the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 31 Aug 2026
Impact of Low-altitude Meteorological Drone Data Assimilation on Convective-Scale Short-Term Rainfall Forecasts: An Observing System Simulation Study
Abstract. This work leverages observing system simulation experiments (OSSEs) to quantify the utility of low-altitude meteorological drone (MD) measurements for convective-scale analyses and short-term rainfall forecasts over the Beijing-Tianjin-Hebei region. Synthetic MD observations of temperature, specific humidity, and horizontal wind are generated from a free-running truth simulation and assimilated into the Weather Research and Forecasting (WRF) model using the National Severe Storms Laboratory three-dimensional variational data assimilation (DA) system. Five sets of sensitivity experiments are conducted to evaluate the impacts of horizontal resolution, observation height, spatial distribution, joint assimilation of thermodynamic and wind observations, and observation errors of MD data, respectively. The results show that assimilation of MD temperature and humidity observations improves both thermodynamic analyses and precipitation forecasts, with the magnitude of benefit strongly dependent on MD network design. Denser MD networks more effectively reduce thermodynamic analysis and forecast errors, leading to better rainband placement, rainfall intensity, and higher quantitative precipitation skill. Among the tested configurations, the 5- and 10-km networks provide the most robust and consistent forecast benefits. Multi-level MD data yield the most balanced improvement, while observations extending to higher levels within the planetary boundary layer are generally more beneficial than those confined to the lowest level alone. Restricting observations to the plain area degrades forecast performance, highlighting the importance of upstream mountainous observations where convection is initiated. In addition, joint assimilation of thermodynamic and wind observations further improves quantitative precipitation forecasts by substantially reducing lower-tropospheric wind errors. Short-term forecast skill is also sensitive to the observation error standard deviations, with inflated wind observation error producing a larger degradation than inflated thermodynamic errors. Overall, it is demonstrated that MD observations have considerable potential to improve convective-scale numerical weather prediction, particularly when the observing network is sufficiently dense, vertically resolved, and capable of constraining both thermodynamic and dynamical structures within the planetary boundary layer.
- Preprint
(4375 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 26 Oct 2026)
- CC1: 'Comment on egusphere-2026-3350', Xin Li, 01 Sep 2026 reply
-
RC1: 'Comment on egusphere-2026-3350', Anonymous Referee #1, 06 Sep 2026
reply
This study uses simulated meteorological‑UAV observations to investigate the impacts of assimilating meteorological UAV data on short‑range numerical weather prediction. The manuscript systematically analyzes how assimilation performance is affected by different observation data types, observation locations, and prescribed observation errors. It further highlights the value of assimilating observations within the planetary boundary layer. This work is of great significance for improving numerical forecasts of high‑impact weather, and also provides valuable references for designing meteorological UAV observation strategies. The research results demonstrate good scientific and application values. Nevertheless, it is confusing that the UAV observations are only performed at three fixed altitude levels. The authors are suggested to clarify in the data description whether this configuration arises from technical limitations of UAV platforms or is an empirical setup.
Specific comments:
Line 197-198: “a MD assimilation module is developed and incorporated into the DA system.”
Why is a separate assimilation module constructed for MD observations, rather than directly adopting the existing assimilation module for conventional observations? The authors are suggested to provide relevant explanations in the manuscript.
Line 216-217: “The chosen correlation scales are proportional to the observation spacing, reflecting the characteristic scales resolved by the MD network.”
The authors are suggested to discuss the relationship between the relevant scales and the MD network. It should be clarified whether the 5‑km configuration represents the optimal setting for the MD network, so that readers can better understand the purpose of this setup.
Line 262-267: “ These deficiencies indicate considerable … highly sensitive to thermodynamic and kinematic conditions in the PBL”
This paragraph is difficult to follow. Why does the inconsistency in the spatial distribution of precipitation indicate that precipitation is related to the thermodynamic and dynamic characteristics of the planetary boundary layer? The authors are suggested to elaborate on this point.
Line 306: “Gaussian random errors with a mean of zero are added”
It is suggested to clarify which scenario the Gaussian‑distributed errors correspond to. For instance, some instruments tend to exhibit larger errors in high‑temperature regions and smaller errors in low‑temperature regions. Could such characteristics violate the assumption of Gaussian distribution?
