response to Reviewer1

The authors evaluated global changes in dryland aridity using data from meteorological sites during 2003-2019, with the goal of reducing scale-related uncertainty. They obtained a comprehensive understanding of the multifaceted characteristics of these changes and identified a decoupling between atmospheric, hydrological, and vegetation aridity. The manuscript is well-written, and the results are intriguing.

Response:

Thank you for your positive comments. Thank you also for your time spent reviewing our manuscript. We will revise this manuscript based on your insightful suggestions.

Some specific points:

Method: the scale mismatch between the site observations and the gridded data used in the RF model may introduce uncertainty. The authors may want to consider addressing this uncertainty in their modeling practice, or at the very least, discuss it in Section 4.

Response: Thank you for your insightful comments.

Predictor data extractions in our study were at the 500 m scale, which can be matched to the flux observation scale of the stations used in this study. We will discuss possible uncertainties regarding the degree of match between the footprint area of the flux station and the scale of the predictor data used.

Figure 2(a): why is the performance of one site exceptionally poor with a negative correlation (Rcorr<0)?

Response: Thank you for your insightful comments.

This site is ES-AMO and the low accuracy of ET simulations at this site may be induced by its hydrogeologic characteristics and groundwater table dynamics (López-Ballesteros et al., 2017). The predictors used in this study may have difficulty explaining the contribution of groundwater dynamics and subterranean ventilation (López-Ballesteros et al., 2017) to ET at the site location. The biological crust at this site (Chamizo et al., 2016) may also control ET by influencing surface soil moisture.

In addition, since our accuracy evaluation is based on the leave-one-site-out cross-validation, the validation accuracy may be relatively low when there are no stations with similar environmental conditions in the training set. The RF model that we finally applied to the weather stations included all stations (i.e., no flux station was left), the accuracy may be improved when applied to weather stations with similar environmental conditions to this flux station.

In the Discussion section, we will elaborate on this issue and clarify possible limitations and uncertainties.

Figure 3: the Antarctic continent could be omitted from the figure to enhance the clarity of the focal information.

Response:

The Antarctic continent will be removed from the figures.

Lines 28-33 and Lines 279-283: the content provided is repetitive. Please rephrase and avoid redundancy.

Response: Thank you for your insightful comments. We will rephrase and avoid redundancy.

References

Chamizo, S., Cantón, Y., Rodríguez-Caballero, E., and Domingo, F.: Biocrusts positively affect the soil water balance in semiarid ecosystems, Ecohydrology, 9, 1208–1221, https://doi.org/10.1002/eco.1719, 2016.

López-Ballesteros, A., Serrano-Ortiz, P., Kowalski, A. S., Sánchez-Cañete, E. P., Scott, R. L., and Domingo, F.: Subterranean ventilation of allochthonous CO2 governs net CO2 exchange in a semiarid Mediterranean grassland, Agricultural and Forest Meteorology, 234–235, 115–126, https://doi.org/10.1016/j.agrformet.2016.12.021, 2017.

This is an interesting study. Combining the station observation and the Random Forest (RF) model, the authors investigated the global dryland aridity changes from 2003 to 2019. They found an evident decoupling between atmospheric, hydrological, and vegetation aridity.

Compared to the prior studies, I think this work creatively assessed global aridity with station data, instead of reanalysis dataset or numeric model outputs. The results are helpful to our understanding of the influence of climate change on the global aridity.

The results are convincing and also highlight the added values of using station scale data to assess dryland change as a complement to the results based on coarse resolution reanalysis data and land surface models.

The manuscript was technically well-written and easy to follow in logic. However, the following minor and technical revisions should be addressed.

Response:

Thank you for your positive comments. Thank you also for your time spent reviewing our manuscript. We will revise this manuscript based on your insightful suggestions.

Lines 38-39

Change to “In the context of global warming, the global dryland is expected to expand due to potential higher atmospheric water demand.”

Response: Thank you for your insightful comments.

It will be modified.

Figure 1 Please add some related references to support your classification of AI levels.

Response:

Thank you for your insightful comments. It will be added. Here, aridity is classified by the index of aridity (AI) as the average annual precipitation divided by potential evapotranspiration (Programme, 1997). Hyperarid: AI < 0.05, Arid: 0.05 < AI < 0.20, Semiarid: 0.20 < AI < 0.50, Dry sub-humid: 0.50 < AI < 0.65, and Humid: AI > 0.65.

