Remote Sensing of Environment 194 (2017) 219–229 Contents lists available at ScienceDirect Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse Area-wide evapotranspiration monitoring at the crown level of a tropical mountain rain forest Brenner Silva a,⁎, Paulina Álava-Núñez a, Simone Strobl b, Erwin Beck b, Jörg Bendix a a b Faculty of Geography, Philipps University of Marburg, Deutschhausstr. 10, 35032 Marburg, Germany Bayreuth Center of Ecology and Environmental Research and Department of Plant Physiology, University of Bayreuth, Universitaetsstr. 30, 95447 Bayreuth, Germany a r t i c l e i n f o Article history: Received 5 August 2016 Received in revised form 14 March 2017 Accepted 25 March 2017 Available online xxxx Keywords: Individual trees Forest Ecosystem function Indicator Evapotranspiration Scintillometry Climate change Operational satellite High-granularity a b s t r a c t Ecosystem water regulation couples energy and water balance, depends on the integrity of the ecosystem, and responds to changes in climate. Changes in tree-water relationships in the biodiversity hotspot of the tropical Andes in southern Ecuador might be potentially observed at the level of individual trees, thus providing an efficient ecosystem monitoring method with applications in forest management and conservation at the tree and landscape levels. In this study, we combine area-average measurements from a laser scintillometer above the forest with optical satellite data at high spatial resolution to obtain area-wide evapotranspiration data. The processing of field data includes the calculation of energy storage in forest biomass and the partitioning of evapotranspiration into transpiration and evaporation. Satellite-based estimates are calibrated by using tower flux measurements and meteorological data within periods of humid and less-humid atmosphere. The annual evapotranspiration was 1316 mm, of which 1086 mm per year corresponds to the forest transpiration at the study site. Average values of 4.7 and 4.1 mm d−1 per tree crown are observed under humid and less-humid atmospheric conditions, respectively, when applying high-resolution area-wide evapotranspiration in individual crown analysis. Approximately 24% of the observed crowns show a positive monthly change in ET, and 51% of the crowns show a significant change in the daily ET, which can be considered sensitive individuals concerning water relationships. The limitations in the area-wide evapotranspiration at the crown level can be explained by considering the spectral responses of the crown individuals. The presented method can be robustly deployed in the ecological monitoring of mountain forests. © 2017 Elsevier Inc. All rights reserved. 1. Introduction Tropical mountain forests harbor a unique diversity of tree species and complex structural patterns, which have not yet been fully understood (Bendix et al., 2013). While functional traits and tree diversity were found to correlate with edaphic conditions (e.g., eastern Andes, Homeier et al., 2010), the response of vegetation to climate was mostly investigated from a broad perspective and rarely reached an individualor species-level approach (Condit et al., 2013). At the same time, land use change threatens biodiversity and modifies the environment, so the functional variability within a forest is forced to adapt or restitute the optimal natural conditions. Consequently, the conservation and management of natural forests depend on an efficient assessment of ecosystem dynamics (e.g., growth, acclimation), which in turn requires monitoring ecosystem indicators in a systematic manner across a range of scales (Lawley et al., 2016). A key indicator of an ecosystem's state and function is evapotranspiration (ET) (Nagler et al., 2009a) because ET represents both the water ⁎ Corresponding author. E-mail address: [email protected] (B. Silva). http://dx.doi.org/10.1016/j.rse.2017.03.023 0034-4257/© 2017 Elsevier Inc. All rights reserved. flow and energy balance of an ecosystem (Müller, 2005). Two consolidated methods are used to estimate ET by remote sensing. The first approach calculates ET as the residual of the surface energy balance, where the solar radiation is obtained by astronomical calculations, the ground heat flux is empirically based on the surface albedo, and the sensible heat flux depends on the radiometric surface temperature. The surface temperature requires thermal sensors, which exhibit coarse granularity and are thus applied to investigate the hydrological regulation of crop