Studies on summer monsoon rainfall using Regional Climate Model PRECIS

Transcription

1 Studies on summer monsoon rainfall using Regional Climate Model PRECIS SMRC No. 22 SAARC METEOROLOGICAL RESEARCH CENTRE (SMRC) E-4/C, Agargaon, Dhaka-1207, BANGLADESH SMRC Publication AUGUST

2 Author: Dr. Md. Nazrul Islam Head, Synoptic Division SAARC Meteorological Research Centre (SMRC) Plot No. E-4/C, Agargaon, Dhaka-1207, Bangladesh Assisted by: Nazlee Ferdousi, Senior Research Assistant, SMRC, Dhaka Md. Ashfaqur Rahman, Senior Research Assistant, SMRC, Dhaka Md. Nazmul Ahsan, Research Assistant, SMRC, Dhaka Sk. Md. Abubakar Abdullah, Research Assistant, SMRC, Dhaka Address of Correspondence: Dr. Md. Nazrul Islam Head, Synoptic Division SAARC Meteorological Research Centre (SMRC), Plot No. E-4/C, Agargaon, Dhaka-1207, Bangladesh CC: and Associate Professor Department of Physics Bangladesh University of Engineering & Technology (BUET), Dhaka-1000, Bangladesh CC: Published by: SAARC Meteorological Research Centre (SMRC), E-4/C, Agargaon, Dhaka-1207, Bangladesh. Printed by: 1

3 INDEX Contents FOREWORD 3 PREFACE 4 List of Abbreviations 5 List of Figures 6 List of Tables 7 SUMMARY 8 1. Introduction 9 Page 2. Model description and Methodology Model Description Methodology Results and Discussion Rainfall in SAARC domain Calibration of Rainfall in Bangladesh Validation of Rainfall in Bangladesh Rainfall Projection for SAARC region in Rainfall Projection for Bangladesh in Conclusions 38 Acknowledgments 39 References 40 2

4 FOREWORD The SAARC is the most vulnerable region to climate change that affect agricultural production, crippling vital infrastructures, diminishing natural resources and limiting development options for the future in the region. According to the World Bank climate change expert, the poorest of the poor in South Asia are the most affected by climate change. Climate change is recognized as the greatest long-term threat to the SAARC region. The economic impact of climate change, rising food prices and assessment of food security are the key issues to address. Long-term planning is impossible without any idea of change of climate to be anticipated in the future. Climate models are the main tools available for developing projections of climate change. Rainfall is one of the most important climate parameters; especially summer monsoon rainfall, which controls the socio-economy of the SAARC region. The 13 th Governing Board meeting of SMRC recommended this research project in realizing the importance of this issue. I am happy to state that the project is completed within the stipulated time and SMRC is going to add one more publication in its series. The study has come up with calibration and validation of the rainfall in Bangladesh and the projection is made for the year The study may be extended for all member states in near future. I express my sincere thanks to Dr. B. R. S. B. Basnayake, Scientist, Theoretical Division of SMRC for critical review of the report upon request by the Editorial Board of the Centre. Arjumand Habib Director SMRC 3

5 PREFACE The 13 th Governing Board (GB) Meeting of SMRC recommended the research project entitled Studies on summer monsoon rainfall using Regional Climate Model PRECIS under the program of Theoretical Division. The project objective was for validation and development of future scenarios in South Asian domain. The objective of the project scientifically relevant to the present day context of climate change and its impact studies. The project was not pursued as the in-house capacity was not present to run PRECIS modeling system at the SMRC. Therefore the work was supposed to perform with the new professionals joined later in the year. The author joined the Synoptic Division of SMRC on 04 November 2007, the date of the 13 th GB meeting was held and upon the request of the Director of the Centre, the author had agreed to perform the job besides his own research projects. Generation of climate change scenarios is very much demandable in the present world for long-term planning and it is remarkably valuable for densely populated and poverty stricken SAARC region. The economy of the SAARC countries is mainly agriculture based and the agriculture is one of the most climate sensitive sectors. In this perception present study bears significance and continuation of such type of research work is indispensable for detail understanding of the climatology of SAARC region. The work may be considered as the entry point of the SMRC in using regional climate model for forecasting purposes. Author 4

6 List of Abbreviations AOGCM Atmosphere-Ocean General Circulation Model BMD Bangladesh Meteorological Department DEFRA Department for Environment, Food and Rural Affairs DFID Department for International Development ECHAM European Centre Hamburg Model LBC Lateral Boundary Conditions PRECIS Providing Regional Climate for Impact Studies RCM Regional Climate Model SAARC South Asian Association for Regional Cooperation SMRC SAARC Meteorological Research Centre SRES Special Report on Emission Scenarios TRMM Tropical Rainfall Measuring Mission UNDP United Nations Development Program JAN January FEB February MAR March APR April JUN June JUL July AUG August SEP September OCT October NOV November DEC December DJF December to February MAM March to May JJAS June to September ON October and November 5

