Remote Sensing-Based Assessment of Evapotranspiration Patterns in a UNESCO World Heritage Site Under Increasing Water Competition

Abstract

In water-scarce regions, natural ecosystems and agriculture increasingly compete for limited water resources, intensifying stress during periods of drought. To assess these competing demands, we applied a modified PT-JPL model that incorporates the thermal inertial approach as a substitute for relative humidity (RH) in estimating soil evaporation—a method that significantly outperforms the original PT-JPL formulation in Mediterranean semi-arid irrigated areas. This remote sensing framework enabled us to quantify spatial and temporal variations in water use across both natural and agricultural systems within the UNESCO World Heritage site of Doñana. Our analysis revealed an increasing evapotranspiration (ET) trend in intensified agricultural areas and rice fields surrounding the National Park (R = 0.3), contrasted by a strong negative ET trend in wetlands (R < −0.5). These opposing patterns suggest a growing diversion of water toward irrigation at the expense of natural ecosystems. The impact was especially marked during droughts, such as the 2011–2016 period, when precipitation declined by 16%. In wetlands, ET was significantly correlated with precipitation (R > 0.4), highlighting their vulnerability to reduced water inputs. These findings offer crucial insights to support sustainable water management strategies that balance agricultural productivity with the preservation of ecologically valuable systems under mounting climatic and anthropogenic pressures typical of semi-arid Mediterranean environments.

1. Introduction

The growing global population and rising food demand is placing increasing pressure on natural ecosystems, many of which depend on the same water resources as agricultural lands [1,2,3,4]. Climate change is intensifying these challenges. The increasing frequency and severity of extreme weather events, such as droughts, floods, and heatwaves, are threatening crop production and food security worldwide [5,6,7]. These climatic extremes not only disrupt agricultural productivity but also impair the functionality of ecosystems globally, underscoring the urgent need for adaptive strategies to safeguard both food security and ecosystem health [7,8].

A prime example of this is the Mediterranean region [9,10,11]. The Iberian Peninsula is projected to be one of the regions in the world most severely impacted by droughts [12,13] and as a result, competition for water resources is expected to intensify significantly [14]. These effects will be especially acute in the river basins of southern Spain [15], where irrigated agriculture accounts for 80% of total water extraction [16]. In the Guadalquivir River basin in southern Spain, water consumption is expected to increase up to 20% by 2050 [17], a trend that could further compromise the ecological integrity of dependent natural ecosystems [18]. A significant case within this basin is the Doñana Biosphere reserve located at the river’s mouth and protected under multiple international frameworks, including the Ramsar Convention on Wetlands.

Given the projected decrease in precipitation across southern Europe and the Mediterranean basin, including the Guadalquivir River basin [19,20], there is an urgent need to understand how natural ecosystems respond to meteorological drivers—particularly precipitation—under increasingly arid conditions [21,22]. In regions like Doñana, where ecosystems are already sensitive to water availability, declining rainfall is expected to exacerbate existing stressors. These include intensive agricultural irrigation and groundwater abstraction, which further reduce water accessibility for wetlands and other natural habitats [23,24]. Understanding ecosystem responses to climatic variability is critical for anticipating tipping points, identifying thresholds, and informing adaptive water management strategies. Such strategies must account for both natural climate fluctuations and increasing anthropogenic pressures [25,26], which further constrain water availability for wetlands and other natural habitats [23].

This urgency is underscored by recent hydrological observations in the Doñana region, which reveal that streamflow rates have nearly halved and the water table has dropped by more than 6 m due to over-extraction, primarily for agriculture [27,28,29]. These changes are already placing considerable stress on the region’s natural ecosystems, notably the wetlands. Recognizing the severity of the situation, the World Heritage Committee has called for a comprehensive assessment of Doñana’s ecohydrological conditions to curb illegal water abstraction and prevent the site’s designation as a World Heritage in Danger [30]. Since 2016, the Doñana area has been increasingly affected by anthropogenic pressures, such as intensified groundwater extraction, agricultural expansion, and policy changes, culminating in environmental and legal actions, including the European Commission’s 2019 infringement procedure against Spain for non-compliance with the EU Habitats Directive (Case C-559/19) [31].

Providing scientific evidence through enhanced monitoring and accurate modeling of water use is essential for informing policymakers, supporting regulatory action, and adapting water demands to safeguard these threatened ecosystems [32].

A central component of this modeling effort is evapotranspiration (ET), which represents the largest water flux from land to atmosphere and is key to understanding historical and spatial dynamics of water consumption by both crops and natural vegetation [33,34]. Spatially distributed ET estimates are fundamental for quantifying historical water use patterns and for identifying critical areas of overuse or imbalance.

PT-JPL is a widely adopted model for estimating ET across diverse biomes and underpins NASA’s ECOSTRESS mission [35]. However, it shows reduced accuracy in semi-arid agroecosystems, particularly due to its handling of soil evaporation (ETs) [36,37]. This limitation stems from its reliance on the moisture content of the overlying atmosphere to simulate soil evaporation dynamics [38]. High values of vapor pressure deficit (VPD) may reflect the surrounding arid conditions but fail to capture fine-scale soil moisture dynamics, reducing ET reliability [39,40,41,42,43]. To address these shortcomings, remote sensing (RS)-based proxies, like apparent thermal inertia (ATI) and SWIR indices, have been integrated into PT-JPL [36,37,44], offering better spatial resolution and more direct sensitivity to surface moisture. PT-JPL-thermal is a modified version of the PT-JPL model [45], which incorporates thermal remote sensing data as a proxy to the soil moisture status, outperforming other RS models, such as the PM model adapted by Leuning et al. [46] and the MOD16 ET algorithm used by Mu et al. [47] and Mu et al. [48] in semi-arid regions. PT-JPL-thermal was benchmarked at EC flux sites [44] and later regionalized by Moyano et al. [49] to compare VPD-based approaches in semi-arid areas, validating ATI’s effectiveness in capturing ET dynamics under Mediterranean conditions, specifically in Doñana.

