central and western Himalayas, respectively. Model precipitation (temperature) field from RegCM4.7 (REMO) and corresponding observed precipitation (temperature) field from nine (eight) stations of UGRB (SRB) are considered. Empirical quantile mapping (inverse function method) method is used. It is seen that each model has a distinct systematic bias relating to both precipitation and temperature means with respect to their corresponding observed means. Applying bias correction methods to the model fields resulted in reducing these mean biases and other errors. These findings illustrate handling and improving the model fields for hydrology, glaciology studies, etc.
Bias correction using empirical quantile mapping (inverse function method) is employed on both precipitation and temperature model.
Applying bias correction methods to the model fields resulted in reducing the mean biases and other errors.
These findings illustrate handling and improving the model fields for hydrologist, glaciologist, etc.
The Himalayas is the source of many north Indian rivers, namely, Indus, Satluj, Ganges, Brahmaputra, etc. These rivers support millions of downstream populations which are dependent on agriculture and pasture usages, water resources for domestic usages, hydro-electric dams, industries, etc. These rivers are mainly fed by monsoonal precipitation and spring snowmelt from high elevation glaciers and the cryosphere (Eriksson et al. 2009; Bookhagen & Burbank 2010; Palazzi et al. 2013; Norris et al. 2015, 2017; Bannister et al. 2019). Most of the rivers originating from the western Himalayas are mainly fed by the Indian winter monsoon (Dimri et al. 2016) embedding western disturbances (Dimri et al. 2015); and rivers originating from the central and eastern Himalayas are mainly fed by the Indian summer monsoon (Maharana & Dimri 2019).
Most of the Indian Himalayan river basins have variable topography and heterogeneous landuse/landcover. These basins are data sparse as well. Usage of dynamical or statistical or statistical–dynamical downscaled model outputs at finer resolutions are greatly needed for climate impact studies for hydrological, glaciological and water resources (Akhtar et al. 2008; Narula & Gosain 2013; Sanjay et al. 2017). However, over river basins having complex topography and landuse, these finer resolution model inputs are still not sufficient as they have systematic biases due either to model physics or parameterization. Improving these model fields further by reducing these biases/errors is important. Choudhary & Dimri (2019) have discussed various bias correction methods, their strengths and weaknesses. Bannister et al. (2019) have carried out bias correction in two of the western Himalayan river basins using in-situ observations and made an assessment of precipitation extremes. These regions have strong precipitation and temperature gradients which are difficult to produce in a model environment and, at times, are elusive as well. Hence, representation or usage of these model fields directly for impact studies is not suitable or ideal. There are some studies on precipitation and temperature bias corrections over complex topographical mountainous regions (Christensen et al. 2008; Piani et al. 2010; Bordoy & Burllando 2013; Shrestha et al. 2017). Yet, these studies fall short in providing comprehension regarding the Indian Himalayan river basins.
In the present paper, we have shown precipitation bias correction over the Upper Ganga River Basin (UGRB), in central Himalayas, and Satluj River Bain (SRB), in western Himalayas.
Upper Ganga River Basin (UGRB)
The Ganga is the longest river of India and ranks among the world's top 20 rivers by amount of water discharge. The Ganga River Basin covers an area of ∼981,371 km2 shared by India, Nepal, China and Bangladesh. It originates at the confluence of the Bhagirathi and Alaknanda rivers, at Devprayag in Tehri Garhwal district of Uttarakhand State of India – within the mountainous region of the lower Himalayas. The river travels a distance of ∼2,520 km before merging with the Bay of Bengal. Throughout its length, a number of tributaries join it. The principal tributaries joining are the Yamuna, Ramganga, Ghaghra, Gandak, Kosi, Mahananda and Son. The Ganga River is divided into three distinct zones, i.e., Upper Ganga Basin, Middle Ganga Basin and Lower Ganga Basin.
The focus of the present study is on the UGRB – which is delineated and shown in Figure 1(a) using the 90 m SRTM digital elevation map. The total area of the UGRB is ∼21,698.05 km2. The basin has a significant amount of glacier cover including debris cover glaciers. The elevation in the UGRB ranges from 7,500 m in the Himalayan mountain region to 100 m in the lower plains. Around 60% of the basin is occupied by agriculture (main crop types include wheat, maize, rice, sugarcane, bajra and potato), while 20% is covered by forests, mostly in the upper mountainous regions, and approximately 2% of the mountain peaks are permanently snow covered. The annual average rainfall in the UGRB ranges between 550 and 2,500 mm (Bharati & Jayakody 2010), and a major part of the rain is due to the Indian summer monsoon.