Lines 126-128 Provide the full name of RMSE. What does the feature importance mean?

Additionally, I recommend adding a detailed introduction to your Random Forest model and leave-one-site-out cross-validation method in the Methodology Section.

Response: Thank you for your insightful comments.

Descriptions of accuracy metrics, feature importance, and random forest models are added to the manuscript to provide more details to the reader. In terms of the leave-one-out cross-validation method, we referred to previous studies (Tramontana et al., 2016; Zhang et al., 2021).

Figure 2  Could you explain the lower Rcorr value of Shrubland in Figure 2c, compared to other land use types?

Response: Thank you for your insightful comments.

The low accuracy of shrubland sites such as ES-AMO may be induced by its hydrogeologic characteristics and groundwater table dynamics (López-Ballesteros et al., 2017), and also associated limitations in the quantifying the belowground hydrological process. The predictors used in this study may have difficulty explaining the contribution of groundwater dynamics and subterranean ventilation (López-Ballesteros et al., 2017) to ET at the site location. The inclusion of cumulative soil water deficit (Giardina et al., 2023) during the growing season and regional-scale drought severity has the potential to better characterize belowground water availability. In addition, in arid shrub ecosystems, the biological crust (Chamizo et al., 2016) may also control ET by influencing surface soil moisture.

In the Discussion section, we will elaborate on this issue and clarify possible limitations and uncertainties.

Lines 136-137 and Figure 3

Why did you separate the entire study period into the two stages (i.e., 2003-2010 and 2011-2019)? Does the 2003-2009 and 2010-2019 work?

Response:

Thank you for your insightful comments. 2003-2010 and 2011-2019 have similar year lengths, which can somewhat reduce the impact of extreme years on the results and understanding of the period scale. Due to the temporal limitation of MODIS data and predictors derived from it (such as the radiation-related predictor RSDN starts from 2002), the year span of this study is not very long, thus the finding of changes in aridity should be treated with more caution. We will elaborate on the discussion.

I recommend adding a column plot in Figure 3 to summarize the changes in ET, TA, P, VPD, LAI, and P-ET.

Response:

Thank you for your suggestion. We will add column plots to show changes in these variables.

References

Chamizo, S., Cantón, Y., Rodríguez-Caballero, E., and Domingo, F.: Biocrusts positively affect the soil water balance in semiarid ecosystems, Ecohydrology, 9, 1208–1221, https://doi.org/10.1002/eco.1719, 2016.

Giardina, F., Gentine, P., Konings, A. G., Seneviratne, S. I., and Stocker, B. D.: Diagnosing evapotranspiration responses to water deficit across biomes using deep learning, New Phytologist, 240, 968–983, https://doi.org/10.1111/nph.19197, 2023.

López-Ballesteros, A., Serrano-Ortiz, P., Kowalski, A. S., Sánchez-Cañete, E. P., Scott, R. L., and Domingo, F.: Subterranean ventilation of allochthonous CO2 governs net CO2 exchange in a semiarid Mediterranean grassland, Agricultural and Forest Meteorology, 234–235, 115–126, https://doi.org/10.1016/j.agrformet.2016.12.021, 2017.

Programme, U. N. E.: World Atlas of Desertification: Second Edition, 1997.

Tramontana, G., Jung, M., Schwalm, C. R., Ichii, K., Camps-Valls, G., Ráduly, B., Reichstein, M., Arain, M. A., Cescatti, A., Kiely, G., Merbold, L., Serrano-Ortiz, P., Sickert, S., Wolf, S., and Papale, D.: Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms, Biogeosciences, 13, 4291–4313, https://doi.org/10.5194/bg-13-4291-2016, 2016.

Zhang, C., Luo, G., Hellwich, O., Chen, C., Zhang, W., Xie, M., He, H., Shi, H., and Wang, Y.: A framework for estimating actual evapotranspiration at weather stations without flux observations by combining data from MODIS and flux towers through a machine learning approach, Journal of Hydrology, 603, 127047, https://doi.org/10.1016/j.jhydrol.2021.127047, 2021.