Although several studies have analyzed ET dynamics within specific natural ecosystems of the Doñana region, such as wetlands, shrublands, and forested areas [23,50], there is still a lack of comprehensive, long-term assessments that integrate both the irrigated agricultural zones surrounding the park and the protected ecosystems within Doñana National Park. This gap in spatial integration and thematic scope has limited our understanding of the broader ecohydrological dynamics at the landscape scale and impedes the detection of regional water dynamics. In light of projected precipitation declines across the Mediterranean basin [21,22] and rising irrigation demands in the Guadalquivir River basin [51], it is increasingly important to assess how both natural and agricultural systems respond to meteorological drivers under growing anthropogenic pressures.

Simultaneous analysis of ET dynamics in relation to water use patterns across land use types, as well as its relationship with precipitation, can reveal areas of vulnerability, enhance understanding of ecosystem resilience, and support more adaptive and sustainable water resource planning, contributing to safer operating spaces for long-term conservation [27].

This study aims to quantify the spatio-temporal dynamics of evapotranspiration (ET) and assess the influence of meteorological drivers on water use across contrasting land systems in the UNESCO World Heritage Site of Doñana and its surrounding irrigated agricultural areas from 2003 to 2016. In this region, where water resources are increasingly contested between ecological conservation and intensive agriculture [52], comprehensive long-term assessments remain scarce. Notably, existing studies often overlook the integrated analysis of both the protected ecosystems within Doñana National Park and the adjacent agricultural zones, despite the area’s critical ecological importance.

To achieve this goal, we set two specific objectives:

• To analyze the intra- and inter-annual dynamics of ET and the water deficit index (WDI) in the Doñana region, with a particular focus on differences between irrigated croplands and natural protected ecosystems within the park, while also examining the relationships between land cover types and meteorological drivers, especially precipitation.
• To assess the spatial patterns and temporal trends of ET and WDI over the study period, with special attention given to years of significant water stress, such as 2005, characterized by severe precipitation deficits affecting natural ecosystems, and 2007, when irrigation supply was notably insufficient.

2. Study Area and Data

2.1. Study Area

The study area covers Doñana National Park (54,251 ha) and the adjacent Natural Park (53,835 ha), located above the Almonte-Marismas aquifer in Andalusia, southern Spain (36.98°N, 6.37°W). Within the park, the wetland landscape forms a mosaic of distinct types, including the seasonally flooded marshes, temporary dune ponds, and ecotone freshwater habitats lying between dunes and marshes [30]. These are complemented by shrublands and coniferous forests. Surrounding the protected areas are agriculturally developed lands, notably rice fields [18] along both banks of the Guadalquivir River and mixed irrigated croplands established on former marshland areas (Figure 1).

Historically, Doñana’s marshes extended across approximately 150,000 ha, but have undergone substantial transformation over the last century [53]. Currently, only about 30,000 ha of seasonal marshes—fed by rainfall and surface streams—and 3000 ha of temporary dune ponds remain within the National Park. These ponds rely on the annual fluctuations of the Almonte-Marismas aquifer, which is nearly five times larger than the park itself [23,54]. The area is a crucial refuge for over half a million migratory birds and endangered species, including the Iberian lynx and the Spanish imperial eagle [55].

In the 1960s, the remaining ancient marshes were converted into arable land, causing a deep landscape change [52] and defining the actual landscape of the region. To counteract this trend, it was declared a National Park in 1969, included in the Ramsar Convention as one of the largest wetlands in Europe [28], and designated a World Heritage Site by UNESCO in 1995. The National Park has also been buffered by a Natural Park, which entered the endangered Montreux Record of Ramsar sites in 1990 [56].

In recent decades, a more gradual yet profound transformation has been reshaping the landscape of Doñana. The same groundwater that sustains the wetland’s natural flooding cycles is increasingly being tapped for uses beyond the boundaries of the National Park.

Conflicts between environmental conservation and economic development in the Doñana region have grown significantly [53]. Since the 1970s, agricultural intensification and rising urban water demand have driven substantial groundwater extraction, leading to historical drawdowns of up to 20 m in some areas [28,57]. This overexploitation threatens not only the long-term sustainability of the aquifer but also the stability of groundwater-dependent ecosystems, including ponds and marshes [23]. Given their strong reliance on groundwater, these wetlands are particularly vulnerable to fluctuations in aquifer levels. If the current management of the aquifer is continued over time, the sustainability of its associated ecosystems will be compromised [58]. Additionally, high concentrations of nitrates from agrarian origin in the aquifer [59] have led to many of the croplands, rice fields, and wetlands in the area to be designated as nitrate vulnerable zones [60] under the European Commission nitrates directive [61].

