Search Results
You are looking at 1 - 10 of 21 items for
- Author or Editor: Michael J. Pavolonis x
- Refine by Access: All Content x
Abstract
Infrared measurements can be used to obtain quantitative information on cloud microphysics, including cloud composition (ice, liquid water, ash, dust, etc.), with the advantage that the measurements are independent of solar zenith angle. As such, infrared brightness temperatures (BT) and brightness temperature differences (BTD) have been used extensively in quantitative remote sensing applications for inferring cloud composition. In this study it is shown that BTDs are fundamentally limited and that a more physically based infrared approach can lead to significant increases in sensitivity to cloud microphysics, especially for optically thin clouds. In lieu of BTDs, a derived radiative parameter β, which is directly related to particle size, habit, and composition, is used. Although the concept of effective absorption optical depth ratios β has been around since the mid-1980s, this is the first study to explore the use of β for inferring cloud composition in the total absence of cloud vertical boundary information. The results showed that, even in the absence of cloud vertical boundary information, one could significantly increase the sensitivity to cloud microphysics by converting the measured radiances to effective emissivity and constructing effective absorption optical depth ratios from a pair of spectral emissivities in the 8–12-μm “window.” This increase in sensitivity to cloud microphysics is relative to BTDs constructed from the same spectral pairs. In this article, the focus is on describing the physical concepts (which can be applied to narrowband or hyperspectral infrared measurements) used in constructing the β data space.
Abstract
Infrared measurements can be used to obtain quantitative information on cloud microphysics, including cloud composition (ice, liquid water, ash, dust, etc.), with the advantage that the measurements are independent of solar zenith angle. As such, infrared brightness temperatures (BT) and brightness temperature differences (BTD) have been used extensively in quantitative remote sensing applications for inferring cloud composition. In this study it is shown that BTDs are fundamentally limited and that a more physically based infrared approach can lead to significant increases in sensitivity to cloud microphysics, especially for optically thin clouds. In lieu of BTDs, a derived radiative parameter β, which is directly related to particle size, habit, and composition, is used. Although the concept of effective absorption optical depth ratios β has been around since the mid-1980s, this is the first study to explore the use of β for inferring cloud composition in the total absence of cloud vertical boundary information. The results showed that, even in the absence of cloud vertical boundary information, one could significantly increase the sensitivity to cloud microphysics by converting the measured radiances to effective emissivity and constructing effective absorption optical depth ratios from a pair of spectral emissivities in the 8–12-μm “window.” This increase in sensitivity to cloud microphysics is relative to BTDs constructed from the same spectral pairs. In this article, the focus is on describing the physical concepts (which can be applied to narrowband or hyperspectral infrared measurements) used in constructing the β data space.
Abstract
Two algorithms for detecting multilayered cloud systems with satellite data are presented. The first algorithm utilizes data in the 0.65-, 11-, and 12-μm regions of the spectrum that are available on the Advanced Very High Resolution Radiometer (AVHRR). The second algorithm incorporates two different techniques to detect cloud overlap: the same technique used in the first algorithm and an additional series of spectral tests that now include data from the 1.38- and 1.65-μm near-infrared regions that are available on the Moderate Resolution Imaging Spectroradiometer (MODIS) and will be available on the Visible/Infrared Imager/Radiometer Suite (VIIRS). VIIRS is the imager that will replace the AVHRR on the next generation of polar-orbiting satellites. Both algorithms were derived assuming that a scene with cloud overlap consists of a semitransparent ice cloud that overlaps a cloud composed of liquid water droplets. Each algorithm was tested on three different MODIS scenes. In all three cases, the second (VIIRS) algorithm was able to detect more cloud overlap than the first (AVHRR) algorithm. Radiative transfer calculations indicate that the VIIRS algorithm will be more effective than the AVHRR algorithm when the visible optical depth of the ice cloud is greater than 3. Both algorithms will work best when the visible optical depth of the water cloud is greater than 5. Model sensitivity studies were also performed to assess the sensitivity of each algorithm to various parameters. It was found that the AVHRR algorithm is most sensitive to cloud particle size and the VIIRS near-infrared test is most sensitive to cloud vertical location. When validating each algorithm using cloud radar data, the VIIRS algorithm was shown to be more effective at detecting cloud overlap than the AVHRR algorithm; however, the VIIRS algorithm was slightly more prone to false cloud overlap detection.
