Abstract
The conceptual rainfall–runoff (HBV model) is applied to evaluate impacts of future climate changes on the hydrological system of the Richmond River catchment, Australia. Daily observed rainfall, temperature and discharge and long-term monthly mean potential evapotranspiration from the hydro-meteorological stations within the catchment over the period 1972–2014 were used to run, calibrate and validate the HBV model before the simulation. Future climate signals were extracted from a multi-model ensemble of eight global climate models (GCMs) of the CMIP5 under three scenarios (RCP2.6, RCP4.5 and RCP8.5). The calibrated HBV model was forced with the downscaled rainfall and temperature to simulate future streamflow at catchment outlet for the near-future (2016–2035), mid (2046–2065) and late (2080–2099) 21st century. A baseline run, with baseline climate period 1971–2010, was used to represent current climate status. Almost all GCMs’ scenarios predict slight increase in annual mean rainfall during the beginning of the century and decrease towards the mid and late century. Modelling results also show positive trends in annual mean streamflow during the near-future (13–23%), and negative trends in the mid (2–6%) and late century (6–16%), under all scenarios compared to the baseline-run. Findings could assist in managing future water resources in the catchment.
INTRODUCTION
Water scarcity resulting from climate change impact is a growing problem, especially in areas experiencing precipitation reduction and temperature increase. Water scarcity could also result from the continuous economic and population growth and the enlargement in industry and agriculture fields which consume high quantities of water (Chartres & Varma 2010; Liu et al. 2017). The combined impact of climate change and human development significantly affects the availability of future water resources. Since the mid-1970s, a noticeable climate shift in many parts of Australia has resulted in rising temperature and decreasing rainfall trends across the continent (Barron et al. 2011). Accordingly, this has badly impacted the available local water resources in terms of quantity and quality. Change in hydrological behaviour across many Australian local catchments has been reported by many researchers (Chiew et al. 1995, 2009; Bari et al. 2010; McFarlane et al. 2012; Silberstein et al. 2012; Cheng et al. 2014; Islam et al. 2014; Al-Safi & Sarukkalige 2017a, 2017b, 2017c). Also, Pittock (2003) and CSIRO & Australian Bureau of Meteorology (BoM) (2007) reported that the rainfall trends in most parts of south-eastern Australia are expected to decline as a result of climate change. Southeast Australia has experienced a prolonged dry spell similar to the one that began around 1970 in the south-west of Western Australia (SWWA) (Indian Ocean Climate Initiative 2002; Power et al. 2005). Furthermore, most of Australia's population and agricultural activities are extremely concentrated in the south-east of the continent (Murphy & Timbal 2008). Thus, the problem of below average rainfall calls for special attention from the hydrologic research community to formulate an efficient water resources management to overcome the problem of water deficiency in the region.
Hydrological modelling is a common procedure used in most climate change impact studies to simulate the future discharge at a catchment scale. It is also used to predict the combined impact of climate change and other components on the hydrological status of a catchment. Normally, the hydrological simulation requires weather predictions to simulate the future streamflow. Future climate series of temperature and rainfall can be extracted from the analysis of global climate model (GCM) results. Zorita & Von Storch (1999) and Solomon et al. (2007) pointed out that GCMs represent a suitable data source to extract the regional and global future climate signals. However, the spatial resolution of the GCM outputs is too coarse to be used directly in catchment scale hydrological modelling and needs to be downscaled before the modelling process (Fowler et al. 2007). A plethora of hydrologic research studies in different environments across the world has been conducted to address the problem of climate change and its influence on future water demands (e.g., Kundzewicz et al. 2007; Praskievicz & Chang 2009; Whitehead et al. 2009; Driessen et al. 2010). In Australia, a wide range of hydrological studies with a diversity of GCMs, warming scenarios and hydrologic models have warned of the inevitable decline in future rainfall and streamflow trends in many parts of the continent (Charles et al. 2010). In short, the concerns regarding low future water availability in many parts of Australia need to be carefully addressed to achieve a consistent and sustainable water management to meet the future water demands in the continent.
The climate data of the Coupled Model Intercomparison Project phase 5 (CMIP5) (Taylor et al. 2012) and the Hydrologiska Byrans Vattenbalansavdelning (HBV) conceptual model (SMHi 2012) have been widely used in climate change impact assessment studies around the world, at catchment and global scales. For instance, Sperna Weiland et al. (2013) investigated the impact of climate change informed by a multi-model ensemble of the CMIP5 on the extremes streamflow of the Meuse River Basin. They utilized the HBV model to formulate the future streamflow of the river which provides a comprehensive insight into the variations in discharge extremes. Bouaziz et al. (2014) assessed the impact of future climate changes on the annual water yield of the Rhine River basin. They also forced the HBV model with the downscaled climate outputs extracted from an ensemble of 31 GCMs of CMIP5 to simulate the future discharge of the river. Szépszó et al. (2014) and Photiadou et al. (2016) also utilized the HBV lumped-parameter model forced by a multi-model ensemble of the CMIP3 and CMIP5 to study the hydrological response of the Rhine River basin under the impacts of future climate change. In light of the present study, a multi-model ensemble of eight CMIP5-CGMs, which effectively represents the Australian future climate, is incorporated into the HBV hydrological model for a comprehensive climate impact assessment research.
The key objective of the current work is to investigate the hydrological response of the Richmond River catchment in New South Wales (NSW) to the future climate change impact using the hydrological modelling procedure. There are many motivations behind the selection of the study area for this research project. First, the catchment comprises popular tourist places such as Ballina, and second, it supports a continuously growing population attracted by the region's coastal lifestyle. Furthermore, it holds extensive agricultural and wetlands which consume high quantities of water. Hence, assessing the impact of future climate change on the hydrological system of the catchment is highly beneficial to formulate efficient and sustainable water management strategies in the area. A multi-model ensemble of eight GCMs of the CMIP5 which belongs to the Intergovernmental Panel on Climate Change Fifth Assessment Report (i.e., IPCC-AR5) (Stocker 2014) under three representative concentration pathways (RCPs) (2.6, 4.5 and 8.5) was used to represent the global-scale future climate signals across the catchment. The LARS-WG5.5 stochastic weather generator was employed to extract the local-scale future daily rainfall and temperature from each GCM of the eight CMIP5-GCMs by incorporating the daily observed climate data. The ensemble mean of the daily downscaled climate data was then derived and used to force the HBV conceptual model to simulate the future daily streamflow at Casino gauging station on Richmond River for three periods including the start (2016–2035), middle (2046–2065) and end (2080–2099) of the 21st century. Daily future streamflow forecasting can provide a comprehensive image about the availability of future water resources in the catchment. Thus, the outcomes of this research study could deliver effective water management policies for the study area to overcome the problem of low water accessibility in the future.
