the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Aug 2026
LESSONS report: Fine-scale representation of consecutive dry days
Abstract. The number of consecutive dry days (CDD) is a widely used metric for assessing meteorological drought. This study investigates spatial CDD patterns over Germany during the drought year of 2018, comparing a convection-permitting model (CPM) at 3 km resolution with ERA5 reanalysis at 31 km, gauge-based interpolated observations at 1 km, and radar measurements at 1 km. The analysis reveals a surprise: high-resolution CPM and radar data produce strikingly heterogeneous, “brushstroke-like” spatial structures that differ fundamentally from the smooth fields of reanalysis and gridded observations.
Our report confirms that these sharp spatial discontinuities are not computational artifacts but a physical consequence of the sensitivity to the fixed daily dryness threshold of the CDD metric. We find that the CPM can well reproduce these fine-scale patterns that are also detected by radar measurements, which is quantified by a spatial autocorrelogram and the radially averaged power spectral density.
While averaging CDD over multiple years smooths out the brushstroke patterns, such temporal averaging does not resolve the underlying shortcoming: multi-year CDD is built upon individual annual CDD values that are, as shown here, highly variable and strongly dataset-dependent. A statistic whose single-year realizations are dominated by threshold-crossing noise provides an unstable foundation for climatological analysis regardless of the aggregation period applied. As a lesson learnt, we recommend caution when applying CDD in any context — including drought monitoring, impact assessment, and climate model evaluation. This study reports on the surprising fine-scale spatial patterns and resulting shortcomings of the CDD metric and is therefore submitted as a LESSONS report, a paper category dedicated to documenting Limitations, Errors, Surprises, Shortcomings, and Opportunities for New Science.
- Preprint
(1519 KB) - Metadata XML
-
Supplement
(494 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-4014', Jonathan Wille, 03 Sep 2026
-
CC1: 'Comment on egusphere-2026-4014', Sivarajah Mylevaganam, 15 Sep 2026
Natural disasters have been occurring since the inception of this planet, manifesting in various ways. Among these, drought stands out as a unique phenomenon, demonstrating to the Earth's inhabitants how the absence of a single drop of water can devastate humanity. Its widespread implications span multiple domains, ranging from critical dehydration in human biology to severe disruptions in global agricultural food production.
Drought can be subdivided into categories based on the specific field or objective of interest. For example, it is commonly classified as surface water drought, groundwater drought, or meteorological drought. Although these subdivisions share the root term "drought," their operational contexts and foundational definitions differ significantly based on how they are perceived and reported.
In this research, the authors focus specifically on meteorological drought, which they define based on the ETCCDI (the maximum number of consecutive dry days with precipitation below 1 mm within one year) framework and characterize as the absence of rainfall. Although this constraining threshold of 1 mm is not fully aligned with the measurement precision of modern instruments like rain gauges and other advanced precipitation measurement systems, the authors guide the reader through four distinct pathways to comprehensively investigate meteorological drought in Germany. These pathways utilize various datasets across different spatial resolutions. Finally, the authors draw multiple conclusions based on the 2018 rainfall events in Germany, as they consider this specific year to be the best reflection of severe drought conditions across Europe.
Based on the current version of this research article, the following points are brought forward for consideration:
1. Lines 26–27: The authors state that meteorological drought—defined here as the absence of rainfall—can induce a cascading effect leading to agricultural, surface water, and groundwater droughts. While the authors cite a manuscript to support this claim, the definition provided contradicts the consensus on meteorological drought in current literature. A clear distinction between "rainfall" and "precipitation" must be established to ground the definition used in the current version of the manuscript.
2. The study's analysis relies on the ERA5 reanalysis, convection-permitting model (CPM) simulation, the radar-based precipitation reanalysis RADKLIM, and HYRAS station-based gridded rain gauge measurements [Lines 46-50]. If one considers radar observations, for instance, these datasets capture total atmospheric precipitation rather than liquid rainfall exclusively—a fact correctly indicated by the dataset names themselves (e.g., RADKLIM precipitation reanalysis). Therefore, attempting to align a strict "absence of rainfall" definition[Line 26] with datasets that measure total precipitation suggests a fundamental oversight. The authors must address these hydrological nuances rather than applying these terms interchangeably without considering their underlying technical definitions.
3.Moreover, given that the authors cite a specific manuscript to justify this unusual definition of meteorological drought, it becomes the responsibility of the handling editor to review the cited work [Line 27]. It is critical to verify whether that referenced study actually supports the authors' statement or if it has been misinterpreted in the current manuscript.
4. Lines 46–47: The authors state that the convection-permitting simulations of COSMO-CLM are driven by the boundary conditions of ERA5 at a 3 km resolution. However, based on the current version of the manuscript, ERA5 is natively produced at a spatial resolution of approximately 31 km, whereas COSMO-CLM operates at 3 km[Lines 46-47]. It is currently unclear how the 31 km ERA5 data was processed or downscaled to serve as boundary conditions for the 3 km CLM simulation.
