| 11 Aug 2026
A 30-Year 1-km Daily Precipitation and Air Temperature Dataset for the Po River District (Italy)
Abstract. High-resolution daily climate data remain scarce for complex Alpine terrain, where coarse gridded products often fail to resolve orographic precipitation gradients and elevation-dependent temperature variability. Here we present a 1-km gridded daily precipitation and temperature dataset for the Po River District (Northern Italy) covering the 1991–2020 climatological period. The dataset is based on a harmonized multi-source observational network comprising 1,583 precipitation and 1,555 temperature stations after quality control. Spatial interpolation was performed using Ordinary Kriging for precipitation and Detrended Kriging for temperature, with elevation as the trend variable, and daily-adaptive semivariogram models selected from five candidate functions via Particle Swarm Optimisation. Leave-one-out cross-validation indicates strong overall performance. For precipitation, the mean Kling–Gupta Efficiency (KGE) across all station-wise leave-one-out cross-validations exceeds 0.84 under All-Days conditions and 0.82 under Wet-Days conditions, with mean absolute errors of 1.28 mm and 3.05 mm, respectively. Temperature interpolation achieves a mean KGE of 0.88 and a mean absolute error of 1.14 °C, with negligible bias. Spatial diagnostics reveal higher precipitation errors in high-relief Alpine sectors and along basin boundaries, while temperature performance remains comparatively uniform. Interpolation skill decreases with altitude, particularly for precipitation during wet events, due to decreasing station density. The resulting dataset provides spatially continuous daily climate fields suitable for hydrological modelling, climate variability assessment, and environmental analysis in one of Europe’s most topographically diverse river basins.
The paper “A 30-Year 1-km Daily Precipitation and Air Temperature Dataset for the Po River District (Italy)” by H. Salehi et al. presents new high-resolution daily fields for precipitation and temperature for the Po River District for the 30-year period 1991-2020.
The topic addressed by the paper is very important because there is strong needing of high-resolution fields of meteorological variables for many applications. However, the paper has three fundamental deficits that, in my opinion, make it not acceptable for publication. The first concerns the actual spatial resolution of the temperature and precipitation fields, the second concerns the adopted methodology that does not allow sufficient robustness to missing data, the third concerns the lack of comparisons with the best data sets available for the study area.
Actual spatial resolution of the temperature and precipitation fields
The nominal resolution of the produced dataset is 1 km, but the actual resolution of it is not clear from the paper. The number of available stations within the considered area is between 1100 and 1200, but considering that about half of the data are missing, the interpolation methods works, on average, with fewer than 600 stations. Considered that the area has an extension of 90000 km2, this means that there is about 1 station per 150 km2, which correspond to an actual resolution greater than 10 km. However, in my opinion this is an optimistic assumption because for precipitation the actual resolution is probably further reduced by the smoothing introduced by the interpolation method. For temperature, the use of elevation helps better capturing the local values, even if deriving only one lapse rate for all points and elevations is probably not the best possible choice.
The rather low actual resolution of the dataset is clearly documented by the first panel of figure 8, reporting the mean annual precipitation over the considered area. This panel shows that areas characterized by very strong spatial gradients in rainfall climatologies (for example the Aosta Valley region or the Valtellina area - see Crespi et al., 2018) present spatial distributions strongly smoothed by the interpolation methodology. Regarding this panel, I am also very surprised by the average annual rainfall indicated, which is 559.7 mm. This is about half the value suggested in the scientific literature for the Po River basin! A so low value of the average annual rainfall could not sustain the observed discharge of the Po river (about 1500 m3/s at the Pontelagoscuro station).
