The Post Oak Savannah ecoregion in east-central Texas covers over 55 000 km² and spans 31 counties, with its western boundary encompassing much of the Carrizo–Wilcox Aquifer (Fig. 1A). Historically, this region supported an open-canopy savanna characterized by diverse grasses and forbs interspersed with stands of post oak (Quercus stellata) and blackjack oak (Quercus marilandica) (Wasowski and Wasowski, 1988). Positioned between the East Texas Piney Woods – dominated by dense evergreen forests – and the Central Texas Blackland Prairie – characterized by black, calcareous, alkaline, clay-rich soils – this landscape functions as an ecological transition zone (Diggs et al., 1999; Schmidly, 2002).

During the study period (2008–2023), both precipitation and air temperature displayed pronounced spatial variability, with annual precipitation ranging from approximately 1400 mm in the northeastern portion of the ecoregion to about 600 mm in the southwest (Fig. 1b). Mean annual air temperature exhibited a similar gradient, decreasing from roughly 22 °C in the northeast to 17 °C in the southwest (Fig. 1d). Canopy cover over this interval showed substantial fragmentation, with densely wooded stands interspersed among open, grass-dominated areas (Fig. 1e). In contrast, the southern portion exhibited a more continuous mosaic of cover types, with less abrupt transitions between wooded and non-wooded patches (Fig. 1e).

Consistent with the four objectives of this study, we (1) validate the MOD16 ET product against water-balance estimates (WBET) at the subbasin (HUC8) scale; (2) analyze monthly and seasonal ET differences as they relate to canopy cover; (3) use linear regression to examine the coupling and decoupling of woody plant metrics (canopy cover and canopy height), climatic factors (precipitation and air temperature), and ET within different precipitation zones; and (4) evaluate excess water (P–ET) at the water-year scale – including an ET / P analysis – to assess broader trends in water availability over the study period (Fig. 2).

Comparisons of the MOD16 product with WBET estimates yielded an R2 of 0.65, a bias of -7 mm wyr⁻¹ (-0.8 %), and an RMSE of 103 mm wyr⁻¹ (11.6 %) (Fig. 4). Among individual HUC8s, R2 ranged from 0.11 to 0.70, bias spanned -79 to 85 mm wyr⁻¹, and RMSE varied between 63 and 104 mm wyr⁻¹. Examined by water year, R2 ranged from 0.04 to 0.80, bias extended from -125 to 117 mm wyr⁻¹, and RMSE ranged from 57 to 127 mm wyr⁻¹ (Fig. 4).

Monthly ET increases from January to June, peaking at 133.6 mm (averaged across all canopy classes), before dropping to 32.2 mm in December (Fig. 5). The highest single ET value, 168.0 mm, occurs in June within the ≥ 81 % canopy cover class, while the lowest single value, 30.5 mm, is observed in December within the 0 %–10 % cover class. The ≥ 81 % cover class exhibits the highest ET values for 6 months (April–September), whereas the 61 %–80 % cover class dominates in the remaining months (Fig. 5).

ET exhibited a seasonal cycle, with average lows of 38.5 mm in winter (December–February), moderate levels of 98.7 mm in spring (March–May), and peak levels of 112.1 mm in summer (June–August), before declining to 61.0 mm in fall (September–November) (Fig. 6). Notably, the drought year 2011 showed substantially lower ET values relative to other years – averaging 78.7 mm in spring, 42.6 mm in summer, and 36.1 mm in fall – coinciding with the low precipitation totals. Conversely, higher precipitation levels led to greater distinction between seasonal ET averages (Fig. 6).

The two climatic variables P and Ta exhibited notably different relationships with ET. Precipitation showed a moderate positive correlation, with R2 values ranging from 0.23 (1001–1200 mm) to 0.61 (600–800 mm) and slopes from 0.29 (1001–1200 mm) to 0.55 (≥ 1201 mm) (Fig. 7). By contrast, air temperature demonstrated a weak negative relationship with ET, with R2 values varying from 0.09 (801–1000 mm) to 0.27 (600–800 mm) and slopes between -82.02 (600–800 mm) and -45.72 (801–1000 mm) (Fig. 7).

