Investigating Rainfall Concurrence Based on Radar Measurements for the Eastern Iowa Flood of 2008
In summer of 2008 eastern Iowa experienced record flooding described as a 500-year event. Analysis of the storm that led to this extreme flooding in the Iowa and the Cedar River basins is crucial for the better understanding of hydrologic responses as well as the mitigation of similar disasters in the future. Using radar measurements to describe main aspects of rainfall fields, instead of sparsely distributed rain gauges, provides the benefit of higher resolution information in space and time. Although data from four WSR-88D weather radars operating in Minneapolis (MN), La Crosse (WI), Des Moines (IA), and Davenport (IA) can be used to capture rainfall features of these extreme events, none of them fully covers the entire Iowa and Cedar River basins. Furthermore, since the operation of these radars are not synchronized and they all 'look' over the basin from different directions, their measurements might be affected by calibration biases and different strength of range-dependent biases. Thus, unknown factors that cause significant measurement and estimation differences between individual radar observations need to be characterized or quantified before the data from the four radars are combined to cover the entire basins. The authors present comparison results based on the reflectivity measurements between individual radars and demonstrate how these differences affect the quantitative estimation of extreme, flood-producing rainfall.
Sensitivity of simulated land surface states to different GPM-proxy precipitation inputs
It is expected that GPM-era satellite precipitation estimates will improve a suite of environmental applications, which utilize routinely satellite rainfall data. Advances in satellite precipitation estimates might result in improved understanding of regional/global energy and water cycles. Hence, it is important to quantify sensitivities of land surface states simulated by land surface models (LSM) to potential uncertainties in satellite precipitation. The suite of GPM proxy precipitation data was used to evaluate sensitivity of different LSMs to variations in the precipitation forcing over the Southern Great Plains during spring-fall 2008. Noah and Mosaic LSMs available within NASA's Land Information System (LIS) framework were used to simulate surface and subsurface state variables. Both models were configured at 0.1x0.1 latitude-longitude grid over the domain (with 200x80 grid points) covering south-central United States. To produce realistic soil moisture fields, both LSMs were integrated for 2.5 year period (from 1 Jan. 2005 to 1 June 2007) using NLDAS forcing fields including precipitation. Initially, a constant 30 % volumetric soil moisture content was assigned at all grid points. After the 2.5 year spin-up time, the Noah and Mosaic LSMs were additionally integrated for 16 months (until 31 Oct. 2008) using eight different GPM-proxy precipitation products and the NLDAS atmospheric fields (except for precipitation). Typical seasonal uncertainties in simulated soil moisture and surface fluxes resulted from precipitation variations will be examined and compared over semi-arid and relatively humid regions. Relative impact/importance of precipitation errors, having different characteristics (amount and frequency deviations) on soil moisture and surfaces fluxes, will be also explored.
A NeuroEvolution Approach to Reconstructing Missing Daily Precipitation Data
The Caribbean region consists of several Small Island Developing States (SIDS). Many of these islands are extremely vulnerable to the impacts of climate change including sea level rise, changes in rainfall regimes and increasing day and night-time temperatures. In addition, the region is extremely susceptible to natural disasters, the most frequent of which is flooding. As part of its disaster management and mitigation strategy several Caribbean SIDS are developing a parametric insurance policy for flooding. The insurance mechanism is contingent on an adequate precipitation data set. However, missing data within the precipitation time series is one of the issues that is ubiquitous within the region, presenting a challenge to the implementation of such a rainfall-triggered insurance policy. The purpose of this study is to investigate the plausibility of using a class of Evolutionary Computation (EC) algorithms known as a Genetic Algorithms (GAs) for evolving a near-optimal Artificial Neural Network (ANN) to reconstruct missing daily precipitation data. The structure of the artificial neural network used for rainfall imputation in this study has a feedforward Multi-layer Perceptron (MLP) architecture. The predictive capability of this novel hybrid ANN-GA precipitation data infilling technique is tested by degrading 10%, 20% and 30% of a daily precipitation data set, from the period 1997-2000, in a stochastic manner and estimating the missing observations. Tests performed using the proposed approach show the method predicts the randomly selected missing precipitation data with reasonable reliability for a rain gauge in Region 3, Guyana. Research is currently on going to assess the efficacy of the artificial neural network technique for longer time series, which is required to support climate change impact on the Caribbean region.
