We sincerely thank you for your thoughtful and constructive comments on our manuscript. We consider them fair, direct, and very helpful for strengthening the paper. Below, we address each point in turn.

Q1 - low-accuracy years and sub-basins:

We recognize that including years and sub-basins with weak validation can bias the analyses. Accordingly, we will recompute all temperature-, precipitation-, canopy-cover-, and canopy-height–ET relationships with 2009–2011, 2018, and 2022, as well as sub-basins 1–3, excluded, and will revise Figure 7 so that the updated results appear alongside the full data set for reference. The discussion will be expanded to explain the poor performance in 2009 and 2010: although basin-average precipitation was close to the long-term mean, rainfall was strongly concentrated in the northern catchments and deficient in the south, creating spatial inconsistencies between the water-balance validation and the MOD16 ET data. We will also clarify why model performance is lower in sub-basins 1–4; these catchments contain a higher density of small wetlands and ponds, and, at MODIS's 500 m resolution, many pixels remain mixed even after wetland and open-water masks are applied, leading to systematic overestimation of ET.

Q2 - Figure 7 Modification:

We agree and will modify the figure to a 2 × 2 panel, incorporating the additional changes noted above.

Q3 - Figure references in discussion:

Thank you for pointing this out. We will update the figure references accordingly

Q4 - Vegetation differences, inclusion of GAM analysis:

Regarding the non-linear analysis, we will implement the GAM as suggested. Would you find it more informative to incorporate the resulting analysis directly into Figure 7—perhaps as an additional panel overlaying the curve on the scatterplot—or would you prefer to keep Figure 7 in its current form and place the GAM diagnostics in the Supplement? Because our temperature–ET relationship is derived from annual aggregates, the frequent extreme summer temperatures in east-central Texas can dominate the annual mean and suppress total-year ET through moisture limitation and stomatal closure; we suspect this aggregation effect underlies the negative annual slope you noted. Would a brief clarification of this mechanism strengthen Section 4.3? Finally, we propose to add text outlining the contrasting thermal and stomatal strategies of the dominant evergreen (Pinus taeda) and deciduous oaks (Quercus stellata, Q. marilandica)—evergreens limiting transpiration above ~32 °C and deciduous species above ~35 °C—drawing on Novick et al., 2016 and Oren et al., 1999.



Thank you again for your helpful feedback.

We appreciate the reviewer’s thorough and positive assessment of our manuscript. We are encouraged that the study’s objectives, methods, and relevance to woody-plant encroachment were clear and that the work is considered a valuable contribution within the scope of HESS. We will carefully incorporate the recommendation to provide a more critical discussion of MODIS ET performance in water-limited savannas, using this context to better explain our results. Thank you again for the thoughtful feedback; it will help us strengthen the paper. Below we address each specific comment and suggestion:

Introduction

L27: I suggest to use another acronym for temperature. T is often used to refer to transpiration in the ET modeling community and I also suggest to be more specific throughout the text and refer to air temperature (Ta) rather than just temperature, which could be confused with land surface temperature (LST)

Response: We will revise the manuscript and change the temperature acronym to (Ta), signifying we are referring to air temperature and not land surface temperature or transpiration.

L38-39: ‘[…] observed ET decreases of 31.9 mm and 110 mm, respectively’. I assumed this is at annual scale? If so, add mm/year.

Response: This is correct. We will add (mm/year).



L71: I don’t think it is correct to use the term ‘validate’ when you only compare the MOD16 product with water balance method. This is more of a comparison rather than any kind of validation since no observed benchmark values are used, since the water balance method is subjected to uncertainties in the precipitation/runoff products and assumptions made about other processes at annual scales (groundwater recharge, storage etc).

Response:  We agree that “validation” should be reserved for comparisons against independent ground-truth data. Because the annual water-balance estimate carries its own uncertainties (e.g., precipitation bias, runoff parameterization, groundwater storage), it cannot serve as a definitive benchmark. We will therefore replace “validation” with “evaluation” throughout the manuscript, and we will explicitly frame the analysis as a comparative performance assessment rather than a formal validation.

Materials and methods

Figure 1. Please add the data sources for each of variables (precipitation, air temperature and canopy cover). Although it is specified in section 2.2, figure captions should be interpretable as much as possible without refereeing to the text. Also please add details about the air temperature as similarly done with mean annual precipitation. It is annual daytime average? Also mean annual precipitation is calculated using which years?

Response: Thank you for the observation. We will revise every relevant figure caption so that it cites the data source and time period for each variable: precipitation and air temperature will come from the Daymet V4 product, and canopy cover will come from the Rangeland Analysis Platform (RAP v4). The captions will state that mean annual temperature (MAT) is the multi-year average of the daily mean air temperature, where the daily mean is calculated as (Tmax + Tmin)/2, and that mean annual precipitation (MAP) is the multi-year average of the summed daily precipitation totals. In each case the averaging window will be 2008 – 2023, matching the Daymet record used in the study, so the captions can be interpreted without referring back to the main text.

2.3.1 MOD16 ET validation: Again, please consider changing the sub-title since a validation is not actually done. At most, it can be considered an ‘evaluation’ or ‘benchmarking’.