Abstract
Two algorithms for detecting multilayered cloud systems with satellite data are presented. The first algorithm utilizes data in the 0.65-, 11-, and 12-μm regions of the spectrum that are available on the Advanced Very High Resolution Radiometer (AVHRR). The second algorithm incorporates two different techniques to detect cloud overlap: the same technique used in the first algorithm and an additional series of spectral tests that now include data from the 1.38- and 1.65-μm near-infrared regions that are available on the Moderate Resolution Imaging Spectroradiometer (MODIS) and will be available on the Visible/Infrared Imager/Radiometer Suite (VIIRS). VIIRS is the imager that will replace the AVHRR on the next generation of polar-orbiting satellites. Both algorithms were derived assuming that a scene with cloud overlap consists of a semitransparent ice cloud that overlaps a cloud composed of liquid water droplets. Each algorithm was tested on three different MODIS scenes. In all three cases, the second (VIIRS) algorithm was able to detect more cloud overlap than the first (AVHRR) algorithm. Radiative transfer calculations indicate that the VIIRS algorithm will be more effective than the AVHRR algorithm when the visible optical depth of the ice cloud is greater than 3. Both algorithms will work best when the visible optical depth of the water cloud is greater than 5. Model sensitivity studies were also performed to assess the sensitivity of each algorithm to various parameters. It was found that the AVHRR algorithm is most sensitive to cloud particle size and the VIIRS near-infrared test is most sensitive to cloud vertical location. When validating each algorithm using cloud radar data, the VIIRS algorithm was shown to be more effective at detecting cloud overlap than the AVHRR algorithm; however, the VIIRS algorithm was slightly more prone to false cloud overlap detection.
Abstract
This paper demonstrates that the split-window approach for estimating cloud properties can improve upon the methods commonly used for generating cloud temperature and emissivity climatologies from satellite imagers. Because the split-window method provides cloud properties that are consistent for day and night, it is ideally suited for the generation of a cloud climatology from the Advanced Very High Resolution Radiometer (AVHRR), which provides sampling roughly four times per day. While the split-window approach is applicable to all clouds, this paper focuses on its application to cirrus (high semitransparent ice clouds), where this approach is most powerful. An optimal estimation framework is used to extract estimates of cloud temperature, cloud emissivity, and cloud microphysics from the AVHRR split-window observations. The performance of the split-window approach is illustrated through the diagnostic quantities generated by the optimal estimation approach. An objective assessment of the performance of the algorithm cloud products from the recently launched space lidar [Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation/Cloud-Aerosol Lidar with Orthogonal Polarization (CALIPSO/CALIOP)] is used to characterize the performance of the AVHRR results and also to provide the constraints needed for the optimal estimation approach.
Abstract
This paper demonstrates that the split-window approach for estimating cloud properties can improve upon the methods commonly used for generating cloud temperature and emissivity climatologies from satellite imagers. Because the split-window method provides cloud properties that are consistent for day and night, it is ideally suited for the generation of a cloud climatology from the Advanced Very High Resolution Radiometer (AVHRR), which provides sampling roughly four times per day. While the split-window approach is applicable to all clouds, this paper focuses on its application to cirrus (high semitransparent ice clouds), where this approach is most powerful. An optimal estimation framework is used to extract estimates of cloud temperature, cloud emissivity, and cloud microphysics from the AVHRR split-window observations. The performance of the split-window approach is illustrated through the diagnostic quantities generated by the optimal estimation approach. An objective assessment of the performance of the algorithm cloud products from the recently launched space lidar [Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation/Cloud-Aerosol Lidar with Orthogonal Polarization (CALIPSO/CALIOP)] is used to characterize the performance of the AVHRR results and also to provide the constraints needed for the optimal estimation approach.
