Hong, T.; White, P.A. 2015 Rainfall recharge estimation based on a nonlinear Bayesian technique with a dynamic state-space formulation in the Canterbury Plains. Lower Hutt, N.Z.: GNS Science. GNS Science report 2014/37 43 p.
Abstract: The “SMB-SMC” model described in this report is a newly developed method to calculate rainfall recharge to groundwater. The model combines a soil moisture balance (SMB) with the sequential Monte Carlo (SMC) filtering technique to estimate actual evapotranspiration (AET) from potential evapotranspiration (PET). The SMC filtering technique was developed with a SMB in a dynamic Bayesian state-space formulation that aims to improve the estimation of rainfall recharge to groundwater. This model was developed and tested using rainfall recharge measurements (including rainfall, rainfall recharge, and PET at a daily time step) that were made at four lysimeter monitoring sites (Christchurch Airport, Hororata, Lincoln, and Winchmore) in the Canterbury Plains region, during the period 1 July 2000 to 30 June 2004. Generally, the SMB-SMC model performed reasonably well in the estimation of rainfall recharge. The model simulated rainfall recharge patterns at sites that were very similar to observations, i.e., plots of cumulative rainfall recharge commonly show steps in winter and spring. Total rainfall recharge calculated by the model at the sites were generally similar to observations, i.e.: Christchurch Airport was 696 mm (623 mm observed), Hororata 871 mm (919 mm observed), Lincoln 380 mm (270 mm observed) and Winchmore 815 mm (849 mm observed). Rainfall recharge estimates from the SMB-SMC model were compared with the SOILMOD/DRAIN model, which is used in New Zealand to estimate groundwater recharge at the sub-regional scale and to calculate sub-regional groundwater allocation limits (Scott, 2004). The SMB-SMC model performed similarly to SOILMOD/DRAIN in predictions of cumulative rainfall recharge for the 2000-2004 period. For example, the SMB-SMC model estimates were closest to observed values at Christchurch Airport, Hororata, and Winchmore, whereas SOILMOD/DRAIN estimates were closest to observed values at Lincoln. Generally, the SMB-SMC model provided better quality seasonal estimates of rainfall recharge than the SOILMOD/DRAIN model as measured by differences between model calculations and observations, and the standard deviation of these differences. However, the SMB-SMC model provided relatively poor estimates of winter rainfall recharge at Lincoln. (auth)