2.2. Remote Sensing Data

Multiple remote sensing observations were acquired from the Moderate Resolution Imaging Spectroradiometer (MODIS), using sensors from both the Aqua and Terra satellites. The satellite data were used in combination with in situ meteorological data (air temperature and radiation) as inputs for the PT-JPL-thermal model. MODIS land products were retrieved from the Earth Observing System Data and Information System (EOSDIS), a core capability in the National Aeronautics Space Administration (NASA) Earth Science Data Systems Program. All the selected MODIS products (version 5) were acquired at a 1 km pixel resolution for the 2003–2016 study period. The temporal resolutions of the data sets were (1) daily for the land surface temperature (LST), land emissivity (ε_S), and MODIS overpass time from MOD11A1 and MYD11A1, (2) 8-day composites of the land surface temperature (LST) from MOD11A2 and MYD11A2, (3) 8-day composites of the leaf area index (LAI) and fraction of photo-synthetically active radiation (f_APAR) from MOD15A2, as well as the broadband surface albedo (α) acquired from MCD43B3, and (4) 16-day composites of the normalized difference vegetation index (NDVI) retrieved from MOD13A2. To interpolate the daily values from the 8- and 16-day composite variables, each composite value was assumed to be constant over its respective period.

To ensure long-term consistency in the satellite data, all MODIS products used in this study were sourced from Collection 5 (version 5), which benefits from retrospective calibration and standardized processing algorithms across the full time series. We used data from both the Terra and Aqua platforms, enhancing temporal coverage and enabling cross-sensor consistency checks. While minor inconsistencies due to sensor degradation or algorithm updates cannot be entirely excluded, several measures were implemented to mitigate these effects. These include the application of product-specific quality control flags to retain only high-quality observations, and the integration of in situ meteorological data (air temperature and radiation) into the PT-JPL-thermal model, thereby reducing dependence on satellite-derived inputs alone.

2.3. Meteorological Data

Meteorological data were obtained from the agro-climatic station Lebrija I (36.98°N, 6.13°W, https://www.juntadeandalucia.es/agriculturaypesca/ifapa/ria/, (accessed on 15 June 2024)), which is a station site of the Agroclimatic Information Network of Andalusia (RIAA). This station is controlled by a CR10X datalogger with sensors to measure T_air (T_max, T_min, and T_mean), relative humidity (RH) (RH_max, RH_min, and RH_mean), solar radiation R_s, precipitation P, wind speed and direction, and reference evapotranspiration (ET_o) on a daily basis. The data are transferred by GSM modems for quality control and data validation [62].

2.4. Land Use Maps

Land use maps were retrieved from the 2013 update Land Cover and Use Information System of Spain (SIOSE) database in the Doñana region [63]. SIOSE is accessible at a 1:10,000 scale and follows the European Terrestrial Reference System 1989 (ETRS89) under the European Directive INSPIRE (2007/2/CE) requirements.

3. Methods

3.1. Remote Sensing ET Based on a Thermal and Optical RS Model

In this work, the PT-JPL-thermal model by García et al. [44], based on the approach by Fisher et al. [45], was spatially distributed over the study area for a 14 year-period at a daily time scale. The model was previously validated in the semi-arid region of Doñana and surrounding irrigated areas [49].

To facilitate comprehension of the model used in this work, all the equations and variables used to retrieve the ET as the combination of canopy transpiration and soil evaporation are detailed in

Table 1. Equations and variables used to estimate the daily values of ET in the PT-JPL-thermal model. α_PT = 1.26, the Priestley–Taylor coefficient; Δ is the slope of the saturation-to-vapor pressure curve (Pa K⁻¹); γ is the psychrometric constant (0.066 kPa C⁻¹); G is the soil heat flux (Wm⁻²), negligible at the daily scale as in Fisher et al. [45]; k_Rn = 0.6 [64]; LAI is from MOD15A2; σ is the Stefan–Boltzmann constant (5.67 × 10⁻⁸ Wm⁻²); c and d are constants (0.261 and 7.77 × 10⁻⁴, respectively); T_max and T_min (°C) are the max and min climatic T_air [65]; N (h) is the time lag between sunrise and sunset (https://www.esrl.noaa.gov/gmd/grad/researchp.html, (accessed on 2 February 2020)); d is the time lag for the maximum temperature before sunset (1.86 h); MODIS time is the time from MYD11A1 or MOD11A1 depending on clouds; c is the time lag for the minimum air temperature after sunrise (−0.17 h); εS is the average of emissivity bands 31 and 32; LST is from MYD11A1 or MOD11A1 (depending on clouds); R↓Sday (MJ/m²/day) is the climatic data [65]; αBSA and αWSA are the broadband black-sky and white-sky albedos; t (h) is the time lag between the sunrise time from NOAA and MODIStime; f_APAR is the fraction of absorbed photosynthetic active radiation; f_IPAR is the fraction of intercepted photosynthetic active radiation as a function of the NDVI acquired from MOD13A2 [42]; T_opt is the optimum temperature for plant growth (25 °C) as in García et al. [44]; T_am is the daily mean T_air (°C); (LSTDay−LSTNight) is the maximum daily LST oscillations from 8-day MYD11A2 and MOD11A2 data; ϑ is the latitude; and φ is the solar declination factor.

