Hydrologic Model Calibration in the Cloud
A key concept in achieving an accurate model in a timely manner for large
and distributed watersheds is through the use of parallel calibration algorithms.
Cloud computing offers quick and easy access to shared pools of configurable
computing resources being a good candidate for computing needs as it offers
a pay-for-use model. We used the Dynamically Dimensioned Search (DDS) calibration
method to introduce the Cloud calibration tool for hydrologic models using Microsoft
Windows Azure. Later, I evaluated the Cloud calibration tool for a small watershed
and quantified the speedup and cost of calibration against different sized watersheds,
model durations and number of cores used in the Cloud. For the Neuse watershed,
the calibration time reduced from 207 hours (8.6 days) on a personal computer to 3.4
hours using 256 cores for the cost of $69.94 on the Cloud. We also presented a relationship
to estimate the calibration time and cost for any SWAT model based on the number of
Hydrologic Respond Units, time steps, and compute cores.
Figure: The Cloud Speedup for Different Size Watersheds and Time Spans*
*Ercan, M. B., J. L. Goodall, A. M. Castronova, M. Humphrey and N. Beekwilder (2014),
Calibration of SWAT models using the Azure Cloud.,
Environmental Modelling & Software, 62, 188-196.