Abstract
Since 2019, National Weather Service (NWS) offices have been able to issue 360-character Wireless Emergency Alert (“WEA360”) messages for tornadoes. NWS is now considering changing from a “deterministic” to a “probabilistic” warning paradigm. That change could possibly influence how WEA360 messages for tornado are issued in the future. Recent experimental studies have found that probabilistic hazard information (PHI) forecast graphics improve consumers’ risk perception for tornadoes, but findings from these studies concerning whether PHI forecast graphics improve people’s protective action decision-making are mixed. The present study therefore investigated how mock PHI-enhanced WEA360 messages might influence people’s risk perception and protective action decision-making. Analysis of qualitative data gathered from a combination of questionnaire and focus group interview methods conducted in collaboration with 31 community members in Denver, Colorado, indicated that inclusion of PHI forecast graphics within WEA360 messages elicited high levels of understanding and message believability but did not consistently lead to appropriate precautionary intent. Because warning response is a complex social phenomenon, PHI may not significantly improve protective action decision-making if PHI forecast graphics are eventually presented to consumers via the Wireless Emergency Alerts system. Factors that PHI stakeholders should consider before the adoption of PHI-enhanced WEA360 messages for consumers are discussed.
Significance Statement
This study examines how consumers respond to and talk about mock WEA360 messages for tornadoes that contain embedded PHI forecast graphics. As NWS considers moving to a probabilistic warning paradigm, stakeholders will need to determine how PHI forecast graphics might be communicated directly to consumers, if at all. Our findings suggest that combining WEA360 messages with PHI forecast graphics creates challenges and complexities related to consumers’ assessment of personal risk and protective action decision-making. Overall, the study suggests that any future PHI-enhanced WEA360 messages provided directly to consumers, if at all, must avoid discrepancies (even subtle) between the level of risk represented by the PHI forecast graphic and the protective action guidance included in the text of the messages.
Abstract
Since 2019, National Weather Service (NWS) offices have been able to issue 360-character Wireless Emergency Alert (“WEA360”) messages for tornadoes. NWS is now considering changing from a “deterministic” to a “probabilistic” warning paradigm. That change could possibly influence how WEA360 messages for tornado are issued in the future. Recent experimental studies have found that probabilistic hazard information (PHI) forecast graphics improve consumers’ risk perception for tornadoes, but findings from these studies concerning whether PHI forecast graphics improve people’s protective action decision-making are mixed. The present study therefore investigated how mock PHI-enhanced WEA360 messages might influence people’s risk perception and protective action decision-making. Analysis of qualitative data gathered from a combination of questionnaire and focus group interview methods conducted in collaboration with 31 community members in Denver, Colorado, indicated that inclusion of PHI forecast graphics within WEA360 messages elicited high levels of understanding and message believability but did not consistently lead to appropriate precautionary intent. Because warning response is a complex social phenomenon, PHI may not significantly improve protective action decision-making if PHI forecast graphics are eventually presented to consumers via the Wireless Emergency Alerts system. Factors that PHI stakeholders should consider before the adoption of PHI-enhanced WEA360 messages for consumers are discussed.
Significance Statement
This study examines how consumers respond to and talk about mock WEA360 messages for tornadoes that contain embedded PHI forecast graphics. As NWS considers moving to a probabilistic warning paradigm, stakeholders will need to determine how PHI forecast graphics might be communicated directly to consumers, if at all. Our findings suggest that combining WEA360 messages with PHI forecast graphics creates challenges and complexities related to consumers’ assessment of personal risk and protective action decision-making. Overall, the study suggests that any future PHI-enhanced WEA360 messages provided directly to consumers, if at all, must avoid discrepancies (even subtle) between the level of risk represented by the PHI forecast graphic and the protective action guidance included in the text of the messages.