Variable Description	PT-JPL-Thermal Equations	Reference
Evapotranspiration	$ET={ET}_c+{ET}_s$	[45]
Canopy transpiration	${ET}_c={ETp}_c{·f}_g·f_m·f_t$	[45]
Potential canopy transpiration	${ETp}_c={\alpha }_{PT}·{\displaystyle \frac{\Delta }{\Delta +\gamma }}·(R_{nc}-G)$	[45]
• Net canopy radiation	$R_{n_c}= R_n-R_{n_s}$	[45]
• Net soil radiation	$R_{n_s}=R_n·e^{\left(-k_{Rn}·LAI\right)}$	[64]
• Net radiation	$R_n=R_L^{\downarrow }-R_L^{\uparrow }+R_S^{\downarrow }-R_S^{\uparrow }=R_L{+R}_S$	[45]
Instant. incoming longwave radiation	$R_{L_{inst}}^{\downarrow }=\sigma ·({T_{{air}_{MODIS-time}}+273.15)}^4· \left[1-\left(c·e^{\left(-d \left({T_{{air}_{MODIS-time}}}^2\right)\right)}\right)\right]$	[66]
Air temperature at MODISpass-time	$T_{{air}_{MODIS-time}}=\left(T_{max}-T_{min} \right)·\mathit{sin}\left({\displaystyle \frac{\pi ·m}{N+2·d}}\right)+T_{min}$	[67]
Number of hours from Tmin until sunset	$m={MODIS}_{time}-\left(12-\left({\displaystyle \frac{N}{2}}\right)+c\right)$	[67]
Instant. outgoing longwave radiation	$R_{L_{inst}}^{\uparrow }=-{\epsilon }_S·\sigma ·T_S^4$	[68]
Daily shortwave radiation	$R_{S_{day}}=R_{S_{day}}^{\downarrow }· \left(1-\alpha \right)$	[68]
Albedo	$\alpha =0.8 · {\alpha }_{BSA}+0.2 · {\alpha }_{WSA}$	[69]
Instant. shortwave radiation	$R_{S_{inst}}={\displaystyle \frac{R_{S_{day}}}{J}}·{\displaystyle \frac{24}{N}}$	[70]
Conversion factor day-inst	$J={\displaystyle \frac{2}{sin\left(\frac{pi·t}{N}\right)}}$	[70]
Instantaneous net radiation	$R_{n_{inst}}=R_{L_{inst}}+R_{S_{inst}}$
Daily net radiation	$R_{n_{day}}=R_{n_{inst}}·J·{\displaystyle \frac{N}{24}}$	[71]
Canopy transpiration constraints
• Green canopy fraction	$f_g=f_{APAR}/f_{IPAR}$	[45]
• Plant moisture constraint	$f_m={\displaystyle \frac{f_{APAR}}{f_{APARmax}}}$	[45]
• Plant temperature constraint	$f_t=1.1814/\left[\left(1+e^{0.2·\left(T_{opt}-10-{Ta}_m\right)}\right)\right]/\left[\left(1+e^{0.3·\left(-T_{opt}-10+{Ta}_m\right)}\right)\right]$	[68]
Soil evaporation	${ET}_S={ETp}_s·f_{sm}$	[45]
Potential soil evaporation	${ ETp}_s={\alpha }_{PT}·{\displaystyle \frac{\Delta }{\Delta +\gamma }}·(R_{n_s}-G)$	[45]
Soil evaporation constraints
• Soil moisture constraint	$f_{sm}={\displaystyle \frac{ATI-{ATI}_{min}}{{ATI}_{max}-{ATI}_{min}}}$	[44]
Apparent thermal inertia	$ATI=C·{\displaystyle \frac{1-\alpha }{\left({LST}_{Day}-{LST}_{Night}\right)}}$	[44]
Solar flux correction factor	${{C=sin\vartheta ·sin\phi ·(1-tan}^2\vartheta ·tan}^2\phi )+cos\vartheta ·cos\phi ·arccos(-tan\vartheta ·tan\phi )$	[72]

.

We calculated the daily ET derived from the PT-JPL-thermal model, as well as the water deficit index (WDI), defined as ET/ETp [73], in the region on a pixel basis. ETp was calculated as the aggregation of the potential evapotranspiration of the canopy and that of the soil (see Table 1).

The daily results obtained from the model over the period from 2003 to 2016 were spatially aggregated by land cover class, distinguishing between irrigated areas (including mixed-irrigation BXII, mixed-irrigation left bank, rice-right, and rice-left) and natural ecosystems (wetland, shrubland, and coniferous forest). For each class, we calculated the spatial mean of all constituent pixels on a daily basis.

3.2. Intra- and Inter-Annual Dynamics of Evapotranspiration in the Doñana Region and the Influence of Meteorological Drivers

Daily model outputs were aggregated to monthly and annual values at the pixel level to assess the ET and WDI intra-annual and inter-annual dynamics for all land cover classes.

To assess the inter-annual dynamics of ET and their relationship with meteorological drivers, particularly precipitation, standardized anomalies of precipitation $STD{A}_P$ and evapotranspiration $STD{A}_{ET}$ [74] for the period 2003–2016 were calculated (Equations (1) and (2)).

$${STDA}_{P_i}={\displaystyle \frac{P_{i-}\underset{\_}{P}}{{\sigma }_P}}$$

(1)

$${STDA}_{{ET}_i}={\displaystyle \frac{{ET}_{i-}\underset{\_}{ET}}{{\sigma }_{ET}}}$$

(2)

where $P_i$ and ${ET}_i$ represent the annual precipitation and evapotranspiration values for year i, $\underset{\_}{P}$ and $\underset{\_}{ET}$ are their respective time series averages, and ${\sigma }_P$ and ${\sigma }_{ET}$ are the respective standard deviations of the time series. The cumulative Gaussian distribution functions of the standardized anomalies ${\Phi }_{0,1}\left({STDA}_{ET, P}\right)$ were used to indicate associated probability levels. Values of ${\Phi }_{0,1}\left({STDA}_{ET, P}\right)$ > 0.95 and ${\Phi }_{0,1}\left({STDA}_{ET, P}\right)<$ 0.05 indicate significantly humid and dry years, respectively.