Abstract
Surface cloud radiative forcing from the newly extended Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder (APP-x) dataset and surface cloud radiative forcing calculated using cloud and surface properties from the International Satellite Cloud Climatology Project (ISCCP) D-series product were used in this 9-yr (1985–93) study. On the monthly timescale, clouds were found to have a warming effect on the surface of the Antarctic continent every month of the year in both datasets. Over the ocean poleward of 58.75°S, clouds were found to have a warming effect on the surface from March through October in the ISCCP-derived dataset and from April through September in the APP-x dataset. Net surface fluxes from both datasets were validated against net surface fluxes calculated from measurements of upwelling and downwelling shortwave and longwave radiation at the Neumayer and Amundsen–Scott South Pole Stations in the Antarctic. The net all-wave surface flux from the ISCCP-derived dataset was found to be within 0.4–50 W m−2 of the net all-wave flux at the two stations on the monthly timescale. The APP-x net all-wave surface flux was found to be within 0.9–24 W m−2. Model sensitivity studies were conducted to gain insight into how the surface radiation budget in a cloudy atmosphere will change if certain cloud and surface properties were to change in association with regional and/or global climate change. The results indicate that the net cloud forcing will be most sensitive to changes in cloud amount, surface reflectance, cloud optical depth, and cloud-top pressure.
Abstract
Surface cloud radiative forcing from the newly extended Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder (APP-x) dataset and surface cloud radiative forcing calculated using cloud and surface properties from the International Satellite Cloud Climatology Project (ISCCP) D-series product were used in this 9-yr (1985–93) study. On the monthly timescale, clouds were found to have a warming effect on the surface of the Antarctic continent every month of the year in both datasets. Over the ocean poleward of 58.75°S, clouds were found to have a warming effect on the surface from March through October in the ISCCP-derived dataset and from April through September in the APP-x dataset. Net surface fluxes from both datasets were validated against net surface fluxes calculated from measurements of upwelling and downwelling shortwave and longwave radiation at the Neumayer and Amundsen–Scott South Pole Stations in the Antarctic. The net all-wave surface flux from the ISCCP-derived dataset was found to be within 0.4–50 W m−2 of the net all-wave flux at the two stations on the monthly timescale. The APP-x net all-wave surface flux was found to be within 0.9–24 W m−2. Model sensitivity studies were conducted to gain insight into how the surface radiation budget in a cloudy atmosphere will change if certain cloud and surface properties were to change in association with regional and/or global climate change. The results indicate that the net cloud forcing will be most sensitive to changes in cloud amount, surface reflectance, cloud optical depth, and cloud-top pressure.
Abstract
Data from the National Oceanic and Atmospheric Administration’s (NOAA’s) Advanced Very High Resolution Radiometer (AVHRR) instrument are used to provide the mean July and January global daytime distributions of multilayer cloud, where multilayer cloud is defined as cirrus overlapping one or more lower layers. The AVHRR data were taken from multiple years that were chosen to provide data with a constant local equator crossing time of 1430–1500 local time. The cloud overlap detection algorithm is used in NOAA’s Extended Clouds from AVHRR (CLAVR-x) processing system. The results between 60°N and 60°S indicated that roughly 20% of all clouds and roughly 40% of all ice clouds were classified as cirrus overlapping lower cloud (cirrus overlap). The results show a strong July–January pattern that is consistent with the seasonal cycle in convection. In some regions, cirrus overlap is found to be the dominant type of cloud observed. The distributions of overlapping cirrus cloud presented here are compared with results from other studies based on rawinsondes and manual surface observations. Comparisons are also made with another satellite-derived study that used coincident infrared and microwave observations over the tropical oceans during a 6-month period
Abstract
Data from the National Oceanic and Atmospheric Administration’s (NOAA’s) Advanced Very High Resolution Radiometer (AVHRR) instrument are used to provide the mean July and January global daytime distributions of multilayer cloud, where multilayer cloud is defined as cirrus overlapping one or more lower layers. The AVHRR data were taken from multiple years that were chosen to provide data with a constant local equator crossing time of 1430–1500 local time. The cloud overlap detection algorithm is used in NOAA’s Extended Clouds from AVHRR (CLAVR-x) processing system. The results between 60°N and 60°S indicated that roughly 20% of all clouds and roughly 40% of all ice clouds were classified as cirrus overlapping lower cloud (cirrus overlap). The results show a strong July–January pattern that is consistent with the seasonal cycle in convection. In some regions, cirrus overlap is found to be the dominant type of cloud observed. The distributions of overlapping cirrus cloud presented here are compared with results from other studies based on rawinsondes and manual surface observations. Comparisons are also made with another satellite-derived study that used coincident infrared and microwave observations over the tropical oceans during a 6-month period
Abstract
Cloud properties from the newly extended Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder (APP-x) dataset were incorporated into the atmospheric component of the Arctic Regional Climate System Model (ARCSyM) in order to improve the simulation of the Antarctic surface energy balance. A method for using the APP-x cloud properties in 48-h model simulations is presented. In the experiments, the model cloud fields were altered via the water vapor mixing ratio using cloud properties from the APP-x dataset. Significant improvements in monthly mean downwelling longwave radiation at the surface were observed relative to surface measurements. In the austral summer, the use of the APP-x dataset resulted in improvements as large as 30 W m−2 at the South Pole when compared to model results without APP-x clouds. However, only a very small improvement was seen in the turbulent heat fluxes and the surface temperature. It was also found that the satellite data can be used to shorten the model “spinup” time and may be useful in model initialization for short duration forecasts.
