the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 May 2026
Radar range dependent validation of spaceborne cloud profiling radar precipitation detection: lessons from CloudSat and the Canadian C band network for the EarthCARE era
Abstract. Ground based validation of spaceborne precipitation detection algorithms depends critically on the quality of the reference radar observations, which degrades in a range dependent fashion due to ground clutter contamination and beam geometry effects. This study quantifies that dependence using five years (2006–2010) of coincident CloudSat Cloud Profiling Radar (CPR) overpasses and observations from the King City C band dual polarisation radar in southern Ontario, Canada, supplemented by automated and human METAR observations from twelve Environment and Climate Change Canada (ECCC) weather stations. Across 75,239 matched profiles, the CloudSat precipitation occurrence product achieves a probability of detection (POD) of 54.6 %, a critical success index (CSI) of 48.9 %, and a false alarm ratio (FAR) of 17.7 %. A systematic range bin decomposition reveals anomalously elevated Network Radar Precipitation (NRP) algorithm detection frequencies within 70 km of the King City radar, attributable to ground clutter passing the NRP 480 m vertical extent filter before Doppler discrimination becomes reliable. Excluding profiles within this range raises the POD to 66.8 % and the CSI to 56.3 %. Combining this range filter with the ZCPR > -10 dBZ threshold recommended for the Great Lakes winter precipitation regime further reduces the FAR to 13.8 % (CSI = 58.7 %). The 519 independent METAR comparisons yield a POD of 55.8 % and a low FAR of 9.2 %, confirming the radar based findings through a fully instrument independent pathway and demonstrating that the dominant CloudSat error mode is precipitation detection failure rather than false alarming. The performance degradation relative to the winter only benchmark of Hudak et al. (2008) is shown to arise primarily from near range clutter contamination and year round sampling rather than from algorithmic limitations. Critically, the 70 km threshold is governed by antenna beam geometry rather than radar wavelength and is therefore transferable to validation frameworks employing the new Canadian S band dual polarisation network. Implications for the design of ground validation campaigns for the recently launched EarthCARE Cloud Profiling Radar are discussed.
- Preprint
(1404 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 10 Nov 2026)
-
RC1: 'Comment on egusphere-2026-1753', Anonymous Referee #1, 27 Aug 2026
reply
-
AC1: 'Reply on RC1', Farrukh Chishtie, 14 Sep 2026
reply
I thank the referee for a constructive review. I agree the 70 km threshold and the difference from Hudak et al. (2008) need physical, not only statistical, support. My responses:
1. Definition of precipitation in the ground radar. I will fully describe the NRP algorithm: precipitation is flagged from the full 24-tilt volume above each CloudSat footprint when a contiguous echo ≥480 m deep exceeds the range-dependent minimum detectable signal, with Doppler clutter rejection within 120 km. The range-bin table shows the near-range excess (99.5% precipitation frequency at 0–10 km vs. ~20% for CloudSat) lies in the ground radar, not CloudSat. I will add a range-resolved figure of POD, FAR and CSI in 10 km bins so the 70 km break is shown directly, a beam-height/vertical-extent figure explaining why residual clutter satisfies the 480 m criterion near range, and the near-range NRP false-positive rate on overpasses that are both CPR-clear and METAR-dry. I will also quantify the complementary CloudSat effect: shallow echo seen by the ground radar at low elevation near the site but removed by the CPR clutter mask (~1 km). Results will be stratified by regime using 2B-CLDCLASS and CPR echo-top height.
2. Case examples. Individual hit, miss and false-alarm overpasses inside and beyond 70 km will be selected from the matchup dataset. For each I will show the CPR reflectivity curtain (2B-GEOPROF) with the CloudSat and NRP flags plotted along the track against range from King City, the co-timed METAR report where a station is nearby (Buttonville, 19.5 km; Toronto, 32 km), and, where available, a NEXRAD (KBUF/KCLE/KDTX) Level II cross section as an independent ground-radar view of the same overpass. The near-range false-alarm case will show NRP flagging precipitation where CPR and the surface report see none; the near-range miss will show shallow CPR echo truncated by the clutter mask.
