چكيده به لاتين
Adaptive management and operation is considered a beneficial approach in the field of water resource management. The goal of this research is to develop a forecast-based operational framework to support the adaptive management procedure (specifically reservoir management, as one of the important available water resources) in watersheds. One of the key features of adaptive management strategies is its ability to adapt to changing system conditions in future periods. This capability relies on access to updated forecasts of the system's future state, which informs decision-makers. Therefore, employing approaches that provide real-time hydrological forecasts due to the new observations becoming available over time enhances the possibility of success in adaptive management. Data assimilation (DA) is a modern approach to providing real-time forecasts. In this study, a simulation-data assimilation module was specifically developed to deliver real-time forecasts of various variables (particularly the inflow to the Mahabad reservoir and the downstream reservoir’s water demand). The framework utilized the EnKF data assimilation algorithm and the SWAT simulation model. These forecasts are provided to the reservoir optimization model, informing it and enabling the adaptation of decisions based on new forecasts. The optimization module, based on dynamic programming (DP), was developed and performed in real-time. The Mahabad watershed was selected as the study area. Among the 13 watersheds feeding Urmia Lake, the Mahabad watershed is the third most water-abundant. Additionally, the Mahabad reservoir, the Boukan dam, and the Shaharchai dam are the largest reservoirs at the Lake Urmia basin. Implementing the simulation-data assimilation module requires that the hydrological simulation model provides extensive access of its initial conditions (ICs) for the data assimilation algorithm. The SWAT model, like many (semi-)distributed models, does not provide this capability due to its complexity and the multitude of variables and parameters. This limitation has restricted research in the field of data assimilation within the SWAT model. To address this limitation, structural modifications were made to the model's reference codes (introducing SWAT_DA) to enable effective and accurate integration with the data assimilation algorithm. In addition to structural modifications, the model's reference codes were also enhanced and upgraded in terms of simulating processes related to snow and groundwater. The simulation-data assimilation module, based on the multivariate assimilation of runoff observations, snow cover, surface soil moisture, leaf area index, and groundwater storage, provides real-time forecasts of various hydrological variables (particularly reservoir inflow and downstream water demand). Innovative multivariate data assimilation scenarios (innovative in terms of the type and number of observations) were defined and executed, and their results were compared. The use of multivariate assimilation of field runoff observations and MODIS remote sensing snow cover data in the upstream region improved reservoir inflow simulations and increased the Nash-Sutcliffe efficiency (NS) coefficient from 0.53 to 0.58 compared to single-variable runoff assimilation. Additionally, considering orographic precipitation effects in the multivariate data assimilation process improved runoff simulations (increasing NS from 0.58 to 0.83) and snow cover simulations (increasing NS from 0.3 to 0.79) compared to a multivariate assimilation scenario without orographic precipitation effects. The improvement in results when considering orographic precipitation (due to the elevation band capability in the SWAT model) is attributed to the presence of high-altitude areas and elevation variations in the upstream region of the Mahabad Dam. Due to the concentration and expansion of agriculture in the downstream area of Mahabad, the hydrological regime in this region is influenced by human activities. The irrigation water demand in the downstream area decreased from approximately 140 million cubic meters (in 2013) to about 90 million cubic meters (in 2017). Given these factors, a synthetic experiment approach was used in the downstream region. In this area, multivariate data assimilation (synthetic runoff observations, surface soil moisture, leaf area index, and groundwater storage) improved the accuracy of water demand simulations from 60% to 97%, compared to a scenario involving the assimilation of two variables (surface soil moisture and groundwater storage). Based on the downstream results, it was observed that while the runoff generation process is correlated with soil moisture and leaf area index processes, the relationship between runoff and soil moisture is significantly stronger than that between runoff and leaf area index. Overall, the use of multivariate data assimilation enhanced the robustness of simulations, preventing deviations in the face of increased simulation uncertainties, and provided a solution to the equifinality problem. The optimization scenario based on the proposed framework in this study (DP-DA) achieved results similar to the optimal reservoir operation scenario (DP-OP), which assumes that the optimization model has access to 100% accurate forecasts of inflow and water demand variables. For example, the average shortage index (for meeting demand) during the optimization period under the DP-DA and DP-OP scenarios was estimated at 16.1 and 22 million cubic meters, respectively. In contrast, running the optimization model without the forecast-informed module (DP-OL) resulted in a shortage index of 49.8 million cubic meters. Similar comparisons apply to reliability indices and the average objective function. Additionally, the time series of reservoir releases (storage) under the DP-DA and DP-OP scenarios were similar to each other and significantly different from the DP-OL scenario.