Satluj River Basin (SRB)
The SRB, one of the main tributaries of the Indus river system, is also considered for the study. The river originates from the lakes of Mansarover and Rakastal in the Tibetan Plateau at an elevation of about 4,572 m. The Indian part of the SRB covers an area of ∼22,305 km2 (Figure 1(b). The topography varies from 350 m to as high as 6,735 m, and is shown with the 90 m resolution DEM from Shuttle Radar Topographic Mission (SRTM). About 57% of the basin area lies between 3,600 and 5,400 m altitude; only a small fraction of the area is above 6,000 m. The topographical setting and the abundant availability of water provide huge hydropower generation potential, and hence several hydropower schemes exist and/or are planned on this river. Bhakra Dam, the oldest dam in India, is situated on this river, in the foothills of the Himalayas. Due to the wide range of altitudes and precipitation patterns, a diverse climate is experienced in the basin. The lower part of the basin has tropical and warm temperate climate, whereas the middle part has a cold temperate climate. In the upper part, the climate is very cold and in the uppermost part, which is a perpetually frozen area (permafrost), the climate is similar to that of polar regions. The upper part of the basin mostly receives solid precipitation by the Indian winter monsoon embedding western disturbances, whereas the middle part receives both rain and snow and the lower part of the basin receives only rain. The average annual rainfall in the lower, middle and greater Himalayan basin regions is about 1,300, 700 and 200 mm, respectively (Singh & Kumar 1997).
METHOD: BIAS CORRECTION
The bias corrections are performed on the regional climate model (RCM) outputs. For precipitation field, RegCMver4.7 setup was used under three different land surface characterization schemes, namely, CONTROL, CLM4.5 and SUBGRID BATS (details are provided in the Supplementary material and Table S1). Corresponding model topography over the UGRB is shown in Figure 2. For temperature, CORDEX-REMO model outputs are chosen. (For model precipitation and temperature fields different RCMs are chosen. Deliberation and discussion on using and picking different models is beyond the scope of this paper and hence not provided. Please refer to Choudhary & Dimri (2019). In addition, there is limitation of the in-situ observations, hence over the UGRB precipitation, and over the SRB, temperature fields are considered).
Precipitation bias correction
The precipitation bias correction is performed using nine in-situ observations distributed across the UGRB (Figure 1(a) and Table 1). The data range and stations’ details are shown in the table. It is imperatively important to mention that precipitation variability over the varying topographic region is always a key issue as complex topographic regions within the model physics and processes thereof are not truly captured (Norris et al. 2017; Bannister et al. 2019).
Temperature bias correction
RESULTS AND DISCUSSION
In the following sections firstly, precipitation bias correction over UGRB is demonstrated using RegCMver 4.7 model precipitation field. It is followed by temperature bias correction over SRB using REMO model temperature field. (Model fields chosen for bias correction are from different RCMs. Deliberation and discussion on using and picking different model sets is beyond the scope of this paper and hence not provided. In addition, there is limitation of in-situ observations, hence over the UGRB precipitation, and over the SRB temperature fields are considered).
Precipitation bias correction over UGRB
Many authors (Ali et al. 2015; Li et al. 2016; Nepal 2016; Bannister et al. 2019) have worked on precipitation bias correction. However, it is important to improve upon and to include the role of model topography and associated physical processes to be captured in bias correction method. Here, precipitation percentage bias (PBIAS) is a measure of estimating the average tendency of the model precipitation field to be larger or smaller than the corresponding in-situ precipitation observation. Figure 3 illustrates this precipitation percentage bias at the nine stations in the UGRB. It is clearly seen that at six (three) stations, precipitation percentage bias is positive (negative). It corresponds that at six (three) stations average tendency of model precipitation field is larger (smaller) than the corresponding in-situ observation. It is interesting to note that as we move from lower to higher elevation stations, the measure of percentage bias tendency starts decreasing and finally reverses over the highest elevation stations. This corresponds to the role of precipitation forming mechanism and associated physical processes such that, over the lower elevation regions, in the model environment, leads to a drift away of the amount of precipitation from the actual than compared over the higher elevation regions. To further look into the precipitation and associated complexities over the mountainous region, wet periods (days) in logarithmic frequency distribution with bias correction at one of the station, Devprayag, is shown in three model simulations: CONTROL (Figure 4(a)), CLM4.5 (Figure 4(b)) and SUBGRID (Figure 4(c)). It is necessary to mention that in all the three simulation experiments, longer wet spell periods are not well captured.
Still employing bias correction method with corresponding in-situ observation at Devprayag, improvement in the wet spell periods, although in lower length, is distinctly seen. Lower wet spell period counts mainly have higher frequencies. To improve upon this further, quantile–quantile (q-q) bias correction method is employed upon the model daily precipitation field from three simulations – CONTROL, CLM4.5 and SUBGRID – using corresponding in-situ daily precipitation at nine stations in the UGRB and presented in Figure 5. Comparison on scatter distribution of model and in-situ precipitation fields shows higher deviation for higher intensity precipitation events. Comparison on q-q scale of non-bias corrected precipitation distribution shows comparatively better distribution along the slope of distribution. However, q-q scale of bias corrected precipitation fields shows best distribution along the slope of the distribution. This improved bias corrected model precipitation field is important for river basins like the UGRB. q-q distributions in figures illustrate distinct improvement in the model precipitation field after employing bias correction. It is quite clear that over a smaller river basin such efforts of bias correction will be needed for bias corrected model fields. Non-bias corrected data show that model precipitation fields have lower values than that of the corresponding observed precipitation field, which after employing the bias correction, shows that bias corrected model precipitation fields are closer to observed precipitation fields.