compartments or forest patches at the lowest level (Bastiaanssen et al., 1998; Gonzalez-Dugo et al., 2009). The second approach relates a vegetation index (VI) to a reference or potential ET. This reference or potential ET requires the field meteorology and assumes a known reference or water-saturated surface. One advantage of this approach is that the surface reflectance that is used to calculate the VI relies on optical sensors with high radiometric performance (Glenn et al., 2010; Yebra et al., 2013), thus reaching lower levels than forest patches. Global ET products are available at spatial resolutions of 0.05 and 1.0 degrees (Jung et al., 2010; Liu et al., 2011; Mu et al., 2011) and can be estimated at up to 100-m ground sampling (Roy et al., 2014). However, none of the available products can be used to directly extract information at the individual crown level. The large number of tree crown 220 B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 detection approaches in the literature reveals the interest of forestry and ecology communities in deploying emerging remote sensing technologies at the crown level (Chadwick and Asner, 2016; Dalponte et al., 2014; Hyyppä et al., 2008; Wulder, 1998). Common to these studies is that remote sensing improves the regionalization of biophysical and forest inventory elements (e.g., density and biomass) and that trees can be analyzed individually in terms of crown traits or their physiological response to environmental changes. Consequently, knowledge at the crown level empowers forest management institutions in high biodiversity environments by combining taxonomic and indicator-specific analyses at the landscape level. In addition to costly field and aerial campaigns (Chadwick and Asner, 2016), solid advancement in mapping spectral traits at the tree-crown level can be achieved by using currently operating satellites at high granularity (McGraw et al., 1998). To our knowledge, no research exists on monitoring evapotranspiration (ET) at the level of individual trees by satellite remote sensing. In terms of the VI-based approach, any uncertainties in area-wide ET are mainly related to the illumination geometry and radiometric processing of satellite data and to the constraints and parameterization of the potential ET model. Different VI-based approaches have been published in recent decades, with the potential ET model incorporating either only atmospheric terms or also including aerodynamic and surface resistance terms (Choudhury et al., 1994; Glenn et al., 2011; Nouri et al., 2016). Following these authors, a calibration coefficient is used to adjust the potential ET model to a specific vegetation surface to obtain the actual ET. However, none of these previous studies has been implemented at the level of individual tree crowns. The overall aim of this work is to develop a method to monitor subtle changes in canopy water relationships by operational remote sensing at the crown level for a tropical mountain forest. In addition, we identify potential sensitive tree crowns as early warning indicators for a change in ecosystem water regulation. The developed method combines stateof-the-art optical remote sensing data and recent field meteorology techniques (scintillometer). The area-wide ET at the crown level is assessed by considering the radiometric and spectral variability, the realistic meteorological input, and the effects of local climate variability. Finally, the applicability of the area-wide ET at the crown level is tested by its capacity to track climate variations over one year based on potential indicator trees. 2. Materials and methods This study's approach combined area-averaged measurements of evapotranspiration (ETSci) with a surface layer scintillometer (SLS) to calibrate a potential evapotranspiration (ETPM) that was adjusted by satellite reflectance data, thus obtaining the actual area-wide evapotranspiration (ETsat). First, the SLS measurements were processed to consider biomass energy storage as a solution for the surface energy balance above the forest. The Penman-Monteith model was used to calculate the potential evapotranspiration (ETPM), which requires parameters for the surface and atmospheric resistances. A particular case is provided by precipitation events when transpiration is suppressed while the surface resistance increases until the canopyintercepted water is released in the atmosphere. Thus, a second processing step was considered to identify canopy evaporation, i.e., data that are directly influenced by precipitation