Abstract
Recent severe droughts, extreme floods, and increasing differences between seasonal high and low flows on the Amazon River may represent a twenty-first-century increase in the amplitude of the hydrologic cycle over the Amazon Basin. These precipitation and streamflow changes may have arisen from natural ocean–atmospheric variability, deforestation within the drainage basin of the Amazon River, or anthropogenic climate change. Tree-ring reconstructions of wet-season precipitation extremes, substantiated with historical accounts of climate and river levels on the Amazon River and in northeast Brazil found in the Brazilian Digital Library, indicate that the recent river-level extremes on the Amazon may have been equaled or possibly exceeded during the preinstrumental nineteenth century. The “Forgotten Drought” of 1865 was the lowest wet-season rainfall total reconstructed with tree-rings in the eastern Amazon from 1790 to 2016 and appears to have been one of the lowest stream levels observed on the Amazon River during the historical era according to first-hand descriptions by Louis Agassiz, his Brazilian colleague João Martins da Silva Coutinho, and others. Heavy rains and flooding are described during most of the tree-ring-reconstructed wet extremes, including the complete inundation of “First Street” in Santarem, Brazil, in 1859 and the overtopping of the Bittencourt Bridge in Manaus, Brazil, in 1892. These extremes in the tree-ring estimates and historical observations indicate that recent high and low flow anomalies on the Amazon River may not have exceeded the natural variability of precipitation and streamflow during the nineteenth century.
Significance Statement
Proxy tree-ring and historical evidence for precipitation extremes during the preinstrumental nineteenth century indicate that recent floods and droughts on the Amazon River may have not yet exceeded the range of natural hydroclimatic variability.
Abstract
Recent severe droughts, extreme floods, and increasing differences between seasonal high and low flows on the Amazon River may represent a twenty-first-century increase in the amplitude of the hydrologic cycle over the Amazon Basin. These precipitation and streamflow changes may have arisen from natural ocean–atmospheric variability, deforestation within the drainage basin of the Amazon River, or anthropogenic climate change. Tree-ring reconstructions of wet-season precipitation extremes, substantiated with historical accounts of climate and river levels on the Amazon River and in northeast Brazil found in the Brazilian Digital Library, indicate that the recent river-level extremes on the Amazon may have been equaled or possibly exceeded during the preinstrumental nineteenth century. The “Forgotten Drought” of 1865 was the lowest wet-season rainfall total reconstructed with tree-rings in the eastern Amazon from 1790 to 2016 and appears to have been one of the lowest stream levels observed on the Amazon River during the historical era according to first-hand descriptions by Louis Agassiz, his Brazilian colleague João Martins da Silva Coutinho, and others. Heavy rains and flooding are described during most of the tree-ring-reconstructed wet extremes, including the complete inundation of “First Street” in Santarem, Brazil, in 1859 and the overtopping of the Bittencourt Bridge in Manaus, Brazil, in 1892. These extremes in the tree-ring estimates and historical observations indicate that recent high and low flow anomalies on the Amazon River may not have exceeded the natural variability of precipitation and streamflow during the nineteenth century.
Significance Statement
Proxy tree-ring and historical evidence for precipitation extremes during the preinstrumental nineteenth century indicate that recent floods and droughts on the Amazon River may have not yet exceeded the range of natural hydroclimatic variability.
Abstract
This study reveals that South China precipitation (SCP) anomalies tend to persist well from winter to the following spring after the late 1990s, favoring long-lasting drought or flood events over South China. Mechanism analysis indicates that the interdecadal changes in El Niño–Southern Oscillation (ENSO) and the preceding November central Asian snow cover could contribute to the increased persistence of winter-to-spring SCP anomalies. ENSO has a stable impact on winter SCP, whereas its impact on spring SCP is significantly enhanced after the late 1990s. With a weakened intensity and faster decay rate in the recent two decades, the ENSO-related spring SST anomalies over the tropical Pacific are relatively weaker, inducing a weakened and more southward-located western North Pacific anticyclone. This further leads to an interdecadal migration of the spring rainfall belt anomaly, consequently favoring the persistence of winter-to-spring SCP anomalies after the late 1990s. Additionally, the impacts of November central Asian snow cover on winter and spring SCP are both strengthened after the late 1990s. In the most recent two decades, the snow-cover-related cooling effect has become stronger, which induces winter cyclonic anomalies over Lake Baikal, favoring increased winter SCP. In addition, increased snow cover excites upward-propagating waves from the troposphere to the stratosphere, consequently weakening the stratospheric polar vortex. In spring, the stratospheric polar vortex signals propagate downward and result in a negative Arctic Oscillation in the troposphere, favoring more spring SCP. Therefore, central Asian snow cover is also conductive to the persistence of winter-to-spring SCP anomalies after the late 1990s.
