Breakout Summary Report

The scanning strategy for ARM radars can be adapted based on field campaign objectives or user requirements. For instance, during TRACER, precipitation echoes from CSAPR2 were collected using an adaptive scanning strategy based on the Multisensor Agile Adaptive Sampling (MAAS) framework (Kollias et al. 2023, Lamer et al. 2023). Additionally, numerous cell tracking algorithms have been used in recent studies to track convective clouds sampled by ARM radars or external radars near ARM sites. For instance, Gupta et al. (2024) used an object tracking algorithm called tobac (Tracking and Object-Based Analysis of Clouds) to track isolated deep convective clouds using an S-band radar during the GoAmazon2014/5 field campaign in Brazil. Similarly, Feng et al. (2022) and Sharma et al. (2024) have used PyFLEXTRKR (Python FLEXible object TRacKeR) to track shallow and deep convection sampled by the CSAPR2 during CACTI and TRACER, respectively.

There is a growing need for the development of optimized and adaptive radar scanning strategies at long-term mobile observatories such as the AMF3 site at Bankhead National Forest, which can meet the science objectives of field campaigns proposed by ARM/ASR scientists. With the growing number of tracking algorithms, it is critical to develop frameworks that guide new users on which algorithm is best suited under different conditions. The goal of this session is to discuss the merits and drawbacks of existing radar scanning strategies, bring together new and existing users of cell tracking algorithms applied to ARM radar data, solicit ideas for adaptive radar scanning strategies that optimize the use of different tracking algorithms, and advertise past and current activities conducted by ARM staff and ASR scientists to create tracking datasets for existing field campaigns and radar datasets.

Pavlos Kollias outlined plans for synchronized scanning between phased-array radars (PARs) and the UAH radar near BNF. This will be a testbed for improving synchronized scanning across multiple radars and coordinating the targeted sampling of meteorological features from different radars. He described a project funded by the NSF Cyberinfrastructure for Sustained Scientific Innovation program to develop software that aims to transition other research radars to have the MAAS capability used during TRACER. His team plans to make this an open-source software within 2 years. In response to concerns associated with the data volume from PARs, he replied PAR data will be sorted per the target feature rather than an entire volume scan which would then be a relatively sparse dataset. The real time cell-tracking algorithm for PAR scanning can be integrated within the MAAS framework to guide dish antenna radars to scan convective cells undergoing rapid updraft intensification. Rule-based scanning strategies can be designed to sample specific atmospheric phenomena. There was feedback from modelers that such data should be collected and formatted following practices that encourage model-observation intercomparison efforts.

Bobby Jackson discussed how tracking capabilities can be enhanced using edge computing and high-performance computing workflows. He emphasized developing a generalizable infrastructure for adaptive scanning with Doppler lidars using a case study on improving wind forecasts for offshore wind farms. ANL’s wireless sensor platform ‘Waggle AI’ is being used for computing at the sensor nodes to quickly analyze real-time observations and adapt a scanning strategy for Doppler lidars to characterize the environment. He showed how an AI algorithm was trained to automatically track a lake breeze front in real-time. Ensuing discussions identified Observing System Simulation Experiments (OSSEs), digital twins, and instrument simulators as objective methods for training AI algorithms with minimal human involvement. AI techniques could be integrated with these methods to create adaptive scanning strategies. Edge computing at the sensor can be used for simple radar antenna moment calculations from Doppler spectra, while science missions aimed at maximizing information content may require storing complex Doppler spectra for post-processing. Techniques like 3DVAR multi-Doppler analyses, developed from OSSEs, have been immensely useful, and can be extended to PARs. Since technology is available for conducting testbed experiments, it is important to pursue such advancements.

Science questions to be addressed if such technologies were available for ARM campaigns? The audience generally agreed convective dynamics could be studied in much greater detail. Do ARM radars have the capability to leverage these techniques, or should new systems be considered? It was noted that KAZR and XSACR volume scans are often data-sparse, making them highly valuable for studying shallow convection. In such cases, Doppler lidars can be used to analyze boundary layer convective eddies in the sub-cloud layer, while cloud and precipitation radars provide insights into in-cloud dynamics.