Given that COSMO-CLM is one of the four core datasets used to investigate meteorological drought in Germany, the authors must provide a few clarifying lines to explicitly explain the nesting or downscaling procedure implemented during the production of this dataset. Furthermore, this methodological clarification should be contextualized alongside lines 31–33, noting what has been widely reported in the literature: that ETCCDI extreme climate indices are largely smoothed out or lost at coarser resolutions.
5.In the text, the authors state that CDD values span a wide range of 7 to 100 days across the datasets (Line 68). However, the colorbar scale reported in Figure 1 is limited to a maximum value of 60. This artificial constraint conceals the exact geographic locations where the highest CDD values (above 60 days) occur across the datasets. Therefore, the scale in Figure 1 should be adjusted up to 100 days to align with the text statement in Line 68 and properly reflect the full spatial variability of extreme values.
6. A closer examination of the subplots in Figure 1 reveals stark discrepancies in the areas just outside the primary domain of interest (the digitized polygon boundary), particularly toward the Northwest (NW) region. Because the colorbar scale is currently capped at 60, it is difficult to trace the exact numerical variation in these boundary grid cells. However, the coarser-resolution datasets—ERA5 (31 km) and COSMO-CLM (3 km)—appear to exhibit exceptionally high values in this zone, likely reaching or exceeding 60 days. In sharp contrast, the finer-resolution datasets, HYRAS (1 km) and RADKLIM (1 km), show virtually zero CDD in the exact same peripheral region, appearing completely white. The authors need to address what physical mechanism, data processing step, or domain-masking artifact has introduced this pronounced, distinct variation among the datasets in the NW boundary region.
Citation: https://doi.org/10.5194/egusphere-2026-4014-CC1 -
CC2: 'Comment on egusphere-2026-4014', Sivarajah Mylevaganam, 15 Sep 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-4014/egusphere-2026-4014-CC2-supplement.pdf
-
RC2: 'Comment on egusphere-2026-4014', Eleonora Dallan, 22 Sep 2026
The paper presents an analysis of the spatial structure of a threshold-based drought metric, the number of consecutive dry days (CDD), for four different precipitation products. While the 31-km resolution ERA5 shows limited spatial variability, the three km-scale products show substantially higher variability at short distances. In particular, the authors show how the surprising “brush-stroke” pattern emerging from the CPM and radar product is related to the threshold-based definition of CDD rather than to artifacts in these products. As a main lesson, they suggest caution when using this type of metric for drought assessment.
Based on my reading, the paper is clearly written and presents useful insights into the use of CDD (and similar threshold-based climate indices). I find it suitable for publication as LESSONS report after addressing the following minor comments.
- Figure 2: The radar product also shows “brush-stroke” patterns. Is this due to the same mechanism illustrated for the CPM in Figure 2? Adding a similar example based on two radar grid points could help clarify this point.
- Lines 114–116: You mention here the possible underestimation of CDD by ERA5. Since “evaluation of the CDD magnitudes is not the focus of the study” (line 74), perhaps the lesson learned could focus more specifically on the spatial structure and the implications of the threshold-based definition. The same applies to lines 128–130. Alternatively, if the magnitude of CDD is relevant to the discussion, you could add some consideration of the average CDD over the domain when presenting Figure 1.
- Lines 117–119: Could you mention the typical inter-station distance? The steep gradients also depend on how close the stations are to each other when their CDD values “differ substantially”.
- Line 133: On what basis do you conclude that the CPM reproduces the fine-scale spatial patterns of CDD? Similarly, at line 143, on what basis do you conclude that the station-based product misses them? In other words, what is considered the reference or “truth” for these fine-scale patterns?
Citation: https://doi.org/10.5194/egusphere-2026-4014-RC2 -
RC3: 'Comment on egusphere-2026-4014', Marc Lemus-Canovas, 22 Sep 2026
General comment
This manuscript investigates the fine-scale spatial representation of CDD using four precipitation datasets over Germany. The authors show that CDD exhibits pronounced fine-scale spatial variability in the convection-permitting simulation and radar product, with sharp spatial gradients that arise from small differences in precipitation around the 1 mm/day threshold. The comparison using spatial autocorrelation and spectral analysis provides an interesting perspective on the scale dependence of CDD. In my opinion, the topic is relevant and the results potentially interesting for the interpretation of high-resolution precipitation datasets and dry-spell metrics. However, I think that two aspects of the manuscript require further consideration before the work can be considered for acceptance.
Specific comments
- The manuscript shows that COSMO-CLM and RADKLIM exhibit similar fine-scale spatial structures, whereas HYRAS and particularly ERA5 produce smoother CDD fields. The similarity between COSMO-CLM and RADKLIM is an interesting result. However, a benchmark dataset is still missing. To me, this is particularly relevant as the authors themselves highlighted that RADKLIM has uncertainties in the representation of low-intensity precipitation, including an underestimation of low-intensity precipitation events. Since CDD is defined using a threshold of 1 mm/day, these uncertainties are directly relevant to the metric investigated here.