STUDY AREA: RICHMOND CATCHMENT
The Richmond catchment, with an approximate area of 7,000 km2, is located in the distant north part of NSW, Australia. It extends from the Border Ranges in the north to the Richmond Ranges in the west and south with a variable elevation range between more than 1,000 m above sea level near the Border Ranges to a few metres above sea level near the coastal floodplain. The catchment includes a diversity of natural sceneries such as world heritage, rainforest, agricultural lands and coastal estuaries. The area also comprises popular tourist places such as Ballina and supports a continuously growing population attracted by the region's coastal lifestyle. In the present work, the area upstream to the Casino gauging station (Figure 1) will be taken into consideration. It extends over an approximate drainage area of 1,790 km2 and stretches from latitude 28.22°–29.05°S and longitude 152.15°–153.15°E. The catchment has Mediterranean climatic conditions with a relatively warm dry summer, approximately ranged between 27 and 30°C and a moderate winter, ranged between 19 and 20°C (CSIRO & BoM 2007). The period between November and April holds the peak rainfall, which is approximately ranged between 1,350 and 1,650 mm/year in the catchment's coastal areas. However, the interior areas receive the lowest amount of precipitation, which is under 800 mm/year at Armidale (CSIRO & BoM 2007).
Richmond River catchment with the hydro-meteorological stations (Source:Department of Primary Industries, Water, NSW 2016).
Richmond River catchment with the hydro-meteorological stations (Source:Department of Primary Industries, Water, NSW 2016).
METHODOLOGY
Observed data
The selected study area has a good continuous hydro-meteorological record for 43 years (1972–2014) from seven weather stations and one discharge station (Table 1). The locations of the hydro-meteorological stations are illustrated in Figure 1. The recorded data were obtained from the Australian Bureau of Meteorology, and the quality of data has been checked as a high priority. Daily observed mean values of rainfall and temperature and the high-quality streamflow record from Casino gauging station on Richmond River were used to calibrate and validate the HBV model prior to the streamflow prediction. In addition, the observed average monthly mean values of potential evapotranspiration (PET) from the Tabulam (Muirne) weather station were also included in the hydrological simulation. The average areal precipitation over the catchment was obtained from a Thiessen polygon method. Rainfall and temperature data are further regionalized vertically over the catchment using gradients reflecting the changes in rainfall and temperature with elevation.
Locations of the hydrological and meteorological stations
. | Station no. . | Elevation (m) . | Latitude (S°) . | Longitude (E°) . | Observed parameter(s) . |
---|---|---|---|---|---|
Meteorological stations | |||||
Bentley | 58078 | 29 | 28.78 | 153.11 | Rainfall |
Green Pigeon | 58113 | 210 | 28.47 | 153.09 | Rainfall |
Loadstone | 58141 | 160 | 28.41 | 152.98 | Rainfall |
Old Bonalbo | 57085 | 290 | 28.57 | 152.59 | Rainfall |
Tabulam post office | 57018 | 130 | 28.89 | 152.57 | Rainfall |
Tabulam (Muirne) | 57095 | 555 | 28.76 | 152.45 | Rainfall, temperature and evapotranspiration |
Murwillumbah | 58158 | 8 | 28.34 | 153.38 | Temperature |
Hydrological stations | |||||
Richmond River – Casino | 203004 | 20 | 28.86 | 153.05 | Discharge |
. | Station no. . | Elevation (m) . | Latitude (S°) . | Longitude (E°) . | Observed parameter(s) . |
---|---|---|---|---|---|
Meteorological stations | |||||
Bentley | 58078 | 29 | 28.78 | 153.11 | Rainfall |
Green Pigeon | 58113 | 210 | 28.47 | 153.09 | Rainfall |
Loadstone | 58141 | 160 | 28.41 | 152.98 | Rainfall |
Old Bonalbo | 57085 | 290 | 28.57 | 152.59 | Rainfall |
Tabulam post office | 57018 | 130 | 28.89 | 152.57 | Rainfall |
Tabulam (Muirne) | 57095 | 555 | 28.76 | 152.45 | Rainfall, temperature and evapotranspiration |
Murwillumbah | 58158 | 8 | 28.34 | 153.38 | Temperature |
Hydrological stations | |||||
Richmond River – Casino | 203004 | 20 | 28.86 | 153.05 | Discharge |
Future climate data
Climate scenarios derived from the coupled atmosphere–ocean general circulation models can be used in line with the process-based models for global- and local-scale impact assessment studies. GCMs climate projections always involve uncertainties that result from using different climate scenarios (Fu et al. 2007). Therefore, an ensemble analysis combining multiple GCM projections and quantifying the probability of future climate is usually adopted to produce more reliable future regional climate change scenarios. For the present study, the global-scale future climate signals (monthly mean outputs) were extracted from a multi-model ensemble of eight GCMs of the CMIP5 under three RCPs (RCP2.6, RCP4.5 and RCP8.5). Next, the global-scale monthly outputs of rainfall and temperature from each GCM were downscaled into local-scale daily climate projections (point-specific data) suitable for regional impact assessment studies by using the LARS-WG5.5. The ensemble mean of the eight GCMs was then derived and adopted for streamflow simulation. Table 2 provides a detailed description of the eight GCMs incorporated into the multi-model ensemble. The CSIRO and the Australian Bureau of Meteorology explained that these models represent the best eight GCMs out of 40 GCMs of the CMIP5 model, that have been selected according to specific criteria to effectively represent Australian future climatic conditions (CSIRO & BoM 2015). The basis for selecting these models as the best among the CMIP5 models can be found in (https://www.climatechangeinaustralia.gov.au/en/support-and-guidance/faqs/eight-climate-models-data/). RCPs define plausible future green house gas concentration trajectories, and associated socio-technological development, leading to radiative forcing of 2.6, 4.5 and 8.5 W/m2 at the end of the 21st century (Moss et al. 2010). The future data have spanned the 21st century into three continuous periods, start (2016–2035), mid (2046–2065) and late (2080–2099). A baseline climatic period of 40 years (1971–2010) was also extracted from the multi-model ensemble to be compared with the observed climate.