events. Then, the area-wide product (ETsat) was calculated based on the calibration coefficient and monthly averages of meteorological data (e.g., temperature and humidity). The calibration of the ETsat was demonstrated for each period (humid and less-humid) by using the daily course of flux measurements on the two closest days to the corresponding day of the year of the satellite overpass. Afterwards, the proposed calibration was analyzed along one year of measurements by considering sub-daily precipitation interference on ETsci measurements. Finally, the area-wide product ETsat for humid and less-humid periods was applied in an object-oriented analysis by using individual tree crowns. The response of each individual to a change to a drier atmosphere helped us to identify sensitive crowns or potential indicator individuals in the mountain forest. 2.1. Study site The study area (Fig. 1) is located in the Reserva Biológica San Francisco (RBSF) in South Ecuador. The RBSF is located in the San Francisco River Valley, which breaches the main eastern Andean cordillera in the ecozone of the humid tropics and has elevations between 1600 and 3140 m a.s.l. While the south-facing slopes of the valley are extensively deforested and now covered by pasture land, the north-facing slopes are covered by a pristine tropical mountain forest. The forest types in the RBSF are spatially distributed according to the topography and altitudinal gradient. Altitudes from 1900 to 2100 m a.s.l. are covered by evergreen lower mountain forest with considerably different species between valleys or lower slopes and upper slopes or ridges. The research station San Francisco (ECSF) is located at S 3° 58′ 25″ W 79° 4′ 31″ at 1850 m a.s.l., with an average temperature of 15.5 °C and annual precipitation of 2050 mm. Diverging evapotranspiration observations indicate annual evapotranspiration between 540 and 1580 mm at the ECSF (Beck et al., 2008). Higher values were obtained by measuring the inflow and outflow of three micro-catchments (954 and 1580 mm), while sap flux and gas exchange measurements showed annual evapotranspiration between 561 and 654 mm at the forest-stand level (Motzer et al., 2005). In a subsequent review, annual transpiration between 919 and 1281 mm was reported for the study site (Bruijnzeel et al., 2011). 2.2. Field measurements Two towers (30 and 36 m high) were erected in the forest to the west and east to measure the flux above the forest canopy. The western tower was located at 1975 m a.s.l., 36 m above ground, and 90 m (220° SW) from the eastern tower. The SLS-40 transmitter (Scintec AG, Germany) was set up in the western tower and the SLS-40 receiver in the eastern tower, both ~13 m above the forest canopy. The SLS is an optical instrument that uses the covariance in a dual-beam laser (wavelength = 670 nm) to estimate the dissipation rates of the measurements. The dissipation rates are used alongside the Monin-Obukhov similarity theory (MOST) to calculate atmospheric turbulence, heat and momentum flux (Nakaya et al., 2006; Odhiambo and Savage, 2009). The calculation of the latent heat, which is converted to evapotranspiration, is based on the energy balance equation and requires additional meteorological data. An automatic weather station was installed in the eastern tower to provide additional energy balance terms (net radiation and soil heat flux) alongside the air temperature, air pressure, and relative humidity. Net radiation sensors (8111; Schenk GmbH, Austria; and CNR01; Kipp & Zonen, Netherlands) were installed at 20 m above ground, 5 m above the canopy. The air temperature and relative humidity were measured with a shield protected thermometer and hygrometer (HC2S3; Campbell Sci. Inc., USA) at 20 m above ground. At the same level, the air pressure was measured with a barometric pressure sensor (61302V; RM Young, USA), the precipitation was measured with a tipping bucket rain gauge (52203; RM Young, USA), and the wind direction was measured with a wind vane (W200P; Vector Instruments, Ltd., UK). Additionally, the air temperature was measured at 25 m above ground. The soil heat flux was measured by two soil heat plates (HFP01; Campbell Sci., USA) that were installed 5 cm below the soil surface. Data from a second weather station 150 m from the eastern tower was available to analyze the soil moisture and temperature (Rollenbeck and Peters, 2009). These data were stored with two data loggers (CR1000; Campbell S-ci. Inc., USA), which used a 10-min interval. The observation towers