Evaluation and improvement of quantitative precipitation estimates for use in hydrological applications
MetadataShow full item record
Accurate precipitation estimation at the catchment scale is required for various hydrological applications. Spatial interpolation using gauges as a single source often proves insufficient to obtain a distributed precipitation field with accurate spatial variation adequately. In this aspect, remote sensing (weather radar and satellite) and numerical weather prediction (NWP) model-based data hold promise as they provide precipitation estimates over a large area with high spatial and temporal resolution. However, these data sources other than gauges are still not widely used in hydrological applications due to the presence of significant errors and uncertainties in them. This thesis aims to improve the precipitation estimates with a high spatiotemporal resolution by using different data sources for hydrological applications. The primary focus was to utilise the long-term radar precipitation rates, available from the Norwegian national weather radar network. The work is mainly based on statistical methods and tools as well as hydrological model. Given the availability of a large amount of radar rain rates data, earlier studies showed that nonparametric approaches produce more reliable radar-rainfall estimates compared to a traditional parametric Z – R relationship. In paper 1, a nonparametric k-nearest neighbour (k-nn) technique was employed to estimate radar precipitation estimates. Since air temperature is intrinsic to the occurrence of different phases of the precipitation in cold climates, the study investigated the use of air temperature as an additional covariate within the k-nn to improve the radar precipitation estimates in cold climates. Four years (2011 – 2015) of hourly radar precipitation rates from the Norwegian national radar network over the Oslo region, hourly gauge precipitation from the available gauges and gridded observational air temperature were used to formulate the k-nn predictive model for the investigation. In paper 2, the study formulates a framework to merge the two sources of precipitation to generate an hourly spatial precipitation field. For that, the study adopted the k-nn model presented in paper 1 and the dynamic forecast combination method, prevalent in combining seasonal forecasts from multiple climate models. In this paper, the k-nn estimates were merged with Thin Plate Spline (TPS) interpolated gauge precipitation on a regular grid. In the dynamic combination, radar precipitation and air temperature were used as two variables in the algorithm to identify similar events to ascertain the optimal combination weight. In paper 3, the study evaluated the merger of satellite-based precipitation with radar precipitation to get improved precipitation estimates for the study region. First, the study utilised the radar-gauge merged estimates in paper 2 as a ground reference to assess the Global Precipitation Measurement (GPM) data. Then the study investigated to merge the k-nn radar precipitation estimates with GPM employing the forecast combination method presented in paper 2. Paper 4 evaluated the performance of NWP model-based meteorological reanalysis (precipitation and air temperature) data, comparing them with gauge and gauge-based gridded datasets and as input for calibrating a precipitation-runoff model used in operational inflow predictions for three hydropower systems in middle Norway. The study employed a Monte Carlo approach of model calibration to assess the uncertainty in the model parameters and the simulated response depending on atmospheric forcing datasets. The use of air temperature as a covariate in a k-nn model presented in paper 1 significantly improves the radar precipitation estimation compared to the model with the radar precipitation rate as a sole predictor. However, the temperature effects became insignificant for temperatures warmer than 10_ C. The proposed merging framework in paper 2 considerably decreased the bias in the radar precipitation while it reduced the mean squared error by one-fourth of the errors associated with the original hourly radar precipitation rates. Air temperature plays a significant role in the k-nn model to ascertain radar precipitation estimates; however, it contributes marginally in the dynamic combination. The performance of GPM precipitation in high latitudes is encouraging; however, merging of hourly GPM did not contributes to improving the radar precipitation field. Paper 4 showed that model-based data can be a preferable alternative for calibrating hydrological models used for inflow predictions. Paper 1: Estimating radar precipitation in cold climates: the role of air temperature within a non-parametric framework. Paper 2: Merging radar and gauge information within a dynamical model combination framework for precipitation estimation in cold climates. Paper 3: Evaluation of Global Precipitation Measurement (GPM) as a potential precipitation source in a high latitude area. Paper 4: Can model-based data products replace gauge data as input to hydrological model?