Finally, we calculated average intra-annual dynamics of ET, NDVI, and WDI over the study period, focusing on three years with contrasting moisture conditions: 2005 (anomalously dry), 2010 (anomalously wet), and 2007 (marked by notably low actual ET values in irrigated areas relative to the long-term mean), derived from the $\left({STDA}_{ET, P}\right)$ analysis.

3.3. Spatial Patterns of Evapotranspiration Dynamics Across Land Cover Types (2003–2016) and During Years of Significant Water Constraints

Pixel-based average values of ET and WDI were calculated for the full study period (2003–2016). Years with significant negative deviation over the average were identified as experiencing strong water-limited conditions, resulting from either meteorological drought (precipitation deficits) or hydrological stress related to water management practices.

For these years, we analyzed the spatial distribution of both ET and the water deficit index (WDI) by land cover class to explore differential ecosystem responses under water-limited conditions. This analysis allowed us to identify spatially differentiated responses across land cover types and assess their vulnerability to hydrological stress.

Finally, a pixel-level correlation analysis of the daily ET values over time was performed to evaluate the temporal ET trends of the various land cover types throughout the study period. This analysis was further refined by dividing the temporal data into multi-year intervals categorized as wetter or drier periods, with separate correlation analyses conducted for each.

4. Results

4.1. Intra- and Inter-Annual Dynamics of Evapotranspiration in the Doñana Region and the Influence of Meteorological Drivers

The daily ET time series derived from the PT-JPL-thermal model revealed clearly differentiated spatio-temporal patterns determined by the various land cover classes over the period spanning from 2003 to 2016.

In natural ecosystems (Figure 2a), the maximum average monthly ET values occurred during late spring and early summer. Specifically, the wetlands reached their peak ET in May (109 mm/month), the shrublands in June–July (98 mm/month), and the coniferous forests in July (180 mm/month). In the wetlands, the WDI values in May generally remained above 0.5, except in 2005 and 2012, when significant water deficits were recorded (WDI < 0.25). Decreasing WDI values were consistently observed from spring around April–May until the end of summer in September across the study period. In the shrublands, the July WDI values averaged 0.4 but dropped below 0.29 in 2005, reflecting intensified water stress during that year.

In the irrigated areas (Figure 2b), ET consistently peaked during the irrigation season (June–August), with July showing the highest values across the series—198 mm/month in rice fields and 125 mm/month in mixed-irrigation areas. During the years 2006 to 2008, although July remained the month with the maximum ET, we found a clear reduction compared to all the years in the study period. Specifically, ET in July dropped by up to 50% in rice fields and 30% in mixed-irrigation areas, standing out clearly within the time series. These reductions were also reflected in the WDI values: while summer WDI typically exceeded 0.8, indicating low water stress, the values fell below 0.5 in rice fields during this period. A similar trend was observed in mixed-irrigation areas, where the WDI decreased from the average 0.5 to 0.35 in 2008, suggesting marked deficit conditions during these peak ET months.

Secondly, we assessed the inter-annual dynamics of evapotranspiration and the WDI. To identify specific years in which these variables presented significant variations, the daily model results were aggregated to annual values per pixel and discretized by land cover class (Figure 3).

In the natural ecosystems, the average annual values of ET over the period 2003–2016 reached 680 mm/year in the wetland, 620 mm/year in the shrubland, and 1020 mm/year in the coniferous forest. The lowest ET values were recorded in 2005, coinciding with the driest year (240 mm of annual precipitation), with 430 mm/year in the wetland and 450 mm/year in the shrubland. Persistently low ET values were also observed from 2011 to 2016, averaging 507 mm/year for the wetland and 550 mm/year for the shrubland. Regarding the WDI, the highest deficits were similarly observed in 2005 (0.25 for the wetland, 0.24 for the shrubland, and 0.49 for the coniferous forest), and during 2011–2016, when values remained below 0.33 for both the wetland and the shrubland, compared to their average annual WDI values of 0.43 and 0.35, respectively.

In the irrigated areas, the minimum ET values occurred in 2007 and 2008, reaching 605 mm/year in the rice fields and 530 mm/year in the mixed-irrigation sites, (compared to their respective long-term averages of 900 mm/year and 620 mm/year. During the period from 2007 to 2008, ET represented only 40% of potential evapotranspiration (ETp) in all the irrigated areas, and the WDI values were notably low, averaging 0.38 in the rice fields compared to a long-term average of 0.47.

The $STD{A}_P$ analysis (Figure 4) showed that the year 2005 was anomalously dry, with ${\Phi }_{0,1}\left({STDA}_P\right)$ values below $-$1.64, corresponding to a probability of occurrence P $[{STDA}_P\le -1.64]=0.05.$ The year 2010 was anomalously humid, with ${\Phi }_{0,1}\left({STDA}_P\right)$ values over 1.64, corresponding to a probability of occurrence P $[{STDA}_P\ge 1.64]=0.05$.

We then compared these precipitation anomalies with corresponding ET anomalies, disaggregated by natural ecosystems and irrigated areas, to assess the influence of precipitation variability on ET dynamics. The analysis of ${STDA}_{ET}$ revealed distinct inter-annual patterns between natural ecosystems and irrigated areas.

