An official website of the United States government
Abstract Producing high-quality forecasts of key climate variables, such as temperature and precipitation, on subseasonal time scales has long been a gap in operational forecasting. This study explores an application of machine learning (ML) models as postprocessing tools for subseasonal forecasting. Lagged numerical ensemble forecasts (i.e., an ensemble where the members have different initialization dates) and observational data, including relative humidity, pressure at sea level, and geopotential height, are incorporated into various ML methods to predict monthly average precipitation and 2-m temperature 2 weeks in advance for the continental United States. For regression, quantile regression, and tercile classification tasks, we consider using linear models, random forests, convolutional neural networks, and stacked models (a multimodel approach based on the prediction of the individual ML models). Unlike previous ML approaches that often use ensemble mean alone, we leverage information embedded in the ensemble forecasts to enhance prediction accuracy. Additionally, we investigate extreme event predictions that are crucial for planning and mitigation efforts. Considering ensemble members as a collection of spatial forecasts, we explore different approaches to using spatial information. Trade-offs between different approaches may be mitigated with model stacking. Our proposed models outperform standard baselines such as climatological forecasts and ensemble means. In addition, we investigate feature importance, trade-offs between using the full ensemble or only the ensemble mean, and different modes of accounting for spatial variability. Significance StatementAccurately forecasting temperature and precipitation on subseasonal time scales—2 weeks–2 months in advance—is extremely challenging. These forecasts would have immense value in agriculture, insurance, and economics. Our paper describes an application of machine learning techniques to improve forecasts of monthly average precipitation and 2-m temperature using lagged physics-based predictions and observational data 2 weeks in advance for the entire continental United States. For lagged ensembles, the proposed models outperform standard benchmarks such as historical averages and averages of physics-based predictions. Our findings suggest that utilizing the full set of physics-based predictions instead of the average enhances the accuracy of the final forecast.
more »
« less
Multidecadal Sea Level Prediction Using Neural Networks and Spectral Clustering on Climate Model Large Ensembles and Satellite Altimeter Data
Abstract Sea surface height observations provided by satellite altimetry since 1993 show a rising rate (3.4 mm yr−1) for global mean sea level. While on average, sea level has risen 10 cm over the last 30 years, there is considerable regional variation in the sea level change. Through this work, we predict sea level trends 30 years into the future at a 2° spatial resolution and investigate the future patterns of the sea level change. We show the potential of machine learning (ML) in this challenging application of long-term sea level forecasting over the global ocean. Our approach incorporates sea level data from both altimeter observations and climate model simulations. We develop a supervised learning framework using fully connected neural networks (FCNNs) that can predict the sea level trend based on climate model projections. Alongside this, our method provides uncertainty estimates associated with the ML prediction. We also show the effectiveness of partitioning our spatial dataset and learning a dedicated ML model for each segmented region. We compare two partitioning strategies: one achieved using domain knowledge and the other employing spectral clustering. Our results demonstrate that segmenting the spatial dataset with spectral clustering improves the ML predictions. Significance StatementLong-term projections are needed to help coastal communities adapt to sea level rise. Forecasting multidecadal sea level change is a complex problem. In this paper, we show the promise of machine learning in producing such forecasts 30 years in advance and over the global ocean. Continued improvements in prediction skills that build on this work will be vital in sea level rise adaptation efforts.
more »
« less
- Award ID(s):
- 2153040
- PAR ID:
- 10566597
- Publisher / Repository:
- American Meteorological Society
- Date Published:
- Journal Name:
- Artificial Intelligence for the Earth Systems
- Volume:
- 3
- Issue:
- 4
- ISSN:
- 2769-7525
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Mass loss from the Antarctic ice sheet is projected to continue over the coming century. The resultant sea level change will have a regional pattern that evolves over time as the ocean adjusts. Accurate estimation of this evolution is crucial for local communities. Current state-of-the-art climate models typically do not couple ice sheets to the atmosphere–ocean system, and the impact of ice sheet melt has often been studied by injecting meltwater at the model ocean surface. However, observational evidence suggests that most Antarctic meltwater enters the ocean at depth through ice shelf basal melt. A previous study has demonstrated that the regional sea level pattern at a given time depends on meltwater injection depth. Here, we introduce a 2.5-layer model to investigate this dependence and develop a theory for the associated adjustment mechanisms. We find mechanisms consistent with previous literature on the ocean adjustment to changes in forcing, whereby a slower Rossby wave response off the eastern boundary follows a fast response from the western boundary current and Kelvin waves. We demonstrate that faster baroclinic Rossby waves near the surface than at depth explain the injection depth dependence of the adjustment in the 2.5-layer model. The identified Rossby wave mechanism may contribute to the dependence of the ocean’s transient adjustment on meltwater injection depth in more complex models. This work highlights processes that could cause errors in the projection of the time-varying pattern of sea level rise using surface meltwater input to represent Antarctica’s freshwater forcing. Significance StatementSea level rise is expected to be larger in some locations than in others. Accurate projections of the pattern of sea level change, which changes in time as the ocean adjusts, are essential information for local communities. One of the factors that leads to uncertainty in the local sea level change due to Antarctic melt is the depth at which this meltwater is input into an ocean model. We propose a mechanism for a faster response of sea level around the basin when meltwater is injected at the ocean surface compared to when it is injected well below the surface. This mechanism has implications for projections of the regional sea level response to Antarctic melt.more » « less