Abstract
Cloud properties from the newly extended Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder (APP-x) dataset were incorporated into the atmospheric component of the Arctic Regional Climate System Model (ARCSyM) in order to improve the simulation of the Antarctic surface energy balance. A method for using the APP-x cloud properties in 48-h model simulations is presented. In the experiments, the model cloud fields were altered via the water vapor mixing ratio using cloud properties from the APP-x dataset. Significant improvements in monthly mean downwelling longwave radiation at the surface were observed relative to surface measurements. In the austral summer, the use of the APP-x dataset resulted in improvements as large as 30 W m−2 at the South Pole when compared to model results without APP-x clouds. However, only a very small improvement was seen in the turbulent heat fluxes and the surface temperature. It was also found that the satellite data can be used to shorten the model “spinup” time and may be useful in model initialization for short duration forecasts.
Abstract
A comparison is made between a new operational NOAA Advanced Very High Resolution Radiometer (AVHRR) global cloud amount product to those from established satellite-derived cloud climatologies. The new operational NOAA AVHRR cloud amount is derived using the cloud detection scheme in the extended Clouds from AVHRR (CLAVR-x) system. The cloud mask within CLAVR-x is a replacement for the Clouds from AVHRR phase 1 (CLAVR-1) cloud mask. Previous analysis of the CLAVR-1 cloud climatologies reveals that its utility for climate studies is reduced by poor high-latitude performance and the inability to include data from the morning orbiting satellites. This study demonstrates, through comparison with established satellite-derived cloud climatologies, the ability of CLAVR-x to overcome the two main shortcomings of the CLAVR-1-derived cloud climatologies. While systematic differences remain in the cloud amounts from CLAVR-x and other climatologies, no evidence is seen that these differences represent a failure of the CLAVR-x cloud detection scheme. Comparisons for July 1995 and January 1996 indicate that for most latitude zones, CLAVR-x produces less cloud than the International Satellite Cloud Climatology Project (ISCCP) and the University of Wisconsin High Resolution Infrared Radiation Sounder (UW HIRS). Comparisons to the Moderate Resolution Imaging Spectroradiometer (MODIS) for 1–8 April 2003 also reveal that CLAVR-x tends to produce less cloud. Comparison of the seasonal cycle (July–January) of cloud difference with ISCCP, however, indicates close agreement. It is argued that these differences may be due to the methodology used to construct a cloud amount from the individual pixel-level cloud detection results. Overall, the global cloud amounts from CLAVR-x appear to be an improvement over those from CLAVR-1 and compare well to those from established satellite cloud climatologies. The CLAVR‐x cloud detection results have been operational since late 2003 and are available in real time from NOAA.
