General comments

This paper presents a new global 1km fractional urban change dataset for the 2015-2100 period. It is the first global fractional urban change dataset that I am aware of, which should make it of high interest to the Earth and Environmental Science community. The methodology used was adapted from previous methods developed by the authors to allow for fractional (rather than binary) urban change modelling. The model validation results (e.g., RMSE = 0.08) are encouraging, although I request the authors to explain the calibration/validation procedure in more detail (see my Specific comments 4-5). The manuscript is generally well structured and readable, but could use a language and grammar check. Overall, I believe this paper is a valuable addition to the scientific literature on urban change, and that it could be publishable after further revisions and clarifications by the authors.

Specific comments

1. Page 3, line 10: “Although several global datasets of urban extent dynamic with conversions from non-urban to urban have been proposed, there is still limited effort to characterize the gradual urban fractional change (i.e., ISA) within each grid when projecting future global urban sprawl (Potere et al., 2009; Huang et al., 2021; Herold et al., 2003; Seto et al., 2012; Li et al., 2017)”.

I suggest to include more information on some of these other global urban extent datasets, e.g., their spatial resolution, the data used to calibrate/validate the model, the years for which the data is available (e.g., to 2050? 2100?). Also, you may want to note if any are not freely available for download. This additional information can help to highlight the other advantages of your dataset (aside from its mapping of fractional cover)

2. Page 4, line 15: It would be beneficial to readers if you can explain why the global artificial impervious area (GAIA) dataset was used for this model calibration and validation. For example, are there no appropriate ~1km fractional urban cover maps that could have been used for this? My concern is that it GAIA a binary urban/non-urban map that was resampled to 1km, and not a “true” fractional cover map.

3. Page 5, line 5. More information is needed on these spatial proxies, and how they were considered in the model. I suggest to add the references for each dataset used in Table 1, as well as how the spatial proxies were derived from these datasets (e.g., based on distance to the features like city centers/roads/protected areas/MODIS land cover types?).

4. Page 7, line 14: “That is, the continuous values can be divided into binary maps using different 15 thresholds to measure the agreement between threshold-derived results and the referenced urban extent. In this way, the area under the curve (AUC) is commonly used to quantitatively evaluate the performance of derived global suitability (Hosmer et al., 2013).”

Is this binary validation necessary, considering the purpose of the model was to generate fractional urban cover estimates? If so, I suggest to explain why.

5. Page 7, Section 3.2 (Calibration): What is the time period of the GAIA data used for the model calibration and validation? It’s not clear if there was an independent calibration and validation period, or if the calibration/validation were both based on the entire 1985-2015 dataset.

6. Page 8, line 1. “Given that the GAIA data were derived from satellite observations with good quality and fine resolution, we harmonized future urban growth trends (2015-2100) from LUH2 under different SSP-RCP scenarios with the derived urban areas from GAIA in 2015.”

Do the GAIA data and the LUH2 data use the same definition of “urban” land? It may be another reason for the difference between the urban area extents of the two datasets in 2015.

7. Page 13, Data availability. This data on fractional urban changes from 2015-2100 will be of much interest to researchers around the world, so I appreciate that you have made the data openly available. Considering all of the data you have generated in this study, another suggestion is that you may want to also share the development probability (Pdev) dataset, which contains the probability of urban development in each 1km grid cell(?). Using this data, readers could potentially generate their own future urban (fractional) change maps, e.g., based on national urban development/land demand scenarios.

The paper takes on the very substantial challenge of developing a global dataset of urban fractional changes at a 1km resolution from 2020 to 2100 under eight scenarios of socioeconomic pathways and climate change. The newly developed fractional urban land dataset is quite valuable and is helpful to assess the environmental impact of future urbanization. The manuscript is generally well structured, but the discussion and conclusion need to be reorganized. For example, the conclusion section is too wordy and repeats some information from the method and result sections. Though the method used in this work has been published, I still have some concerns when the model is applied to global-scale modeling. There are some missing details in the method section, and I have provided detailed comments below. Overall, I believe this paper could be publishable after major revision.

Specific comments:

1. P6, Line 1-2. “We characterized urban fractional change across different states in each country, using the long-term (1985-2015) urban extent data (i.e., GAIA) and the sigmoid growth model.”

The sigmoid growth model is fitted in the state level, and the estimate parameters (a, b, c, d) are also shown in Fig. 6. How the sigmoid growth model performs in each state, because the urbanization stage is different among developing countries and developed countries. I suggest adding a fit performance of the sigmoid growth model to prove the model reliability and explain the model uncertainty.

2. P6, Line 12-15. “We incorporated the ISA-based growth mechanism with the Logistic-Trend-CA model (He et al., 2023), which incorporates temporal contexts of urban sprawl into the neighborhood configuration….”

The sigmoid growth model is estimated at the state level, so did you also train the logistic regression model at the state level? The logistic regression is a binary regression model, but input data (GAIA) is fractional type. How did you implement the model training? I also see the methods in the supplementary materials, so only three spatial proxies (i.e., DEM, Distance to city centers, Distance to major road) are used to training the logistic regression model? And how many samples were selected to train the model in each state?

3. P7, Line 12. “We calibrated the Logistic-Trend-ISA-CA model at the state level using historical urban extent time series data (i.e., GAIA) from satellite observations”. So how did you validate the sigmoid growth model?

4. P7, Line 13-15. “First, we evaluated the performance of derived suitability surface using the receiver operating characteristic (ROC) method (Sunde et al., 2014). That is, the continuous values can be divided into binary maps using different thresholds to measure the agreement between threshold-derived results and the referenced urban extent.”

The binary urban land map was used to evaluate the performance of the derived suitability surface. How the thresholds were determined to extract the binary urban maps for each state. Are there large differences in the thresholds among states?

If you evaluate the model performance for each state, I suggest that Fig.7 (i.e., model performance at the country level) could show the AUC values at the state level.

5. P8, Line 5-6. Data harmonization should be cautious because the data source and definitions of GAIA and LUH2-urban are different. It is also simple to use equations (6) and (7) to harmonize the urban land from GAIA and LUH2. As I know, the overlap period for GAIA and LUH2 is 1985-2020. Did the harmonization rate change a lot during the overlap period? I suggest adding an uncertainty analysis for the data harmonization.

6. P8, Line 20-24. Global South countries located in middle Asia, south America, and Africa, would likely experience more noticeable urban growth than Global North countries in the future, e.g., the growth rates of the United States of America (USA) and China are 3.95 and 1.05 times in 2100 under SSP2-RCP4.5, respectively, relative to the base year of 2015. These two sentences are contradictory, you made me confused.

7. P12, Section 4.3. In this section, you mainly compared the spatial pattern of the newly developed urban fractional dataset and previous datasets. I suggest adding a comparison analysis of future urban land area between the harmonized data and other available datasets.

8. Fig. S6. I note that there will be no low-density ISA area in the city you show after 2060, and it seems that most of the urban area have the same ISA fraction. It also existed in other metropolitan areas (e.g., Fig. 10 and 11, New York city). so, I suggest to explain why.