In my opinion, a dataset like the one proposed is of little use without a clear and documented discussion of the actual resolution of the proposed fields. I understand that addressing this topic extensively is not easy, but I believe that an in-depth analysis of the kriging weights could be very helpful in better understanding this important aspect. An analysis of this type could, for example, identify—for each variable, day, and location—the distances at which the weights decrease to 2/3, 1/2, and 1/3, and subsequently present and discuss a series of distributions to show the values obtained in different areas and—given the significant increase in available data from the early years to the later ones—across different time periods. Moreover, other analyses could concern case studies (events) for which high-resolution fields could be acquired. For precipitation these fields can e.g. be acquired from radar measurements. The variances of the radar spatial fields can then be compared with those of the files obtained by kriging and this can permit to understand how much the radar fields must be smoothed to get spatial variability comparable to those of the kriging fields. The topic of the actual resolution of the fields is not covered at all in the paper, but from i) the method used (ordinary kriging for precipitation), ii) the number of stations, iii) the high fraction of missing data and iv) some of the figures presented in the paper, I believe I can deduce that, at least for precipitation, the effective resolution is well above 10 km. In my opinion, the actual resolution of the precipitation fields in not higher than that of the ARCIS data set (Pavan et al., 2019), which has the relevant benefit of covering a longer period (it starts in 1961) and being regularly updated.
I believe it is nonetheless useful for many applications to provide datasets at a nominal resolution higher than the actual resolution; however, it is essential to address the issue of actual resolution so that users understand whether the value provided for a specific point is truly representative of that point or, instead, represents an average value for an area of hundreds of square kilometres.
Evaluating the effective resolution of the fields produced is, of course, only part of the problem. Alongside this assessment, it would be very important to identify methods to improve this resolution. This can be done by modelling the relationship between meteorological variables and the physiographic characteristics of the territory by means of the PRISM (Parameter-elevation Regressions on Independent Slopes Model) approach. In this paper this approach is used only for temperature and also for this variable it is used in a rather trivial way as the link between temperature and elevation is estimated considering all stations in the study area without taking into account that the temperature-elevation relation can have relevant spatial differences and can be different for different atmospheric layers. Moreover, it can depend on the aspect, on the slope, on the presence of valleys, etc. All this features can of course be considered also in the kriging following the elevation effect. This step depends, however, on the availability of stations. If a point is in a valley and the closest stations are on the surrounding slopes the estimation can be strongly biased in case of significant temperature inversions that could be better captured by means of more distant stations but with more similar physiographic features.
Robustness to missing data of the adopted methodology
High-resolution fields of meteorological variables are often produced by means of the anomaly method, which is based on the combination of fields of climate normals (usually defined over standard 30-year periods) and anomalies from them. The former depend only on space, whereas the latter depend on space and time. The benefit of this combination is that the definition of the climate normals and the anomalies can be performed with different methods and, when useful, also with different data sets. A well-known example of this approach is the daily Alpine Precipitation Grid Data Set (Isotta et al., 2014), which is based on fields of monthly normals from Schwarb (2000) and anomalies from them. In this case, the fields of the monthly normals were derived with a PRISM approach allowing to model (at the local level) the link between the physiographic features of the territory and the precipitation normals. The same approach to define spatial fields of monthly precipitation normals has been adopted by Crespi et al. (2018) for Italy and more recently by Manara et al. (2026) for northern Italy and the northern part of central Italy. For the Italian territory, such spatial fields of climate normals are available also for temperature (Brunetti et al., 2014). Once the spatial fields of the climate normals (e.g. over the 30-year period 1991-2020) are defined, any time series of observations can be converted into an anomaly record and these anomaly records can be interpolated by means of many methods. The most important benefit of this approach is that it can be applied also to data sets with a relevant fraction of missing data, provided that the issue of missing data is correctly managed when the station climate normals are defined. This because the anomalies have much greater spatial coherence than the corresponding absolute values, which makes the interpolation of the anomalies robust to missing data. However, the same robustness is not guaranteed when the decomposition into normal values and anomalies is not used. For example, if we consider a point in the Aosta Valley between the city of Aosta and the Great S. Bernard Pass, it is located near a very dry station (Aosta) (about 500 mm of rainfall per year) and at the same time near a very wet station (Gran San Bernardo) (over 2000 mm of rainfall per year). It is quite clear that obtaining with kriging the rainfall for that point produces a record that is more affected by the missing data at one of the two stations than by any other factor linked to climate variability and change.