Abstract
A comparison is made between a new operational NOAA Advanced Very High Resolution Radiometer (AVHRR) global cloud amount product to those from established satellite-derived cloud climatologies. The new operational NOAA AVHRR cloud amount is derived using the cloud detection scheme in the extended Clouds from AVHRR (CLAVR-x) system. The cloud mask within CLAVR-x is a replacement for the Clouds from AVHRR phase 1 (CLAVR-1) cloud mask. Previous analysis of the CLAVR-1 cloud climatologies reveals that its utility for climate studies is reduced by poor high-latitude performance and the inability to include data from the morning orbiting satellites. This study demonstrates, through comparison with established satellite-derived cloud climatologies, the ability of CLAVR-x to overcome the two main shortcomings of the CLAVR-1-derived cloud climatologies. While systematic differences remain in the cloud amounts from CLAVR-x and other climatologies, no evidence is seen that these differences represent a failure of the CLAVR-x cloud detection scheme. Comparisons for July 1995 and January 1996 indicate that for most latitude zones, CLAVR-x produces less cloud than the International Satellite Cloud Climatology Project (ISCCP) and the University of Wisconsin High Resolution Infrared Radiation Sounder (UW HIRS). Comparisons to the Moderate Resolution Imaging Spectroradiometer (MODIS) for 1–8 April 2003 also reveal that CLAVR-x tends to produce less cloud. Comparison of the seasonal cycle (July–January) of cloud difference with ISCCP, however, indicates close agreement. It is argued that these differences may be due to the methodology used to construct a cloud amount from the individual pixel-level cloud detection results. Overall, the global cloud amounts from CLAVR-x appear to be an improvement over those from CLAVR-1 and compare well to those from established satellite cloud climatologies. The CLAVR‐x cloud detection results have been operational since late 2003 and are available in real time from NOAA.
Abstract
Three multispectral algorithms for determining the cloud type of previously identified cloudy pixels during the daytime, using satellite imager data, are presented. Two algorithms were developed for use with 0.65-, 1.6-/3.75-, 10.8-, and 12.0-μm data from the Advanced Very High Resolution Radiometer (AVHRR) on board the National Oceanic and Atmospheric Administration (NOAA) operational polar-orbiting satellites. The AVHRR algorithms are identical except for the near-infrared data that are used. One algorithm uses AVHRR channel 3a (1.6 μm) reflectances, and the other uses AVHRR channel 3b (3.75 μm) reflectance estimates. Both of these algorithms are necessary because the AVHRRs on NOAA-15 through NOAA-17 have the capability to transmit either channel 3a or 3b data during the day, whereas all of the other AVHRRs on NOAA-7 through NOAA-14 can only transmit channel 3b data. The two AVHRR cloud-typing schemes are used operationally in NOAA’s extended Clouds from AVHRR (CLAVR)-x processing system. The third algorithm utilizes additional spectral bands in the 1.38- and 8.5-μm regions of the spectrum that are available on the Moderate Resolution Imaging Spectroradiometer (MODIS) and will be available on the Visible–Infrared Imaging Radiometer Suite (VIIRS). The VIIRS will eventually replace the AVHRR on board the National Polar-Orbiting Operational Environmental Satellite System (NPOESS), which is currently scheduled to be launched in 2009. Five cloud-type categories are employed: warm liquid water, supercooled water–mixed phase, opaque ice, nonopaque high ice (cirrus), and cloud overlap (multiple cloud layers). Each algorithm was qualitatively evaluated through scene analysis and then validated against inferences of cloud type that were derived from ground-based observations of clouds at the three primary Atmospheric Radiation Measurement (ARM) Program sites to help to assess the potential continuity of a combined AVHRR channel 3a–AVHRR channel 3b–VIIRS cloud-type climatology. In this paper, “validation” is strictly defined as comparisons with ground-based estimates that are completely independent of the satellite retrievals. It was determined that the two AVHRR algorithms produce nearly identical results except for certain thin clouds and cloud edges. The AVHRR 3a algorithm tends to incorrectly classify the thin edges of some low- and midlevel clouds as cirrus and opaque ice more often than the AVHRR 3b algorithm. The additional techniques implemented in the VIIRS algorithm result in a significant improvement in the identification of cirrus clouds, cloud overlap, and overall phase identification of thin clouds, as compared with the capabilities of the AVHRR algorithms presented in this paper.
