Abstract. Forest fires in Croatia inflict substantial economic and ecological damage and frequently pose a threat to infrastructure and human lives. The southern part of the Croatian Adriatic, belonging to the Mediterranean basin, is the most severely affected region. To evaluate fire risk, the Canadian Fire Weather system was applied, and indices based on Fire Weather Index (FWI) – Seasonal Severity Rating (SSR), the number of days with FWI > 30 (FWI30), the 90-th percentile of FWI (FWIp90), and Length of Fire Season (LOFS) were derived. This study investigates the extent to which climate change has influenced the variability of latter indices across Croatia during June-September season. The analysis covers the period 1961–2020, revealing upward trends and predominantly positive anomalies in the evaluated indices. The most favourable fire weather conditions occur in the southern part of the Croatian Adriatic, which also exhibits the strongest increasing trends in SSR and FWI30. Although the continental parts of Croatia have historically been less susceptible to wildfires, the observed trends in the analysed indices suggest that conditions conducive to ignition and spread of wildfires are gradually emerging in these areas as well.
Received: 23 Dec 2025 – Discussion started: 04 Jan 2026
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
Anić et al. developed a 1-km dataset for fire-weather and related meteorological variables and indices based on station observations and digital elevation model (DEM) data. The results could be of interest to the local scientific community. My main concern is regarding the validity of the developed 1-km dataset and its added value.
Specifically, this dataset is essentially using 35 meteorological stations (24 Croatian + 11 Slovenian), 18 climatological stations, 105 rain gauges, interpolated to a 1 km grid by considering the elevation, distance from the sea, latitude, and longitude. Does the created dataset actually contain 1-km information? The network density is definitely much coarser than 1 km. On the other hand, even if information is added by elevation, this would mainly be relevant for temperature, while the relationship between elevation and precipitation/wind/cloud cover/sunshine duration/relative humidity is more complicated. It would be more convincing if the authors could add other sources of high-resolution data for climate and meteorological variables, such as satellite data. This is a common practice used for generating high-resolution observation datasets, such as for E-OBS and MSWEP.
The question of validity set aside, what added value has the developed high-resolution dataset provided? What part of the conclusion is not what we’ve already known based on other observation, satellite, reanalysis, or model datasets?
Specific comments
Lines 39-40: What are the causes of wildfire ignition in JJAS (the second peak)?
Lines 126-129: Please provide references for the definition of daily severity rating. Why is it defined this way? How are the values for the coefficient and exponent chosen? What is the physical meaning of this quantity? How is its physical meaning different from FWI?
Line 145: Is the leave-one-out cross-validation for leaving out one station or one or more years? Please provide more details on what (which stations/years) are used for model training, and what is left for validation.
Lines 296-298: Why does a larger number of stations lead to higher error? Wouldn’t more information be added? Could this larger error be simply due to the larger variability at the seasonal timescale?
Lines 332-334: The negative correlation between temperature and precipitation is merely a statistical observation. I don’t think it is sufficient to be the driver of the positive land-atmospheric feedback. Please provide physical explanations for the positive feedback mechanism.
Technical corrections
Line 278: “the FWI30 and FWIp90 trend results” → “the FWI30 trend and FWIp90 anomalies”
Line 360: “…, could potentially…” → “…, which could potentially…”
This study provides a high-resolution assessment of changes in fire weather conditions and wildfire risk in Croatia during period 1961–2020. Significant warming, increased hot extremes, and drying trends have led to rising Fire Weather Index-based indices and an extension of the fire season. Besides coastal hotspots, emerging fire-risk areas are identified in continental regions, highlighting new challenges for wildfire prevention and risk management.
This study provides a high-resolution assessment of changes in fire weather conditions and...
General comments
Anić et al. developed a 1-km dataset for fire-weather and related meteorological variables and indices based on station observations and digital elevation model (DEM) data. The results could be of interest to the local scientific community. My main concern is regarding the validity of the developed 1-km dataset and its added value.
Specifically, this dataset is essentially using 35 meteorological stations (24 Croatian + 11 Slovenian), 18 climatological stations, 105 rain gauges, interpolated to a 1 km grid by considering the elevation, distance from the sea, latitude, and longitude. Does the created dataset actually contain 1-km information? The network density is definitely much coarser than 1 km. On the other hand, even if information is added by elevation, this would mainly be relevant for temperature, while the relationship between elevation and precipitation/wind/cloud cover/sunshine duration/relative humidity is more complicated. It would be more convincing if the authors could add other sources of high-resolution data for climate and meteorological variables, such as satellite data. This is a common practice used for generating high-resolution observation datasets, such as for E-OBS and MSWEP.
The question of validity set aside, what added value has the developed high-resolution dataset provided? What part of the conclusion is not what we’ve already known based on other observation, satellite, reanalysis, or model datasets?
Specific comments
Lines 39-40: What are the causes of wildfire ignition in JJAS (the second peak)?
Lines 126-129: Please provide references for the definition of daily severity rating. Why is it defined this way? How are the values for the coefficient and exponent chosen? What is the physical meaning of this quantity? How is its physical meaning different from FWI?
Line 145: Is the leave-one-out cross-validation for leaving out one station or one or more years? Please provide more details on what (which stations/years) are used for model training, and what is left for validation.
Lines 296-298: Why does a larger number of stations lead to higher error? Wouldn’t more information be added? Could this larger error be simply due to the larger variability at the seasonal timescale?
Lines 332-334: The negative correlation between temperature and precipitation is merely a statistical observation. I don’t think it is sufficient to be the driver of the positive land-atmospheric feedback. Please provide physical explanations for the positive feedback mechanism.
Technical corrections
Line 278: “the FWI30 and FWIp90 trend results” → “the FWI30 trend and FWIp90 anomalies”
Line 360: “…, could potentially…” → “…, which could potentially…”