The eight CMIP5 GCMs of the IPCC AR5 integrated into the LARS-WG in the present study
CMIP5 model ID (name) . | Institute . | Atmosphere resolution (km) . |
---|---|---|
ACCESS1.0 | CSIRO-BOM, Australia | 210 × 130 |
CanESM2 | CCCMA, Canada | 310 × 310 |
CNRM-CM5 | CNRM-CERFACS, France | 155 × 155 |
GFDL-ESM2M | NOAA, GFDL, USA | 275 × 220 |
CESM1-CAM5 | NSF-DOE-NCAR, USA | 130 × 100 |
HadGEM2-CC | MOHC, UK | 210 × 130 |
MIROC5 | JAMSTEC, Japan | 155 × 155 |
NorESM1-M | NCC, Norway | 275 × 210 |
CMIP5 model ID (name) . | Institute . | Atmosphere resolution (km) . |
---|---|---|
ACCESS1.0 | CSIRO-BOM, Australia | 210 × 130 |
CanESM2 | CCCMA, Canada | 310 × 310 |
CNRM-CM5 | CNRM-CERFACS, France | 155 × 155 |
GFDL-ESM2M | NOAA, GFDL, USA | 275 × 220 |
CESM1-CAM5 | NSF-DOE-NCAR, USA | 130 × 100 |
HadGEM2-CC | MOHC, UK | 210 × 130 |
MIROC5 | JAMSTEC, Japan | 155 × 155 |
NorESM1-M | NCC, Norway | 275 × 210 |
Calculation of PET
The HBV hydrological model
The Integrated Hydrological Model System (IHMS 6.3, HBV 7.3 version) (SMHI 2012) is used in the present study to investigate the hydrologic response of the Richmond catchment under future climate conditions. The HBV model can be classified as a semi-distributed conceptual rainfall–runoff model of catchment hydrology which uses daily rainfall, air temperature and monthly PET as input data to simulate the daily streamflow at catchment outlet (Singh 1995; Seibert 2005). The purpose of adopting the HBV model in the present study was due to two main incentives. First, the simplicity of the input data and the robust and flexible model structure have demonstrated the reliable performance of the model in solving water resource problems (SMHI 2012). Second, daily streamflow prediction offers a comprehensive idea about the hydrological changes and the status of future water resources in the study area. This could help decision-makers to formulate efficient water management strategies for the Richmond River catchment. Lindström et al. (1997) reported that the HBV model had proven its applicability in many regions around the world with a diversity of climatic conditions. In addition to hydrological forecasting, the model was also effectively used in many water resources-related fields such as data quality control, broadening of runoff records, estimation of missing data and groundwater simulation (SMHI 2012). More information about the HBV model can be found in Seibert (2005) and SMHI (2012).
Model structure and parameter description
The principal structure and parameters of the HBV model when applied on catchments without snow (Lidén & Harlin 2000).
The principal structure and parameters of the HBV model when applied on catchments without snow (Lidén & Harlin 2000).
The Richmond River catchment can be considered as a snow-free area, therefore, the precipitation routine will be characterized by the rainfall only. Soil moisture routine, which provides a hint about the water content in the soil, can be represented by three main parameters: field capacity (FC), beta (β) and the limits of potential evaporation (LP) (Abebe et al. 2010). FC refers to the extreme soil storing capacity of the catchment. The parameter β governs the relative participation of rainfall to the volume of runoff for a specified soil dampness deficiency. The parameter LP governs the shape of the PET curve. The surplus water of the soil moisture routine is transformed through the response routine to be released into catchment storage through two connected boxes ( and
). These boxes are connected together by a filtration rate (PERC) in which water percolates from the
to the
at a constant proportion (Abebe et al. 2010). The channel flow hydraulics (runoff) can be described by the transformation function parameter (MAXBAZ) which calculates the simulated outflow from the catchment.
RESULTS AND DISCUSSION
HBV model calibration, validation and parameter estimation
To calibrate and validate the HBV model, the daily observed discharge data should be available for a given period (in line with the observed climate) with a variety of hydrological regimes. Vaze et al. (2010) reported that the recent discharge record from the south-eastern Australian catchments could be used effectively to calibrate the rainfall–runoff models to represent the current prolonged drought across the region and to predict the future climate change impact on the local catchments. Daily observed streamflow data from Casino gauging station was available for 43 years (1972–2014). Before calibrating the model, the whole catchment area was divided into five elevation zones of different areal fractions to lapse rainfall and temperature with elevation (using the parameters pcalt and tcalt). It was proposed that rainfall will increase by 10% and temperature will decrease by 0.61°C with each 100 m elevation increment (Seibert 2005; SMHI 2012). Since there was only one gauging station with high-quality streamflow records available in the study area, the whole catchment was treated as a single spatial unit during the calibration and validation periods as well as during the simulation of future streamflow. The HBV model was first run for an initial state of one year (1972–1973) to initialize the system. Next, the manual calibration was used to calibrate and validate the model for the periods 1973–2000 and 2001–2014, respectively. Driessen et al. (2010) suggested that long-term calibration periods of hydrological models could be useful to simulate large data sets of future scenarios. Therefore, a calibration period of two-fold the validation period is utilized in this study.
Eleven parameters are included in the calibration and validation processes (Table 3). The modelling performance was evaluated using the three criteria of efficiency described in Equations (3)–(5). A satisfactory modelling performance was acquired during the calibration and validation processes (Table 4) which indicates that the model can be used effectively to simulate the future discharge in the catchment. Figure 3 displays a graphical comparison between the observed and simulated discharges at Casino gauging station for the calibration and validation periods. Through visual inspection of Figure 3, it can be seen that the simulated discharge fairly captured the observed discharge for the calibration and validation periods.
