Date of Award

Spring 1-1-2016

Document Type


Degree Name

Doctor of Philosophy (PhD)

First Advisor

Balaji Rajagopalan

Second Advisor

Guillermo Podestá

Third Advisor

William Kleiber

Fourth Advisor

Ben Livneh

Fifth Advisor

Joseph Kasprzyk


This dissertation presents three statistical models and applies them to the predominantly rain-fed Argentine Pampas, one of the most productive agricultural regions in the world. The Argentine Pampas experienced an upward trend in annual precipitation since the 1960s; global soybean prices surged shortly thereafter, which provided an optimal combination of climate, economics, and technology, and motivated vast agricultural expansion to semi-arid regions. Annual precipitation totals have declined since the turn of the century, which begs the question: "Are the existing agricultural production systems viable in a drier future?" Stochastic weather generators have long been used to produce synthetic daily weather series to drive process based models, which in turn are used to assess likely impacts on climate-sensitive sectors of society, and to evaluate the outcomes of alternative adaptive actions. Unfortunately, many traditional approaches of stochastic weather generation are limited in their ability to generate space-time weather (i.e., at unobserved locations), or values outside the range of the historical record, which is particularly important for climate change applications in rural agricultural regions, such as the Argentine Pampas.

To this end, we developed a coupled stochastic weather generator (GLMGEN), which takes advantage of the flexibility of generalized linear models (GLMs) to model skewed and discrete variables (i.e., precipitation intensity and occurrence, respectively). Spatial process models estimate the GLM parameters in space to simulate at arbitrary locations, such as on a regular grid. Subsequent application of GLMGEN within a nonstationary context, such as climate change studies, is presented for the Salado A sub-basin of the Argentine Pampas. The inclusion of large-scale climate indices as covariates enables the simulation of daily weather ensembles that exhibit the traits and trends of seasonal forecasts and climate model projections. Regional climate model, experiment RCP8.5, and two IRI seasonal forecasts are used to condition the output of GLMGEN, thus translating this coarse scale climate information into more salient information for decision makers. In addition, we present a Bayesian stochastic weather generator (BayGEN), which quantifies and preserves the uncertainty associated with all model parameters. Uncertainty will subsequently propagate to synthetic daily weather ensembles and their respective uses, such as to drive crop simulation and hydrologic models, properly quantifying risk for decision making and climate change adaptation strategies. Direct comparison of BayGEN with GLMGEN will illustrate the benefit of propagating this uncertainty to simulation space. Finally, a statistical space-time hierarchical metamodel for monthly actual evapotranspiration (ET) and monthly water table depth (WTD) was developed as a complementary tool for near real-time decision support. In the first level of hierarchy, ET is modeled as a function of climate and land use decision variables; the second level models WTD as a function of climate and predicted ET. The metamodel was conditioned on and validated by a calibrated hydrologic model (i.e., MIKE-SHE) for the Salado A sub-basin, and is shown to adequately capture the dominant mechanisms of spatial and temporal variability. Use of the metamodel with output from a weather generator, as well as with ensembles of different land uses, can identify regions of high risk by producing distributions of WTD and its response to climate and land use change scenarios.