Abstract
This study reveals that South China precipitation (SCP) anomalies tend to persist well from winter to the following spring after the late 1990s, favoring long-lasting drought or flood events over South China. Mechanism analysis indicates that the interdecadal changes in El Niño–Southern Oscillation (ENSO) and the preceding November central Asian snow cover could contribute to the increased persistence of winter-to-spring SCP anomalies. ENSO has a stable impact on winter SCP, whereas its impact on spring SCP is significantly enhanced after the late 1990s. With a weakened intensity and faster decay rate in the recent two decades, the ENSO-related spring SST anomalies over the tropical Pacific are relatively weaker, inducing a weakened and more southward-located western North Pacific anticyclone. This further leads to an interdecadal migration of the spring rainfall belt anomaly, consequently favoring the persistence of winter-to-spring SCP anomalies after the late 1990s. Additionally, the impacts of November central Asian snow cover on winter and spring SCP are both strengthened after the late 1990s. In the most recent two decades, the snow-cover-related cooling effect has become stronger, which induces winter cyclonic anomalies over Lake Baikal, favoring increased winter SCP. In addition, increased snow cover excites upward-propagating waves from the troposphere to the stratosphere, consequently weakening the stratospheric polar vortex. In spring, the stratospheric polar vortex signals propagate downward and result in a negative Arctic Oscillation in the troposphere, favoring more spring SCP. Therefore, central Asian snow cover is also conductive to the persistence of winter-to-spring SCP anomalies after the late 1990s.
Abstract
The Indian monsoon is of utmost concern to agriculture, the economy, and the livelihoods of billions in South Asia. However, little attention has been paid to the possibility of distinct subseasonal episodes phase-locked in the Indian monsoon annual cycle. This study addresses this gap by utilizing the self-organizing map (SOM) method to objectively classify six distinct subseasonal stages based on the 850-hPa wind fields. Each subseasonal stage ranges from 23 to 90 days. The Indian summer monsoon (ISM) consists of three substages, the ISM-onset, ISM-peak, and ISM-withdrawal, altogether contributing to 82% of the annual precipitation. The three substages signify the rapid northward advance, dominance, and gradual southward retreat of southwesterlies from mid-May to early October. The winter monsoon also comprises three substages (fall, winter, and spring), distinguishable by the latitude of the Arabian Sea high pressure ridge and hydrological conditions. This study proposes two compact indices based on zonal winds in the northern and southern Arabian Sea to measure the winter and summer monsoons, respectively. These indices capture the development and turnabouts of the six SOM-derived stages and can be used for subseasonal monsoon monitoring and forecasts. The spring and the ISM-onset episodes are highly susceptible to compound hazards of droughts and heatwaves, while the greatest flood risk occurs during the ISM-peak stage. The fall stage heralds the peak season for tropical storms over the Arabian Sea and the Bay of Bengal. The annual start and end dates of the ISM-peak are highly correlated (0.6–0.8) with the criteria-based dates proposed previously, supporting the delineation of the Indian monsoon subseasonal features.
Significance Statement
This research explores the existence of subseasonal features in the Indian monsoon annual cycle. Through the use of machine learning, we discover that the Indian summer monsoon and winter monsoon each consist of three substages. These substages’ evolution can be measured by two compact indices proposed herein, which can aid in subseasonal monsoon monitoring and forecasts in South Asia. Pertaining to hazard adaptations, this work pinpoints the subseasonal episodes most susceptible to droughts, heatwaves, floods, and tropical storms. High correlations are obtained when validating the substages’ yearly start and end dates against those documented in the existing literature, offering credibility to the subseasonal features of the Indian monsoon.
