Development of a stochastic storm generator using high-resolution precipitation records

Authors: REVUELTA-ACOSTA, JOSEPT - Purdue University; Flanagan, Dennis; ENGEL, BERNARD - Purdue University

Submitted to: Applied Engineering in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/11/2019
Publication Date: 9/4/2019
Citation: Revuelta-Acosta, J.D., Flanagan, D.C., Engel, B.A. 2019. Development of a stochastic storm generator using high-resolution precipitation records. Applied Engineering in Agriculture. 35(4):461-473. https://doi.org/10.13031/aea.13259.
DOI: https://doi.org/10.13031/aea.13259

Interpretive Summary: Many natural resource problems are often evaluated using computer simulation models, because models can quickly and efficiently produce estimated predictions related to processes such as surface runoff, plant growth, soil erosion, sediment delivery, and pollutant transport. Measurement in the field at a large number of locations can be very expensive, time-consuming, and often impractical; models can help target locations where field monitoring may be worthwhile. One of the most important inputs to natural resource models is climate, and if observed weather information is not available or not complete, climate generator software is often used. This typically relies upon statistical values determined from measurements at weather stations, and then creation of strings of daily synthetic values for precipitation, temperatures, etc. In this study we developed a new storm generator using detailed weather station rainfall intensity measurement information. The new generator can capture the important features of rainstorms, including total storm depth, storm intensities, storm patterns, and extreme events. This impacts scientists, natural resource model users, decision-makers, and others involved in application of models and/or use of their results. Improved prediction of within-storm rainfall intensities should improve predictions of other processes relying upon those (infiltration, runoff, soil detachment, etc.).

Technical Abstract: Sophisticated field and watershed scale environmental models for runoff, erosion control, environmental, and global-change investigations require detailed continuous temporal and spatial inputs of precipitation to drive the hydrologic processes. For accurate estimates of these processes, the resolution of the input data must allow the representation of the variability of precipitation as it represents a major source of variability in the model outputs. Currently, the use of stochastic weather generators is wide spread to generate continuous series of meteorological data at gauged and ungauged locations. These weather simulators are designed to replicate the statistical properties of real weather data at monthly or daily time resolutions. However, daily values of precipitation do not represent the variability of storm parameters within a day, which is assumed to significantly influence the predictions of environmental or agricultural models where processes are sensitive to sub-daily values. This research proposes a parsimonious stochastic storm generator based on 5-minute time resolution and correlated non-normal Monte Carlo-based numerical simulation. The model considers correlated non-normal random rainstorm characteristics such as time between storms, duration, and amount of precipitation, as well as the storm intensity structure. The accuracy of the model was verified by comparing the generated rainfall with rainfall data from a randomly selected 5-minute weather station in North Carolina. Current results have shown that the proposed storm generator can capture the essential statistical features of rainstorms as well the patterns followed by their intensities, preserving the first four moments of monthly storm events, good annual extreme event correspondence, and the correlation structure within each storm. Finally, as the proposed model depends on statistical properties at a site, this may allow the use of the synthetic storms in ungauged locations provided relevant information from a regional analysis is available.