Sean Freeman presented on cloud tracking using tobac and highlighted the need to make cell track output more accessible to the wider user community. His group is working on a product that tracks convective cells near BNF using (i) MRMS 3D reflectivity, (ii) MRMS 4-km AGL reflectivity, (iii) composite 3D reflectivity from southeast US NEXRAD data, and (iv) individual radar sites. In response to a question about incorporating other remote sensing platforms (GOES-16/18 satellites) to achieve greater coverage - Sean’s team is currently developing a multivariate threshold version of tobac, which could later be adapted to integrate multimodal observations. 

Zhe Feng presented results from a comparative study to understand biases in modeled convective processes at different grid spacings from LASSO CACTI simulations by evaluating them against observations. Going forward with BNF-specific modeling comparison with observations and radar-based cloud-tracking, it would be great if the parameters used for tracking algorithms and target variables used for tracking are shared with the community to encourage reproducibility. The scanning strategy for CSAPR2, if different from the KHTX WSR-88D (e.g., RHI-mode only), could make it problematic for users to track cells exclusively from CSAPR2 data and thereby losing significant details regarding the cell life cycle.

Die Wang introduced a simplified workflow to help ARM data users quickly familiarize themselves with various open-source cloud-tracking tools. The Community Cloud Model Evaluation Toolkit (CoCoMET) aims to lower the entry barrier for running cell tracking by offering utilities that standardize input data within a consistent framework and enable multiple trackers to run simultaneously using a single configuration file. There was strong community interest in incorporating this tracking toolkit for different models (e.g., RAMS, WRF) used in TRACER-MIP and different trackers (e.g., PyFLEXTRKR). 

Explore ARM field campaign or testbed opportunities whereby ARM radars/lidars are operated using a common scanning software which could be integrated with AI/ML models that help determine the scan strategy based on the target feature selected for tracking. AI algorithms have been trained to automatically track a lake breeze front in real-time. This type of intelligent cyberinfrastructure can support targeted measurements of atmospheric phenomena across a range of spatial and temporal scales, addressing the diverse interests of different groups.

OSSEs provide an opportunity to compare observations and model output by sampling model output in the same way as observations. Given the expertise within ARM/ASR in fundamental convective processes, experimental strategies leveraging real observations from testbeds may represent a more exciting and practical frontier than OSSEs.

Convective cells simulated in LASSO runs were found to diverge from observations the most during the early stages of their life cycle. Do we need additional radars to capture the early stages of the convective cloud development? Would XSACR and KAZR help address some of these concerns?

N/A

N/A

N/A

N/A

Pavlos invited the ARM radar community to participate in testing the MAAS cyberinfrastructure (a Java app) that allows users to remotely control PAR scanning strategy. User feedback will help developers in enhancing feature detection, cell tracking, and decision-making workflows when multiple features of interest are present. The feedback will also lend insights into sampling requirements for PI-specific research goals. 

Leverage AI/ML to automate feature tracking techniques. OSSE models used to finalize scanning strategies should be shared with the modeling community to help them understand the design of these strategies and how to implement them for their model output for comparison with observations.

The community highlighted the utility of a cell-track database to analyze the fraction of shallow, deep, and organized convection within the domain at any given time. Assessing the contribution of these convective systems to the total domain-wide precipitation would provide insights into precipitation variability. Add supplementary metadata with cell tracking output that contains information about the ambient environment (from reanalysis). Sean’s group currently runs tobac in near real-time. The modular design of tobac makes it easy to adapt to LASSO simulations, as well.

It was emphasized that, once publicly released, the CoCoMET tracking package should be maintained as a community-supported effort, rather than relying on the original developers to accommodate individual user needs, which would be unsustainable in the long run.