3. Difference from Hudak et al. (2008). I will isolate the three differences: season (statistics restricted to Sept–Apr), range window (the 70 km cut already raises POD 55→67% and CSI 49→56%), and manual vs. automated quality control (quantified via the clear-sky false-positive rate). Winter stratiform cases will be shown alongside summer convective and shallow cases from Comment 2.
4. METAR comparison. I will screen the 22 CloudSat-yes/METAR-no cases for virga using the CPR low-level reflectivity gradient and ECMWF-AUX sub-cloud relative humidity, reporting FAR with and without probable virga, and noting station–footprint separation.
5. Temporal variability. I will report POD, FAR, CSI and bias by year and by season, with and without the 70 km cut.
All proposed analyses use the existing matchup dataset together with publicly available CloudSat, NEXRAD and METAR data.
Citation: https://doi.org/10.5194/egusphere-2026-1753-AC1
-
AC1: 'Reply on RC1', Farrukh Chishtie, 14 Sep 2026
reply
-
RC2: 'Comment on egusphere-2026-1753', Anonymous Referee #2, 19 Oct 2026
reply
This paper describes validation of precipitation detection of CloudSAT CPR using King City C-band radar.
Major revision is needed for acceptance of this paper.
1.
This paper only analyzes precipitation detectability by both ground radar and spaceborne CPR. For spaceborne radar validation, measurement accuracy of rain rate or snowfall rate should be discussed. Precipitation detectability analysis is only the first step, and further analysis results are required for publishing this paper.
2.
Range dependency of POD revealed in this paper is just the feature of King City C-band radar and NRP algorithm. These features of precipitation detectability are not general features for the other radars. This paper lacks generality and usefulness for other radar users.
3.
POD validation using METAR includes stations over 100 km from CloudSAT overpass.
For small scale precipitation system, it seems too large distance to validate.
4.
Ground clutter contamination near the surface bin of CloudSAT CPR is well-known feature, and precipitation detection using old version of data processing seems doubtful. I think the new revision of the CloudSAT 2C-PRECIP-COLUMN product is improved for precipitation detection.
5.
In caption of Figure1, 70, 120, 180 km circles from King City radar are mentioned, but I cannot find those circles. There is no explanation on EarthCARE nominal track (blue and green lines). METAR validation sites names in this figure are difficult to read.
In caption of Figure2, 70, 120 km vertical are mentioned, but they are not drawn.
6.
Title “Radar range dependent validation of spaceborne cloud profiling radar precipitation detection: lessons from CloudSat and the
Canadian C band network for the EarthCARE era” should be modified.
General readers who check this title will expect that there are some validation results for EarthCARE CPR data in this paper, but none of data analysis using EarthCARE CPR are shown. As to reduce misleading, you should omit EarthCARE words from the title at least.
Citation: https://doi.org/10.5194/egusphere-2026-1753-RC2 -
AC2: 'Reply on RC2', Farrukh Chishtie, 20 Oct 2026
reply
I thank the referee for the comments. Before responding point by point I note that the manuscript is, by design, a validation of precipitation detection, the same quantity validated by Hudak et al. (2008) in the same configuration, and is submitted to the special issue on early EarthCARE results as a methodological contribution on how ground-based radar references should be used for spaceborne CPR validation. Comments 1 and 2 largely ask for a different paper; I respond to them on their merits but have not changed the scope.
1. Rain/snow rate accuracy should be discussed. I disagree that detection analysis is insufficient for publication. Detection is the logically prior quantity: rate validation statistics are conditional on the detection decision, and a reference contaminated at near range (the central result here) would corrupt any rate comparison built on it. Hudak et al. (2008) was published in JGR on detection alone. Rate validation of W-band CPR against a C-band ground radar beyond 70 km is a separate problem: it requires vertical-profile-of-reflectivity correction to bring the 1–2 km ground-radar beam to the surface, treatment of W-band attenuation in the CPR rate products (2C-RAIN-PROFILE, 2C-SNOW-PROFILE), and matching of C-band and W-band reflectivity–rate relations, each with error budgets larger than the detection question addressed here. I will add a paragraph in the Discussion stating these requirements and why they were deliberately excluded.
