Director, National Institute of Hydrology
Roorkee-247667 (U.P.), INDIA
The Mahaweli Development Programme in Sri Lanka was implemented with the aims of providing water to the dry zone of the country through a massive diversion scheme and also generating hydropower.
b The Mahaweli is the longest river in Sri Lanka, covering a catchment area of 10,400 sq. km. The upper catchment is defined as the area above 150 m elevation from mean sea level encompassing an area of 3124 sq.km. The Mahaweli Development Programme, a major undertaking in the Upper Mahaweli Catchment Area (UMCA) was implemented with the aims of providing water to the dry zone of the country through a massive diversion scheme and also generating hydropower. Under this development programme, a series of large reservoirs was constructed across the main water course at Kotmale, Polgolla, Victoria, Randenigala and Rantembe. The construction of these reservoirs inundated a considerable agricultural land area (5400 ha) and the accompanied developments were responsible for the displacement of several settlements (14000 families).
The changes made on the catchment environment are very prominent due to both the short time span in which they occurred and the high magnitude of the impact. The diplaced peasants have been settled in the unsuitable terrain. Steep slopes have been put under cultivation without proper conservation measures. Agricultural objectives have been derived to obtain short-term benefits without any concern on the sustainability of the land and water resources.
Present Problems and a Future Crisis
Critics now say that the hydrological regime has been adversely affected due to indiscriminate land use changes. It is also pointed out that the river flows have diminished significantly during the last two decades. Productivity of the agricultural land has declined. This has directly affected the income status of the farmers thus creating social problems. Further, reservoir siltation and eutrophication demand considerable maintenance expenditure. Frequent landslides have threatened human life and infrastructure. If these problems are not properly resolved, a future crisis is inevitable and that will jeopardise the expectations of the Mahaweli Development Programme.
A large-scale afforesttion/ reforestation scheme has been introduced considering the environmental benefits of the forest cover in terms of the hydrological stability and the sustainability of the catchment resources. However, the hydrological consequences of reforestation and the other land resource conservation measures could lead to high water losses from the catchment system. In contrast, trees at proper locations can intercept water from clouds and improve the water yield. Hence, it is required to find strategic locations for tree plantations to ensure positive water budget in the catchment while providing extensive benefits in the other sectors. This created the need for a comprehensive model to simulate the hydrological dynamics of the catchment to answer ‘what if’ questions for different conservation scenarios.
Hydrological Modelling and GIS
Most of the hydrological models are numerical and computer based, and assume some form of spatially averaging process for parameter definitions. The lack of recognition of spatial diversity at catchment sale has been a serious constraint hampering the simulation, validation and practical application of the hydrological modelling results.
In contrast, GIS is well suited for spatial modelling with large and complex databases. However, the present GIS have an inherent limitation of representing time in its spatial data structures. Hence, GIS and hydrological modelling can be considered as complimentary. GIS could benefit from the temporal modelling capabilities of hydrological models and hydrologic models can benefit from the spatial modelling capabilities of GIS. Within this conceptual framework, this study was focussed on developing a spatiotemporal hydrological model in GIS to investigate and analyse the hydrological dynamics and behaviour of UMCA in Sri Lanka.
Data and Software for Hydrological and Spatial Modelling
Daily rainfall data for a period of 30 years (1964 – 1993) from 64 gauging locations in the UMCA or its close proximity were collected and formatted for Lotus 1-2-3 and Approach database. Fog interception data were retrieved from recent research results carried out in Horton Plane and Kundasale, Sri Lanka. Pan evaporation data were also collected for seven (7) locations within UMCA to generate open-water evaporation parameters. Flow data were also collected for nine (9) stations for the same 30-year period.
The TYDAC SPANS GIS Ver 5.3 (1993) and Ver. 5.4 (1994) running on IBM Operating System 2 (OS/2) was the GIS software environment used for hydrological modelling. In addition, GIS and image processing facilities available in ERDAS Ver. 7.5 and IDRISI operating in DOS environment were also used for data conversion, formatting and processing.
Hydrological model parameters were derived from the supervised classification of IRS LISS II imagery for different land use categories based on their hydrological significance. An extensive GPS survey was carried out to determine the location information for gauging stations and ground truth sampling frames.
The UMCA hydrological model is a simplified version of a set of water balance equations to calculate runoff response from the catchment. It calculates daily runoff depending on the daily precipitation while taking water losses and soil moisture fluctuations into account.
The model includes the basic hydrological processes as shown in Fig. 1. The total precipitation includes rainfall and fog interception in natural forests and forest plantations where elevation is 1000 m from mean sea level. The interception is estimated by means of an exponential stochastic interception model. Evaporation is approximated using soil moisture and moisture stress moderator . A water balance was simulated for each day. Depending on the available water status which was derived from hydrologically important land use categories, runoff predictions were made for the individual day.
