Greetings from the National Weather Center.
Visiting the South Central Climate Science Center at Norman (OK) and getting ready to travel to Fort Worth for the annual meeting with the rest of our partners !!
Attended at the ISI 2015 60th World Statistics Congress in Rio de Janeiro, Brazil.
On Wednesday 29th I led a discussion about Climate change impacts on society and economy and on Friday the 31st I gave a talk in the Agriculture and food security under climate change session. The title of the talk was: Statistical Downscaling and the effect of the time-invariance assumption on growing season lengths, heat wave durations and other climate indices
The South-Central Climate Science Center highlights our work on statistical downscaling in their Annual Report. Here is a section of the report:
Evaluation of Statistical Downscaling Techniques
GCM results typically lack fine-scale detail and may contain biases that make it inappropriate to use the raw GCM output in studies of projected regional or local-scale climate impacts. Informed by observational data sets, statistical downscaling (SD) techniques are often applied to refine GCM output in an attempt to account for shortcomings in a GCM's simulation of local climate. A variety of SD methods
exist and can provide dramatically different results from the same source data (right). It typically
is assumed that the skill exhibited by a SD method during the historical period will be retained in the future even as the climate is changing - an untested assumption with potentially large
implications for the quality of SD output used as input to climate impacts analyses. Dr. Carlos Gaitán, a SC-CSC research scientist, is working with collaborators at NOAA’s GFDL to test this
assumption. Carlos is working on a novel experimental design, known as a “Perfect Model” design, to compare the downscaled methods’ performance for both historical and future periods through the
use of synthetic data.
Check out my newest article featured in the first page of Climate Dynamics (43)-12 3201-3217. Thanks to my co-authors at UBC and to the reviewers for their constructive comments.
Working on a comparison of quantile mapping methods for the South Central Region of the U.S.
Check out my newest article:
The Journal of Agricultural Science, FirstView Articles
My new journal article about evaluation of different linear and nonlinear regression methods applied to statistical downscaling of temperatures in southern Quebec and Ontario is available to download here: http://www.tandfonline.com/toc/tato20/current#.UpDrx6USN4M
The figure on the left shows multiple linear regression methods with white symbols and nonlinear artificial neural networks models with black symbols. Four different predictors were analyzed a) ALL (using a combination of upper level and surface atmospheric variables), B) T (using only temperature at 2 m., C) PC (using the leading principal components) and D) HD, using the 500 hPa - 850 hPa geopotential height differences.
The results show that the nonlinear BNNALL model outscored the other 5 models in terms of MAE (y axis) and in terms of the average index of agreement from 6 different climate indices (x axis) derived from the STARDEX project.
Thanks Oklahoma for the great time. I enjoyed visititing OU and the South Central Climate Science Center.
Just attended the 2013 Climate Informatics Workshop, in Boulder, CO. Thanks NCAR Mesa lab for hosting us.
This month, I found out that the northern temperature range of the Northern Cardinal overlaps with the region I studied during my PhD. The image on the left shows in red the Northern Cardinal (approximate) range. The purple squares indicate the 10 locations I studied in southern Ontario and Quebec, Canada.
Attending the NCPP Qualitative Evaluation of Downscaling Workshop at NCAR Foothills Laboratory, Boulder, Colorado.
Please see the presentation about the Perfect Model evaluation framework to familiarize yourself with our current research goals.
Going back to UBC to defend my doctoral dissertation " Comparison of linearly and nonlinearly statistically downscaled atmospheric variables in terms of future climate indices and daily variability.