Abstract
Three multispectral algorithms for determining the cloud type of previously identified cloudy pixels during the daytime, using satellite imager data, are presented. Two algorithms were developed for use with 0.65-, 1.6-/3.75-, 10.8-, and 12.0-μm data from the Advanced Very High Resolution Radiometer (AVHRR) on board the National Oceanic and Atmospheric Administration (NOAA) operational polar-orbiting satellites. The AVHRR algorithms are identical except for the near-infrared data that are used. One algorithm uses AVHRR channel 3a (1.6 μm) reflectances, and the other uses AVHRR channel 3b (3.75 μm) reflectance estimates. Both of these algorithms are necessary because the AVHRRs on NOAA-15 through NOAA-17 have the capability to transmit either channel 3a or 3b data during the day, whereas all of the other AVHRRs on NOAA-7 through NOAA-14 can only transmit channel 3b data. The two AVHRR cloud-typing schemes are used operationally in NOAA’s extended Clouds from AVHRR (CLAVR)-x processing system. The third algorithm utilizes additional spectral bands in the 1.38- and 8.5-μm regions of the spectrum that are available on the Moderate Resolution Imaging Spectroradiometer (MODIS) and will be available on the Visible–Infrared Imaging Radiometer Suite (VIIRS). The VIIRS will eventually replace the AVHRR on board the National Polar-Orbiting Operational Environmental Satellite System (NPOESS), which is currently scheduled to be launched in 2009. Five cloud-type categories are employed: warm liquid water, supercooled water–mixed phase, opaque ice, nonopaque high ice (cirrus), and cloud overlap (multiple cloud layers). Each algorithm was qualitatively evaluated through scene analysis and then validated against inferences of cloud type that were derived from ground-based observations of clouds at the three primary Atmospheric Radiation Measurement (ARM) Program sites to help to assess the potential continuity of a combined AVHRR channel 3a–AVHRR channel 3b–VIIRS cloud-type climatology. In this paper, “validation” is strictly defined as comparisons with ground-based estimates that are completely independent of the satellite retrievals. It was determined that the two AVHRR algorithms produce nearly identical results except for certain thin clouds and cloud edges. The AVHRR 3a algorithm tends to incorrectly classify the thin edges of some low- and midlevel clouds as cirrus and opaque ice more often than the AVHRR 3b algorithm. The additional techniques implemented in the VIIRS algorithm result in a significant improvement in the identification of cirrus clouds, cloud overlap, and overall phase identification of thin clouds, as compared with the capabilities of the AVHRR algorithms presented in this paper.
Abstract
Lightning strikes pose a hazard to human life and property, and can be difficult to forecast in a timely manner. In this study, a satellite-based machine learning model was developed to provide objective, short-term, location-specific probabilistic guidance for next-hour lightning activity. Using a convolutional neural network architecture designed for semantic segmentation, the model was trained using GOES-16 visible, shortwave infrared, and longwave infrared bands from the Advanced Baseline Imager (ABI). Next-hour GOES-16 Geostationary Lightning Mapper data were used as the truth or target data. The model, known as LightningCast, was trained over the GOES-16 ABI contiguous United States (CONUS) domain. However, the model is shown to generalize to GOES-16 full disk regions that are outside of the CONUS. LightningCast provides predictions for developing and advecting storms, regardless of solar illumination and meteorological conditions. LightningCast, which frequently provides 20 min or more of lead time to new lightning activity, learned salient features consistent with the scientific understanding of the relationships between lightning and satellite imagery interpretation. We also demonstrate that despite being trained on data from a single geostationary satellite domain (GOES-East), the model can be applied to other satellites (e.g., GOES-West) with comparable specifications and without substantial degradation in performance. LightningCast objectively transforms large volumes of satellite imagery into objective, actionable information. Potential application areas are also highlighted.
Significance Statement
The outcome of this research is a model that spatially forecasts lightning occurrence in a 0–60-min time window, using only images of clouds from the GOES-R Advanced Baseline Imager. This model has the potential to provide early alerts for developing and approaching hazardous conditions.
Abstract
Lightning strikes pose a hazard to human life and property, and can be difficult to forecast in a timely manner. In this study, a satellite-based machine learning model was developed to provide objective, short-term, location-specific probabilistic guidance for next-hour lightning activity. Using a convolutional neural network architecture designed for semantic segmentation, the model was trained using GOES-16 visible, shortwave infrared, and longwave infrared bands from the Advanced Baseline Imager (ABI). Next-hour GOES-16 Geostationary Lightning Mapper data were used as the truth or target data. The model, known as LightningCast, was trained over the GOES-16 ABI contiguous United States (CONUS) domain. However, the model is shown to generalize to GOES-16 full disk regions that are outside of the CONUS. LightningCast provides predictions for developing and advecting storms, regardless of solar illumination and meteorological conditions. LightningCast, which frequently provides 20 min or more of lead time to new lightning activity, learned salient features consistent with the scientific understanding of the relationships between lightning and satellite imagery interpretation. We also demonstrate that despite being trained on data from a single geostationary satellite domain (GOES-East), the model can be applied to other satellites (e.g., GOES-West) with comparable specifications and without substantial degradation in performance. LightningCast objectively transforms large volumes of satellite imagery into objective, actionable information. Potential application areas are also highlighted.
