The results of the comparison are presented in the figure below. On the left-hand side you can see the gridded observations of daily rainfall characteristics in Florida. These four columns represent the four different ways the data was downscaled. The final column is the Bias Corrected Statistic Analog (BCSA) that we developed. Some of the other versions do not do a good job of reproducing daily rainfall variations. Here though, you can see BCSA does do a good job reproducing the daily rainfall variations throughout the year. The first two methods also do not do a great job of capturing the variability between wet seasons and dry spells.

Comparison of downscaling methods.

Another important aspect driving Tampa Bay’s integrated hydrologic model was the spatial variability of the rainfall. The four plots in the next figure measure how variable this rainfall was over space with further distances. The top lines represent observations.

Wet season spatial variability.

In the figure below, the graphs show the eight GCMs over the Tampa Bay region and the whole state of Florida. The gray envelope is the variation across the eight GCMs, and these are mean values each month. The data must be bias-corrected in order to reproduce the respective hydrology. Once that is complete, the GCM data perfectly matches the historical data.

Comparison of raw and downscaled, bias-corrected retrospective P and ET₀

The hydrologic model we are using is a couple of HSPF mod-flow models that have been calibrated to local rainfall weather station data. The figure below is an example of an Integrated Hydrologic Model (HM).

Integrated hydrologic model (IHM).

These next graphs show data for the Hillsboro and Alafia Rivers, which are the two that the Tampa Bay Watershed draws from for water supply. These are the mean monthly observations over the retrospective period, which was the calibration period (I believe) 1982 to 2006. The black line shows the observations. The green line is the model that was calibrated with the local weather station data. The red line is the model driven by NLDAS-2 data, which is the national gridded product of weather data over the entire continent. The blue envelope is the eight GCMs to predict retrospective hydrology.

Comparison of retrospective monthly streamflow predictions.

The image below shows the projected future changes in monthly precipitation, potential evapotranspiration, and water availability not put through the hydrologic model yet. These results are for the entire state of Florida. The blue line is Future 1, which is 2030 to 2060. The red line is Future 2, which is 2070 to 2100. Both models are predicting slightly higher monthly mean precipitation. You can see there is quite a bit of variability between the GCMs and ET method.

Projected changes in monthly precipitation, potential evapotranspiration and water availability.

A Global Sensitivity Analysis was projected across the GCMs, ET method, and water use scenarios to understand and identify the biggest drivers of uncertainty. When looking at streamflow projections, the GCMs are driving the range in the box and whisker plot. The ET method and water use scenarios are much smaller. The results are in the table below.

Global sensitivity analysis.

In this second analysis (results below), the GCM and water use scenarios are more equivalent. This is largely because the groundwater prediction is more local, located near a wetland, and is influenced by urban pumping. Comparatively, the streamflow is an integrated quantity over the entire watershed.

Global sensitivity analysis where GCM and water use scenarios are more equivalent.