Spatial Modelling in GIS
The lumped hydrological model was required to run on a spatial platform to estimate the river flow status. In selecting the modelling functionality in SPANS GIS, the main criterion was the capability of direct spatial data processing in the modelling exercise. Map modelling allows to use spatially distributed and tabulated data through various mathematical processes to derive associated data and information. It was noted that the accuracy of model representation was dependent on the quad resolution of the data and data analysis process. Hence, the selection of the best quad level was made carefully after considering the accuracy of data representations, storage requirements, and computational efficiency. The quad level 11 was rated as the best in terms of these variables.
Table 1. Statistical Summary of Spatiotemporal Modelling Results
Sub Staistical Parametres Catchments
|Coefficient of Determination||0.84||0.92||0.83||0.90||0.96|
|Cross Correlation Coefficient||0.71||0.84||0.69||0.81||0.92|
|Lag 01 Correction||0.49||0.23||0.22||0.33||0.46|
|Coefficient of Efficiency||0.15||0.69||0.22||0.62||0.82|
|Residual Mass Curve Cof.||-0.40||-1.20||0.21||0.32||0.35|
Figure 1 Generic Structure of the Hydrological Model
Several interpolation methods were attempted to represent spatial distribution of rainfall in GIS. Thiessen polygon method was adopted due to its computational efficiency with the time invariant spatial boundary demarcation.
Saptial Model Structure
A series of equations were formulated using map modelling language codes of SPANS GIS. Each sub model of the UMCA hydrological model was represented in a set of equations and additional equations were required to increment the file pointer along the columns of the data tables for daily rainfall.
In the case of Thiessen polygons, representative areas for each gauging station are directly identified from the morton numbers of the gauging locations. Morton numbers are hexa-decimal numbers used for spatial referencing in SPANS GIS. For a particular day, the hydrological model reads the relevant column of the rainfall data tables and calculate the fog interception according to the season. The total precipitation is then assigned to the corresponding Thiessen polygons. Based on sub models, it calculates the spatial distribution of interception and evaporation losses according to the hydrological parameters assigned for each land use. It also takes into account the spatial variation of antecedent moisture and the soil moisture stress. Finally, the model calculates the daily runoff and changes in soil moisture regime through the water balance equations. It then updates spatial coverage for cumulative runoff, soil moisture, cumulative interception and evaporation, stored in thematic maps. The updated map of soil moisture provides antecedent moisture status for the water balance calculations of the following day. The entire model structure was set-up to run on actual numerical values of each quad cell of thematic coverage.
Data Output Formats
The distributed modelling approach adopted for the study was capable of estimating daily upstream flow at any desired point of the drainage network. However, the model calibration required flow to be predicted at the flow gauging locations in order to make comparison with the historical flow records. Further, provisions were made in the model to estimate composite flow values at the identified 32 sub catchments in UMCA. In addition to the real time series of flow data generated from the model, it provided the display facilities representing the spatial distribution of runoff on thematic maps at any desired time period of interest.
Spatiotemporality in GIS
SPANS GIS menu driven functions can be run using equivalent command mode codes. The advantage is that a series of SPANS functions can be programmed into a batch file recognised as an audit file and run on the command mode. In order to include the temporal dimension into hydrological modelling, command mode functions were used extensively. The entire methodology depends on the format and thematic details of the input data and map files. Having prepared daily rainfall data in monthly tables with a column of data series for each day, it was possible to use only one set of equations for a month, incrementing the file pointer to read the data in different columns. One equation file was designed for each year incorporating a series of monthly equations. In addition to the equation files, command files were required to call the relevant equations for map modelling. The command filing system was organised in such a way that each file contains executable files for each month. The REXX procedures, the available programming language in OS/2 were set up so that they could produce monthly values of weighted average of spatial distribution of runoff at each subcatchment. They also created maps showing spatial distribution of monthly runoff on the thematic scale according to the user-defined classification scheme. Cumulative monthly totals of the other hydrological parameters such as evaporation, interception and soil moisture were also calculated whenever required.
Limitations for Spatiotemporal Modelling in GIS Hydrological modelling efforts in GIS are generally hampered by the limitations of time representation in spatial data structures. As such, it is not possible to readily model the evolution through time of spatial variations in a phenomenon with GIS and such variations are often needed in hydrology.
However, the continuous development of the conceptual framework for spatiotemporal modelling confirms that the goal of fully functional temporal GIS is close to realisation. Nevertheless, it was found that provisions are made within the existing software architecture for the time varying modelling at discrete temporal resolutions through iterative procedures. This study shows how time dimension could be implicitly incorporated into the existing GIS modelling algorithms in order to employ time variant modelling while maintaining the integrated spatial dimension.
A comprehensive statistical evaluation was made to compare the observed flow data with the simulated flow series of the modelling exercise. The statistical summary of the modelling results is listed in Table 01. It is apparent that there is a great deal of agreement between the measured and simulated flow time series. In addition, the sensitivity of the model for the defined hydrological parameters, spatial resolution and land use changes were also assessed. The model is obviously sensitive to land use changes in the catchment and it shows 15 – 35% increase of annual runoff when forests are converted to grasslands.