Trivariate Copula for Extreme Rainfall Events
Estimation of design storms and associated risk requires detailed knowledge of the three major storm characteristics: peak intensity, volume, and duration. To date, only little work has been done to apply copula modeling to trivariate problems in the context of extreme rainfall analysis. No work has been done to apply trivariate copula modeling at fine (15 min or less) temporal resolution. In this study, we examine trivariate copula modeling for extreme rainfall analysis using 15-min time series rainfall data from 12 stations in Connecticut. Our focus is particularly on metaelliptical copulas, since they offer convenience and flexibility. A major issue in this application is that the sample size at most stations are small, ranging from 10 to 33, because the 15-minute precipitation data are only available fairly recently. For each station, we propose to estimate the model parameters by maximizing a weighted likelihood, which assigns weight to data at stations nearby, borrowing strengths from them. The weights are assigned by some kernel function whose bandwidth is chosen by cross-validation in terms of predictive log likelihood.
Evaluating Satellite Rainfall Products and Their Impacts in Hydrologic Model Simulations
High-resolution rainfall products across the globe are increasingly becoming available from space-based platforms. However, examples of the operational uses of these products are rare. In this study, we evaluate the accuracy of high-resolution satellite rainfall products by comparing them with high-quality and dense ground-based rainfall observations, and assess their hydrological impacts by comparing observed streamflows with hydrologic model simulations forced by satellite rainfall products. We focus our work on the Little Washita watershed (610 km2) in central Oklahoma, which contains high-quality radar datasets, a dense network of rain gauges (42 rain gauges), five stream gauges, and three meteorological stations. In the first part of the study, we describe the accuracy of the satellite rainfall products on different space-time resolutions, using a variety of performance metrics. In the second part, we describe the performance of the hydrologic model forced by satellite rainfall products with respect to reproducing observed streamflow data. We used a physically-based, fully-distributed hydrologic model, known as the TIN-based Real-time Integrated Basin Simulator (tRIBS) for this investigation. The results of this study will shed light on the accuracy of satellite rainfall products, both from meteorological and hydrological perspectives.
Evaluation of a Satellite-based Near Real-time Global Flood Prediction System
Satellite-based rainfall and geospatial datasets are potentially useful for cost effective detection and early warning of natural hazards, such as floods, specifically for regions of the world where local data are sparse or non-existent. Recently, our group has implemented an initial satellite-based near real-time global flood prediction system that is operationally available. In this system, a relatively simple hydrologic model, based on the runoff curve number (CN) and antecedent precipitation index (API) methods, transforms rainfall into runoff. Runoff is then routed grid-to-grid to estimate flow. The key input to the current system is the near real-time rainfall estimates from the NASA-based Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA; 3 hourly, 0.258 x 0.258 degree). In this study we will present an in-depth testing/evaluation of this current flood prediction system, discuss its strengths and limitations and point toward potential improvements necessary for increasing its near real-time global flood prediction reliability and accuracy. This evaluation study focuses on the severe flooding events and will include comparison of the current product with observed runoff and inundation data at global and watershed scale as well as with other available remotely sensed products, such as those from Dartmouth Flood Observatory. Initial evaluation suggests that current global near-real time flood predictions provide valuable information related to spatial extent and onset time of extreme flooding events. However the accuracy diminishes in tracking the later stages of the flood event. This behavior suggests that one way to improve the current system is a new (possibly finer scale) routing component.
Dependence of the global statistics of cloud properties on the averaging resolution
Analyses of the atmospheric water cycle, the impact of clouds on climate and theoretical predictions of the feedbacks between the hydrological cycle and the atmosphere require the knowledge of statistical distributions of water vapor and cloud characteristics. We use data from MODIS on Aqua to show how the global probability distribution functions of cloud fraction, cloud water path, cloud effective radii and water vapor are a function of the resolution of the averaging size. We discuss the implications for our description of of the hydrological cycle.