7 List of Figures Fig. 1. The BMD observational site (plus mark) with elevation in m (below plus 16 mark). Fig. 2. PRECIS generated annual rainfall averages for Fig. 3. The same as Fig. 2 except for rainfall during monsoon period (JJAS). 18 Fig. 4. Seasonal distribution of rainfall obtained from Observation and PRECIS 19 model. Fig. 5. Annual rainfall (mm/day) derived from TRMM V6 3B42 (upper panel) 23 and PRECIS simulation (lower panel). Fig. 6. The same as Fig. 5 except for monsoon (JJAS) season. 24 Fig. 7. Annual rainfall (mm/day) distribution obtained from observation, model 25 simulation (Scenario), model projection (Estimated) and observed baseline period (Normal). Fig. 8 (a-g). Validation of PRECIS simulated rainfall (mm/d) over Bangladesh 29 for (a) 2000, (b) 2001, (c) 2002, (d) 2003, (e) 2004, (f) 2005 and (g) Fig. 9. Validation of PRECIS simulated annual rainfall with observed amount for 30 Bangladesh. Fig. 10. Time sequences of monthly rainfall (mm/d) obtained from model 31 simulation and rain-gauge data during for Bangladesh. Fig. 11. Scatter plot of monthly model rainfall (estimated) versus observed 32 rainfall at different months for Fig. 12. Projection of annual rainfall (mm/d) in the SAARC region for using PRECIS with the SRES A2 scenarios as the model input. Fig. 13. The same as Fig. 11 except for monsoon season (JJAS). 34 Fig. 14. Rainfall projection over Bangladesh for the year 2009 with normal. 38 6

8 List of Tables Table 1. Regression constants in different months and at different observational sites over Bangladesh. 20 Table 2. Regression slopes in different months and at different observational sites over Bangladesh. 21 Table 3. Validation of PRECIS with model performance. 31 Table 4. Model simulated rainfall (mm/d) (without calibration) at different location over Bangladesh in Table 5. Projected rainfall (mm/d, with calibration) at different location over Bangladesh in The SRES A2 scenario background: For the A2 emissions scenario the main emphasis is on a strengthening of regional and local culture, with a "return to family values" in many regions. The A2 world "consolidates" into a series of roughly continental economic regions, emphasizing local cultural roots. In some regions, increased religious participation leads many to reject a materialist path and to focus attention on contributing to the local community. Elsewhere, the trend is towards increased investment in education and science and growth in economic productivity. Social and political structures diversify, with some regions moving towards stronger welfare systems and reduced income inequality, while others move towards "lean" government. Environmental concerns are relatively weak, although some attention is paid to bringing local pollution under control and maintaining local environmental amenities. The A2 world sees more international tensions and less cooperation than in A1 or B1. People, ideas and capital are less mobile so that technology diffuses slowly. International disparities in productivity, and hence income per capita, are maintained or increased. With the emphasis on family and community life, fertility rates decline only slowly, although they vary among regions. Hence, this scenario family has high population growth (to 15 billion by 2100) with comparatively low incomes per capita relative to the A1 and B1 worlds, at US$7,200 in 2050 and US$16,000 in Technological change is rapid in some regions and slow in others as industry adjusts to local resource endowments, culture, and education levels. Regions with abundant energy and mineral resources evolve more resource intensive economies, while those poor in resources place very high priority on minimizing import dependence through technological innovation to improve resource efficiency and make use of substitute inputs. The fuel mix in different regions is determined primarily by resource availability. And divisions among regions persist in terms of their mix of technologies, with high-income but resource-poor regions shifting toward advanced post fossil technologies (renewable in regions of large land availability, nuclear in densely populated, resource poor regions) and low-income resource-rich regions generally relying on older fossil technologies. With substantial food requirements, agricultural productivity is one of the main focus areas for innovation and RD efforts in this future. Initially high levels of soil erosion and water pollution are eventually eased through the local development of more sustainable high-yield agriculture. Although attention is given to potential local and regional environmental damage, it is not uniform across regions. For example, sulfur and particulate emissions are reduced in Asia due to impacts on human health and agricultural production but increase in Africa as a result of the intensified exploitation of coal and other mineral resources. The A2 world sees high energy and carbon intensity, and correspondingly high GHG emissions. Its CO2 emissions are the highest of all four scenario families. 7