Abstract
The Indian monsoon is of utmost concern to agriculture, the economy, and the livelihoods of billions in South Asia. However, little attention has been paid to the possibility of distinct subseasonal episodes phase-locked in the Indian monsoon annual cycle. This study addresses this gap by utilizing the self-organizing map (SOM) method to objectively classify six distinct subseasonal stages based on the 850-hPa wind fields. Each subseasonal stage ranges from 23 to 90 days. The Indian summer monsoon (ISM) consists of three substages, the ISM-onset, ISM-peak, and ISM-withdrawal, altogether contributing to 82% of the annual precipitation. The three substages signify the rapid northward advance, dominance, and gradual southward retreat of southwesterlies from mid-May to early October. The winter monsoon also comprises three substages (fall, winter, and spring), distinguishable by the latitude of the Arabian Sea high pressure ridge and hydrological conditions. This study proposes two compact indices based on zonal winds in the northern and southern Arabian Sea to measure the winter and summer monsoons, respectively. These indices capture the development and turnabouts of the six SOM-derived stages and can be used for subseasonal monsoon monitoring and forecasts. The spring and the ISM-onset episodes are highly susceptible to compound hazards of droughts and heatwaves, while the greatest flood risk occurs during the ISM-peak stage. The fall stage heralds the peak season for tropical storms over the Arabian Sea and the Bay of Bengal. The annual start and end dates of the ISM-peak are highly correlated (0.6–0.8) with the criteria-based dates proposed previously, supporting the delineation of the Indian monsoon subseasonal features.
Significance Statement
This research explores the existence of subseasonal features in the Indian monsoon annual cycle. Through the use of machine learning, we discover that the Indian summer monsoon and winter monsoon each consist of three substages. These substages’ evolution can be measured by two compact indices proposed herein, which can aid in subseasonal monsoon monitoring and forecasts in South Asia. Pertaining to hazard adaptations, this work pinpoints the subseasonal episodes most susceptible to droughts, heatwaves, floods, and tropical storms. High correlations are obtained when validating the substages’ yearly start and end dates against those documented in the existing literature, offering credibility to the subseasonal features of the Indian monsoon.
Abstract
The diurnal cycle of precipitation and precipitation variances at different time scales are analyzed in this study based on multiple high-resolution 3-hour precipitation datasets. The results are used to evaluate nine CMIP6 models and a series of GFDL AM4.0 model simulations, with the goal of examining the impact of SST diurnal cycle, varying horizontal resolutions, and different microphysics scheme on these two precipitation features. It is found that although diurnal amplitudes are reasonably simulated, models generally generate too early diurnal peaks over land, with a diurnal phase peaking around noon instead of the observed late afternoon (or early evening) peak. As for precipitation variances, irregular subdaily fluctuations dominate the total variance, followed by variance of daily mean precipitation and variance associated with the mean diurnal cycle. While the spatial and zonal distribution of precipitation variances are generally captured by the models, significant biases are present in tropical regions, where large mean precipitation biases are observed. The comparisons based on AM4.0 model simulations demonstrate that the inclusion of ocean coupling, adoption of a new microphysics scheme, and increasing of horizontal resolution have limited impacts on these two simulated features, emphasizing the need for future investigation into these model deficiencies at the process level. Conducting routine examinations of these metrics would be a crucial first step towards better simulation of precipitation intermittence in future model development. Lastly, distinct differences in these two features are found among observational datasets, highlighting the urgent need for a detailed evaluation of precipitation observations, especially at suddaily time scales, as model evaluation heavily relies on high-quality observations.