In natural ecosystems (Figure 4a), the ${STDA}_{ET}$ values were generally positive from 2003 to 2011, suggesting stable or above-average ET conditions during this period, except in 2005, when ${STDA}_{ET }$ dropped close to −1.64, corresponding to a substantial reduction in ET in wetlands and shrublands. From 2012 to 2016, the $STD{A}_{ET}$ values were predominantly negative, suggesting a consistent decline in ET.

In the irrigated areas (Figure 4b), the most pronounced anomalies were observed during 2007 and 2008, when rice fields showed ${STDA}_{ET }$ values below −1.64, with marked low values in actual ET relative to the long-term mean.

The $STD{A}_{ET}$ patterns in natural ecosystems significantly mirrored those of the precipitation $STD{A}_P$, as illustrated in Figure 4a. To quantify this relationship, we performed a correlation analysis between the annually aggregated ET and precipitation P, disaggregated by land cover class. The results reveal a significant correlation in natural ecosystems, with coefficients exceeding R > 0.4 in both the wetland and shrubland. Conversely, no meaningful relationship was observed in the irrigated areas, where ET dynamics appeared largely decoupled from inter-annual variations in precipitation (Figure 4b).

Finally, we calculated the average intra-annual dynamics of ET, NDVI, and WDI over the study period, focusing on three years with contrasting moisture conditions: 2005 (anomalously dry), 2010 (anomalously wet), and 2007 (marked by notably low actual ET values in irrigated areas relative to the long-term mean) (Figure 5a,b).

In all land cover types except wetlands, the beginning of the year is characterized by high WDI values, mainly due to low temperatures and limited vegetative activity, as phenological development has not yet started. As temperatures rise in late winter and early spring, vegetation begins to transpire, gradually reducing the ET deficit.

4.2. Spatial Patterns and Trends of Evapotranspiration Across Land Cover Types (2003–2016) and During Years of Significant Water Constraints

To examine the spatial dynamics of ET and WDI under water-limited conditions across irrigated areas and natural ecosystems, we first calculated the average ET per pixel for the entire study period (2003–2016) to establish a reference baseline (Figure 6b). We then analyzed the years 2005 and 2007, each characterized by distinct water constraints affecting different land systems. In 2005, the driest year of the period, with annual precipitation of 240 mm, natural ecosystems showed marked reductions in ET and lower WDI values (Figure 6c,d). In contrast, in 2007, limited water availability for irrigation, particularly in the rice fields, resulted in sharp ET declines and significantly reduced the WDI values in those zones (Figure 6e,f).

Finally, we performed a pixel-wise correlation analysis of the daily ET values over time to evaluate the temporal ET trends across various land cover types throughout the study period. The resulting correlation coefficients highlight distinct differences among the land cover categories (Figure 7b). The strongest negative correlations (R < −0.5) were concentrated in natural ecosystems, particularly within the wetlands (indicated by red frames in Figure 7b). Conversely, positive trends were predominantly found in irrigated areas, with the highest coefficients (up to R = 0.30) observed in rice fields, the central zone of the western Natural Park, and scattered patches in the northwestern sector of the National Park (highlighted by blue frames in Figure 7b).

The pixel-wise correlation was further stratified by the prolonged wetter (2005–2011) and drier (2011–2016) periods identified in Figure 4. Irrigated areas were excluded from this analysis, as previous results (Figure 4b) indicate a lack of direct correspondence between ET and precipitation in those zones. The most notable shift in ET trends between the periods occurred in the wetlands, where the correlation values changed from positive (R > 0.3) during the wetter phase (Figure 8b) to strongly negative (R < −0.5) during the drier phase (Figure 8c). Shrubland and coniferous forest areas exhibited more moderate changes in ET trends between the two periods. This analysis highlights the varying degrees of ecosystem sensitivity to precipitation variability.

5. Discussion

In this study, we analyzed spatio-temporal monthly and annual evapotranspiration (ET) time series derived from the PT-JPL-thermal model for the period 2003–2016. This timeframe was selected to maintain consistency with prior applications of the PT-JPL-thermal model in the Doñana region, which were validated against in situ data up to 2012, showing strong agreement (ρ₁ month-lag = 0.94), as reported by Moyano et al. [49]. Extending the analysis to 2016 allowed us to capture a broader range of hydro-climatic variability beyond the validation period, including notable drought years, such as 2005 and 2015, which significantly impacted the Iberian Peninsula [75,76]. Incorporating more recent years would require additional validation efforts to ensure that the model remained accurate under increasingly altered land and water use regimes.

Significant differences in the temporal dynamics of ET and WDI were observed over the 2003–2016 study period, with distinct patterns emerging across different land cover types. Intra- and inter-annual dynamics provided key information to assess the importance of the balance between energy provision and water availability for ET, as well as the impact of droughts on the ET of different land covers.

The PT-JPL-thermal model effectively captured the diverse intra-annual ET dynamics of natural ecosystems over the study period, reflecting their varying capacities to access water, as previously reported by Moyano et al. [49]. These presented peak ET in early spring and summer (Figure 5a), with an average of 96 mm/month across wetlands, shrublands, and coniferous forests. Specifically, wetlands showed the highest ET fluxes and NDVI values in spring until April–May, indicating increased importance of vegetation transpiration, coinciding with post-rainfall water availability and the onset of vegetation growth. These dynamics align with previously reported intra-annual ET patterns by Drexler et al. [77]. In shrublands, peak ET occurs slightly later at the beginning of the summer, around June–July, while the WDI reaches its minimum, indicating water availability and vegetation growth conditions. As the season transitions into summer, the NDVI values decline slightly (Figure 5a), coinciding with reduced precipitation and increasing moisture deficits driven, consistent with observations in semi-arid shrublands by Nagler et al. [78]. For coniferous forests, the maximum ET was observed later in the summer, around July–August.

