JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 2000 / 145 REGIONAL DROUGHT ANALYSIS BASED ON NEURAL NETWORKS By Hyun-Suk Shin 1 and Jose D. Salas, 2 Member, ASCE ABSTRACT: The main objective of the research reported herein has been to develop an approach to analyze and quantify the spatial and temporal patterns of meteorological droughts based on annual precipitation data. By using a nonparametric spatial analysis neural network algorithm, the normalized and standardized precipi- tation data are classified into certain degrees of drought severity (for example, extreme drought, severe drought, mild drought, and nondrought) based on a number of truncation levels corresponding to specified quantiles of the standard normal distribution (the 15%, 35%, and 50% quantiles were used here for illustration). Then posterior probabilities of drought severity at any given point in the region are determined and the point is assigned a Bayesian Drought Severity Index depending on whether the maximum posterior probabilities corre- spond to extreme, severe, mild, or nondrought. This index may be useful for constructing drought severity maps that display the spatial variability of drought severity for the whole region on a yearly basis. Furthermore, the severity of the drought event for the region as a whole and the sequence and duration of drought episodes through time can be determined. The proposed regional drought analysis approach was applied to analyze and quantify regional droughts for the southwestern region of Colorado. The results were useful for deriving maps of precipitation fields for the entire region, maps of posterior probability of drought severity, and maps of drought severity indices. They were useful for visualizing the spatial pattern of droughts and for deriving other drought properties such as duration. The results obtained suggest that the proposed approach is a viable tool for analyzing and synthesizing droughts on a regional basis. INTRODUCTION Knowledge of droughts has been an important aspect in the planning and management of water resources systems. Res- ervoirs are often planned so that they will be able to supply the expected water demands during a drought of a certain mag- nitude, and water supply systems are often evaluated to see whether they will be able to withstand a T-year drought (Frick et al. 1990). In any case, determining drought properties at a point and in space (or region) is an important aspect of water planning and management activities. Drought analysis may be made based on single site data (Yevjevich 1967; Dracup et al. 1980) and multisite data (Tase 1976; Santos et al. 1983; Gutt- man et al. 1992; Soule 1992), depending on the specific pur- pose of the study at hand. In this paper, we are concerned with regional droughts, so our analysis will be based on data mea- sured at several sites in space. To analyze droughts statistically, it is necessary to specify the following factors: (1) the climatic or hydrologic variable that will be used for defining droughts; (2) the characterization of the spatial distribution of the underlying variable; (3) the truncation levels for classifying the severity of a drought at a point; and (4) the quantification of the regional drought. A number of climatic and hydrological variables, such as precip- itation, streamflow, soil moisture, ground-water levels, mois- ture content in the air, and similar other variables, have been widely used in the literature for characterizing regional droughts. Precipitation has been commonly used for meteor- ological drought analysis (Pinkayan 1966; Tase 1976; Santos et al. 1983; Chang 1991; Eltahir 1992), while streamflow data have been widely applied for hydrologic drought analysis (Dracup et al. 1980; Sen 1980; Zelenhasic and Salvai 1987; Wang and Salas 1989; Chang 1990; Frick et al. 1990; Clausen and Pearson 1995). In this study, annual precipitation will be considered as the key variable for drought analysis. 1 Asst. Prof., Dept. of Civ. Engrg., Pusan Nat. Univ., Pusan, South Korea. 2 Prof., Dept. of Civ. Engrg., Colorado State Univ., Fort Collins, CO 80523. Note. Discussion open until September 1, 2000. To extend the closing date one month, a written request must be filed with the ASCE Manager of Journals. The manuscript for this paper was submitted for review and possible publication on January 20, 1998. This paper is part of the Jour- nal of Hydraulic Engineering, Vol. 5, No. 2, April, 2000. ASCE, ISSN 1084-0699/00/0002-0145–0155/$8.00 + $.50 per page. Paper No. 17408. In general, historical precipitation data have different record lengths and are observed at a limited number of gauging sta- tions in the area of interest. Variability in space may be taken into account by using an interpolation technique so as to es- timate precipitation quantities at any ungauged location in the area. Several interpolation methods such as Thiessen polygons, inverse distance, multiquadric, polynomial, and kriging have been used for this purpose (Tabios and Salas 1985). For in- stance, Tase (1976) used the polynomial method to regionalize precipitation data, Chang (1991) applied kriging for investi- gating monthly precipitation droughts, and Kingery (1992) used the Thiessen polygon method for interpolating monthly and annual precipitation data for drought analysis. In the pres- ent study, a nonparametric spatial analysis combined with a neural network computational scheme is used to estimate the spatial distribution of the precipitation field. Once the precip- itation field is characterized, it is subjected to further analysis for deriving drought properties. Various drought severity indices based on precipitation data have been introduced in the literature. Gibbs and Maher (1967) used deciles of precipitation for characterizing droughts in Australia. Monthly (or annual) precipitation totals are ranked from highest to lowest, and decile ranges are determined from the cumulative frequency distribution. The Palmer drought se- verity index, developed by Palmer (1965), has become a widely used meteorological drought index (Karl and Quayle 1981; Guttman et al. 1992; Akinremi et al. 1996). It relates drought severity to the accumulated weighted differences be- tween actual precipitation and evaporation. McKee et al. (1993) introduced the standardized precipitation index, which is designed to quantify the precipitation deficit for multiple time scales. In the present study, after estimating the precipi- tation field in the region of concern, the posterior probabilities of drought severities at any point in the region, such as ex- treme drought, severe drought, mild drought, and nondrought, are determined. From such point drought probabilities, drought severity probability maps can be obtained. For determining droughts of various degrees of severity, appropriate truncation (demand) levels of the underlying precipitation data must be specified. In principle, truncation levels may vary in space and time, but such data may not be readily available. The problem may become too cumbersome because of the variety of geo- morphic or climatic conditions over the region and over time. That is why the historical annual precipitation data are usually