HBV model parameters and their optimal values resulting from calibration and validation periods
Parameter . | Symbol . | Unit . | Optimal value . |
---|---|---|---|
Rainfall correction factor | rfcf | – | 1.1 |
Elevation correction factor for precipitation | pcalt | 1/100 m | 0.1 |
Temperature lapse | tcalt | °C/100 m | 0.6 |
Maximum of soil moisture zone | FC | mm | 500 |
Limit for potential evaporation | Lp | – | 0.5 |
Shape coefficient | Beta | – | 1.5 |
General correction factor for potential evaporation | ecorr | – | 0.8 |
Recession coefficient for upper response box | Khq | 1/day | 0.8 |
Recession coefficient for lower response box | K4 | 1/day | 0.1 |
Maximum percolation capacity | Perc | mm/day | 3 |
Routing parameter | Maxbaz | day | 1 |
Parameter . | Symbol . | Unit . | Optimal value . |
---|---|---|---|
Rainfall correction factor | rfcf | – | 1.1 |
Elevation correction factor for precipitation | pcalt | 1/100 m | 0.1 |
Temperature lapse | tcalt | °C/100 m | 0.6 |
Maximum of soil moisture zone | FC | mm | 500 |
Limit for potential evaporation | Lp | – | 0.5 |
Shape coefficient | Beta | – | 1.5 |
General correction factor for potential evaporation | ecorr | – | 0.8 |
Recession coefficient for upper response box | Khq | 1/day | 0.8 |
Recession coefficient for lower response box | K4 | 1/day | 0.1 |
Maximum percolation capacity | Perc | mm/day | 3 |
Routing parameter | Maxbaz | day | 1 |
HBV model performance during the calibration and verification periods
Process . | NSE . | VE % . | r2 . |
---|---|---|---|
Calibration | 0.92 | 4.0 | 0.84 |
Validation | 0.90 | 4.8 | 0.81 |
Process . | NSE . | VE % . | r2 . |
---|---|---|---|
Calibration | 0.92 | 4.0 | 0.84 |
Validation | 0.90 | 4.8 | 0.81 |
Calibration (a) and validation (b) results at Casino gauging station on Richmond River.
Calibration (a) and validation (b) results at Casino gauging station on Richmond River.
Data downscaling (LARS-WG5.5)
The spatial and temporal scales of the outputs resulting from the multi-model ensembles are too coarse to be applied directly for local-scale impact assessment studies. Thus, GCM data need to be downscaled to a finer scale before using them as input to the process-based impact models. For the present study, the Long Ashton Research Station Weather Generator version 5.5 (LARS-WG5.5) (Semenov & Stratonovitch 2010), is used to extract the local-scale daily rainfall and temperature from each GCM of the eight CMIP5-GCMs for the start, mid and late periods of this century as well as the baseline period (1971–2010). LARS-WG5.5 is a statistical downscaling model (Wilks & Wilby 1999) used to generate local-scale daily weather data required for climate change impact studies. Semenov & Barrow (1997) explained that the magnitude and periodic sequence of the main climate features had been effectively simulated by the LARS-WG model. This downscaling technique provides a cross-validation for the generated data, which has significantly improved the simulation of extreme weather events (Semenov & Stratonovitch 2010). Accordingly, it has been successfully applied in many local impact assessment studies in diverse climates and has proven its applicability and its high performance, where bias corrections or any other adjustments are not required (Semenov & Stratonovitch 2010; Gunawardhana et al. 2015).
The process of weather generation using the LARS-WG5.5 includes three key stages (Semenov & Barrow 2002):
Model calibration: the model analyses the daily observed weather parameters (rainfall, min and max temperature and solar radiation) of a specified location during a baseline period to determine their statistical characteristics. Then, it creates a set of calibrated probability distribution parameters for that site to be stored in two parameter files.
Model validation: the created parameter files are then used to generate synthetic climate data having the same statistical characteristics as the original observed data. The validity of the model is examined by comparing the statistical characteristics of the observed and synthetic data to evaluate the suitability of the LARS-WA to simulate future weather data for that site. Then, the model calculates relative change factors for each month considering the data in the baseline and GCMs projections.
Climate scenarios’ generation: the relative change factors are then used with the calibrated parameters to produce daily climate scenarios for the site in consideration that are compatible with the GCM projections and the observations (Wilks & Wilby 1999).
Therefore, by using the average daily observed weather parameters, from seven stations, in line with the monthly scale climate outputs resulting from each GCM, the LARS-WG can produce a local-scale daily climate series which is statistically similar to the CMIP5 climate projections. By treating each GCM prediction from the CMIP5 ensemble as an equally possible evolution of climate, we can explore the uncertainty in impact assessment resulting from the uncertainty in climate predictions.
The model utilizes a semi-empirical probability distribution (SED) to estimate probability distributions of dry and wet series of daily climate parameters (Semenov & Barrow 2002). SED is defined as a separate histogram that has a constant number of intervals of flexible lengths. The wet days are expressed as the days with precipitation of more than 0.0 mm. The LARS-WG5.5 uses 23 intervals to describe the shape of the SED compared to the ten intervals of the earlier versions (Semenov & Stratonovitch 2010). This offers diverse distributions of weather statistics (rainfall and temperature) to be simulated more accurately. The simulation of daily temperature statistics (min and max) is governed by the status of the day whether it is wet or dry. A good record of daily observed weather (minimum of 20 years) is required to obtain robustly site-specific weather parameters which are used later to produce the synthetic future data (Semenov & Barrow 1997). In the current study, 40 years (1972–2011) of observed daily rainfall, min and max temperature from seven weather stations (sites) are utilized as a baseline period to create the site-specific weather parameters. Next, the grid climate outputs of each GCM which are covering the catchment were incorporated with the average site parameters to generate catchment scale future daily time series of rainfall and temperature. Finally, the ensemble mean of the local-scale climate outputs was derived and used to force the calibrated HBV model to simulate the future daily streamflow at Casino gauging station on Richmond River.
Performance of the LARS-WG5.5 model (calibration and validation)
Before generating the future rainfall and temperature climate series, the ability of the LARS-WG model to capture the observed climate signals should be checked. As mentioned earlier, 40 years (1972–2011) of observed daily rainfall, minimum and maximum temperature were used to calibrate and validate the LARS-WG model. The modelling performance was assessed by relating the probability distributions of the generated (synthetic) climate data with those resulting from the observations. For the rainfall time series, two characteristics were used including monthly mean precipitation and standard deviation (Figure 4). While for the temperature time series, the min and max monthly mean statistics were taken into account (Figure 5). In addition, the Kolmogorov–Smirnov (K–S) test was performed to compare the seasonal probability distributions of the wet/dry period lengths (Table 5). The K–S goodness-of-fit test was also adopted to assess the equality of the daily distributions of rainfall, min and max temperature calculated from the observed and simulated data series (Tables 6 and 7). The test computes a p-value which gives an indication of the possibility that the observed and generated data sets may have come from the same distribution. A very low p-value (corresponding to a high K–S value) indicates that the synthetic data belong to a distribution different from that of the observed climatic data, and therefore, should be rejected. While a large p-value means that the differences between the observed and generated climate statistics for the variable in consideration are too small and there is no indication to reject the generator. Semenov & Barrow (2002) recommended that a p-value of 0.01 can be used as the acceptable significance limit of the model results.