The analysis of monthly ET series over the study period in the irrigated areas revealed that the intra-annual ET dynamics derived from the PT-JPL-thermal model closely followed on-site irrigation management practices, with peak values occurring in the summer months and averaging 162 mm/month across rice fields and mixed-irrigation areas (Figure 2b). In mixed-irrigated areas in particular (Figure 5b), the maximum ET, peak NDVI, and minimum WDI typically aligned between June and August in years with sufficient water availability. Although rice fields share characteristics with wetlands due to the presence of standing water, they display distinct temporal dynamics. High WDI values were observed early in the year before field flooding, followed by a decline after crop establishment. During the initial growth phase, ET increases more rapidly than the NDVI, indicating a predominance of evaporation; both variables reach peak values around July, when evaporation and transpiration jointly contribute to the total ET. This strong alignment with irrigation practices underscores the effectiveness of using apparent thermal inertia (ATI) as a proxy for soil evaporation in the PT-JPL-thermal model, where replacing relative humidity (RH) with ATI significantly improves performance in semi-arid irrigated areas with active water management. As reported by Zhang et al. [38], even when driven by in situ RH measurements, the original PT-JPL model consistently underperforms in agroecosystems, an issue particularly evident in actively managed irrigated landscapes, as also noted by Marshall et al. [37]. Our findings support the value of thermal-based indicators for capturing spatial and temporal variability in soil moisture.

Figure 4 further illustrates the relationship between observed ET and precipitation patterns from 2003 to 2016, and additionally highlights years identified as anomalously dry, wet, or exhibiting significant deviations in ET dynamics.

In particular, the year 2005 was classified as anomalously dry (Figure 4a) and coincided with one of the most severe drought episodes recorded both globally and in the Iberian Peninsula [75].

The assessment of inter-annual ET variations in natural ecosystems revealed pronounced reductions in 2005 (Figure 3a), a year marked by severe drought conditions (Figure 4a). The extent of the impact varied by vegetation type, with wetlands being the most affected. In recent decades, the wetland has shifted from relying primarily on alluvial inputs to depending mainly on rainfall [50], leading to near-desiccation of ponds during dry years [79], especially in this region where precipitation is steadily decreasing [23,80]. ET in the wetland declined by up to 55%, accompanied by persistently low NDVI values, underscoring their strong dependence on precipitation. Additionally, intra-annual dynamics (Figure 5a) reveal that in 2005, the ET peak was delayed until July, compared to the usual April–May maximum. This shift suggests a diminished contribution from vegetation transpiration during that period. Shrublands experienced a 27% reduction in ET, while coniferous forests were largely unaffected. This ecosystem maintained relatively stable ET levels throughout the year, demonstrating greater resilience to drought conditions, possibly owing to their deep-rooted water access strategies (Figure 5a) [81]. These findings align with those of Garcia et al. [82], who also identified wetlands as the most sensitive to rainfall variability, with coniferous forests showing the highest resilience.

Regarding irrigated areas, no significant ET reduction was observed in 2005, Figure 3b, as these systems are less directly dependent on precipitation. In the irrigated areas, the effects of the 2005 drought manifested with a time lag, starting from 2007 (Figure 3b and Figure 4b). Reservoir and groundwater reserves were heavily depleted in 2005, leading to insufficient water supplies for irrigation in the following hydrological years [80], and consequently causing significant ET reductions in irrigation-dependent areas. Irrigated areas exhibited notable ET decreases during the period from 2006 to 2008, with the most evaporative month, July, showing reductions up to 50% in rice fields and 30% in mixed-irrigation areas (Figure 2b and Figure 5b) when compared to the officially recorded period between 2002 to 2005 [80].

The distinct impact of the 2005 drought across different land cover types is spatially illustrated in Figure 6c,d. Reddish-orange pixels indicate areas with the lowest ET values (<600 mm/year) and highest water deficits (WDI < 0.4), most notably within the wetlands. Conversely, greenish pixels represent areas with the highest ET values (around 900 mm/year) and lowest water deficits (WDI > 0.6), primarily corresponding to rice fields and coniferous forests, which were not immediately affected by the severe precipitation decline in 2005. As shown in Figure 6e, particularly for 2007, all areas cultivated with rice presented ET values of around 600 mm, even reaching minimum values up to 400 mm, very far from the average ET of 900 mm observed during the entire study period (Figure 6b). Similar patterns were found in WDI dynamics, with average values in the rice fields of around 0.4 (Figure 6f), much higher water deficits in comparison to those years not affected by irrigation restrictions.

In contrast to 2005, the year 2010 was identified as anomalously humid, with a probability of occurrence below 5% (Figure 4a,b). Particularly in the wetlands, which are highly dependent on precipitation, the WDI is lower at the beginning of the year (Figure 5a), suggesting that ET is predominantly driven by surface evaporation rather than plant transpiration. In these ecosystems, ET peaks in April–May, concurrently with the NDVI, indicating an increase in the importance of vegetation transpiration.