2. The range dependence is specific to King City and NRP; the paper lacks generality. The 70 km value is specific; the result is not. King City was one of 30 identically specified C-band radars in the Canadian network, and NRP was the operational precipitation product, so the finding applies directly to the network used in the only prior statistical validation of CloudSat detection. More importantly, the methodological point is general: any validation of a spaceborne CPR against a scanning ground radar must first characterise the range dependence of the reference’s own false-positive and miss rates, because skill scores integrated over range mix the satellite’s performance with the reference’s. The revision will (a) present the range-resolved skill scores and the clear-sky false-positive rate as a reusable procedure, (b) relate the threshold to beam height, clutter-filter range, and the vertical-extent criterion so that users of other radars can derive their own, and (c) generalise the framing in the Abstract and Conclusions accordingly. This is precisely the procedure EarthCARE validation teams working with ground radar networks will need.
3. METAR stations more than 100 km from the overpass. The distances in Table 1 are from the King City radar, not from the CloudSat ground track, and are listed to show that the METAR comparison is independent of radar range. Matching of METAR reports to CloudSat was done by distance from the footprint to the station and by time offset from the overpass [state the criteria used, e.g. ≤ X km and ≤ Y min]. I will make this explicit in Section 2.3 and, to address the concern about small-scale systems, add a distance-stratified version of the METAR contingency table so the sensitivity to the matching radius is visible.
4. CPR near-surface clutter is well known; old product version; 2C-PRECIP-COLUMN is improved. The referee’s comment does not bear on the paper’s result. The range dependence reported here is a property of the ground radar reference, not of the CPR: the NRP precipitation frequency rises from ~20% beyond 70 km to 99.5% within 10 km of the radar while the CloudSat frequency is flat with range. No change of CloudSat product version can alter that signature, because the satellite detection is not what varies with range. The CPR near-surface clutter mask is treated in the paper as a known, fixed property, and it is referenced as such.
The detection algorithm used is the one validated by Hudak et al. (2008), applied to the same radar and the same precipitation product, and retained deliberately so that the 2006–2010 statistics are directly comparable to that benchmark. Substituting a later product would break that comparability without addressing the reference problem. If a newer CPR product detects precipitation more accurately, the effect would be to make the near-range disagreement attributable to NRP more, not less, evident, since the CloudSat side of the contingency would improve while the reference contamination would not. I will state the product version and release explicitly in Section 2.1, and add a sentence noting that the current 2C-PRECIP-COLUMN release still bases its decision on the lowest clutter-free bin (~1 km AGL), so the structural inability of any CPR to see shallow precipitation visible to a nearby ground radar is unchanged.
5. Figure 1 and Figure 2 captions. Agreed. The 70, 120 and 180 km range rings and the station labels in Figure 1, the EarthCARE track description, and the 70 and 120 km vertical lines in Figure 2 were lost in the final rendering. All will be restored and the captions checked against the figures.
6. Title. I accept that the current title could lead readers to expect EarthCARE data analysis. The revised title will be: “Radar range dependent validation of spaceborne cloud profiling radar precipitation detection: lessons from CloudSat and the Canadian C-band network.” The relevance to EarthCARE, which is the reason for submission to this special issue, will be stated in the Abstract and developed in the Discussion rather than in the title.
Citation: https://doi.org/10.5194/egusphere-2026-1753-AC2
-
AC2: 'Reply on RC2', Farrukh Chishtie, 20 Oct 2026
reply
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 348 | 104 | 41 | 493 | 50 | 43 |
- HTML: 348
- PDF: 104
- XML: 41
- Total: 493
- BibTeX: 50
- EndNote: 43
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents an interesting investigation of radar-range-dependent validation of spaceborne precipitation detection using CloudSat CPR observations and a ground-based C-band radar. I appreciate that the author use a substantially larger dataset than previous studies, which provides a useful statistical basis for the analysis. The topic is also timely because the results may provide useful guidance for the validation of EarthCARE CPR observations.
However, I currently have difficulty fully understanding the validation methodology and, particularly, the physical and observational reasons for the large disagreement between CloudSat and the ground radar within approximately 70 km of the radar site. Although the manuscript presents extensive statistical results, I do not think the physical interpretation of these results is sufficiently demonstrated. In particular, the proposed explanation for the 70-km threshold would be much more convincing if it were supported by actual radar examples rather than primarily by statistical evidence. I therefore recommend that the author conduct additional analyses to clarify these issues.