In comparison, the two woody-vegetation metrics showed more consistent positive relationships with ET than the climatic variables. Canopy height exhibited a moderate positive correlation, with R2 values between 0.48 (600–800 mm and ≥ 1201 mm) and 0.54 (801–1000 mm) and slopes ranging from 35.05 (1001–1200 mm) to 38.66 (801–1000 mm) (Fig. 7). Similarly, canopy cover displayed a weaker but still positive association, with R2 values ranging from 0.12 (600–800 mm) to 0.20 (1001–1200 mm) and slopes between 3.61 (600–800 mm) and 4.32 (1001–1200 mm) (Fig. 7).

The generalized additive model (GAM) results – shown by the magenta curves in Fig. 7 – largely confirmed the linear patterns but captured subtle non-linearities, especially at the extremes of the predictor ranges. GAM pseudo-R2 values were highest for canopy height (0.51) and precipitation (0.49), modestly exceeding the corresponding aggregated linear R2 values (0.50 and 0.43, respectively). For canopy cover and air temperature, the pseudo-R2 values were low (0.15 and 0.13), mirroring the weak to moderate linear associations.

The ratio of ET to P (ET / P) was generally stable, averaging 90 % and falling between 70 % and 100 % in most years (Fig. 8); the only exceptions were the drought years 2011 and 2022, when ET / P rose slightly above 100 %. Both years were characterized by above-average air temperatures and below-average precipitation (Fig. 8).

Areas exhibiting ET / P ratios above 100 % in the Post Oak Savannah predominantly coincided with low-elevation, forested river basins and their tributaries (Fig. 9). Notable examples include the Sabine, Trinity, Navasota, Guadalupe, and San Antonio River basins, where forested riparian zones consistently displayed ET / P values exceeding 100 % (Fig. 9).

Total excess water varied substantially across the Post Oak Savannah ecoregion, with most values ranging from -5000 to 7500 mm (Fig. 10a). Over the entire study period (2009–2023), the mean excess water was 2422 mm, or 161 mm yr⁻¹ (Fig. 10b). Temporally, excess water ranged from a low of -22 394 455 mm in 2022 to a high of 167 853 812 mm in 2016 (Fig. 10c). On average, the ecoregion totaled 47 971 635 mm of excess water per year. Only two years exhibited negative excess water: 2011 (-21 968 413 mm) and 2022 (-22 394 455 mm) (Fig. 10c).

These contrasting totals reflect different hydrometeorological conditions. In 2011, precipitation was relatively low at 137 953 517 mm, whereas in 2022, it reached 221 613 882 mm (Fig. 10c). Conversely, evapotranspiration (ET) in 2011 was also low, returning only 158 624 385 mm of water to the atmosphere compared with 242 295 663 mm in 2022 (Fig. 10c).

When integrating both woody vegetation cover and overall aridity into the analysis, increases in woody cover consistently reduced excess water totals across every precipitation zone (Fig. 11). The highest annual excess water (414.87 mm) occurs in areas with 0 %–10 % woody cover that receive ≥ 1200 mm of precipitation (Fig. 11). In contrast, the lowest annual excess water (-122.87 mm) occurs in areas with ≥ 80 % woody cover in the 600–800 mm precipitation zone (Fig. 11). Notably, none of the 0 %–10 %, 11 %–20 %, 21 %–40 %, or 41 %–60 % woody cover classes exhibited negative excess water values (Fig. 11). Conversely, in the 61 %–80 % and ≥ 80 % woody cover categories, all precipitation zones had negative values except the ≥ 1200 mm zone, which remained positive (Fig. 11).

The MOD16 ET model used in this study showed moderate-to-strong agreement with WBET measurements at the HUC8 scale, with an overall RMSE of 103 mm wyr⁻¹ and an R2 of 0.65 (Fig. 4). The bias of -7 mm wyr⁻¹ indicates that the model generally neither overestimates nor underestimates ET. These findings align with other validations of the MOD16 ET product, performing better in some cases and worse in others (Aguilar et al., 2018; Nadzri and Hashim, 2014; Du and Song, 2018; Ruhoff et al., 2013; Miranda et al., 2017). Because the WBET evaluation itself contains error, we note four key sources of uncertainty: (1) gauge-based precipitation grids can be biased by under-catch and sparse station density; (2) not all of the subbasins were within the Post Oak Savannah's boundary, so lateral inflows/outflows can distort basin averages, (3) long-term soil- and groundwater-storage changes are assumed negligible, yet seasonal drought–recharge cycles can shift storage by several cm; and (4) small reservoirs and irrigation withdrawals remain the streamflow record, potentially inflating inferred ET during dry years.