9 SUMMARY Summer monsoon rainfall is studied using regional climate model (RCM) called PRECIS (Providing REgional Climates for Impact Studies) for the SAARC domain. PRECIS is a PC-based RCM developed by Hadley Centre of United Kingdom Meteorological Office. The version of PRECIS is available at SMRC to run for SRES A2 and B1 scenarios during The model control run ( ) is performed in 50 km horizontal grid resolution for the SAARC domain. In this work PRECIS rainfall is calibrated with ground-based rainfall measurement at 27 observational sites throughout Bangladesh. During the course of calibration, regression coefficients are obtained for 27 observational sites which are the key factors for the validation of model outputs. The model is again run for to carryout validation work. It is found that PRECIS over-performed by only 4.471% in estimating rainfall over Bangladesh. This excellent performance of PRECIS encourages using it for projection of rainfall in the SAARC region. Ii is important to note that without calibration PRECIS generated rainfall scenarios are unrealistic with observational rainfall pattern. Calibration makes the model outputs more realistic with the historical rainfall blueprint. Through this research, the rainfall forecast for the SAARC domain in 2009 is experimentally prepared. Model simulated rainfall in 2009 is calibrated for Bangladesh and the rainfall projection is found to be surplus 2.03% and 14.02% during monsoon (JJAS) and post-monsoon (ON) periods respectively. It is deficit to about 2.08% and 1.44% during pre-monsoon (MAM) and dry periods (DJF) respectively. The seasonal rainfall forecasting approach using downscaling of regional climate model output is quite new in SMRC and also in Bangladesh. The work is expected to extend for the entire SAARC domain through proper calibration and validation in a consecutive research project. The ultimate goal is post seasonal rainfall forecast for entire SAARC region in the SMRC website Homepage. The rainfall forecast is very much important for the SAARC member countries, as they are densely populated and most vulnerable to the climate change. The forecasted rainfall will be benefited to the planners of the SAARC member states to utilize in their application in different sectors. 8

10 1. Introduction Rainfall forecast is very much important for the food security of any Agricultures dependent country. SAARC (South Asian Association for Regional Cooperation) region is amongst the most densely populated areas of the world where about one fifth of the total population is living. It is accounted that in SAARC countries 21% of world population resides on only 4% of the world's total physical area. In this connection proper planning and judicious management of water resources are essential for this region. Monsoon rain helps in making money and producing food for the peoples of SAARC countries. It plays a vital role in water resources of this region. Model simulated climate scenarios can play an important role in forecasting summer monsoon rainfall. SAARC region is considered as one of the most vulnerable zone under climate change and Bangladesh is in the top of the list amongst climate change impact countries. In this region, climate change is likely to exacerbate frequently occurring climatic hazards such as flood, cyclone, storm surge, drought, and heavy rain (Huq et al., 1998; Karim et al., 1998; Ali, 1999). Since the region is primarily agrarian, the projection of rainfall and its effects on water-related hazards and subsequent implications for peoples lives and livelihoods are very important (Ahmed, 2000; Ahmed et al., 1998). Climate models are the main tools available for developing projections of climate change in the future (Houghton et al., 2001; Houghton et al., 1995). In recent years, Atmosphere-Ocean General Circulation Models (AOGCMs) have been used to predict the climatic consequences of increasing atmospheric concentrations of greenhouse gases (McCarthy et al., 2001; McGuffie and Henderson-Sellers, 1997). These predictions may be adequate for areas where the terrain is reasonably flat, uniform and away from coasts. However, in the SAARC regions where coasts and mountains have significant effects on weather and therefore scenarios based on global 9

11 models are unable to capture the local-level details needed for assessing impacts at national and regional scales. The highest horizontal resolution of any AOGCM published is around 300 km (Murphy and Mitchell, 1995). Yet in order to assess potential impacts of climate change, regional information at a scale of 100 km or finer is needed (Robinson and Finkelstein, 1991). Regional Climate Model (RCM), therefore, is the best tool for dynamical downscaling of climate features in case of obtaining detailed information for a particular region (Jones et al., 2004; Giorgi and Hewitson, 2001). A regional model generally covers a limited area of the globe at a higher resolution (typically around 50 km) for which conditions at its boundary is specified from an AOGCM (Dickinson et al., 1989; Hack et al., 1993; Grell et al., 1994). The RCM is better able to resolve mesoscale forcing associated with coastlines, mountains, lakes and vegetation characteristics that exert a strong influence on local climate (Giorgi and Mearns, 1991; Vernekar, 1995; Pal et al., 2000;). In particular, previous investigations (Giorgi et al., 1994; Jones et al., 1995) have shown that the precipitation distributions simulated by RCMs contain a strong orographically related component on scales not resolved by the AOGCM. The RCM simulates a strong precipitation signal, which appears to represent an orographic component of the response to circulation anomalies associated with the intra-seasonal oscillation (ISO), whereas this precipitation signal is absent in the AOGCM (Bhaskaran et al., 1998). Several observational studies have been carried out to understand the spatial structure and phase propagation of the day mode (e.g., Yasunari, 1980; Yasunari, 1981; Krishnamurti and Subrahmanyam, 1982) of the ISO. Bhaskaran et al. (1996) demonstrated the superior ability of an RCM to capture fine scale details of the observed rainfall distribution. The spatial patterns of precipitation over Europe are well simulated by RCM and are validated against observed climatology for Great Britain 10