Abstract
The diurnal cycle of precipitation and precipitation variances at different time scales are analyzed in this study based on multiple high-resolution 3-hour precipitation datasets. The results are used to evaluate nine CMIP6 models and a series of GFDL AM4.0 model simulations, with the goal of examining the impact of SST diurnal cycle, varying horizontal resolutions, and different microphysics scheme on these two precipitation features. It is found that although diurnal amplitudes are reasonably simulated, models generally generate too early diurnal peaks over land, with a diurnal phase peaking around noon instead of the observed late afternoon (or early evening) peak. As for precipitation variances, irregular subdaily fluctuations dominate the total variance, followed by variance of daily mean precipitation and variance associated with the mean diurnal cycle. While the spatial and zonal distribution of precipitation variances are generally captured by the models, significant biases are present in tropical regions, where large mean precipitation biases are observed. The comparisons based on AM4.0 model simulations demonstrate that the inclusion of ocean coupling, adoption of a new microphysics scheme, and increasing of horizontal resolution have limited impacts on these two simulated features, emphasizing the need for future investigation into these model deficiencies at the process level. Conducting routine examinations of these metrics would be a crucial first step towards better simulation of precipitation intermittence in future model development. Lastly, distinct differences in these two features are found among observational datasets, highlighting the urgent need for a detailed evaluation of precipitation observations, especially at suddaily time scales, as model evaluation heavily relies on high-quality observations.
Abstract
We construct a novel multi-input multioutput autoencoder (MIMO-AE) to capture the nonlinear relationship of Southern California precipitation and tropical Pacific Ocean sea surface temperature. The MIMO-AE is trained on both monthly tropical Pacific sea surface temperature (TP-SST) and Southern California precipitation (SC-PRECIP) anomalies simultaneously. The covariability of the two fields in the MIMO-AE shared nonlinear latent space can be condensed into an index, termed the MIMO-AE index. We use a transfer learning approach to train a MIMO-AE on the combined dataset of 100 yr of output from a historical simulation with the Energy Exascale Earth Systems Model, version 1, and a segment of observational data. We further use long short-term memory networks to assess subseasonal predictability of SC-PRECIP using the MIMO-AE index. We find that the MIMO-AE index provides enhanced predictability of SC-PRECIP for a lead time of up to 4 months as compared with the Niño-3.4 index and the El Niño–Southern Oscillation longitudinal index.
Significance Statement
Traditional El Niño–Southern Oscillation indices, like the Niño-3.4 index, although well predicted themselves, fail to offer reliable subseasonal-to-seasonal predictions of western U.S. precipitation. Here, we use a machine learning approach called a multi-input, multioutput autoencoder to capture the relationship between tropical Pacific Ocean and Southern California precipitation and project it onto a new index, which we call the MIMO-AE index. Using machine learning–based time series predictions, we find that the MIMO-AE index offers enhanced predictability of Southern California precipitation up to a lead time of 4 months as compared with other ENSO indices.
Abstract
We construct a novel multi-input multioutput autoencoder (MIMO-AE) to capture the nonlinear relationship of Southern California precipitation and tropical Pacific Ocean sea surface temperature. The MIMO-AE is trained on both monthly tropical Pacific sea surface temperature (TP-SST) and Southern California precipitation (SC-PRECIP) anomalies simultaneously. The covariability of the two fields in the MIMO-AE shared nonlinear latent space can be condensed into an index, termed the MIMO-AE index. We use a transfer learning approach to train a MIMO-AE on the combined dataset of 100 yr of output from a historical simulation with the Energy Exascale Earth Systems Model, version 1, and a segment of observational data. We further use long short-term memory networks to assess subseasonal predictability of SC-PRECIP using the MIMO-AE index. We find that the MIMO-AE index provides enhanced predictability of SC-PRECIP for a lead time of up to 4 months as compared with the Niño-3.4 index and the El Niño–Southern Oscillation longitudinal index.
Significance Statement
Traditional El Niño–Southern Oscillation indices, like the Niño-3.4 index, although well predicted themselves, fail to offer reliable subseasonal-to-seasonal predictions of western U.S. precipitation. Here, we use a machine learning approach called a multi-input, multioutput autoencoder to capture the relationship between tropical Pacific Ocean and Southern California precipitation and project it onto a new index, which we call the MIMO-AE index. Using machine learning–based time series predictions, we find that the MIMO-AE index offers enhanced predictability of Southern California precipitation up to a lead time of 4 months as compared with other ENSO indices.