I would suggest the authors compare CDD calculated directly from neighbouring in-situ precipitation stations with the corresponding CDD from COSMO-CLM, RADKLIM, HYRAS and ERA5 at the station locations. This would provide an independent observational benchmark and would allow the authors to assess whether the fine-scale variability identified in the CPM and radar product is actually supported by observations. This could be included as a supplementary analysis/figure to strengthen the main message of the work. - The authors state that the observed spatial variability in CDD for short distances has implications for drought monitoring, impact assessment, drought propagation and low-flow analyses,etc, but none of these implications is explicitly discussed. In particular, it remains unclear to me whether large differences in CDD at kilometre scales translate into meaningful differences in drought conditions or impacts. I encourage the authors to discuss this scale dependence more explicitly and, if possible, provide at least one example illustrating whether the fine-scale CDD variability has consequences for a relevant application.
Citation: https://doi.org/10.5194/egusphere-2026-4014-RC3 - The manuscript shows that COSMO-CLM and RADKLIM exhibit similar fine-scale spatial structures, whereas HYRAS and particularly ERA5 produce smoother CDD fields. The similarity between COSMO-CLM and RADKLIM is an interesting result. However, a benchmark dataset is still missing. To me, this is particularly relevant as the authors themselves highlighted that RADKLIM has uncertainties in the representation of low-intensity precipitation, including an underestimation of low-intensity precipitation events. Since CDD is defined using a threshold of 1 mm/day, these uncertainties are directly relevant to the metric investigated here.
Data sets
Data and Code for: LESSONS report: Fine-scale representation of consecutive dry days Benjamin Poschlod https://doi.org/10.5281/zenodo.21219566
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 250 | 101 | 88 | 439 | 48 | 87 | 82 |
- HTML: 250
- PDF: 101
- XML: 88
- Total: 439
- Supplement: 48
- BibTeX: 87
- EndNote: 82
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments
This report presents a case for taking caution when using the longest number of yearly consecutive dry days (CDD) for analyzing drought behavior. Using a combination of radar and station rainfall observations and comparing them with ERA5 and kilometer-scale regional climate models, the authors make a convincing argument that the interpretation of CDD becomes different when measured at a higher resolution.
This discussion does open an important conversation about what CDD is actually measuring when used in hydro-climate studies. My opinion is that CDD has traditionally been used in studies using coarser-grid climate models where simulated precipitation occurrence is stretched across a larger grid cell, thus reducing spatial variability and leading to the drizzle bias as this study noted. This artificially reduces the noise in CDD, making it easier to analyze, and draw conclusions about broad precipitation patterns over a large geographic area. The analysis in Petrova et al. (2024) is a good example of using CDD at the CMIP GCM scale to identify broad regions where dry spells are projected to change in the future.
Kilometer-scale models are essentially simulating precipitation in a scale that feels more realistic to the average person on the ground. As this study highlights, kilometer-scale models can simulate rainfall patterns similar to radar measurements which is a tool we use to often perceive weather at the local scale. For example, imagine during a dry summer, you can watch a convective storm bring heavy rainfall to one half of your city while missing your home in the other half of the city, thus leaving your garden more yellow and parched compared to those who saw the rain. A coarse climate model would likely simulate that a light rain happened over the whole city and that the dry spell ended over your home while a kilometer-scale model would simulate that the dry spell continues for half the city. For you and your dry garden, the continued dry spell more accurately describes your situation.
The point of this long and confusing example is that CDD at a higher resolution is measuring something different than CDD at a lower resolution. At a higher resolution, CDD may become less reliable to convey average precipitation patterns over a particular location, but it does become better at expressing the spatial variability of precipitation in a larger region. Especially as a warmer climate is expected to make precipitation more spatially concentrated and more convective, studying the future change in CDD with a kilometer-scale model can be useful for expressing how spatial distribution of precipitation becomes more scattered in the future. A higher variability in CDD across a region conveys that precipitation becomes more scattered or stochastic which has real consequences for agriculture as rainfall becomes less reliable.
Thus, the current text states to recommend caution when using CDD for drought evaluation, however I feel the analysis is pointing towards a different conclusion that is centered around the resolution at which CDD is being analyzed. The lesson could be refined to say that the intention of using CDD depends on the context and deserves caution depending on which resolution is being used and which precipitation characteristic is being analyzed.
Overall, I feel this report is well written and the analysis is robust and timely. I would support publication after a revision of the main lesson and the following comments are addressed
Specific comments
Minor comments
Line 57: Please briefly define what Moran’s I is for the readers who are not familiar with this term.
Line 139: You can bolster your argument here by saying that other threshold dependent climate indices, like days above a certain temperature threshold, have a higher spatial autocorrelation and are less stochastic, even at a higher resolution. Thus, CDD is uniquely problematic given its high spatial variability when examined at a high resolution.