Performance varied markedly among years. In 2009–2010, annual rainfall totals were near the long-term mean, but precipitation was concentrated in northern catchments and deficient in the south, creating north–south gradients that the basin-wide WBET captured but MOD16 ET did not, thereby showing increased RMSE values. During the extreme droughts of 2011 and 2022, MOD16 overestimated by 77 and 117 mm (Fig. 4), respectively, exposing a known weakness in the algorithm in representing soil-moisture stress when stomatal conductance is modeled from meteorology alone (Hu et al., 2015; Miralles et al., 2016; Majozi et al., 2017). Therefore, although our semiarid basins showed comparatively higher values, the literature shows that MOD16 often struggles in arid and semiarid environments because it lacks an explicit soil-moisture constraint – making overprediction likely when soils are critically dry even within otherwise well-performing regions. By contrast, biases in wetter years – such as 2015 and 2016 (-62 and -80 mm, respectively; Fig. 4) – were modest but still larger than those in average-precipitation years. These residual errors may reflect reduced available energy under persistent cloud cover and/or enhanced flood-plain evaporation that raised actual ET beyond what MOD16 captured.

Spatial patterns echoed these temporal contrasts. The humid northeastern basins (HUC8s 1–4) exhibited the lowest accuracy (R2 = 0.11–0.39; RMSE = 92–104 mm; Fig. 4) because a dense matrix of small wetlands and ponds leaves many 500 m MODIS pixels as mixed water–vegetation surfaces – even after masking open-water and wetland pixels – thereby inflating MOD16 ET relative to WBET. Conversely, the semiarid central and southwestern basins (HUC8s 7–11) showed the best agreement (R2 = 0.54–0.70; RMSE = 63–79 mm; Fig. 4), consistent with earlier findings that MOD16 may perform relatively well where soils are thin and vegetation sparse (Du and Song, 2018; Velpuri et al., 2013) – provided soil moisture is not fully depleted.

Finally, our aggregated statistics (RMSE = 103 mm yr⁻¹; R2 = 0.65; Fig. 4) fall between the high accuracy reported for Brazil's Caatinga (≈ 59 mm yr⁻¹; R2 = 0.82; Miranda et al., 2017) and the lower accuracy in the Brazilian Cerrado (RMSE = 168 mm yr⁻¹; bias =+19 %; Ruhoff et al., 2013). Results from northwest Mexico (R2 = 0.46–0.86; RMSE = 142–168 mm; Aguilar et al., 2018) bracket our values and underscore a consistent limitation: MOD16 normalizes extremes – overestimating ET when soils are parched and underestimating in complex, water-rich mosaics – highlighting the need for soil-moisture constraints in future versions.

We observed peak precipitation in May and October, which aligns with the expected wet (spring and fall) and dry (summer) seasons characteristic of a humid subtropical climate (Fig. 5). Correspondingly, each canopy cover class showed its highest ET rates in June (Fig. 5). This peak likely results from a combination of actively growing vegetation, abundant soil moisture following increased May rainfall, and warm air temperatures that raise the vapor pressure deficit (VPD) and therefore the atmosphere's capacity to hold water vapor (Liu et al., 2017; Sun et al., 2023). In addition, because our study area lies in the Northern Hemisphere, the summer solstice occurs in June, providing heightened solar radiation, which further increases PET (Aschonitis et al., 2017).

Despite July and August being the warmest months, ET declines substantially during this period (Fig. 5). We attribute this decrease to reduced soil moisture storage, which is rapidly depleted via high-VPD-induced transpiration and evaporation (Mondal et al., 2024; Yang et al., 2023; Anav et al., 2018). This trend persists into the cooler months, when air temperatures begin to drop and rainfall increases, such that ET begins to rise again only in February as air temperatures rebound.