Response: We agree and will revise the subsection title to “Evaluation,” as this term more accurately reflects the analysis we performed.

Figure 3. Add units in table in the Total Area column.

Response: We will add the units in the Total Area Column

Results

Figure 7. I suggest to also add the R2 aggregating for all precipitation regimes along with separating them for each (600-800, 800-1000, 1000-1200, >1200nm) as the authors did. This might better depict the general tendencies and contrast better if different precipitation regimes (or eco-regions) show different relationships with each of the variables assessed. For example, the relationship between canopy height and ET has very similar slopes and R2 for all precipitation regimes, while the other variables show large differences.

Response: We agree and will provide the aggregated equation and R² for each independent variable.

L294: ‘[…] when ET/P exceeded 100%’. This directly contradicts the earlier statement when the authors say that ET/P ranged between 70 and 100%.

Response: Thank you for catching that. We will revise the sentence to: “ET/P was 70–100 % in most years, but rose above 100 % in 2011 and 2022.”

Discussion

4.1 MOD16 accuracy: I suggest to mention and discuss the possible uncertainties and limitations of using the water balance method as a benchmark in this section.

Response: Thank you for this suggestion. In the revised Discussion we will add a brief paragraph outlining key sources of uncertainty in the water-balance benchmark. First, the gridded precipitation field is produced by interpolating and extrapolating gauge data, so spatial gaps, gauge under-catch, and the kriging procedure can bias annual totals. Second, although the runoff values come from USGS gages, no single gaged sub-basin lies entirely within the Post Oak Savannah; lateral inflows and outflows across ecoregion boundaries introduce additional error. Third, the method assumes that long-term changes in soil- and groundwater storage are negligible, yet seasonal drought–recharge cycles can shift storage by several centimeters and propagate directly into the ET estimate. Finally, small reservoirs and irrigation withdrawals are not removed from the streamflow record, potentially inflating inferred ET during dry years.

4.1 MOD16 accuracy: many studies have shown that MOD16 does not perform well in arid /semi-arid ecosystems, mostly since the model does not properly capture plant water stress, especially stress related to soil moisture deficit since the MOD16 product models stomatal conductance solely based on meteorological data. Here are some studies, in case it could be relevant to contextualize better the MOD16 evaluation done in this study:

Response: Thank you for highlighting these important studies on MOD16 performance in arid and semi-arid systems. We will incorporate a concise paragraph in Section 4.1 that summarizes how MOD16 tends to overestimate ET under strong soil-moisture stress because it relies on meteorological rather than soil-based controls on stomatal conductance. We will cite Hu et al. (2015), Miralles et al. (2016), and Majozi et al. (2017) and explain how their findings help interpret the occasional overestimation we observe in our savanna study area. This addition will better contextualize our evaluation and underscore the limitations of MOD16 in water-limited ecosystems.

L375-376: How come the authors didn’t relate LAI with ET? This may better capture phenological differences and is more related to how much radiation is intercepted to transpire/photosynthesis than canopy cover.

Response: Thank you for the suggestion. We selected canopy cover instead of LAI because the RAP fractional-cover product offers higher resolution and fewer data gaps than the available MODIS LAI in our study area. Canopy cover is also the metric most familiar to local land managers facing rapid woody plant encroachment, so using it keeps the results practical and easy to communicate. Finally, canopy cover complements canopy height by adding a horizontal dimension to the vertical information already analyzed.

L425: canopy height is also an important indicator of surface roughness which can influence the aerodynamic resistance to water transport from surface to atmosphere. Higher canopy height may enhance turbulent conditions and promote transpiration.

Response: Excellent point. We will make sure to incorporate this into the manuscript.

L436: I suspect the ET/P ratios above 100% may be also due to model uncertainties in the MOD16 product. Indeed, as mentioned in previous comment, MOD16 does not capture very well plant water stress, which likely would have been very high in those severe drought years leading to an overestimated ET and , potentially, higher values than P.

Response: We agree that the >100 % ET/P values in 2011 and 2022 may reflect MOD16’s tendency to overestimate ET during severe drought, when the algorithm does not fully represent plant water stress. In the revised Discussion we will explicitly note this point, cite the studies you referenced on MOD16 performance in water-limited systems, and acknowledge that this overestimation could contribute to ET/P exceeding 1.0.

L465: the authors mention that the future direction should be to use higher spatial resolution ET products to capture fine hydrological processes but I would rather suggest to explore other ET products particularly those based on Land Surface Temperature (LST) from thermal infrared (TIR) remote sensing, which have been shown to better capture plant water stress which is an important issue in water limited savanna ecosystems. See these studies:

Response: We will broaden the “Future work” paragraph to emphasize not only higher spatial-resolution ET products but also approaches that use thermal-infrared land-surface temperature to capture soil-moisture stress in savanna systems. We will cite the recommended studies (Guzinski et al., 2020; González-Dugo et al., 2021; Burchard-Levine et al., 2022; Anderson et al., 2024) and note that LST-based models such as TSEB and Sentinel-2/3 fusion products are promising next steps for improving ET estimates in our region.