12 (Jones et al., 1995). However, there is very little research work carried out so far using climate model in the SAARC region. Since climate scenarios will determine the response options for agricultural and water management for the region in future decades, it is expected that the available RCMs will be employed towards development of such scenarios. Therefore, calibration and validation of an RCM is essential in order to develop future climate scenarios for the region. At present two RCMs named Providing REgional Climates for Impact Studies (PRECIS) and Regional Climate Model (RegCM) are available to run at the SMRC. Due to availability of input data to run RCM, PRECIS is selected for generating climate scenarios of this region. Since the adverse implication of climate change will be of paramount importance to waterand agriculture-sector planning, it is therefore equally important to develop PRECIS generated climate scenarios for this region. In this pursuit the model outputs need to be calibrated with the observational data. Once the calibration is completed and the performance is reasonable, model scenarios can be generated and utilized for application purposes. This work explains the generation of future rainfall scenario for the SAARC region and projected rainfall including calibration and validation for Bangladesh. The calibration and validation work will be extended for the SAARC region in consecutive research work to forecast long-term summer rainfall of this region. 2. Model description and Methodology 2.1. Model description Timely access to detailed climate change scenarios is particularly vital in developing countries, where economic stresses are likely to increase vulnerability to potentially damaging impacts of climate change. In order to help address this need the Hadley Centre of United Kingdom has developed PRECIS (Providing REgional Climates for Impacts Studies), a regional climate modeling system which can be run on 11

13 a cheap, easily available personal computer. The aim of PRECIS is to allow developing countries, or groups of developing countries, to generate their own national scenarios of climate change for use in impacts studies. This will allow transfers of technology and ownership resulting in much more timely and effective dissemination of expertise and awareness than if results are simply handed out from models run in developed countries. In addition, countries using PRECIS are in a better position to validate the model using their own observations. An important aspect of PRECIS is the availability of training and training materials explaining its role and how to make the best use of it. One of the main materials is the technical manual which discusses the steps needed to install, configure and use of PRECIS. It is designed both to guide users of PRECIS and as resource for the PRECIS training course. PRECIS is a hydrostatic, primitive equation grid point model containing 19 levels described by a hybrid vertical coordinate (Simmons and Burridge, 1981; Simon et al., 2004). The Hadley Centre, under contract from the United Kingdom government departments DEFRA and DFID and from the UNDP, has developed PC-based PRECIS to provide non-annex I Parties with a practical tool to make their own predictions of national patterns of climate change and hence assess their vulnerability. PRECIS only works on 32-bit Intel (x86) compatible Linux based systems, and not under Microsoft Windows, and not on other Unix systems. PRECIS only run on a single processor, and cannot be configured for multi-processor running. It may be possible however, to run concurrent jobs on a multi-processor system. To run PRECIS at least 512MB of memory is needed, 768MB is recommended. Higher RAM gives a small increase in performance. At least 100GB of disk space is required. PRECIS works under a single or dual disk system. On a dual disk system, one disk should be a dedicated data disk. Some form of off-line storage is needed both to supply the input data that drives PRECIS and to archive its output data. Either DLT or DAT tapes are recommended; 12

14 hard disks can also be used, but are less robust and more difficult to duplicate. The PRECIS RCM is able to adhere to one of two types of calendar, the standard Gregorian calendar or an artificial calendar consisting of 360 days per year. The choice of calendar is determined by the calendar implicit in the driving data, which in turn is determined by the choice of scenario. The choice of calendar is made automatically by PRECIS and is displayed in the GUI. The RCM's clock is always based on Universal Time (UTC), denoted Z e.g. 0300Z. The 360-day calendar divides a year up into 12 months, each of 30 days in length. It is used in long climate simulations for internal organizational convenience. The introduced distortions of the seasonal cycle are minimized by altering the average date of perihelion, shifting it from 2.5 days after the beginning of the year (0Z 1 st January) to 3.2 days after the beginning of the year. This ensures that monthly and seasonal mean values diagnosed from the RCM are comparable with their equivalent observed quantities. The PRECIS RCM is able to run at two different horizontal resolutions: and , giving grid boxes of approximately 50km 50km and 25km 25km respectively. Whilst more realistic land-sea mask topography is expected at 25km resolution, the time taken to complete a simulation is approximately six times longer than for a 50km resolution run covering the same area. Two-thirds of this increase comes from the fourfold increase in the number of grid points and the rest from a halving of the time step used in solving the dynamical equations. In this case, the time step associated with the physical parameterizations in the model remains the same (five minutes) for both resolutions. This both reduces the cost of the high resolution version of the PRECIS RCM and also ensures that the influence of possible time step dependencies in these parameterizations is removed. PRECIS is made freely available for use by scientists of developing countries involved in vulnerability and 13