-
Abstract Design of coastal defense structures like seawalls and breakwaters can no longer be based on stationarity assumption. In many parts of the world, an anticipated sea‐level rise (SLR) due to climate change will constitute present‐day extreme sea levels inappropriate for future coastal flood risk assessments since it will significantly increase their probability of occurrence. Here, we first show that global annual maxima sea levels (AMSLs) have been increasing in magnitude over the last decades, primarily due to a positive shift in mean sea level (MSL). Then, we apply non‐stationary extreme value theory to model the extremal behavior of sea levels with MSL as a covariate and quantify the evolution of AMSLs in the following decades using revised probabilistic sea‐level rise projections. Our analysis reveals that non‐stationary distributions exhibit distinct differences compared to simply considering stationary conditions with a change in location parameter equal to the amount of MSL rise. With the use of non‐stationary distributions, we show that by the year 2050 many locations will experience their present‐day 100‐yr return level as an event with return period less than 15 and 9 years under the moderate (RCP4.5) and high (RCP8.5) representative concentration pathways. Also, we find that by the end of this century almost all locations examined will encounter their current 100‐yr return level on an annual basis, even if CO2concentration is kept at moderate levels (RCP4.5). Our assessment accounts for large uncertainty by incorporating ambiguities in both SLR projections and non‐stationary extreme value distribution parameters via a Monte Carlo simulation.more » « less
-
Coastal populations are facing increasing environmental stress from coastal hazards including sea level rise, increasing tidal ranges, and storm surges from hurricanes. The East and Gulf Coasts of the United States (U.S.) are projected to face high rates of sea level rise and include many of the U.S.’s largest urban populations. This study proposes modelling land-use change and coastal change between 1996-2019 to project the impacts of intensifying coastal hazards on the U.S. Gulf and East Coast populations and to estimate how coastal populations are growing or retreating from high-risk areas. The primary objective is to develop a multifaceted spatial-temporal (MuST) framework to model coastal change through land-use projections and thorough analysis of the indicators of coastal urban growth or retreat. While urban growth models exist, one that presents an interdisciplinary evaluation of potential growth and retreat due to geographic factors and coastal hazards has not been released. This study proposes modelling urban growth using geospatial metrics including topographic slope, topographic elevation, distance to existing urban areas, distance to existing roads, and distance to the coast. The model will also use historic hurricane data, including storm track and footprint for named storms between 1996-2019 and the associated flood claims data from Federal Emergency Management Agency (FEMA), to account for existing impacts from coastal storms. Additionally, climate change data including sea level rise projections and future tidal ranges will be incorporated to project the impacts of future coastal hazards on urban expansion over the next 30 years (2020-2050). The basis of the urban growth model compares land-use change between 1996-2019 to complete a geospatial analysis of both the areas shifting from rural (agricultural, forest, wetlands) to urban, indicating growth and population data from 2000-2020, to evaluate coastal retreat or abandonment over the next 30 years.more » « less
-
Within the last century, the global sea level has risen between 16 and 21 cm and will likely accelerate into the future. Projections from the Intergovernmental Panel on Climate Change (IPCC) show the global mean sea level (GMSL) rise may increase to up to 1 m (1000 mm) by 2100. The primary cause of the sea level rise can be attributed to climate change through the thermal expansion of seawater and the recession of glaciers from melting. Because of the complexity of the climate and environmental systems, it is very difficult to accurately predict the increase in sea level. The latest estimate of GMSL rise is about 3 mm/year, but as GMSL is a global measure, it may not represent local sea level changes. It is essential to obtain tailored estimates of sea level rise in coastline Florida, as the state is strongly impacted by the global sea level rise. The goal of this study is to model the sea level in coastal Florida using climate factors. Hence, water temperature, water salinity, sea surface height anomalies (SSHA), and El Niño southern oscillation (ENSO) 3.4 index were considered to predict coastal Florida sea level. The sea level changes across coastal Florida were modeled using both multiple regression as a broadly used parametric model and the generalized additive model (GAM), which is a nonparametric method. The local rates and variances of sea surface height anomalies (SSHA) were analyzed and compared to regional and global measurements. The identified optimal model to explain and predict sea level was a GAM with the year, global and regional (adjacent basins) SSHA, local water temperature and salinity, and ENSO as predictors. All predictors including global SSHA, regional SSHA, water temperature, water salinity, ENSO, and the year were identified to have a positive impact on the sea level and can help to explain the variations in the sea level in coastal Florida. Particularly, the global and regional SSHA and the year are important factors to predict sea level changes.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1175/AIES-D-23-0089.1