Kolmogorov–Smirnov test results for seasonal wet/dry series distributions
Season . | Event . | No. of tests (N) . | K-S test . | P-value . | Assessment . |
---|---|---|---|---|---|
Dec/Jan/Feb | Wet | 12 | 0.08 | 1.000 | Perfect fit |
Dry | 12 | 0.218 | 0.644 | Good fit | |
Mar/Apr/May | Wet | 12 | 0.037 | 0.992 | Perfect fit |
Dry | 12 | 0.041 | 1.000 | Perfect fit | |
Jun/Jul/Aug | Wet | 12 | 0.154 | 0.741 | Good fit |
Dry | 12 | 0.031 | 0.994 | Perfect fit | |
Sep/Oct/Nov | Wet | 12 | 0.051 | 1.000 | Perfect fit |
Dry | 12 | 0.084 | 1.000 | Perfect fit |
Season . | Event . | No. of tests (N) . | K-S test . | P-value . | Assessment . |
---|---|---|---|---|---|
Dec/Jan/Feb | Wet | 12 | 0.08 | 1.000 | Perfect fit |
Dry | 12 | 0.218 | 0.644 | Good fit | |
Mar/Apr/May | Wet | 12 | 0.037 | 0.992 | Perfect fit |
Dry | 12 | 0.041 | 1.000 | Perfect fit | |
Jun/Jul/Aug | Wet | 12 | 0.154 | 0.741 | Good fit |
Dry | 12 | 0.031 | 0.994 | Perfect fit | |
Sep/Oct/Nov | Wet | 12 | 0.051 | 1.000 | Perfect fit |
Dry | 12 | 0.084 | 1.000 | Perfect fit |
Kolmogorov–Smirnov test results for daily rainfall distributions (in each month)
Month . | No. of tests (N) . | K-S test . | P-value . | Assessment . |
---|---|---|---|---|
Jan | 12 | 0.111 | 0.980 | Perfect fit |
Feb | 12 | 0.109 | 0.989 | Perfect fit |
Mar | 12 | 0.175 | 0.860 | Very good fit |
Apr | 12 | 0.049 | 1.000 | Perfect fit |
May | 12 | 0.051 | 1.000 | Perfect fit |
Jun | 12 | 0.172 | 0.863 | Very good fit |
Jul | 12 | 0.210 | 0.667 | Good fit |
Aug | 12 | 0.162 | 0.914 | Perfect fit |
Sep | 12 | 0.235 | 0.519 | Good fit |
Oct | 12 | 0.141 | 0.960 | Perfect fit |
Nov | 12 | 0.049 | 0.987 | Perfect fit |
Dec | 12 | 0.120 | 0.999 | Perfect fit |
Month . | No. of tests (N) . | K-S test . | P-value . | Assessment . |
---|---|---|---|---|
Jan | 12 | 0.111 | 0.980 | Perfect fit |
Feb | 12 | 0.109 | 0.989 | Perfect fit |
Mar | 12 | 0.175 | 0.860 | Very good fit |
Apr | 12 | 0.049 | 1.000 | Perfect fit |
May | 12 | 0.051 | 1.000 | Perfect fit |
Jun | 12 | 0.172 | 0.863 | Very good fit |
Jul | 12 | 0.210 | 0.667 | Good fit |
Aug | 12 | 0.162 | 0.914 | Perfect fit |
Sep | 12 | 0.235 | 0.519 | Good fit |
Oct | 12 | 0.141 | 0.960 | Perfect fit |
Nov | 12 | 0.049 | 0.987 | Perfect fit |
Dec | 12 | 0.120 | 0.999 | Perfect fit |
Kolmogorov–Smirnov test results for distributions of the daily min and max temperature (in each month)
Month . | No. of tests (N) . | Daily min temperature . | Daily max temperature . | ||||
---|---|---|---|---|---|---|---|
K–S test . | P-value . | Assessment . | K–S test . | P-value . | Assessment . | ||
Jan | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Feb | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Mar | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Apr | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
May | 12 | 0.106 | 0.998 | Perfect fit | 0.105 | 0.999 | Perfect fit |
Jun | 12 | 0.053 | 1.000 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Jul | 12 | 0.053 | 1.000 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Aug | 12 | 0.053 | 1.000 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Sep | 12 | 0.158 | 0.913 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Oct | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Nov | 12 | 0.105 | 0.999 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Dec | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Month . | No. of tests (N) . | Daily min temperature . | Daily max temperature . | ||||
---|---|---|---|---|---|---|---|
K–S test . | P-value . | Assessment . | K–S test . | P-value . | Assessment . | ||
Jan | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Feb | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Mar | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Apr | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
May | 12 | 0.106 | 0.998 | Perfect fit | 0.105 | 0.999 | Perfect fit |
Jun | 12 | 0.053 | 1.000 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Jul | 12 | 0.053 | 1.000 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Aug | 12 | 0.053 | 1.000 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Sep | 12 | 0.158 | 0.913 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Oct | 12 | 0.106 | 0.998 | Perfect fit | 0.106 | 0.998 | Perfect fit |
Nov | 12 | 0.105 | 0.999 | Perfect fit | 0.053 | 1.000 | Perfect fit |
Dec | 12 | 0.106 | 0.998 | Perfect fit | 0.053 | 1.000 | Perfect fit |
A comparison between the observed and generated rainfall time series.
A comparison between the observed and generated temperature time series.
Figures 4 and 5 present well that the simulated rainfall and temperature statistics strongly agree with those of the recorded statistics. Table 5 also demonstrates the good performance of the LARS-WG model in simulating the seasonal distributions of the wet and dry spells. Also, the daily distributions of rainfall, min and max temperature (in each month) (Tables 6 and 7) verify the excellent modelling performance. It can be seen that all p-values in Tables 6, 7 and 8 are more than 0.01 (i.e., 99% confidence level) and the results of the assessment columns ranged between good and perfect fit. This could be attributed to the fact that the LARS-WG generates random data which are comparable to the observed data in their statistical properties only. Furthermore, the high-quality observed climate data of the baseline period could also be a reason for the reasonable agreement between the observed and synthetic climate series. It is also found that the performance of the LARS-WG was satisfactory through the t and F tests which compared the mean and variance values of two time series. The seasonal distributions of the wet/dry spells in line with the daily rainfall, min and max temperature distributions are very important when using the model results in impact assessment studies (Osman et al. 2014). As these properties were perfectly fitted, the calibrated parameters of the LARS-WG can be incorporated properly with the RCP scenarios to generate the future rainfall and temperature series for climate impact assessment in the Richmond River catchment.