15 adaptation studies. In this study PRECIS runs with 50-km horizontal resolution for the present climate ( ) using baseline lateral boundary conditions (LBCs). It runs for in validation purpose and for 2009 in generating forecasted rainfall scenarios using the special report on emissions scenarios (SRES) A2 of ECHAM4 LBC. The model domain is selected E and 6 35 N to cover entire SAARC region Methodology Depending on the application, the PRECIS RCM can be downscaled to recent (say ) and future (say ) climate states. The climate of a particular region is determined by local and remote processes with external forcing provided by solar radiation. The effect of the radiation is modulated by the composition of the atmosphere and various feedback processes within the global climate system. Thus a regional climate model requires, as input, boundary conditions providing the remote forcing of the regional climate and consistent information on atmospheric composition. More specifically, the former comprises lateral boundary conditions of winds, temperature and humidity and surface boundary conditions of pressure and, over the sea, temperature and sea-ice fractions. The latter is represented by prescribed concentrations of the most important greenhouse gases derived from scenarios of their emissions. In this work rainfall is simulated using PRECIS for the entire SAARC domain. Calibration and validation of PRECIS is considered for Bangladesh to understand the model performance in simulating summer monsoon rainfall. Once the reasonable result is obtained, PRECIS can be used to accomplish similar work for other member countries. The Bangladesh Meteorological Department (BMD) collected surface rainfall throughout the country has been used for the calibration of PRECIS generated 14

16 rainfall. The BMD observation network density is low and the distribution is poor; in some cases observation sites are located at about 25 km apart whereas other cases are about 145 km apart (Fig. 1). When the whole Bangladesh region is gridded at 0.5º 0.5 º resolutions a number of grids are found which do not contain any observation site. For the application of PRECIS for climate change impact studies in Bangladesh, it is important to find out the appropriate calibration method. Having this in mind, analyses have been performed on point-to-point basis (Islam et al., 2008). In this procedure, observed data at a particular site is considered as the representative of that location (Islam and Uyeda, 2006). Grid value of the model data is compared with the observed data representing that grid. If more than one observation sites exist within a grid, average value of all the observational sites is considered as representative value for that grid. Rain-gauge daily rainfall data collected by BMD are processed to obtain monthly, seasonal, annual, decadal and long-term values. The model data of rainfall is extracted at 27 observational sites of BMD and then are converted to monthly, seasonal, annual, decadal and long-term values. Rainfall is simulated by PRECIS for ensemble blsula (baseline with the sulphur cycle and ensemble category a ) during baseline period ( ). Through the regression expression the slope and constant values are assigned from model and observed rainfall for the baseline period. Estimated rainfall is obtained from model generated scenarios with the help of slope and constant value. This estimated rainfall is useful for validation of PRECIS RCM in Bangladesh for the year onward the baseline period. In the next step PRECIS run (50 km 50 km resolution) was completed for the year using ECHAM 4 SRES A2 as the model input. One may consider the one year spin-up time from the beginning of model run for the application purpose of 15

17 model outputs. Using the prepared slope and constant, model generated rainfall is validated for The TRMM (Tropical Rainfall Measuring Mission) version 6 3B42 3-hourly data products are also utilized to compare the PRECIS generated rainfall scenarios with observed rainfall. TRMM obtained rainfall can be considered as ground-truth amount especially for Bangladesh where it was calibrated with rain-gauge data (Islam and Uyeda, 2007). The future scenario obtained by PRECIS for 2009 is used to prepare rainfall forecast in Bangladesh for the year Regression coefficients are utilized with 2009 rainfall scenario for the forecast purpose in different months and at different observational sites Observation Site 25.6 Rangpur Dinajpur INDIA Elevation (m) LATITUDE (N) Bogra Sylhet Mymensingh Rajshahi 18 Srimongal INDIA 17 BANGLADESH INDIA 22 Dhaka Faridpur 5 Comilla 8 Chandpur Jessore MaijdiCourt Khulna Barishal 5 Rangamati 3 4 Hatiya Sandwip Chittagong BAY OF BENGAL CoxsBazar LONGITUDE (E) Fig. 1. The BMD observational site (plus mark) with elevation in m (below plus mark). 16

18 3. Results and Discussion 3.1. Rainfall in SAARC domain PRECIS simulated annual rainfall for present climate ( ) in SAARC region is shown in Fig. 2. The same for monsoon months June-September is shown in Fig. 3. There is a large amount of rainfall belt observed along Western Ghat of India, along Nepal-Bhutan region and along northeastern side of the Bay of Bengal. Large amount of rainfall is also observed in the northeastern part of Bangladesh which is actually located in the slope of Shillong Hill of India. These features are very common and seem quite reasonable with rainfall climatology of this region. Fig. 2. PRECIS generated annual rainfall averages for