Interestingly, the ≥ 81 % canopy cover class does not exhibit the highest ET in every month but only from April to September (Fig. 5). One explanation is that heavily thicketized (≥ 81 %) woodlands may have a more complex vertical structure – comprising both deciduous and evergreen species – such that the overlapping foliage layers produce a more closed canopy during the warmer months (Whitehurst et al., 2013; Arumäe and Lang, 2018; Scott et al., 2014; Jucker et al., 2015). Many thicketized Post Oak Savannah stands consist of oak overstory combined with an understory of evergreen species such as Juniperus virginiana and Ilex vomitoria (Olariu et al., 2024; Basant et al., 2023). In contrast, woodlands with 61 %–80 % canopy cover are generally dominated by evergreen species (Pourrahmati et al., 2023; Arumäe and Lang, 2018; Stephens et al., 2015), which remain active during cooler months – potentially explaining their higher ET from October through March. For instance, extensive Pinus taeda stands are found in the eastern Post Oak Savannah, adjacent to the Piney Woods ecoregion. Owing to their needle-shaped leaves, Pinus taeda woodlands typically range from 60 %–80 % canopy cover, the higher percentages associated with mid-aged stands that include a mix of younger and older trees, eventually forming gaps in older stands (Song et al., 2009; Zeide and Stephens, 2010; Johnson et al., 2021).

Seasonal ET trends closely followed seasonal precipitation (Fig. 6). From 2011 to 2014, monthly precipitation averaged 60 mm, resulting in minimal seasonal variation in ET. Between 2015 and 2021, however, monthly precipitation rose to an average of 86 mm – a 26 mm increase – which widened the seasonal stratification in ET. This increased stratification is attributable to higher transpiration rates during spring and summer, driven by the ample water supply that maintained elevated soil moisture (Fu et al., 2022; Koehler et al., 2023).

Notably, the severe drought of 2011 (Nielsen-Gammon, 2012; Chen et al., 2021) caused summer and fall ET to drop to winter-like levels (Fig. 6). Although spring ET remained near average, this was likely a residual effect of the relatively wet conditions in 2009 and the average precipitation in 2010. The 2011 drought caused an estimated mortality of 65.6 (± 7.3) million trees in East Texas alone – encompassing common Post Oak Savannah species such as Quercus stellata, Quercus falcata, Ulmus alata, and Pinus taeda (Klockow et al., 2018). Additionally, the difference between precipitation and PET in 2011 reached -1206 mm (Schwantes et al., 2017). Widespread wildfires consumed nearly 4 million acres across Texas – 31 453 individual fires – representing 47.3 % of all acreage burned by wildfire in the United States that year (Nielsen-Gammon, 2012; Texas A&M Forest Service, 2011).

Evapotranspiration in the Post Oak Savannah showed a moderate positive relationship with precipitation and a weak negative relationship with air temperature (Fig. 7). This P–ET coupling is consistent with global research findings, which highlight the tight linkage between these two fluxes across diverse ecosystems (Mondal and Mishra, 2024; Mondal et al., 2024; Xi et al., 2023; Zeng et al., 2010). The GAM analysis supports this pattern: precipitation explained nearly half of the variance in ET (pseudo-R2 ≈ 0.49) and followed a gently saturating curve, reinforcing the strength of the linear fit. Notably, the correlation was stronger in the more arid regions of the Post Oak Savannah (600–1000 mm vs. ≥ 1001 mm), where limited water availability acts as the primary constraint instead of energy inputs (e.g., radiation and air temperature) (Nagler et al., 2007; Yu et al., 2021). Consequently, in these drier areas, ET begins soon after precipitation events: soils rapidly absorb incoming rainfall, vegetation responds by increasing transpiration, and overall ET rises (Nielsen-Gammon, 2012).

In contrast, the negative relationship between ET and air temperature may appear counterintuitive. However, many plants operate within an optimal air temperature window for photosynthesis and transpiration (commonly 20–30 °C) (Yamasaki et al., 2002; McGowan et al., 2020; Crous et al., 2022). In the Post Oak Savannah, severe summer heat and lower precipitation often drive plants to close their stomata, thereby reducing transpiration despite high VPD. This negative relationship is particularly strong in the most arid (600–800 mm) and most humid (≥ 1200 mm) areas, whereas it is weaker in the intermediate (801–1200 mm) zone. In the arid region, limited soil moisture readily explains stomatal closure and reduced transpiration. In more humid areas, factors such as persistent cloud cover or higher relative humidity may restrict the vertical movement of water vapor from plant surfaces to the atmosphere (Wang et al., 2018; Dai et al., 1999).