Line 416-418: “especially near the Yanshan and Taihang Mountains, while weak cold biases are found over northeastern Hebei, northern Shandong, and the adjacent coastal waters”
The authors are suggested to provide appropriate physical interpretations. The error characteristics show similar patterns to the terrain height variation. Is such error mainly caused by the model’s inability to accurately resolve mountain locations, which consequently yields poor simulations of weather variations induced by high‑elevation terrain?
Line 427-429: “which suggests that the densest MD network most effectively constrains the lower-tropospheric thermal analysis.”
I acknowledge that it is impractical to conduct an unlimited number of experiments with diverse resolutions. Nevertheless, the authors are suggested to appropriately discuss the matching relationship between observation resolution and model resolution. Higher observation resolution does not necessarily yield better performance. In addition, it is worth exploring whether there exists an upper limit for the improvement achievable via resolution enhancement.
Line 610-611: “This suggests that restricting the MD network to the plain area degrades the temperature analysis relative to the full-network configuration,”
The term “degrade” is not recommended, as it may mislead readers into thinking that adding observations over plain regions would degrade model performance. As the authors mentioned, temperature errors differ between plain and plateau areas. However, identical observation errors are prescribed across all regions in this study. It is suggested that region‑dependent observation errors be considered, which can be listed as one of the directions for future exploration.
Citation: https://doi.org/10.5194/egusphere-2026-3350-RC1 -
RC2: 'Comment on egusphere-2026-3350', Anonymous Referee #2, 27 Sep 2026
reply
The authors investigate the potential value of low-altitude meteorological drone (MD) observations for convective-scale short-term forecasting. In this manuscript, an Observing System Simulation Experiment (OSSE) is conducted for a rainfall event over the Beijing–Tianjin–Hebei region, in which synthetic MD observations of temperature, specific humidity, and horizontal wind are generated from a free-running “truth” simulation. A comprehensive analysis is carried out through five sets of sensitivity experiments, examining the impacts of horizontal resolution, observation height, spatial distribution, joint assimilation of thermodynamic and wind variables, and observation errors. The results indicate that assimilating MD temperature and humidity observations improves both thermodynamic analyses and precipitation forecasts. Furthermore, denser MD networks and observations extending to higher altitudes provide additional forecast benefits, and the inclusion of wind observations yields further improvements.
This research is valuable, and the manuscript is generally well written. However, several issues require clarification and further consideration:
- Vertical levels of MD observations: In the experiments, only three observation levels (120 m, 300 m, and 600 m) are considered. However, Lines 282–283 indicate that MD observations are typically conducted within the lowest 1–3 km of the atmosphere, while Lines 279–295 suggest that future MD development will focus on low-altitude airspace below 1 km. Does this imply that future MD deployments will be limited to the three selected levels? If not, what is the rationale for choosing only these levels? It would be helpful to justify this selection or consider including additional vertical levels to assess their impact.
- Observation density across levels: In the experiments examining different observation heights, are the numbers of observations at each level kept consistent? If not, variations in observation count may influence the results and should be discussed or controlled for.
- Representativeness of results: The study focuses on a single heavy rainfall event. Including additional cases would strengthen the generality of the conclusions. If expanding the dataset is not feasible, the authors should be cautious in presenting broad conclusions. For example, how representative are the findings (e.g., Lines 808–811)? To what extent can they be generalized to other convective events with different synoptic conditions?
- Line 485 states that specific humidity in the analysis and forecast is examined. In Figure 6 and related figures, it is labeled as Qv. Could the authors clarify whether the variable shown is specific humidity or water vapor mixing ratio?
- In Lines 337–340, it is important to explicitly state that the data assimilation (DA) experiments are designed to evaluate the upper bound of the information content provided by MD observations, given that no other observational data are assimilated. This context is critical for interpreting the results. Additionally, it would be valuable to discuss how the impact of MD observations might change when assimilated alongside conventional and remote-sensing observation networks.
- In figure 6 and others, is the statistics for Qv? As stated in Line 485, specific humidity of analysis and forecast is studied. However, in figure 6 and others, it is labeled Qv. Just want to clarify is it specific humidity or water vapor mixing ratio.