19 Fig. 3. The same as Fig. 2 except for rainfall during monsoon period (JJAS) Calibration of Rainfall in Bangladesh Rainfall generated by PRECIS need to be calibrated with the observed data otherwise model performance may not be well understood. In this paper, calibration of rainfall for Bangladesh is taken as a pilot study. In order to calibrate model simulated rainfall with observed rainfall in Bangladesh, model rainfall are processed for 27 stations and compared with rain-gauge amounts in seasonal scales (Fig. 4). The similar result is reported by Islam et al. (2008). Model overestimated pre-monsoon rainfall and underestimated monsoon rainfall with amounts in post-monsoon and dry periods is almost closer. To find an adjustment between model and observed rainfall amounts the following regression equation is proposed: 18

20 RF estimated = constant + slope RF scenario Where RF estimated is the rainfall to project and RF scenario is the PRECIS generated rainfall scenario. The regression coefficients such as constant and slope are obtained with the help of both model and observed rainfall in different months and at different observational sites. The constant and slope are tabulated in Table 1 and Table 2 respectively. The constant values vary from month-to-month and also from place-toplace. Similarly slope values vary from month-to-month and also from place to-place with both positive and negative sign. Rainfall (mm/d) Observed Model DJF MAM JJAS ON Fig. 4. Seasonal distribution of rainfall obtained from Observation and PRECIS model. The model simulated parameters like rainfall is not being able to use directly in application purposes. The reason is that model generated parameters are not free from 19

21 uncertainties. Even uncertainties are there, still to date there are no alternatives to predict meteorological variables earlier without any help from model. Therefore, model is used as one of the prediction tools with considering the limitations. Table 1 and Table 2 are utilized in next sections for model validation and forecasting purposes. Table 1. Regression constants in different months and at different observational sites over Bangladesh. JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Barishal Bhola Bogra Chandpur Chittagong Comilla Cox s Bazar Dhaka Dinajpur Faridpur Hatiya Ishurdi Jessore Khepupara Khulna Kutubdia Maijdi Court Mymensingh Patuakhali Rajshahi Rangamati Rangpur Sandwip Sitakunda Srimongal Sylhet Teknaf

22 Table 2. Regression slopes in different months and at different observational sites over Bangladesh. JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Barishal Bhola Bogra Chandpur Chittagong Comilla Cox s Bazar Dhaka Dinajpur Faridpur Hatiya Ishurdi Jessore Khepupara Khulna Kutubdia Maidi Court Mymensingh Patuakhali Rajshahi Rangamati Rangpur Sandwip Sitakunda Srimongal Sylhet Teknaf Validation of Rainfall in Bangladesh To understand the performance of PRECIS modeling system, simulated outputs are needed to validate with ground-based observed datasets. In previous section the calibration of PRECIS outputs are performed with rain-gauge rainfall for the baseline period ( ) and for Bangladesh. Before validating PRECIS generated rainfall, TRMM V6 3B42 datasets is used to see the overall model performance in TRMM derived rainfall can be considered as the ground-truth amount even it scans from the upper side of the convective systems. The TRMM measured rainfall is already calibrated with ground-based rain-gauge rainfall in Bangladesh (Islam and 21

23 Uyeda, 2007). It is found that TRMM able to catch about 97% of the surface rain. The distribution of TRMM derived annual rainfall (mm/day) in the SAARC domain for 2002 is shown in the upper panel of Fig. 5. The idea is that, how much rainfall could generate by PRECIS, at least in qualitatively, because calibration factor could be utilized to obtain the quantitative amounts. In this connection annual rainfall (mm/day) simulated by PRECIS for 2002 is presented in the lower panel of Fig. 5. The patterns of rainfall obtained by TRMM and PRECIS are somehow quite similar, especially in the NE of Bangladesh, Western Ghat of India and a long band along Nepal and Bhutan. In TRMM derived rainfall the no-rain tongue in the eastern side of Western Ghat of India is not observed. The model is run in 50 km horizontal resolution whereas TRMM product is about 25 km in horizontal resolution. Therefore, high resolution PRECIS run is also may be considered in future work. mm/d 22

24 Fig. 5. Annual rainfall (mm/day) derived from TRMM V6 3B42 (upper panel) and PRECIS simulation (lower panel). Monsoon rainfall is much important for the SAARC region; therefore, rainfall averages from June to September (JJAS) of 2002 is presented in Fig. 6. Monsoon rainfall (mm/d) derived by TRMM V6 3B42 is shown in the upper panel of Fig. 6. The same derived from PRECIS outputs is shown in the lower panel of Fig. 6. Monsoon rainfall simulated by PRECIS is much more consistent with TRMM observed rainfall. It is almost below 1 mm/day in most parts of Afghanistan and Pakistan which are well simulated by the model. Also the signal of strong rainfall in the Western Ghat of India, central part of Nepal and northeastern part of Bangladesh are also well simulated. This indicates that PRECIS can simulate the seasonal rainfall will a better spatial distribution. Because the PRECIS output is calibrated with rain-gauge rainfall therefore, model validation is performed against rain-gauge rainfall, however, TRMM datasets provides clear insight for the whole SAARC domain which is not possible from raingauge datasets. 23