Overview of annual mean sums of rainfall (P), temperature (T) and potential evapotranspiration (PET) across the Richmond catchment (from seven weather stations) for the observed, baseline and future periods
Variable . | Observed 1972–2014 . | Baseline period (1971–2010) . | 2016–2035 . | 2046–2065 . | 2080–2099 . | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|
RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | |||
P (mm/year) | 1,209 | 1,180 | 1,240 | 1,213 | 1,233 | 1,178 | 1,175 | 1,155 | 1,145 | 1,080 | 1,117 |
T (°C) | 17.5 | 18.0 | 17.7 | 17.9 | 17.9 | 18.8 | 18.9 | 18.6 | 19.9 | 19.4 | 19.1 |
PET (mm/year) | 1,553 | 1,601 | 1,614 | 1,646 | 1,647 | 1,674 | 1,694 | 1,671 | 1,738 | 1,696 | 1,683 |
Changes in annual mean values relative to the observations | P (%) | +2.6 | +0.3 | +2.0 | −2.6 | −2.8 | −4.5 | −5.3 | −10.7 | −7.6 | |
T (°C) | +1 | +2.3 | +2.3 | +7.4 | +8 | +6.3 | +14 | +11 | +9 | ||
PET (%) | +4 | +6 | +6 | +8 | +9 | +8 | +12 | +9 | +8 |
Variable . | Observed 1972–2014 . | Baseline period (1971–2010) . | 2016–2035 . | 2046–2065 . | 2080–2099 . | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|
RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | RCP 8.5 . | RCP 4.5 . | RCP 2.6 . | |||
P (mm/year) | 1,209 | 1,180 | 1,240 | 1,213 | 1,233 | 1,178 | 1,175 | 1,155 | 1,145 | 1,080 | 1,117 |
T (°C) | 17.5 | 18.0 | 17.7 | 17.9 | 17.9 | 18.8 | 18.9 | 18.6 | 19.9 | 19.4 | 19.1 |
PET (mm/year) | 1,553 | 1,601 | 1,614 | 1,646 | 1,647 | 1,674 | 1,694 | 1,671 | 1,738 | 1,696 | 1,683 |
Changes in annual mean values relative to the observations | P (%) | +2.6 | +0.3 | +2.0 | −2.6 | −2.8 | −4.5 | −5.3 | −10.7 | −7.6 | |
T (°C) | +1 | +2.3 | +2.3 | +7.4 | +8 | +6.3 | +14 | +11 | +9 | ||
PET (%) | +4 | +6 | +6 | +8 | +9 | +8 | +12 | +9 | +8 |
All the RCPs values represent the ensemble mean of eight GCMs.
Future climate projections
The annual mean values of rainfall, temperature and PET for the observed, baseline and future periods across the catchment are presented in Table 8. Almost all GCMs predict a reduction in annual mean rainfall (Figure 6(b) and 6(c)) and an increase in temperature for all future scenarios during the mid and late century compared to the observations. While, for the start of the century, the downscaled ensemble mean of the eight GCMs shows a slight increase in rainfall amounts (Figure 6(a)) and a rise in temperature values relative to the recorded climate.
Annual mean variations of rainfall (as a mean of the six weather stations) for the observed, baseline and the climate scenarios of future periods: (a) RCP2.6, (b) RCP4.5 and (c) RCP8.5. The simulated rainfall is the ensemble mean of eight GCMs.
Annual mean variations of rainfall (as a mean of the six weather stations) for the observed, baseline and the climate scenarios of future periods: (a) RCP2.6, (b) RCP4.5 and (c) RCP8.5. The simulated rainfall is the ensemble mean of eight GCMs.
Compared to the observed period (1972–2014), rainfall is projected to increase slightly during the start of the 21st century with an annual mean increment of 2.6%, 0.3% and 2.0% for the RCP8.5, RCP4.5 and RCP2.6, respectively. By the mid-century, the annual mean rainfall is projected to decline by 2.6%, 2.8% and 4.5% under the RCP8.5, RCP4.5 and RCP2.6, respectively. Towards the end of the century, the average decline in the annual mean rainfall is anticipated to reach 5.3%, 10.7% and 7.6% for the same scenarios, respectively. The decline of rainfall amounts during the mid and late century could be attributed to the lack of high-intensity rainfall events. For instance, the 90th rainfall percentile for the mid and late 21st century show a negative decline ranged between 5% and 15% (Figure 7(b) and 7(c)). The maximum rainfall values are also expected to decline during the mid and late century with a range of 13–21% (Figure 7(b) and 7(c)) compared to the observations. On the other hand, annual mean temperature values show positive trends for all climate scenarios during the future periods compared to the observed period (1972–2014) (Table 8). This anticipated rise in temperature values will lead to an increase in the annual mean PET by approximately 6%, 9% and 12% for the start, mid and late century across the catchment (Table 8). A possible interpretation for this increment in the future PET values is the use of the modified Blaney–Criddle method which depends directly on the daily mean temperature to derive the PET. As the daily mean temperature is expected to rise in the future, additional energy is available for driving soil water and intercepted water for evaporation or transpiration. Consequently, the combined impact of rainfall reduction and PET increase by the mid and late century could adversely affect the future streamflow across the catchment.
Annual statistics of the 10th, 50th and 90th rainfall percentiles (as a mean of the six weather stations) for the observed, baseline and future periods: (a) for the start, (b) for the mid and (c) for the late century. The simulated rainfall is the ensemble mean of eight GCMs.
Annual statistics of the 10th, 50th and 90th rainfall percentiles (as a mean of the six weather stations) for the observed, baseline and future periods: (a) for the start, (b) for the mid and (c) for the late century. The simulated rainfall is the ensemble mean of eight GCMs.
Future streamflow projections
Future streamflow at Casino gauging station was simulated for start, mid and late periods of the current century by forcing the calibrated HBV model with the downscaled data resulted from the ensemble mean of the eight GCMs. The calibrated HBV model was also forced with the downscaled climate data from the baseline period (1971–2010) to simulate the streamflow at the same station for a baseline run. The differences between the two simulations represent the impact of climate change on the hydrological system. Vaze et al. (2010) pointed out that the rainfall–runoff models calibrated over a period of more than 20 years could be used effectively in impact assessment studies, on condition that the annual mean rainfall in the simulated period should not be more than 15% drier or 20% wetter than the calibration period. As the simulated future annual mean rainfall over the catchment is within the above limits relative to the observed annual mean rainfall over the 28-year calibration periods (Table 8), then the calibrated HBV model can be used efficiently for catchment-scale impact assessment. The annual mean streamflow statistics for the observed, baseline and future periods are presented in Table 9. It clearly shows that for the Richmond River catchment, the response to the anticipated climate change is obvious through the decline in the future annual mean streamflow at Casino gauging station for the mid and late century. Further, Figure 8 illustrates the variations of the annual mean streamflow at Casino gauging station for the observed, baseline and future periods. Figure 9 shows the Q10, Q50 and Q90 annual streamflow percentile statistics for the same periods.