GAM fits (magenta curves in Fig. 7) revealed a non-linear, dome-shaped response of annual ET to mean air temperature. ET climbed steadily to a peak at ≈ 22–24 °C, plateaued, and then declined above ∼ 25 °C; the GAM pseudo-R2 was 0.13, only marginally higher than the aggregated linear R2 (0.11), but it captured the threshold beyond which stomatal regulation suppresses transpiration. This pattern is consistent with the divergent thermal strategies of the region's dominant woody species. The evergreen loblolly pine (Pinus taeda) begins to reduce stomatal conductance at leaf temperatures near 32 °C, whereas drought-deciduous post oak (Quercus stellata) and blackjack oak (Q. marilandica) maintain higher conductance until ≈ 35 °C before closing their stomata (Oren et al., 1999; Novick et al., 2016). Because summer days in the Post Oak Savannah frequently exceed these thresholds, particularly during drought years, elevated mean annual temperatures integrate numerous midday periods of stomatal closure, driving down yearly ET despite higher vapor-pressure deficits.

Both canopy cover and canopy height displayed positive relationships with ET, but canopy height was the stronger predictor (Fig. 7). The GAM analysis confirms this contrast: canopy height achieved the highest pseudo-R2 (0.51), whereas canopy cover accounted for far less variance (pseudo-R2 = 0.15). The canopy-height GAM curve is nearly linear across the sampled range, indicating that taller stands consistently increase ET. By contrast, the canopy cover GAM rises sharply up to ∼ 80 % cover and then plateaus, suggesting diminishing returns once the overstory approaches complete closure. Because height integrates above-ground biomass, LAI, and vertical leaf stratification – all key determinants of transpiration (Bonan, 2008; Baldocchi, 2003) – these relationships remain stable across precipitation zones. Taller canopy structures also increase surface roughness, reducing aerodynamic resistance and enhancing turbulent exchange, thereby promoting more efficient transfer of water vapor from the canopy to the atmosphere. Furthermore, taller trees also develop deeper, more extensive root systems that tap subsurface water during droughts, which are common in the summer in the Post Oak Savannah, sustaining transpiration when upper soil layers are dry. Hydraulic lift, where deep-rooted trees redistribute moisture upward (Norton and Hart, 1998), and the complex root architecture of tall individuals (Jackson et al., 2020) together help maintain a dense canopy and elevated ET.

Over the study period, the average ET / P ratio in the Post Oak Savannah was 90 % (Fig. 8). Globally, the mean ET / P ratio over land surfaces is approximately 65 %, varying by continent. For instance, North America averages around 70 %, whereas Australia – which more closely resembles the Post Oak Savannah's overall conditions – exhibits a higher ratio of 87 % (Reitz et al., 2017; McDonald, 1961). An ET / P ratio of 90 % is therefore plausible when compared with other semiarid or arid regions, where studies have reported ratios between 80 % and 93 % (Fleischmann et al., 2023; Irmak, 2017). Moreover, Althoff and Destouni (2023) suggest that ET / P will continue to rise as agricultural and forestry activities expand, increasing the prevalence of trees – a pattern already observed in the Post Oak Savannah (Olariu et al., 2024). This high ratio also indicates an ET-driven system, with evapotranspiration as the dominant water-budget component (Condon et al., 2020; Reitz et al., 2017), confirming the assumption made by Basant et al. (2023).

The ET / P ratios exceeding 100 % in 2011 and 2022 likely reflect severe drought conditions that forced the ecosystem to draw on limited soil-moisture reserves and, in riparian areas, groundwater. Fleischmann et al. (2023) documented similar behavior in South American riparian zones, consistent with our Post Oak Savannah observations (Fig. 9). Ratios above 100 % may also stem from uncertainties in the MOD16 product, which does not adequately represent plant water stress; under the extreme stress of 2011 and 2022, MOD 16 likely overestimated ET, producing values higher than precipitation.

Excess water (P–ET) exhibited substantial spatial variability, ranging from -5000 mm to over 7500 mm, with an average of ∼ 2500 mm across the entire study period (Fig. 10). However, 2011 and 2022 both showed net negative excess water, explained by the same conditions that led to ET / P ratios exceeding 100 %.

Of note, Fig. 11 illustrates a steep decline in excess water with increasing woody cover. As tree and shrub cover expands, transpiration intensifies, lowering the net water surplus. These findings align with those of Basant et al. (2023), who found that understory shrub thicketization in the Post Oak Savannah substantially reduces groundwater recharge. Consequently, under continued WPE, the Post Oak Savannah will likely experience greater reductions in excess water – especially in its more arid regions, where soil moisture is already limited. Such changes may alter local water availability, affect aquifer recharge, and shift ecosystem functioning, as woody plants increasingly outcompete herbaceous vegetation for scarce moisture.