25 Fig. 6. The same as Fig. 5 except for monsoon (JJAS) season. Through the calibration of model outputs with observed rainfall, regression coefficient slope and constant are obtained in previous subsection. With the help of regression slope and constant, PRECIS outputs are validated with observed rainfall. As an example, model simulated scenario (without calibration), estimated (with calibration), observed and normal (observed baseline period) rainfall for 2002 is shown 24

26 in Fig. 7. Hence, utilization of regression coefficients improved a lot the estimated value that influenced model outputs to be very close to observed amount which follows the historical pattern. Only model generated scenario is not match well with the observed rainfall amount. In annual scale the estimated amount is surprisingly coincided with the observed value. This indicates that rainfall was surplus (observed) in 2002 that information is well captured by the PRECIS model. Such type of information in advance is very much helpful for planners in mitigating food crisis, water scarcity and other kind of disaster management of the region JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Annual Rainfall (mm/d) Scenario 2002 Estimated 2002 Observed 2002 Normal Fig. 7. Annual rainfall (mm/day) distribution obtained from observation, model simulation (Scenario), model projection (Estimated) and observed baseline period (Normal). Figure 8 (a-g) shows the monthly, seasonal and annual rainfall over Bangladesh obtained from both observation and PRECIS outputs. Estimated values are obtained with the help of regression coefficients and observational baseline period (Normal) is the average from According to the model validation from 2000 to 2006, it 25

27 is revealed that estimated values are some times differing from observed one in monthly scale. The difference is reduced in seasonal scale and very small in annual scale. It is therefore concluded that model simulation is very good in estimating seasonal rainfall and of course it is comes close to realistic form with the calibration. It is mentioned again that without calibration, straight forward model simulated amounts are not much realistic with observed amount, and not useful in forecasting purposes (a) Observed 2000 Estimated2000 Normal JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM Rainfall (mm/d) JJAS ON Annual 26

28 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual Rainfall (mm/d) Observed 2001 Estimated2001 Normal (b) JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual Rainfall (mm/d) Observed 2002 Estimated 2002 Normal (c) 27

29 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual Rainfall (mm/d) Observed 2003 Estimated2003 Normal (d) JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual Rainfall (mm/d) Observed 2004 Estimated2004 Normal (e) 28

30 Rainfall (mm/d) (f) Observed 2005 Estimated2005 Normal 5 0 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual (g) Observed 2006 Estimated2006 Normal JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC DJF MAM JJAS ON Annual Rainfall (mm/d) Fig. 8 (a-g). Validation of PRECIS estimated rainfall (mm/d) over Bangladesh for (a) 2000, (b) 2001, (c) 2002, (d) 2003, (e) 2004, (f) 2005 and (g)

31 Model simulated annual rainfall (mm/d) after calibration with baseline data period is shown in Fig. 9 along with observed amount for Model overestimated rainfall in 2000, 2002, 2003, 2005 and 2006 whereas it underestimated in Model rainfall in 2001 is very close to the observed amount. Such type of year-to-year or inter-annual variation in simulating rainfall is not unexpected because model can not simulate any meteorological parameter in 100 percent accurately. The question arises how much perfect the PRECIS simulation and how it can be utilized. To answer these question model performance is calculated and tabulated in Table 3. The performance is that difference of model and observed value divided by observed amount in percentage. Then knowing the uncertainties, the forecast may be utilized with percentage of uncertainties that may helpful for planners to do long-term preparedness. Rainfall (mm/d) Observed PRECIS Year Fig. 9. Validation of PRECIS simulated annual rainfall (mm/d) with observed amount for Bangladesh. 30

