%0 Journal Article %1 2013AGUFMGC23B0922M %A Molina, J. M. %A Zaitchik, B. %B AGU Fall Meeting Abstracts %D 2013 %I American Geophysical Union %J AGU Fall Meeting Abstracts %K Africa Change Climate Downscaling Models Precipitation Prediction Variability myown %P GC23B-0922 %R 10.13140/rg.2.1.4674.1447/1 %T Precipitation Prediction in North Africa Based on Statistical Downscaling %U https://doi.org/10.13140/rg.2.1.4674.1447/1 %V 2013 %X Although Global Climate Models (GCM) outputs should not be used directly to predict precipitation variability and change at the local scale, GCM projections of large-scale features in ocean and atmosphere can be applied to infer future statistical properties of climate at finer resolutions through empirical statistical downscaling techniques. A number of such downscaling methods have been proposed in the literature, and although all of them have advantages and limitations depending on the specific downscaling problem, most of them have been developed and tested in developed countries. In this research, we explore the use of statistical downscaling to generate future local precipitation scenarios in different locations in Northern Africa, where available data is sparse and missing values are frequently observed in the historical records. The presence of arid and semiarid regions in North African countries and the persistence of long periods with no rain pose challenges to the downscaling exercise since normality assumptions may be a serious limitation in the application of traditional linear regression methods. In our work, the development of monthly statistical relationships between the local precipitation and the large-scale predictors considers common Empirical Orthogonal Functions (EOFs) from different NCAR/Reanalysis climate fields (e.g., Sea Level Pressure (SLP) and Global Precipitation). GCM/CMIP5 data is considered in the predictor data set to analyze the future local precipitation. Both parametric (e.g., Generalized Linear Models (GLM)) and nonparametric (e,g,, Bootstrapping) approaches are considered in the regression analysis, and different spatial windows in the predictor fields are tested in the prediction experiments. In the latter, seasonal spatial cross-covariance between predictant and predictors is estimated by means of a teleconnections algorithm which was implemented to define the regions in the predictor domain that better captures the variability of the observed local process. Also, a split-window approach is used in the cross-validation stage for comparison purposes of the monthly regression schemes, and different pre-processing alternatives of the precipitation records are implemented to reduce the strong skewness observed in the periodic distribution functions. Preliminary results show that bootstrapping approaches like those based on K-Nearest Neighbors (K-NN) resampling improves the preservation of the historical variability, for which the GLM methods exhibit important limitations. It has been also observed the important role that plays both the teleconnections analysis and the normalization pre-processing in the prediction skill. It is expected that the methodologies from this research can be extrapolated to other regions and time scales for the study of climate change impact and water resources management.

@article{2013AGUFMGC23B0922M, abstract = {Although Global Climate Models (GCM) outputs should not be used directly to predict precipitation variability and change at the local scale, GCM projections of large-scale features in ocean and atmosphere can be applied to infer future statistical properties of climate at finer resolutions through empirical statistical downscaling techniques. A number of such downscaling methods have been proposed in the literature, and although all of them have advantages and limitations depending on the specific downscaling problem, most of them have been developed and tested in developed countries. In this research, we explore the use of statistical downscaling to generate future local precipitation scenarios in different locations in Northern Africa, where available data is sparse and missing values are frequently observed in the historical records. The presence of arid and semiarid regions in North African countries and the persistence of long periods with no rain pose challenges to the downscaling exercise since normality assumptions may be a serious limitation in the application of traditional linear regression methods. In our work, the development of monthly statistical relationships between the local precipitation and the large-scale predictors considers common Empirical Orthogonal Functions (EOFs) from different NCAR/Reanalysis climate fields (e.g., Sea Level Pressure (SLP) and Global Precipitation). GCM/CMIP5 data is considered in the predictor data set to analyze the future local precipitation. Both parametric (e.g., Generalized Linear Models (GLM)) and nonparametric (e,g,, Bootstrapping) approaches are considered in the regression analysis, and different spatial windows in the predictor fields are tested in the prediction experiments. In the latter, seasonal spatial cross-covariance between predictant and predictors is estimated by means of a teleconnections algorithm which was implemented to define the regions in the predictor domain that better captures the variability of the observed local process. Also, a split-window approach is used in the cross-validation stage for comparison purposes of the monthly regression schemes, and different pre-processing alternatives of the precipitation records are implemented to reduce the strong skewness observed in the periodic distribution functions. Preliminary results show that bootstrapping approaches like those based on K-Nearest Neighbors (K-NN) resampling improves the preservation of the historical variability, for which the GLM methods exhibit important limitations. It has been also observed the important role that plays both the teleconnections analysis and the normalization pre-processing in the prediction skill. It is expected that the methodologies from this research can be extrapolated to other regions and time scales for the study of climate change impact and water resources management.}, 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/2013AGUFMGC23B0922M}, author = {Molina, J. M. and Zaitchik, B.}, biburl = {https://www.bibsonomy.org/bibtex/20693c1a793d1cb2b3d187c225d95657c/jomolina10}, booktitle = {AGU Fall Meeting Abstracts}, doi = {10.13140/rg.2.1.4674.1447/1}, eid = {GC23B-0922}, interhash = {ea97f35cd8bc8277b1254c307314ff5d}, intrahash = {0693c1a793d1cb2b3d187c225d95657c}, journal = {AGU Fall Meeting Abstracts}, keywords = {Africa Change Climate Downscaling Models Precipitation Prediction Variability myown}, month = dec, pages = {GC23B-0922}, publisher = {American Geophysical Union}, timestamp = {2020-09-10T22:25:24.000+0200}, title = {Precipitation Prediction in North Africa Based on Statistical Downscaling}, url = {https://doi.org/10.13140/rg.2.1.4674.1447/1}, volume = 2013, year = 2013 }