operated for one year from March 2014 to March 2015. Previously, SLS measurements were conducted during a week in November 2013 on a pasture site at 1960 m a.s.l. at the opposite B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 221 Fig. 1. Study site near the Estación Científica San Francisco in South Ecuador at 1850 m above sea level (a.s.l.). The map in the top left shows the location of the study site between the cities of Loja and Zamora. The two towers that were constructed for scintillometer flux measurements are depicted at an elevation of 1975 m a.s.l. and are overlaid by the measurement footprint and two ancillary weather stations. slope in the valley. This site was homogeneously covered by cultivated pasture (Setaria sphacelata) and thus adequate for reference measurements. A direct comparison with gas exchange measurements at the leaf level revealed the successful applicability of SLS evapotranspiration over rugged terrain (Silva et al., 2016). The available SLS data from the precipitation-free day of November 19, 2013, which were obtained during the field campaign at the pasture site, were the closest data to the cloud-free satellite overpass on November 5, 2013 in this study (Section 2.5). Correspondingly, these data were used to assess the plausibility of area-wide evapotranspiration. 2.3. Measuring the above-forest evapotranspiration Our assessment of the scintillometer measurements in a pasture site near the RBSF, including high-frequency measurements (10-30-min interval), showed a high correlation between the scintillometer measurements and leaf-level gas exchanges. In the forest, the height of the propagation path (~ 14 m above the canopy), which was almost as high as the average tree height (16 m), offered similarly favorable conditions. Commonly, rugged terrain conditions can break the similarity principle (i.e., disruption of neutral stability or non-constant fluxes with height), which is important for SLS deployment. However, previous studies (Martins et al., 2009) identified that topographic effects are minimized under a certain height range of the propagation path in combination with the Obukhov length (z/L). Consequently, the quality of measurements could be assessed by the inner scale of the refractive index fluctuation (average, range, and number of occurrences above 2 mm). Furthermore, a breakdown of the similarity principle can be neglected for measurements that do not occur beyond very stable and unstable atmospheric conditions (Moraes et al., 2005). At the forest site, 93% of the data met this condition, where z/L averaged 7.5, with the inner scale above 2 mm in 85% of the cases. Unlike in the pasture, the tower flux measurements (ETsci) were expected to be overestimated in the forest. The storage of energy between the ground and the propagation path, i.e., the biomass storage, could not be measured at the forest site and thus was not directly included in the energy balance. Fortunately, a straightforward solution from simultaneous measurements of the stem temperature and radiation balance (Lindroth et al., 2010) could be adapted to the RBSF to consider the biomass storage. This approach consists of estimating the biomass storage with the average net radiation during the first half of the night (from 19 to 24 h), which was assumed to be linearly correlated to the daily incoming thermal radiation. The biomass storage (B) is given by B ¼ αL↓ þ β ð1Þ where L↓ is the incoming longwave radiation (W m−2). The coefficients α and β are determined after regression between the average net radiation from 19 to 24 h and the average L ↓ during the corresponding day. Consequently, the biomass storage was included in the energy balance equation to obtain the evapotranspiration (ETsci): ETsci ¼ cðQ  −G−H−BÞ ð2Þ where c is a constant to convert the energy flux (i.e., latent heat) into the water flux (c = 1.46 10−3 mm J−1); Q* is the net solar and atmospheric radiation (W m− 2); and the remaining terms correspond to the partitioning of energy into the soil heat flux (G), sensitive heat flux (H), and biomass storage flux (B) all in the same unit (W m− 2). Soil heat fluxes are commonly observed at very low values (close to zero) in tropical humid forests. Evapotranspiration measurements are also influenced by the precipitation because of either direct interference in the SLS signal or subsequent evaporation from canopy intercepted water. The latter can be considered strongly decoupled from the canopy physiology (transpiration), especially in the short term (sub-daily), and