32 According to table 3, PRECIS overestimated 12.37%, 1.58%, 10.81%, 4.79 and 13.18% in 2000, 2002, 2003, 2005 and 2006 respectively. It underestimated 0.64% and 10.84% in 2001 and 2004 respectively. In an average PRECIS overestimated rainfall 4.47% of rain-gauge value. It seems that the PRECIS performance is quite reasonable because about 4.47% error may be considered in long-term forecasting using a climate model. Correlation coefficients are obtained from both model and rain-gauge rainfall in individual year and found that it lies within 0.87 to 0.97 which are quite significant. Table 3. Validation of PRECIS with model performance. Year Observed Estimated Model Performance Correlation Coefficient Fig. 10 shows the time sequences of monthly rainfall (mm/d) obtained from PRECIS and observation for It is seen that model simulation follows the annual cycle of rainfall almost in similar pattern of observed data. Exception is only observed in magnitude for a few months, especially during monsoon period Obseved PRECIS Rainfall (mm/d) JAN APR JUL OCT 01JAN APR JUL OCT 02JAN APR JUL OCT 03JAN APR JUL OCT 04JAN APR JUL Month from 2000 Jan to 2006 Dec OCT 05JAN APR JUL OCT 06JAN APR JUL OCT Fig. 10. Time sequences of monthly rainfall (mm/d) obtained from model simulation and rain-gauge data during for Bangladesh. 31

33 Scatter plot of monthly rainfall obtained from both model estimation and observation for is shown in Fig. 11. The slope is very close to 0.9 and the correlation coefficient is 0.9 with R 2 is This indicates that after the calibration is performed, model estimated rainfall is very much consistent with the observed data. Hence, the result encourages in utilizing PRECIS for monthly to seasonal rainfall forecasting in Bangladesh. The work may be extended in SAARC countries for better forecasting seasonal rainfall which is considered as the consecutive research project. Monitoring validation results for more couple of years might be useful in improving the calibration coefficients and finally in enhancing summer monsoon rainfall forecasting for the SAARC domain. The similar work for other ensembles and high resolution model run are also in consideration. In order to use many more ensembles and to prepare long-term forecasts, better computational facilities are essential which are not available at SMRC at present. It is expected in coming days PRECIS Rainfall (mm/d) PRECIS = 0.90 (Rain-gauge) R 2 = 0.81 r = Rain-gaige Rainfall (mm/d) Fig. 11. Scatter plot of monthly model rainfall (estimated) versus observed rainfall at different months for

34 3.4. Rainfall Projection for SAARC region in 2009 PRECIS simulated annual rainfall scenario in the SAARC domain for 2009 is shown in Fig. 12. Heavy rainfall is expected along the belt of Nepal and Bhutan including Shillong Hills of India. As mentioned earlier model generated rainfall is not much accurate without any calibration with ground-based data. Calibration of model simulated rainfall with observed data is very much essential, which is not performed yet for the entire SAARC region; only for Bangladesh is completed through this project. For other SAARC countries it is under consideration in a consecutive research project at SMRC. Fig. 12. Projection of annual rainfall (mm/d) in the SAARC region for 2009 using PRECIS with the SRES A2 scenarios as the model input. Figure 13 shows the seasonal rainfall for monsoon months June-September 2009 simulated by PRECIS. As the distribution of annual rainfall, heavy rain belt is 33

35 forecasted along Nepal and Bhutan in addition to Western Ghat of India and northeast of the Bay of Bengal. From the validation of rainfall obtained by PRECIS model with ground-truth value over Bangladesh, it is observed that PRECIS generated rainfall scenarios are not directly useful in application purposes. Therefore, model simulated rainfall scenario is applicable only after performing the suitable calibration, which is prepared for Bangladesh and explained in the next subsection. Fig. 13. The same as Fig. 11 except for monsoon season (JJAS). 34

36 3.5. Rainfall Projection for Bangladesh in 2009 As learned through validation work in the previous subsection, model simulated values are needed to be calibrated with the ground-based observed data before going to use them in application purposes. In this connection, PRECIS simulated rainfall for 2009 is calibrated at different observational sites over Bangladesh and in different months as tabulated in Table 4. Averages for all stations over Bangladesh are considered as country. The normal (averages from ) amounts are also included to understand the surplus or deficit of rainfall in different months of the year For the country (averages for all stations), model simulated rainfall (uncalibrated, Table 4) amounts are somehow unrealistic with historical normal. After the calibration is performed the projected rainfall (calibrated, Table 5) is very closer to the historical trend. Again the necessity of calibration is well understood through this research work and this is very essential for the use of model generated rainfall in application purposes. Once the calibrated and reliable model outputs are available for the planners, the overall improvement will definitely helpful for the end users of the country. SMRC can play an important role in seasonal rainfall forecasting for SAARC domain in extending this work for all SAARC countries. This will be really helpful in contributing SAARC communities by the centre. Of course, improvement of the climate model and any new development in adapting the model may be considered for upgrading the forecasting technique by SMRC. 35

37 Table 4. Model simulated rainfall (mm/d, without calibration) at different location over Bangladesh in JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC Barishal Bhola Bogra Chandpur Chittagong Comilla Cox s Bazar Dhaka Dinajpur Faridpur Hatiya Ishurdi Jessore Khepupara Khulna Kutubdia Maijdi Court Mymensingh Patuakhali Rajshahi Rangamati Rangpur Sandwip Sitakunda Srimongal Sylhet Teknaf Barishal Bhola Country Normal