Previous work by the author on stochastic relationships between dengue and the climatic water budget has shown significant associations between water budget factors and intra- and interannual variations of unconfirmed cases of dengue in San Juan, Puerto Rico, particularly with regard to the seasonality of dengue. Results from this earlier study suggested that the causal climate mechanisms might precede the dengue response by more than the eight weeks analyzed in the study. The purpose of this study was to analyze the association of unconfirmed dengue cases (1988-1993) with climatic water budget factors over a full six months preceding the dengue response. In addition, alternative data smoothing procedures and part-year regressions were explored with the ultimate goal of improving the explanatory ability of dengue/water budget multivariate regressions. An effective water budget-based dengue predictive model would be advantageous because factors of the water budget are based on routinely collected, widely available, and low-cost weather observations.

The 'all-regressions' approach was applied to 26 weeks of smoothed (over 2 and/or 3 weeks) daily unconfirmed dengue counts and water budget factors (precipitation, potential evapotranspiration, actual evapotranspiration, storage, deficit, surplus, minimum temperature, and maximum temperature, and a variable for each listed factor which represented the change in that factor from the preceding two weeks). This procedure considers all possible combinations of the regressors to determine the model with highest explained variance for each total number of regressors desired. Four models were initially developed: 1) an intrannual model, to explain the average annual change in daily dengue throughout the progression of the year, 2) an interannual model explaining daily dengue variation across the study period, 3) a predictive interannual model based water budget factors at least four weeks in advance of the dengue response, and 4) a interannual model utilizing standardized dengue to determine association of the water budget to relatively high or low dengue accounting for the particular time of year. A fifth model was later developed relating standardized dengue in the months of September-December (generally peak dengue in this period) throughout the study period to water budget factors in the preceding 26 weeks.

Resulting models once again demonstrate the particular importance of energy variables in the seasonality of dengue, and water availability-related factors in association with standardized dengue, although both types of variables were present in each model. In addition, models show significant improvements in explained variance over their predecessors: model 1- 95%, model 2- 62%, model 3- 59%, model 4- 36%, and model 5- 78%. Thus, most of the seasonal variation in dengue was explained by seasonal variations in the water budget, and results from model 5 suggest a water budget-based dengue model might ultimately result in a useful tool for dengue prediction.