Abstract
Accurate streamflow simulations rely on good estimates of the catchment-scale soil moisture distribution. Here, we evaluated the potential of Sentinel-1 backscatter data assimilation (DA) to improve soil moisture and streamflow estimates. Our DA system consisted of the Noah-MP land surface model coupled to the HyMAP river routing model and the water cloud model as backscatter observation operator. The DA system was set up at 0.01° resolution for two contrasting catchments in Belgium: i) the Demer catchment dominated by agriculture, and ii) the Ourthe catchment dominated by mixed forests. We present results of two experiments with an ensemble Kalman filter updating either soil moisture only or soil moisture and Leaf Area Index (LAI). The DA experiments covered the period January 2015 through August 2021 and were evaluated with independent rainfall error estimates based on station data, LAI from optical remote sensing, soil moisture retrievals from passive microwave observations, and streamflow measurements. Our results indicate that the assimilation of Sentinel-1 backscatter observations can partly correct errors in surface soil moisture due to rainfall errors and overall improve surface soil moisture estimates. However, updating soil moisture and LAI simultaneously did not bring any benefit over updating soil moisture only. Our results further indicate that streamflow estimates can be improved through Sentinel-1 DA in a catchment with strong soil moisture-runoff coupling, as observed for the Ourthe catchment, suggesting that there is potential for Sentinel-1 DA even for forested catchments.
Abstract
Accurate streamflow simulations rely on good estimates of the catchment-scale soil moisture distribution. Here, we evaluated the potential of Sentinel-1 backscatter data assimilation (DA) to improve soil moisture and streamflow estimates. Our DA system consisted of the Noah-MP land surface model coupled to the HyMAP river routing model and the water cloud model as backscatter observation operator. The DA system was set up at 0.01° resolution for two contrasting catchments in Belgium: i) the Demer catchment dominated by agriculture, and ii) the Ourthe catchment dominated by mixed forests. We present results of two experiments with an ensemble Kalman filter updating either soil moisture only or soil moisture and Leaf Area Index (LAI). The DA experiments covered the period January 2015 through August 2021 and were evaluated with independent rainfall error estimates based on station data, LAI from optical remote sensing, soil moisture retrievals from passive microwave observations, and streamflow measurements. Our results indicate that the assimilation of Sentinel-1 backscatter observations can partly correct errors in surface soil moisture due to rainfall errors and overall improve surface soil moisture estimates. However, updating soil moisture and LAI simultaneously did not bring any benefit over updating soil moisture only. Our results further indicate that streamflow estimates can be improved through Sentinel-1 DA in a catchment with strong soil moisture-runoff coupling, as observed for the Ourthe catchment, suggesting that there is potential for Sentinel-1 DA even for forested catchments.
Abstract
Climate change has forced the world into a state of emergency, but the urgency can also become an opportunity to strengthen the focus on sustainable development and reduce social vulnerability. For developing economies, the first and foremost challenge regarding climate change is to address the knowledge gap on sustainable development and vulnerability. Besides this, evidence-based inputs are needed for the policies and programs that intend to enhance the adaptive capacity and social capital from the gender perspective in comparatively more disaster-prone districts of the country. The environmental impact in terms of socioeconomic conditions specifically pertaining to rural areas of Pakistan cannot be ignored. Natural events such as floods and droughts have raised the question of the social and socioeconomic vulnerability of the rural communities. This paper is an attempt toward understanding that everyone who is affected will be impacted differently by climate change both within the same gender and between different genders, including gender minorities. In addition, an attempt is made to identify the drivers of gender-disaggregated social vulnerability in selected disaster-prone rural communities of the district of Dadu, Sindh Province, Pakistan. Both quantitative and qualitative techniques are employed to examine the differences in gender perception on climate change, experiences related to climate change, disasters, and impacts on their lives. Women and households headed by them are found to be relatively more vulnerable due to their socioeconomic and social status in the rural areas of Pakistan. The paper gives policy directives that not only address the measures for reduction in climate change impacts but also suggest the development of effective disaster management programs, policies, and strategies.