A summary of the annual mean streamflow statistics for the observed, future and baseline run periods at Casino gauging station (m3/s)
Variable . | Observed (1972–2014) . | Baseline period run (1971–2010) . | 2016–2035 . | 2046–2065 . | 2080–2099 . | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|
RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | |||
Q min | 0.6 | 4.5 | 5.2 | 5.6 | 5.8 | 4.1 | 4.2 | 4.3 | 3.9 | 3.8 | 4.0 |
Q10 | 5.4 | 11.0 | 11.8 | 12.7 | 13.9 | 8.8 | 8.8 | 9.8 | 8.9 | 7.9 | 8.9 |
Q50 | 11.9 | 17.7 | 19.0 | 19.5 | 20.8 | 17.3 | 16.8 | 17.0 | 15.2 | 13.4 | 15.8 |
Q90 | 46.6 | 32.9 | 33.1 | 33.0 | 35.0 | 27.4 | 28.4 | 29.4 | 28.1 | 27.7 | 27.7 |
Q max | 60.0 | 42.0 | 46.0 | 47.0 | 46.0 | 35.0 | 36.0 | 36.0 | 37.0 | 36.0 | 36.0 |
Q mean | 19.8 | 17.9 | 20.6 | 20.8 | 22.1 | 17.1 | 17.4 | 17.8 | 16.7 | 15.2 | 17.1 |
Changes in the annual mean values relative to the baseline run (%) +increase, − decrease | Q min | +16 | +24 | +29 | −9 | −7 | −4 | −13 | −16 | −11 | |
Q10 | +7 | +15 | +26 | −20 | −20 | −11 | −19 | −28 | −19 | ||
Q50 | +7 | +10 | +18 | −2 | −5 | −4 | −14 | −24 | −11 | ||
Q90 | +1 | +1 | +6 | −17 | −14 | −11 | −15 | −16 | −16 | ||
Q max | +10 | +12 | +10 | −17 | −14 | −14 | −12 | −14 | −14 | ||
Q mean | +15 | +16 | +23 | −4 | −3 | −1 | −7 | −15 | −4 |
Variable . | Observed (1972–2014) . | Baseline period run (1971–2010) . | 2016–2035 . | 2046–2065 . | 2080–2099 . | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|
RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | RCP 2.6 . | RCP 4.5 . | RCP 8.5 . | |||
Q min | 0.6 | 4.5 | 5.2 | 5.6 | 5.8 | 4.1 | 4.2 | 4.3 | 3.9 | 3.8 | 4.0 |
Q10 | 5.4 | 11.0 | 11.8 | 12.7 | 13.9 | 8.8 | 8.8 | 9.8 | 8.9 | 7.9 | 8.9 |
Q50 | 11.9 | 17.7 | 19.0 | 19.5 | 20.8 | 17.3 | 16.8 | 17.0 | 15.2 | 13.4 | 15.8 |
Q90 | 46.6 | 32.9 | 33.1 | 33.0 | 35.0 | 27.4 | 28.4 | 29.4 | 28.1 | 27.7 | 27.7 |
Q max | 60.0 | 42.0 | 46.0 | 47.0 | 46.0 | 35.0 | 36.0 | 36.0 | 37.0 | 36.0 | 36.0 |
Q mean | 19.8 | 17.9 | 20.6 | 20.8 | 22.1 | 17.1 | 17.4 | 17.8 | 16.7 | 15.2 | 17.1 |
Changes in the annual mean values relative to the baseline run (%) +increase, − decrease | Q min | +16 | +24 | +29 | −9 | −7 | −4 | −13 | −16 | −11 | |
Q10 | +7 | +15 | +26 | −20 | −20 | −11 | −19 | −28 | −19 | ||
Q50 | +7 | +10 | +18 | −2 | −5 | −4 | −14 | −24 | −11 | ||
Q90 | +1 | +1 | +6 | −17 | −14 | −11 | −15 | −16 | −16 | ||
Q max | +10 | +12 | +10 | −17 | −14 | −14 | −12 | −14 | −14 | ||
Q mean | +15 | +16 | +23 | −4 | −3 | −1 | −7 | −15 | −4 |
All the RCPs values represent the ensemble mean of eight GCMs.
Annual mean streamflow variations of future scenarios relative to the observed and baseline periods: (a) RCP2.6, (b) RCP4.5 and (c) RCP8.5 emission scenarios. The simulated streamflow is the ensemble mean of eight GCMs.
Annual mean streamflow variations of future scenarios relative to the observed and baseline periods: (a) RCP2.6, (b) RCP4.5 and (c) RCP8.5 emission scenarios. The simulated streamflow is the ensemble mean of eight GCMs.
The 10th, 50th and 90th streamflow percentiles (annual statistics) at Casino gauging station for the observed, baseline and the scenarios of future periods: (a) for the start, (b) for the mid and (c) for the late century. The simulated streamflow is the ensemble mean of eight GCMs.
The 10th, 50th and 90th streamflow percentiles (annual statistics) at Casino gauging station for the observed, baseline and the scenarios of future periods: (a) for the start, (b) for the mid and (c) for the late century. The simulated streamflow is the ensemble mean of eight GCMs.
Future streamflow at Casino gauging station shows considerably varied tendencies compared to the baseline run, as illustrated in Table 9. For the near-future period (2016–2035), all scenarios have annual mean rainfall values that are significantly bigger than those of the baseline period (Table 8). Therefore, all streamflow statistics are projected to increase relative to the baseline simulation. The minimum flows (Q min and Q10) show relatively higher positive trends, ranged between 7% and 29%, than the maximum flows (Q max and Q90) which have a range of 1–12%. A possible explanation for this phenomenon is the natural behaviour of the HBV model which sometimes over- or underestimates the minimum and maximum flows. The conceptual structure of the HBV model is relatively simple with only one single groundwater storage responsible for the runoff generation. The median (Q50) and annual mean (Q mean) flows also show positive trends ranged from 7% to 18% and 15% to 23% increase, respectively. This could be explained as a consequence of the relative increase in the annual mean rainfall during the start of the century.