- Line 337-340, it is very important to point out that the DA experiments are designed to evaluate the upper bound of the information content provided by MD observations. Since no other observations are included in the experiments. It will be very interesting to see its impact on top of conventional and remote-sensing observation network.
Citation: https://doi.org/10.5194/egusphere-2026-3350-RC2 -
RC3: 'Comment on egusphere-2026-3350', Anonymous Referee #3, 29 Sep 2026
reply
In this study, the authors investigate the potential of low-altitude meteorological drone MD (called UAV) observations to improve convective-scale short-term weather forecasts using the 3D-Var data assimilation system in the Warn-on-Forecast System. An Observing System Simulation Experiment (OSSE) is conducted for a rainfall event over the Beijing–Tianjin–Hebei region. Synthetic MD observations of temperature, specific humidity, and horizontal wind are generated from a free-running simulation treated as the “truth”. Their results show that assimilating synthetic MD parameters including temperature and humidity observations can improve the thermodynamic analysis and precipitation forecasts.
This study is relevant for improving short-term forecasting of high-impact weather using synthetic MD observation. It also provides useful guidance for the design and deployment of meteorological MD observing systems. Overall, the manuscript addresses an interesting and relevant topic.
I have a few additional minor comments that may help improve the clarity of the manuscript:
- Line 95-100: The manuscript uses the terms “tropical cyclone” and “typhoon” in different places. For consistency, I suggest using the “typhoon” throughout the manuscript, unless there is a specific reason to distinguish between these terms.
- It is not clear that in the assimilation system others used only assimilated synthetic observation or additional observation also used. This is very important to know the actual impact of those observation.
- Lines 180: In the 3D-Var data assimilation framework, the background error covariance matrices (B) are utilized; however, the significance of the B-matrix within the data assimilation system is not explicitly elucidated. It remains unclear whether the B-matrix is fixed or flow-dependent, and whether it varies across assimilation cycles. Providing additional clarification on this aspect would enhance the understanding of the assimilation process.
- Lines 184-187: The control variables incorporated in the analysis system encompass the 3D wind components, pressure, potential temperature, water vapor mixing ratio, and hydrometeor mixing ratios for cloud water, rain, ice, snow, and graupel, consistent with the variables employed by Gao and Stensrud (2012). The influence of these control variables varies depending on the type of observations assimilated, as noted in the referenced study. It would be beneficial to know whether alternative sets of control variables were evaluated. Furthermore, clarification regarding the specific warm rain microphysics scheme implemented in the system is warranted.
- Lines 342-344: Regarding the data assimilation experiment setup, the temporal frequency at which synthetic MD observations are assimilated within the 3D-Var system is not clearly specified. Although the OSSE is setup within a 6-hour cycling framework with analyses updated every 15 minutes, it is ambiguous whether this implies that all observations are assimilated at a 15-minute temporal resolution, corresponding to the assimilation update frequency. Further elaboration on the observation assimilation frequency would be valuable.
Citation: https://doi.org/10.5194/egusphere-2026-3350-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 286 | 117 | 44 | 447 | 52 | 51 |
- HTML: 286
- PDF: 117
- XML: 44
- Total: 447
- BibTeX: 52
- EndNote: 51
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents an OSSE to evaluate the potential impact of assimilating low-altitude meteorological drone (MD) observations on convective-scale analyses and short-term precipitation forecasts over the Beijing-Tianjin-Hebei region. The authors generate synthetic MD observations of temperature, specific humidity, and horizontal wind from a truth run and assimilate them into the WRF model using the NSSL 3DVAR system. Sensitivity experiments are performed to assess the influences of horizontal resolution, observation height, spatial distribution, combined thermodynamic-wind assimilation, and prescribed observation errors. The results indicate that denser and vertically distributed MD networks improve both thermodynamic analyses and QPFs, and that wind observations provide added value. The study addresses a timely topic relevant to emerging low-altitude meteorological observation networks and provides useful guidance for future network design.
The manuscript is generally well written and organized, the experimental design is logical, and the figures are mostly clear. However, several important methodological limitations and scientific questions need to be addressed before the paper can be considered for publication. My detailed comments are provided below.
Major Comments
Minor Comments