Abstract
Climate change has forced the world into a state of emergency, but the urgency can also become an opportunity to strengthen the focus on sustainable development and reduce social vulnerability. For developing economies, the first and foremost challenge regarding climate change is to address the knowledge gap on sustainable development and vulnerability. Besides this, evidence-based inputs are needed for the policies and programs that intend to enhance the adaptive capacity and social capital from the gender perspective in comparatively more disaster-prone districts of the country. The environmental impact in terms of socioeconomic conditions specifically pertaining to rural areas of Pakistan cannot be ignored. Natural events such as floods and droughts have raised the question of the social and socioeconomic vulnerability of the rural communities. This paper is an attempt toward understanding that everyone who is affected will be impacted differently by climate change both within the same gender and between different genders, including gender minorities. In addition, an attempt is made to identify the drivers of gender-disaggregated social vulnerability in selected disaster-prone rural communities of the district of Dadu, Sindh Province, Pakistan. Both quantitative and qualitative techniques are employed to examine the differences in gender perception on climate change, experiences related to climate change, disasters, and impacts on their lives. Women and households headed by them are found to be relatively more vulnerable due to their socioeconomic and social status in the rural areas of Pakistan. The paper gives policy directives that not only address the measures for reduction in climate change impacts but also suggest the development of effective disaster management programs, policies, and strategies.
Abstract
This paper presents a novel application of convolutional neural network (CNN) models for filtering the intraseasonal variability of the tropical atmosphere. In this deep learning filter, two convolutional layers are applied sequentially in a supervised machine learning framework to extract the intraseasonal signal from the total daily anomalies. The CNN-based filter can be tailored for each field similarly to fast Fourier transform filtering methods. When applied to two different fields (zonal wind stress and outgoing longwave radiation), the index of agreement between the filtered signal obtained using the CNN-based filter and a conventional weight-based filter is between 95% and 99%. The advantage of the CNN-based filter over the conventional filters is its applicability to time series with the length comparable to the period of the signal being extracted.
Significance Statement
This study proposes a new method for discovering hidden connections in data representative of tropical atmosphere variability. The method makes use of an artificial intelligence (AI) algorithm that combines a mathematical operation known as convolution with a mathematical model built to reflect the behavior of the human brain known as artificial neural network. Our results show that the filtered data produced by the AI-based method are consistent with the results obtained using conventional mathematical algorithms. The advantage of the AI-based method is that it can be applied to cases for which the conventional methods have limitations, such as forecast (hindcast) data or real-time monitoring of tropical variability in the 20–100-day range.
Abstract
This paper presents a novel application of convolutional neural network (CNN) models for filtering the intraseasonal variability of the tropical atmosphere. In this deep learning filter, two convolutional layers are applied sequentially in a supervised machine learning framework to extract the intraseasonal signal from the total daily anomalies. The CNN-based filter can be tailored for each field similarly to fast Fourier transform filtering methods. When applied to two different fields (zonal wind stress and outgoing longwave radiation), the index of agreement between the filtered signal obtained using the CNN-based filter and a conventional weight-based filter is between 95% and 99%. The advantage of the CNN-based filter over the conventional filters is its applicability to time series with the length comparable to the period of the signal being extracted.
Significance Statement
This study proposes a new method for discovering hidden connections in data representative of tropical atmosphere variability. The method makes use of an artificial intelligence (AI) algorithm that combines a mathematical operation known as convolution with a mathematical model built to reflect the behavior of the human brain known as artificial neural network. Our results show that the filtered data produced by the AI-based method are consistent with the results obtained using conventional mathematical algorithms. The advantage of the AI-based method is that it can be applied to cases for which the conventional methods have limitations, such as forecast (hindcast) data or real-time monitoring of tropical variability in the 20–100-day range.