%0 Journal Article %1 2016AGUFM.H33C1557M %A Molina, J. M. %A Zaitchik, B. F. %B AGU Fall Meeting Abstracts %D 2016 %I American Geophysical Union %J AGU Fall Meeting Abstracts %K Andes Climate Colombia Downscaling Hydroclimatology Hydrology Precipitation Tropics Variability myown %P H33C-1557 %R 10.13140/rg.2.2.22452.30080 %T Assimilating Non-linear Effects of Customized Large-Scale Climate Predictors on Downscaled Precipitation over the Tropical Andes %U https://doi.org/10.13140/rg.2.2.22452.30080 %V 2016 %X Recent findings considering high CO2 emission scenarios (RCP8.5) suggest that the tropical Andes may experience a massive warming and a significant precipitation increase (decrease) during the wet (dry) seasons by the end of the 21st century. Variations on rainfall-streamflow relationships and seasonal crop yields significantly affect human development in this region and make local communities highly vulnerable to climate change and variability. We developed an expert-informed empirical statistical downscaling (ESD) algorithm to explore and construct robust global climate predictors to perform skillful RCP8.5 projections of in-situ March-May (MAM) precipitation required for impact modeling and adaptation studies. We applied our framework to a topographically-complex region of the Colombian Andes where a number of previous studies have reported El Niño-Southern Oscillation (ENSO) as the main driver of climate variability. Supervised machine learning algorithms were trained with customized and bias-corrected predictors from NCEP reanalysis, and a cross-validation approach was implemented to assess both predictive skill and model selection. We found weak and not significant teleconnections between precipitation and lagged seasonal surface temperatures over El Niño3.4 domain, which suggests that ENSO fails to explain MAM rainfall variability in the study region. In contrast, series of Sea Level Pressure (SLP) over American Samoa -likely associated with the South Pacific Convergence Zone (SPCZ)- explains more than 65% of the precipitation variance. The best prediction skill was obtained with Selected Generalized Additive Models (SGAM) given their ability to capture linear/nonlinear relationships present in the data. While SPCZ-related series exhibited a positive linear effect in the rainfall response, SLP predictors in the north Atlantic and central equatorial Pacific showed nonlinear effects. A multimodel (MIROC, CanESM2 and CCSM) ensemble of ESD projections revealed an increased variability and a positive and significant trend in the MAM precipitation mean in the next decades, with accentuated changes and projection uncertainty after 2050. ESD traces (2050-2100) from MIROC presented the highest changes in the precipitation mean ( 60%) when compared with the observations.

@article{2016AGUFM.H33C1557M, abstract = {Recent findings considering high CO2 emission scenarios (RCP8.5) suggest that the tropical Andes may experience a massive warming and a significant precipitation increase (decrease) during the wet (dry) seasons by the end of the 21st century. Variations on rainfall-streamflow relationships and seasonal crop yields significantly affect human development in this region and make local communities highly vulnerable to climate change and variability. We developed an expert-informed empirical statistical downscaling (ESD) algorithm to explore and construct robust global climate predictors to perform skillful RCP8.5 projections of in-situ March-May (MAM) precipitation required for impact modeling and adaptation studies. We applied our framework to a topographically-complex region of the Colombian Andes where a number of previous studies have reported El Niño-Southern Oscillation (ENSO) as the main driver of climate variability. Supervised machine learning algorithms were trained with customized and bias-corrected predictors from NCEP reanalysis, and a cross-validation approach was implemented to assess both predictive skill and model selection. We found weak and not significant teleconnections between precipitation and lagged seasonal surface temperatures over El Niño3.4 domain, which suggests that ENSO fails to explain MAM rainfall variability in the study region. In contrast, series of Sea Level Pressure (SLP) over American Samoa -likely associated with the South Pacific Convergence Zone (SPCZ)- explains more than 65% of the precipitation variance. The best prediction skill was obtained with Selected Generalized Additive Models (SGAM) given their ability to capture linear/nonlinear relationships present in the data. While SPCZ-related series exhibited a positive linear effect in the rainfall response, SLP predictors in the north Atlantic and central equatorial Pacific showed nonlinear effects. A multimodel (MIROC, CanESM2 and CCSM) ensemble of ESD projections revealed an increased variability and a positive and significant trend in the MAM precipitation mean in the next decades, with accentuated changes and projection uncertainty after 2050. ESD traces (2050-2100) from MIROC presented the highest changes in the precipitation mean ( 60%) when compared with the observations.}, added-at = {2020-09-04T19:47:43.000+0200}, adsnote = {Provided by the SAO/NASA Astrophysics Data System}, adsurl = {https://ui.adsabs.harvard.edu/abs/2016AGUFM.H33C1557M}, author = {{Molina}, J. M. and {Zaitchik}, B. F.}, biburl = {https://www.bibsonomy.org/bibtex/27c6432ed71cad07481a73b5414b11f5c/jomolina10}, booktitle = {AGU Fall Meeting Abstracts}, doi = {10.13140/rg.2.2.22452.30080}, eid = {H33C-1557}, interhash = {27f1b7e9f13384075f2c060816975d8e}, intrahash = {7c6432ed71cad07481a73b5414b11f5c}, journal = {AGU Fall Meeting Abstracts}, keywords = {Andes Climate Colombia Downscaling Hydroclimatology Hydrology Precipitation Tropics Variability myown}, month = dec, pages = {H33C-1557}, publisher = {American Geophysical Union}, timestamp = {2020-09-10T22:26:23.000+0200}, title = {{Assimilating Non-linear Effects of Customized Large-Scale Climate Predictors on Downscaled Precipitation over the Tropical Andes}}, url = {https://doi.org/10.13140/rg.2.2.22452.30080}, volume = 2016, year = 2016 }