By the mid century, all climate scenarios show relatively less annual mean rainfall and higher PET compared to the baseline period (Table 8). Hence, all streamflow statistics measured at Casino discharge station show decline tendencies relative to the baseline simulation. The minimum flows are expected to decrease with a range of 4–20%. The same is applicable for the maximum flows which ranged between 11% and 17%. The annual mean streamflow is also anticipated to decline by 4%, 3% and 1% under the RCP2.6, RCP 4.5 and RCP8.5 climate scenarios, respectively. Similarly, towards the end of the century, all streamflow statistics are projected to decline under all scenarios relative to the baseline run. This could be attributed to the fact that the annual mean rainfall under all climate scenarios is expected to decline substantially relative to the baseline period, and the PET is also projected to increase as a result of the relative rise in the annual mean temperature (Table 8). The declining percentage of the minimum and maximum flows is expected to range between 11% and 28% and 12% and 16%, respectively. While the decline in the annual mean streamflow is projected to reach 7%, 15% and 4% under the scenarios RCP2.6, RCP4.5 and RCP8.5 compared to the baseline simulation. The RCP4.5 climate scenario shows a higher reduction in the annual mean streamflow because it corresponds to the higher annual rainfall reduction (8%) relative to the baseline period (Table 8). Therefore, the combined impact of rainfall reduction and PET increase across the catchment by the mid and late century could be considered as the main cause of the streamflow reduction tendencies.
The trend analysis of the observed annual mean streamflow at Casino gauging station (Figure 10(a)) revealed a decreasing tendency over the time. The long-term annual streamflow has almost declined to half, from 25 m3/s to 12 m3/s, since the early 1970s until 2014. The trend analysis confirmed evidence of changes in hydrological responses consistent with observed changes in rainfall over the past decades (Figure 10(b)). Hence, in addition to the global warming, the projected decline in the future streamflow could also be credited to the natural climate variations. Based on the above analysis, the results specify that the potential changes in streamflow due to global warming could be very significant. Therefore, assessing the impacts of climate change on the hydrological system of the Richmond River catchment is highly beneficial since it may influence the seasonal or long-term water availability. As the Richmond catchment holds an increasing population growth in line with the highly intensive agricultural lands and tourist places, the expected streamflow reduction will badly impact the future water resources in the area. Thus, long-term development plans in the area should take into consideration the potential future climate change. This requires sustainable and efficient water management strategies to be applied in the catchment to overcome the problem of water scarcity. The outcomes of the present study could help the local community of the Richmond River catchment to manage the usage of future water resources in the region.
(a) Trend analysis of the annual mean streamflow at Casino gauging station and (b) annual mean rainfall trend analysis (from six weather stations).
(a) Trend analysis of the annual mean streamflow at Casino gauging station and (b) annual mean rainfall trend analysis (from six weather stations).
Based on the above outcomes, the current study also supports other studies which have been performed in other south-east Australian catchments and displayed an evident decline in future streamflow. For instance, Chiew et al. (2009) explained that the runoff trends across many catchments of south-east Australia are expected to decrease in the future. They utilized the conceptual modelling approach (SIMHYD rainfall–runoff model) forced by the downscaled climate data from an ensemble of GCMs to simulate the future streamflow at the study area. Cheng et al. (2014) reported that the future runoff is projected to decrease in Glendon Brook River catchment in the south-east of NSW. They also utilized the hydrological modelling procedure (WAVES eco-hydrological model) forced by the downscaled climate series informed by 12 GCMs to simulate the future discharge at the catchment outlet. In addition, the more recent studies by researchers of the CSIRO and the BoM have confirmed that the rainfall–runoff trends in most parts of south-eastern Australia are projected to decline through the mid and late 21st century (CSIRO & BoM 2015).
SUMMARY AND CONCLUSION
The future climate change impacts on the hydrological behaviour of the Richmond River catchment are evaluated for the start, mid and late periods of the 21st century under three future climate scenarios of low, medium and high emissions. The conceptual lumped-parameter HBV model is used to simulate the catchment hydrological behaviour. The model was calibrated and validated (with acceptable performance) by using the observed streamflow (1972–2014) in line with the observed weather data. The calibrated model was then forced by the downscaled rainfall and temperature to simulate the daily discharge at Casino gauging station for the baseline and future periods. Future climate signals were obtained from a multi-model ensemble of eight GCMs from the CMIP5 of the IPCC-AR5. The LARS-WG5.5 (a stochastic weather generator) is utilized to extract the local-scale future rainfall and temperature from each GCM of the eight GCMs ensemble. The model performed very well in capturing the observed rainfall and temperature climate statistics, and it can be used effectively to generate the daily future climate series for a catchment-scale impact assessment. Overall modelling results of the downscaled GCMs show that the rainfall is anticipated to increase slightly during the start of the 21st century and decrease during the mid and late century for all climate scenarios compared to the observations (1972–2014). PET is also projected to increase for all scenarios during the future periods as a result of the increase in annual mean temperature relative to the observed period. Compared to the baseline streamflow simulations, the annual mean streamflow measured at Casino gauging station is projected to increase for all scenarios during the start of the century with a range of 15–23% following an increase of 0.3–2.6% in the annual mean rainfall. By the mid-century, the annual mean streamflow is projected to decline under all climate scenarios with a range of 1–4% following a decline of 2.6–4.5% in the annual mean rainfall. Towards the end of the century, all scenarios projected a decline in the annual mean streamflow ranged between 4% and 15% following a reduction of 5.3–10.7% in the annual mean rainfall. This reduction in the annual streamflow will possibly badly impact the sufficiency of future surface water resources and influence the aquatic and environmental life of the Richmond River system.
The following conclusions from this investigation study could be drawn as below:
- 1.
The trend analysis of the annually observed streamflow confirmed evidence of changes in hydrological responses consistent with observed changes in climate over the past decades.
- 2.
The outcomes of this assessment study specify that the potential changes in streamflow due to global warming could be very significant.
- 3.
The study highlights the similar outcomes with other previous studies that have been conducted in many south-eastern Australian catchments and revealed noticeable rainfall–runoff reduction tendencies.
- 4.
The findings could assist the authorities and the community of the Richmond River catchment to manage the future water resources in the catchment taking into consideration the low flow situation.
- 5.
The results could also be significant to preserve the extensive wetland complexes in the lower Richmond River, such as Tuckean swamp on the Richmond floodplain and Ballina Nature Reserve which protects wide areas of mangroves and saltmarsh communities, from the risk of streamflow reduction.
ACKNOWLEDGEMENTS
The author(s) would like to acknowledge the financial support of the Higher Committee for Education Development in Iraq (HCED Iraq) Sponsorship to carry out this research study.