requires specific parameterization. Therefore, a partitioning approach was implemented with the canopy intercepted water as a proxy for the rapid evaporation after precipitation to identify ETsci measurements that are influenced by precipitation. A relationship was assumed between the available energy and the canopy-intercepted water for each subsequent time step (Crockford and Richardson, 2000): W ¼ min½P:I=ðP þ IÞ; W max Š−0:408 Q  ð3Þ where W is the canopy-intercepted water (mm); P is the precipitation (mm); Wmax and I are the maximum storage and average rainfall interception, respectively, which are known for the study site (I = 1.91 mm, Wmax = 8.01 mm) (Fleischbein et al., 2006); and Q* is the net solar and atmospheric radiation (W m−2), which is multiplied by 0.408 to obtain the equivalent evaporated water (mm). 2.4. Measurement footprint Footprint analysis was conducted to identify the source of the fluxes that were measured by the scintillometer near the propagation path. 222 B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 The measurement footprint is given by the relative contribution of each section of the source area to the measured flux at the propagation path (Hoedjes et al., 2007). The measurement footprint depends on the atmosphere stability, the wind vector, the instrument height, and the surface roughness (Hsieh et al., 2000). For each day of the analysis, a footprint area was delineated by considering the area within 80% of the source contribution to the flux measurement (Chasmer et al., 2011). The soil moisture was assumed not to limit the heat fluxes within the measurement footprint (Hoedjes et al., 2007) once a thick and moist organic layer was present from around 214 to 315 mm and from 27.4 to 48.7 vol% water, as measured in a nearby plot (Bendix et al., 2013). All the wind vectors were integrated during the day and used for correlation analysis between ETsat and ETsci. The measurement height (=30 m), displacement height (=14 m) and surface roughness (=2) (Cataldo and Zeballos, 2009) were used in the footprint calculation. The wind speed was assumed to be proportional to the square of the friction velocity, as already applied at the study site (Silva et al., 2016). 2.5. Remote sensing data and processing Airborne laser scanning (ALS) data were acquired between March and November 2012 with a Leica ALS50 II system that was operated by Altex Technology S.A. (Geneva, Switzerland) on board a helicopter (Eurocopter AS350B2 Ecureuil). Two flight campaigns were required to overcome cloudy weather conditions and to achieve a ground sampling density of at least ten pulses per square meter, considering steep slopes and valleys. The horizontal accuracy of the laser returns was within 5 cm (based on the control landmark of the national geodetic network, IGM). Ground reflections were detected within the collected point cloud and linearly interpolated into a digital terrain model (DTM) of one-square-meter horizontal resolution. Similarly, a Digital Surface Model (DSM) was constructed by interpolating the highest returns onto a regular grid of the same resolution. A canopy height model (CHM) was then calculated by subtracting the DSM from the DTM. The DTM was used in the geometric processing of the satellite data, while the CHM was used for the visual interpretation of individual crown segments. High-resolution satellite images were acquired from the Worldview-2 satellite after overpasses on June 16, 2012 and November 5, 2013. The former image was taken during continuous precipitation, while the latter image was taken at the end of a relative dry week, when depletions in soil water became relevant. Multispectral Worldview-2 data were calibrated at eight spectral ranges, namely, Coastal, Blue, Green, Yellow, Red, Red-Edge, NIR1, and NIR2, which were centered at 427.3, 477.9, 546.2, 607.8, 658.8, 723.7, 831.3, and 908 nm, respectively. Both images were acquired with a ground sample distance of 2.2 m and resampled onto grids of 2.5-m spatial resolution. Both overpasses had similar off-nadir angles (23.7° and 24.6°) and Sun elevation angles (63.2° and 62.1°). Despite the similar illumination geometry, geometric and radiometric corrections were conducted on these two images. The geometric correction used rational polynomial coefficients (RPC) and the DTM, and the corresponding results were assessed by visually inspecting additional control points. Subsequently, a radiometric correction was conducted to minimize the effects of the atmosphere and topography and to rectify