The Cost and Data categories of the Koycegiz watershed were analyzed and based on observed evidence, following results are presented.
A watershed model even at its best is still a mathematical formulation of watershed and the processes within it; therefore, the simplification of a complex watershed process into a mathematical formulation has an inherent uncertainty in predicting the behavior of the system and its processes. This inherent uncertainty means that the model will never be able to predict a 100% actual effect of an event. This fact is further strengthened by the uncertainty in data used for the application of watershed model. In other words, in terms of model performance, a correlation of 1.0 may not be achievable no matter how refined and representative the date may be. Similarly, there will always be some correlation between model generated and actual data due to parameterization of the model. The relationship between model performance and data can be represented by a curve shown in Figure 7.12. The minimum and maximum achievable model performance is shown as points ‘a’ and ‘b’ on the Model Performance-Data curve. The value of correlation attained by a model calibration between the minimum and maximum achievable level is dictated by the amount and quality of data. For each modeling application, there exists a value of correlation between observed and simulated data corresponding to an absolute minimum amount and quality of data and an acceptable model performance lying between the points ‘a’ and ‘b’ on the curve.
Data M odel P e rf o rm a nce 0 1.0 a b
Figure 7.14: Model Performance–Data Curve
The acquisition of the quantity and quality of data for watershed modeling can be related to costs incurred is directly related to the cost of obtaining that specific amount data. At zero cost, one may have some data available in the form of
simulated data for a specific watershed; however, it may not be of the desired quality and representative of the actual watershed. Furthermore, a reasonable amount of data may be obtained at a relatively reasonable cost; however, acquisition of data will require relatively higher cost with lesser data per unit cost incurred. The relationship between data acquisition and cost is represented by Data–Cost curve shown in Figure 7.15. For a cost of X, Y amount of data may be acquired; however, a further cost of X will only provide Z amount of data where Z is relatively 1/3 of Y. This is explained by the fact that initial data required is of general type like topographic data, land use/land cover data, precipitation, flow etc., which in most of the cases may be obtained through some agency or organization. Further data means more specific data which may not be available and thus require higher cost to obtain. The specific data may be infiltration capacities of soil, agricultural practices, etc.
Cost Da ta Y X X Z
Figure 7.15: Data–Cost Curve
The cost incurred for the application of the model can be correlated with the performance of the model. Even without incurring any costs, an initial correlation may be obtained between the models generated results and observed data. The behavior may be represented by the model performance–cost curve shown in Figure 7.16. The model performance–cost curve represents minimum model performance corresponding to no cost while c represent maximum model performance at a high cost. B represents the situation where acceptable model performance is achieved at a relatively lower cost.
The relationship between data acquisition, cost and model performance can be used to define an optimization curve as shown in Figure 7.17. Data acquisition requires expenses in terms of costs to achieve results regarding model performance. Therefore, for each piece of data, a corresponding amount of money is spent to achieve a benefit in terms of model performance. However, based on the discussion
presented above, at a certain point the amount of data obtained or money spent will start giving a decreasing value of model performance.
Cost M o del Perf orm a nce 1.0 0 a b c
Figure 7.16: Model Performance–Cost Curve
The unit change in model performance per unit cost incurred may therefore be defined as the benefit. The relationship between benefit and cost can be represented by a normal curve as shown in Figure 7.17. The point of inflection obtained provides the amount of money spent or data acquired that will give the best model performance for the cost incurred.
Figure 7.17: Cost Benefit Optimization Curve 7.5.1 Implications for model applicability in developing countries
The application of a watershed model in a developing country should lie in the rising limb of the Cost Benefit optimization curve, because acquisition of data and expenditure of costs beyond point of inflection does not provide the same amount of increase in model performance.
A researcher in a developing country should start with the minimum amount of data required by BASINS/HSPF for analyzing a specific watershed based on the
hydrological modeling using BASINS/HSPF is presented in Figure 7.18. DEM and land use maps are required to perform the initial delineation and characterization of the watershed and the determination of the modeling system boundaries. These maps if availbe in a finer spatial resolution in the temporal window of the model simulation will generate best results. However if not available a coarser spatial resolution may be used to perform the aforementioned tasks. The temporal interval of precipitation and evaporation records is a key factor in obtaining accurate results in the simulation of hydrology and later on simulation of water quality. Normally hourly data is of best quality if not available at least daily records should be obtained to obtain satisfactory results from model simulation.
Figure 7.18: Absolute Minimum Data Requirement for Hydrological Modeling using BASINS/HSPF
The model verification data should be 3-5 years to check the performance of model under different conditions. In the case of the simulation of hydrology atleast daily stream flow data for a period of 2-3 years be available for calibration and validation of model results.
The next step based on available resources should be to acquire data required for enhancing the representation of the watershed in the model by improving parameterization. Data for parameter estimation may than be put on top of the priority list starting from the most sensitive parameter. Figure 7.19 shows the
optimized model applicability pathway for a watershed modeling project in a limited data and costs scenario.