HR: 1330h
AN: H12B-0995    [PDF]
TI: Florida Pre-Drainage Everglades Rainfall-Stage Statistical Model as a Step for Rainfall Driven Operation
AU: * Ali, A
EM: aali@sfwmd.gov
AF: South Florida Water Management District, 3301 Gun Club Rd., West Palm Beach, FL 33406 United States
AU: Baldwin, C
EM: ckbald1@yahoo.com
AF: Vidya Inc, 736 S. Majestic Dr., Logan, UT 84321 United States
AU: Lall, U
EM: ula2@columbia.edu
AF: Vidya Inc, 736 S. Majestic Dr., Logan, UT 84321 United States
AU: Cadavid, L
EM: cadavid@sfwmd.gov
AF: South Florida Water Management District, 3301 Gun Club Rd., West Palm Beach, FL 33406 United States
AU: Obeysekera, J
EM: jobey@sfwmd.gov
AF: South Florida Water Management District, 3301 Gun Club Rd., West Palm Beach, FL 33406 United States
AB: The Comprehensive Everglades Restoration Plan, CERP, is designed to capture, store and redistribute fresh water previously lost to tide and to regulate the quality, quantity, timing and spatial distribution of water flows in South Florida. One of the major CERP components is to improve water deliveries to the Everglades. To improve delivery quantity and timing, the existing calendar based regulation schedules are to be replaced with a "Rainfall Driven Operation" (RDO) procedure applied at a short time scale (e.g. weekly). An important step towards a successful RDO is to estimate the pre-drainage (or "natural") stage (water level) response at multiple control locations to recent and/or projected weekly rainfall. Physical modeling of this process in real time is considered difficult. Extensive data analysis shows that this system is non-linear. An Artificial Neural Network, ANN, model is developed to estimate the relationship between multiple stage and rainfall locations for natural conditions considering eight locations throughout the Everglades, and using the historical (1965 -1995) output of the Natural System Model (NSM) developed by the South Florida Water Management District. The ANN uses a combination of stage and rainfall at different time steps as predictors and stage at the next time step as predictant. Stage thresholds at which the stage-rainfall response changes are identified. The ANN regression procedure captures the multiple thresholds and the multi-stage response in a relatively parsimonious manner. Model performance during both dry and wet spells is tested. A strategy for how the estimated natural system conditions can be used in real-time operation of the system is also discussed.
DE: 1800 HYDROLOGY
DE: 6344 System operation and management
SC: Hydrology [H]
MN: 2003 Fall Meeting