Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles
<p>Hydrometeorological hazards such as droughts and floods can have devastating consequences. Central America is one region at risk of hydrometeorological extremes impacts, which will likely increase under human-induced climate change. Physically-based subseasonal to seasonal (S2S) rainfall fo...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2023
|
| Subjects: |
| _version_ | 1797111923655835648 |
|---|---|
| author | Kowal, K |
| author2 | Slater, L |
| author_facet | Slater, L Kowal, K |
| author_sort | Kowal, K |
| collection | OXFORD |
| description | <p>Hydrometeorological hazards such as droughts and floods can have devastating consequences. Central America is one region at risk of hydrometeorological extremes impacts, which will likely increase under human-induced climate change. Physically-based subseasonal to seasonal (S2S) rainfall forecasts from Atmospheric Oceanic General Circulation Models (AOGCMs) can be used to inform anticipatory actions from multiple weeks to several months ahead. Multiple AOGCMs can be combined into multi-model ensembles (MMEs) to generate ensemble forecasts, which often have higher skill than individual models. There are two leading MMEs that provide publicly available forecasts: the North American Multimodel Ensemble (NMME) and the European multi-model seasonal prediction system (C3S). The skill of these models’ rainfall forecasts is critical. Uncertainties in AOGCM estimates affect preparedness planning, as they are used to drive regional rainfall and hydrological models. Rainfall forecasts from individual AOGCMs and MMEs need more evaluation over Central America at the S2S scale. Relatively few evaluations have compared multiple AOGCMs over the region, making it difficult for stakeholders to choose which models to use and to know when to trust the predictions. More clarity is needed on the performance of S2S rainfall forecasts spatially and temporally, and on optimal postprocessing techniques to enhance the detection of low and high rainfall extremes. </p>
<p>This thesis conducts a comparative assessment of S2S rainfall forecasts across ten AOGCMs that contribute to the NMME and C3S, using statistical and process-based evaluation techniques. AOGCMs are found to generally have higher skill in the late wet season (September and October) compared to the early wet season (May and June), possibly due to the increased strength of the El Niño Southern Oscillation (ENSO) teleconnection in the late wet season. The models often perform better on the Pacific side of the region, which experiences a more distinct wet season compared to the Caribbean. Low and high rainfall extremes are challenging to predict even in the late wet season when the ENSO teleconnection is stronger. Techniques to optimize AOGCM outputs for S2S rainfall prediction are then assessed, including using hybrid forecasts and subsampling. Hybrid forecasting methods use AOGCMs to generate forecasts by statistically relating their predictions of large-scale drivers, such as sea surface temperatures, to rainfall. Findings show some hybrid methods improve AOGCM forecasts at the seasonal scale, such as those using forecasts of Tropical North Atlantic (TNA) sea surface temperatures in the early wet season when TNA is more strongly associated with regional rainfall. A novel post-processing technique is also developed that improves subseasonal detection rates of low and high rainfall extremes by subsampling ensemble members that better represent key drivers of Central American rainfall. The post-processing technique is beneficial for operational forecasters who can leverage their expertise of relevant rainfall-generating processes to subsample better performing ensemble members for their region. These findings underscore the importance of evaluating AOGCM rainfall forecasts regionally due to the spatial and temporal variability in AOGCM skill and showcase how alternative ways to post-process or use these models can significantly improve S2S forecasts over Central America.</p> |
| first_indexed | 2024-03-07T08:17:11Z |
| format | Thesis |
| id | oxford-uuid:6cb282bb-9055-4644-b5fe-d1ddd00ff1dc |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:17:11Z |
| publishDate | 2023 |
| record_format | dspace |
| spelling | oxford-uuid:6cb282bb-9055-4644-b5fe-d1ddd00ff1dc2024-01-08T11:11:48ZImproving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensemblesThesishttp://purl.org/coar/resource_type/c_db06uuid:6cb282bb-9055-4644-b5fe-d1ddd00ff1dcPrecipitation forecastingCentral AmericaNumerical weather forecastingEnglishHyrax Deposit2023Kowal, KSlater, LDadson, SVan Loon, AMuñoz, ÁBirkel, CGarcía-López, ALi, SKelder, THall, KMoulds, S<p>Hydrometeorological hazards such as droughts and floods can have devastating consequences. Central America is one region at risk of hydrometeorological extremes impacts, which will likely increase under human-induced climate change. Physically-based subseasonal to seasonal (S2S) rainfall forecasts from Atmospheric Oceanic General Circulation Models (AOGCMs) can be used to inform anticipatory actions from multiple weeks to several months ahead. Multiple AOGCMs can be combined into multi-model ensembles (MMEs) to generate ensemble forecasts, which often have higher skill than individual models. There are two leading MMEs that provide publicly available forecasts: the North American Multimodel Ensemble (NMME) and the European multi-model seasonal prediction system (C3S). The skill of these models’ rainfall forecasts is critical. Uncertainties in AOGCM estimates affect preparedness planning, as they are used to drive regional rainfall and hydrological models. Rainfall forecasts from individual AOGCMs and MMEs need more evaluation over Central America at the S2S scale. Relatively few evaluations have compared multiple AOGCMs over the region, making it difficult for stakeholders to choose which models to use and to know when to trust the predictions. More clarity is needed on the performance of S2S rainfall forecasts spatially and temporally, and on optimal postprocessing techniques to enhance the detection of low and high rainfall extremes. </p> <p>This thesis conducts a comparative assessment of S2S rainfall forecasts across ten AOGCMs that contribute to the NMME and C3S, using statistical and process-based evaluation techniques. AOGCMs are found to generally have higher skill in the late wet season (September and October) compared to the early wet season (May and June), possibly due to the increased strength of the El Niño Southern Oscillation (ENSO) teleconnection in the late wet season. The models often perform better on the Pacific side of the region, which experiences a more distinct wet season compared to the Caribbean. Low and high rainfall extremes are challenging to predict even in the late wet season when the ENSO teleconnection is stronger. Techniques to optimize AOGCM outputs for S2S rainfall prediction are then assessed, including using hybrid forecasts and subsampling. Hybrid forecasting methods use AOGCMs to generate forecasts by statistically relating their predictions of large-scale drivers, such as sea surface temperatures, to rainfall. Findings show some hybrid methods improve AOGCM forecasts at the seasonal scale, such as those using forecasts of Tropical North Atlantic (TNA) sea surface temperatures in the early wet season when TNA is more strongly associated with regional rainfall. A novel post-processing technique is also developed that improves subseasonal detection rates of low and high rainfall extremes by subsampling ensemble members that better represent key drivers of Central American rainfall. The post-processing technique is beneficial for operational forecasters who can leverage their expertise of relevant rainfall-generating processes to subsample better performing ensemble members for their region. These findings underscore the importance of evaluating AOGCM rainfall forecasts regionally due to the spatial and temporal variability in AOGCM skill and showcase how alternative ways to post-process or use these models can significantly improve S2S forecasts over Central America.</p> |
| spellingShingle | Precipitation forecasting Central America Numerical weather forecasting Kowal, K Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title | Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title_full | Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title_fullStr | Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title_full_unstemmed | Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title_short | Improving subseasonal to seasonal rainfall forecasts in Central America using dynamic model ensembles |
| title_sort | improving subseasonal to seasonal rainfall forecasts in central america using dynamic model ensembles |
| topic | Precipitation forecasting Central America Numerical weather forecasting |
| work_keys_str_mv | AT kowalk improvingsubseasonaltoseasonalrainfallforecastsincentralamericausingdynamicmodelensembles |