Temperature bias correction over SRB
To demonstrate bias correction on model temperature, fields from REMO models are chosen (not discussed in detail due to brevity being beyond the scope of the paper). Temperature bias correction is carried over the SRB where there is long-term (1961–2005) data at eight stations. Figure 6 shows the probability density function (pdf) of bias corrected model temperature field of REMO with the corresponding observations at eight stations (Bhakra, Kasol, Suni, Rampur, Kalpa, Namgia, Rakchham and Kaza) in the SRB during a historical time period (1961–2005). The figure represents the distribution of in-situ observed, model and bias corrected temperature field. All distributions are Gaussian distribution. Almost at all stations (vertically in different numbers of quantile used), distributed quantile mappings (DQMs) follow similar distributions which illustrates that the bias correction method employed is quite good at capturing the distribution of temperature. Pdf distribution at different stations are shown by different peaks and spread indicating the variabilities. Lower range of spread indicates the lesser variability among the temperature values. Also, the low probability of temperature indicates that the temperature has comparatively smoother distribution between its lowest and highest value leading to reduced mean. The flattened pdf at Kaza, which is a high-altitude station, shows lesser mean but higher limited variability between lowest and highest values. On comparison with the observed and model pdf of the temperature field, skewness (mostly towards left) and either decreased or increased mean values are seen. On employing the bias correction method, bias corrected temperature distributions are found closer to the corresponding observations. Table 3 illustrates improvement in mean characteristics after using DQMs at different numbers, thus highlighting its potential. Similar improved distribution is shown in the case of model statistics as well. Further skewness, which a measure of simple symmetry or no-symmetry of the normal distribution is assessed. In most of the cases, distribution is negatively skewed, except at the stations where mean is negative and positively skewed. After bias correction, the model temperature field distribution, skewness of the distribution at most of the stations reverts towards the skewness of the corresponding observed temperature field. In addition, in most of the cases, negative kurtosis is noticed in the model temperature field, which after bias correction is improved towards the corresponding observed temperature field distribution. Figure 7 shows the box-whisker plot distribution for the model, bias corrected and corresponding observed temperature field at eight stations (Bhakra, Kasol, Suni, Rampur, Kalpa, Namgia, Rakchham and Kaza) in the SRB. The line from lowest to highest temperature values represents the overall data length. The shaded box starts towards the lowest temperature values representing the first quartile while the shaded box ends towards highest temperature value representing the third quartile. The cross line shows the median value of each data set. The figure shows that distributed quantile method has a good capability to correct the model temperature fields. However, the division of data sets into different quantile numbers has a minor difference to bias corrected data sets. However, in all the cases, non-biased model temperature fields are closer to the corresponding observations once bias corrected. There mean as well as their spread is closer to the corresponding observations. Further, these model temperature fields and bias corrected temperature fields with corresponding observed temperature fields are shown in a scatter plot in Figure 8. The figure illustrates the scatter plot of the model temperature field before and after bias correction against the observed temperature at eight stations (Bhakra, Kasol, Suni, Rampur, Kalpa, Namgia, Rakchham and Kaza) in the SRB. The model temperature field underestimates the temperature as compared to the corresponding observed temperature field which is almost corrected after the bias correction. Regression line between observed and uncorrected data sets and the regression line between observed and corrected data sets is shown. Regression lines show distinct improvement upon the bias corrected field at all the eight stations and all improvements are also seen with increased elevations (elevation-dependent changes over the region are not discussed). These DQM methods improve the model data once bias correction is employed.
Correct estimate of precipitation and temperature field over any river basin is needed to assess hydrological, glaciological and water stress for the downstream habitat and, in addition, to have future assessments based on model inputs as well. In the present work, bias correction demonstration – for precipitation and temperature – is shown over two important river basins of northern India. The proposition of bias correction is to inculcate possible improvement in model fields based on available observation fields, so that model fields after bias correction can be used for other hydrological, glaciological, etc. studies. In addition, bias correction is a pre-requisite for application on model fields for impact studies. By doing so, it removes systematic errors in the model fields which arise either due to initial and boundary forcing provided to RCMs or inherent RCMs limitation of model physics and parameterization schemes.
Therefore, in the present study, bias correction methods are demonstrated over two of the river basins in central Himalayas (UGRB) and western Himalayas (SRB). After employing bias correction, distinct improvements are found in the model fields. It can be seen that precipitation bias correction over the UGRB and temperature bias correction over the SRB of model fields reduces the systematic errors and, thus, will be of extreme guiding help in using these improved data series for future climate impact studies.
The author acknowledges the Department of Science and Technology (DST), Govt. of India for funding project under National Mission for Sustaining Himalayan Ecosystem (NMSHE) in collaboration with National Institute of Hydrology (NIH), Roorkee, Uttarakhand. Data used in the present study are provided by NIH, Roorkee, Uttarakhand.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.