Major Comments
1. Definition of precipitation in the ground-radar observations
It is not sufficiently clear to me how precipitation occurrence is defined from the ground-based C-band radar observations. The manuscript explains the NRP criteria, including the minimum vertical echo extent, but the actual radar sampling and precipitation-detection procedure need to be described more clearly. For example, is precipitation determined from radar reflectivity at a particular altitude using a CAPPI product (e.g., around 2 km), from the lowest available elevation scan, or from the three-dimensional radar volume? What altitude or vertical range of the radar echo is effectively used to determine precipitation occurrence?
This issue is particularly important because the validation results may depend strongly on the type and vertical structure of the precipitation system. For example, a large frontal precipitation system with deep and vertically continuous radar echoes may produce very different validation statistics from shallow clouds producing weak precipitation near the surface. I therefore suggest separating the analysis according to precipitation regime, such as widespread/frontal precipitation, convective precipitation, and shallow/light precipitation, if possible.
More importantly, I suspect that the apparent 70-km threshold may depend strongly on how precipitation is defined from the ground radar observations. The manuscript attributes the poor agreement within 70 km mainly to ground-clutter contamination, but the relationship between the radar sampling geometry, the precipitation-detection algorithm, and the resulting 70-km threshold needs to be demonstrated more clearly.
2. Examples of successful and unsuccessful CloudSat–C-band radar comparisons
The manuscript provides extensive statistics such as POD, FAR, and CSI, but it would be very helpful to show several actual examples of cases in which CloudSat and the C-band radar agree and disagree.
For example, the author could present representative vertical cross sections or PPI/CAPPI images for (1) a hit, (2) a miss, and (3) a false-alarm case, together with the corresponding CloudSat CPR profile. Examples from both inside and outside the 70-km range would be particularly useful.
At present, it is difficult for the reader to understand exactly what kind of radar echoes are responsible for the disagreement. If ground clutter within 70 km is indeed the dominant cause, representative radar observations should be able to demonstrate how the clutter passes the precipitation-detection criteria. Such examples would substantially strengthen the physical interpretation of the statistical results.
3. Large difference from Hudak et al. (2008)
The validation results in this study differ substantially from those reported by Hudak et al. (2008). The manuscript discusses several possible reasons, including differences in sampling period, precipitation characteristics, and near-range clutter contamination. However, I do not think the current discussion fully explains the magnitude of the difference.
As mentioned above, representative case studies would be particularly valuable here. For example, the author could show cases representative of the winter precipitation conditions considered by Hudak et al. (2008) and compare them with summer convective or shallow precipitation cases included in the present dataset. This would help demonstrate why the POD and CSI obtained in the present study are substantially lower than those reported by Hudak et al. (2008), rather than relying mainly on a qualitative explanation.
4. Interpretation of the comparison with METAR observations
Some additional care is needed when interpreting the comparison with surface METAR precipitation observations. A precipitation echo may be detected by CloudSat or the ground radar aloft but may evaporate before reaching the surface, particularly under a dry sub-cloud layer. In such a case, the satellite radar may correctly detect precipitation or virga aloft while the METAR station reports no precipitation at the surface.
Therefore, some of the apparent false alarms in the CloudSat–METAR comparison may not necessarily represent errors in the CloudSat precipitation detection. I suggest examining the vertical radar-reflectivity structure for these cases and, if possible, the thermodynamic environment below the precipitation echo, particularly relative humidity. This would help distinguish genuine CloudSat false alarms from precipitation that evaporates before reaching the surface.
5. Temporal variability of the validation statistics
Since this study uses a relatively long observational dataset, it would also be interesting to examine how the CloudSat–C-band radar validation statistics vary with time rather than presenting only statistics integrated over the entire analysis period.
For example, the author could examine seasonal or annual variations in POD, FAR, and CSI. Such an analysis could help determine whether the validation performance is stable or whether it changes systematically with precipitation regime and season. This would also provide additional evidence for the proposed explanation of the difference between the present results and the winter-only results of Hudak et al. (2008).