The model’s sensitivity to ET variations due to precipitation constraints or irrigation restrictions is highly valuable in a context of increasing competition for water resources. Understanding how precipitation variability, together with its persistence in time and space, affects the behavior of vegetation and crops can help anticipate possible responses of ecosystems to climate-related restrictions [82] in contexts of increasing water competition.

Two marked pluri-annual periods of ET were observed in the natural ecosystems (Figure 4a), aligning with the corresponding pluri-annual periods of precipitation P. In particular, we distinguished a more humid period (2006–2011), with $STD{A}_P$> 0, and a drier period from 2011–2016, with $STD{A}_P$< 0. Focusing on discretized ET trends for the two pluri-annual periods distinguished in Figure 4, the most significant change in ET between periods corresponded indeed to the wetland, with R values decreasing from positive values of R > 0.3 over 2006 to 2011 (Figure 8b) to negative values of R < −0.5 from 2011 to 2016 (Figure 8c). This means a 35% ET reduction in the wetland between the two periods (from $\underset{\_}{ET}$ = 775 mm/year over 2006–2011 to $\underset{\_}{ET}$ = 507 over 2011–2016). Precipitation trends may partially explain ET reductions observed in the wetland. Indeed, precipitation $\underset{\_}{P}$ decreased from 576 mm/year over 2006–2011 to 487 mm/year over 2011–2016, which corresponds to a 16% reduction between the two periods, thus affecting the contributions of precipitation regimes to the wetland dynamics in the last years of the study series.

The results of the correlation analysis model over the entire period from 2003 to 2016 revealed significant positive ET trends (R > 0.30) in the rice fields and several patches in the northwest area of the National Park (Figure 7b), regions previously identified as undergoing agricultural intensification in recent years [83]. Conversely, negative ET trends were observed across the Doñana natural ecosystems, particularly pronounced in the wetland (R < −0.43). This decline is partly driven by land-use changes favoring more economically profitable activities, which have led to a drop in the aquifer’s water table, posing a threat to the conservation of these natural ecosystems [32]. Unlicensed greenhouses cover over 6000 ha in the region, extracting groundwater through illegal wells [83]. In addition, Custodio et al. [84] estimated an average annual extraction of about 90 hm³, representing 45% of the aquifer’s recharge. This over-extraction reduces groundwater contributions to streamflows feeding the wetland during summer months [83]. Particularly within the network of piezometers across the aquifer, the level of “Caracoles” in the wetland lowered from 1.4 to 2.6 m in the period of study from 2003 to 2016 [58]. Compounding this issue are projected declines in precipitation [85], which, as shown in Figure 4, have immediate adverse effects on natural ecosystems, particularly shallow-rooted wetlands.

In response to mounting challenges, the World Heritage Committee has called for urgent action to address groundwater over-extraction [86], aiming to preserve Doñana’s unique ecosystems while promoting sustainable agricultural practices. This call to action comes amid escalating conflicts between environmental conservation and economic development in the region [52], where Doñana National Park remains a critical global biodiversity hotspot [29]. In this Mediterranean region, characterized by a mosaic of natural ecosystems and irrigated agricultural lands, the PT-JPL model has proven effective in assessing ET dynamics [58] in an area of rapid greenhouse expansion surrounding the Park [29], proving to be a sound tool to inform water management decisions aiming to equilibrate natural ecosystems and irrigation use.

The modeling framework presented in this study involves several sources of uncertainty that warrant consideration. First, precipitation inputs were obtained from a single meteorological station without spatial regionalization. While this ensures consistent temporal coverage, it may not fully represent spatial variability in rainfall, which is particularly important in heterogeneous landscapes, such as the Doñana region. Second, the daily values for variables derived from the 8- and 16-day MODIS composites were interpolated, assuming that each composite value remained constant over its respective period. Although future studies could improve accuracy by integrating higher-resolution and spatially distributed meteorological data, the results presented here provide a robust representation of water use dynamics across contrasting land systems.

6. Conclusions

We assessed the evapotranspiration (ET) responses of natural ecosystems and irrigated lands in the Mediterranean Doñana region using the PT-JPL-thermal model. The daily ET spatio-temporal series revealed distinct and coherent patterns across land cover types. From 2003 to 2016, increased water use, indicated by positive ET trends (R = 0.30), was observed in rice fields and several zones in the northwestern area of Doñana National Park, corresponding to regions of intensified agriculture. In contrast, the strongest negative correlations (R < −0.5) were found in natural ecosystems, particularly within the wetland. The model clearly captured intra-annual dynamics and contrasting responses to wet and dry years, enabling robust differentiation of drought impacts and ecosystem behavior at seasonal scales. Persistent droughts had the greatest impact on wetlands, where ET and precipitation were strongly correlated (R > 0.4). In shrublands, the correlation was moderate, while in coniferous forests, it was not significant, likely due to deeper groundwater access during dry periods. The results align with independent sources, confirming the model’s reliability in tracking water use over time. By jointly analyzing natural and irrigated systems, this study offers novel, spatially explicit insights into ET dynamics in Doñana. A key methodological innovation is the use of apparent thermal inertia (ATI) instead of relative humidity (RH) to estimate soil evaporation, providing a more robust and physically grounded proxy for monitoring ET in semi-arid and irrigated landscapes. These findings significantly advance our understanding of ecohydrological processes in Doñana and support more informed sustainable water management. They reinforce the World Heritage Committee’s call by providing tools for improved monitoring, enabling more effective action to curb unsustainable water use and protect both agricultural and ecological values in this critical region.