Not all high-resolution fields of meteorological variables are naturally produced by means of the anomaly method. Many other methods can be used. However, if the goal of the fields is to highlight climatic variability and change, in this case the station records should not present a relevant fraction of missing data.
Using a methodology robust to missing data is, of course, only part of the problem. Alongside this assessment, it would be very important to assess the effect of missing data on the data set. This can e.g. be investigated by focus on a period of full data availability (e.g. the most recent years) and by obtaining a high number of different data sets by randomly deleting 50 % of the data. Several analyses on these data sets can then permit to estimate the uncertainty introduced by this large fraction of missing data. Another option can be to focus on selected points (e.g. the point previously discussed between Aosta and Great S. Bernard) and by trying to understand the effect of missing data on the corresponding temperature and precipitation records. The point between Aosta and Great S. Bernard will e.g. show clearly biased minimum and maximum precipitation values when Great S. Bernard and Aosta data are, respectively, missing.
Another relevant problem
In addition to the two limitations previously discussed, the paper suffers from a fundamental shortcoming: it fails to compare its results with the best datasets available for the study area. Indeed, the only datasets used for comparison are E-OBS and EMO-1arcmin, both of which lack homogeneous data coverage for the region in question. Conversely, other datasets—which are far more robust for this study area—are ignored simply because they do not provide both temperature and precipitation data. This is e.g. the case of the Alpine Precipitation Grid Data Set (Isotta et al., 2014), the 5 km resolution ARCIS precipitation data set (Pavan et al., 2019) (now this data set is available also for temperature (Pavan et al., 2026), the Manara et al. (2026) 30-arc-second resolution precipitation data set, the Crespi et al. (2018) data set of monthly precipitation normals, the Brunetti et al. (2014), data set of monthly temperature normal, etc. Although they differ in characteristics from the dataset presented in this paper (in terms of spatial resolution, time period covered, and data aggregation), these datasets represent a valuable resource for evaluating its quality. An example is again the spatial distribution of mean annual precipitation presented in the first panel of figure 8. This distribution could be compared with the corresponding distribution from Isotta et al. (2014) (the data are now updated to 2019), Manara et al. (2026), Pavan et al. (2019) (the ARCIS data are regularly updated) and also (even though the investigated period is different) with those from Crespi et al. (2018). These comparisons would be very useful for better understanding the quality of the presented dataset.
Final comment
In light of the described limitations, I believe the paper cannot be accepted for publication in ESSD.
References
Brunetti M. et al., 2014. High-resolution temperature climatology for Italy: Interpolation method intercomparison. Int. J. Climatol. 34, 1278–1296 https://doi.org/10.1002/joc.3764.
Crespi A. et al., 2018. 1961–1990 high-resolution monthly precipitation climatologies for Italy. Int. J. Climatol. 38, 878–895 https://doi.org/10.1002/joc.5217.
Isotta F.A. et al., 2014. The climate of daily precipitation in the Alps: Development and analysis of a high- resolution grid dataset from pan-Alpine rain-gauge data. Int. J. Climatol. 34, 1657–1675 https://doi.org/10.1002/joc.3794.
Manara et al., 2026. A new daily high-resolution gridded precipitation dataset for the Italian Greater Alpine Region (1951–2023): spatio-temporal analysis and hydrological applications. Journal of Hydrology: Regional Studies 67 (2026) 103776 https://doi.org/10.1016/j.ejrh.2026.103776.
Pavan V. et al., 2019. High resolution climate precipitation analysis for north-central Italy, 1961–2015. Clim. Dyn. 52, 3435–3453 https://doi.org/10.1007/s00382-018-4337-6.
Pavan V. et al., 2026. A new operational dataset of gridded minimum and maximum temperature over North-Central Italy 1991 to present. Climate Services Volume 43, August 2026, 100689 https://doi.org/10.1016/j.cliser.2026.100689.
Schwarb, 2000. The Alpine Precipitation Climate: Evaluation of a High-resolution Analysis Scheme using Comprehensive Rain-Gauge Data. ETH, Zurich.