Significance Statement
The outcome of this research is a model that spatially forecasts lightning occurrence in a 0–60-min time window, using only images of clouds from the GOES-R Advanced Baseline Imager. This model has the potential to provide early alerts for developing and approaching hazardous conditions.
Abstract
This paper presents the Thunderstorm Nowcasting Tool (ThunderCast), a 24-h, year-round model for predicting the location of convection that is likely to initiate or remain a thunderstorm in the next 0–60 min in the continental United States, adapted from existing deep learning convection applications. ThunderCast utilizes a U-Net convolutional neural network for semantic segmentation trained on 320 km × 320 km data patches with four inputs and one target dataset. The inputs are satellite bands from the Geostationary Operational Environmental Satellite-16 (GOES-16) Advanced Baseline Imager (ABI) in the visible, shortwave infrared, and longwave infrared spectra, and the target is Multi-Radar Multi-Sensor (MRMS) radar reflectivity at the −10°C isotherm in the atmosphere. On a pixel-by-pixel basis, ThunderCast has high accuracy, recall, and specificity but is subject to false-positive predictions resulting in low precision. However, the number of false positives decreases when buffering the target values with a 15 km × 15 km centered window, indicating ThunderCast’s predictions are useful within a buffered area. To demonstrate the initial prediction capabilities of ThunderCast, three case studies are presented: a mesoscale convective vortex, sea-breeze convection, and monsoonal convection in the southwestern United States. The case studies illustrate that the ThunderCast model effectively nowcasts the location of newly initiated and ongoing active convection, within the next 60 min, under a variety of geographical and meteorological conditions.
Significance Statement
In this research, a machine learning model is developed for short-term (0–60 min) forecasting of thunderstorms in the continental United States using geostationary satellite imagery as inputs for predicting active convection based on radar thresholds. Pending additional testing, the model may be able to provide decision-support services for thunderstorm forecasting. The case studies presented here indicate the model is able to nowcast convective initiation with 5–35 min of lead time in areas without radar coverage and anticipate future locations of storms without additional environmental context.
Abstract
This paper presents the Thunderstorm Nowcasting Tool (ThunderCast), a 24-h, year-round model for predicting the location of convection that is likely to initiate or remain a thunderstorm in the next 0–60 min in the continental United States, adapted from existing deep learning convection applications. ThunderCast utilizes a U-Net convolutional neural network for semantic segmentation trained on 320 km × 320 km data patches with four inputs and one target dataset. The inputs are satellite bands from the Geostationary Operational Environmental Satellite-16 (GOES-16) Advanced Baseline Imager (ABI) in the visible, shortwave infrared, and longwave infrared spectra, and the target is Multi-Radar Multi-Sensor (MRMS) radar reflectivity at the −10°C isotherm in the atmosphere. On a pixel-by-pixel basis, ThunderCast has high accuracy, recall, and specificity but is subject to false-positive predictions resulting in low precision. However, the number of false positives decreases when buffering the target values with a 15 km × 15 km centered window, indicating ThunderCast’s predictions are useful within a buffered area. To demonstrate the initial prediction capabilities of ThunderCast, three case studies are presented: a mesoscale convective vortex, sea-breeze convection, and monsoonal convection in the southwestern United States. The case studies illustrate that the ThunderCast model effectively nowcasts the location of newly initiated and ongoing active convection, within the next 60 min, under a variety of geographical and meteorological conditions.
Significance Statement
In this research, a machine learning model is developed for short-term (0–60 min) forecasting of thunderstorms in the continental United States using geostationary satellite imagery as inputs for predicting active convection based on radar thresholds. Pending additional testing, the model may be able to provide decision-support services for thunderstorm forecasting. The case studies presented here indicate the model is able to nowcast convective initiation with 5–35 min of lead time in areas without radar coverage and anticipate future locations of storms without additional environmental context.