the acquisitions. Similarly to previous works (Masek et al., 2006), an absolute atmospheric correction was conducted with the 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) model, together with reanalysis and field data. Reanalysis data for water and ozone atmospheric profiles were provided by the NOAA-ESRL Physical Sciences Division, Boulder, Colorado, which is available online at bhttp://www.esrl.noaa. gov/psd/N. A visibility sensor (VPF-710, BIRAL, Inc.) that was installed at the ECSF was used to estimate the aerosol optical depth. Afterwards, topography correction was applied to the images by calculating a Lambertian coefficient based on the relationship between the DTM and each available spectral band. In addition, the radiometric rectification of the acquisition from June 2012 was performed on the reference from November 2013 once the latter image's overpass exhibited higher visibility (=46 km). This radiometric rectification (Hall et al., 1991) applied the Kauth–Thomas (KT) transformation with specific coefficients for the Worldview-2 sensor (Yarbrough et al., 2014). Image features were selected with the KT greenness-brightness image components by allowing a variation of b0.1% between acquisitions. Subsequently, linear rectification coefficients were calculated based on the selected pseudo-invariant features and applied on the acquisition from June 2012. The radiometric correction was assessed by first using a set of random samples (n = 100) within the image's extent and then using artifacts on the landscape (i.e., samples on the road and on the roof of the research station). To avoid the effects of illumination geometry at the canopy, maximum reflectance values were applied in the object-based analysis either by using footprints to calibrate the ETsat or by using individual tree crowns for a time comparison. In the latter case, the variation between acquisitions was compared by using the Kolmogorov-Smirnov test (KS-test) for each available spectral range. The analyzed individual crowns were obtained by visually interpreting the CHM alongside ancillary aerial photography and field observations. Two interpreters worked independently in the same area, and the resulting polygons were double-checked to obtain individual tree crown objects. 2.6. Area-wide evapotranspiration According to Choudhury et al. (1994), Glenn et al. (2011), and Nouri et al. (2016), the area-wide evapotranspiration is calculated by multiplying the potential evapotranspiration (ETPM) by a spatially explicit vegetation index. This vegetation index combines the visible and nearinfrared surface reflectance into a scalar that integrates leaf pigmentation and the structural characteristics of the leaves and canopy. According to Glenn et al. (2010), vegetation indices are nearly scale invariant from leaf-level to remote sensing landscape-level measurements and thus adequate for crown-level estimates. Accordingly, the crown-level area-wide ETsat was obtained by ETsat ¼ max a ETPM EVI; Eg
ð4Þ where ETsat (mm h−1) is the actual evapotranspiration; ETPM (mm h−1) is the potential evapotranspiration from the Penman-Monteith model; the EVI (stretched to the [0, 1] range) is the enhanced vegetation index that is used to scale ETPM to the actual vegetation state; and a is a calibration coefficient, which is determined by replacing ETsat with ETsci in Eq. (4) and measuring the linear correlation between ETsat and ETsci. The linear correlation is measured daily and the calibration coefficient is aggregated monthly. ETsat is set to a minimum value from the soil evaporation (Eg) to consider evaporation from non-vegetated surfaces (i.e., EVI ≈ 0), as noted by Glenn et al. (2010) (see Eq. (7)). The evapotranspiration from the Penman-Monteith ETPM (McMahon et al., 2013) is calculated hourly and requires ancillary meteorological data. Both ETPM and EVI are given by the following equations: ETPM ¼ EVI ¼
ΔQ  þ ρa ca ea −ea =r a λðΔ þ γð1 þ r s =r a ÞÞ GðNIR−RÞ NIR þ c1 R−c2 B þ L ð5Þ ð6Þ where Δ, λ, ρa, ca, and ϒ in Eq. (5) are the parameters of the PenmanMonteith equation, and their formulations are provided in the literature (see Eq. (5) in McMahon et al., 2013). Likewise, the constants G, c1, c2, and L in Eq. (6) are provided in the literature (see Eq. (2) in Huete et al., 2002). The aerodynamic resistance (ra; s m− 1) was calculated based on the average wind velocity (Silva et al., 2016). The surface resistance (rs; s m−1) depends on the leaf area index (LAI; m2 m−2), which was set to 6 m2 m−2 for the mountain forest (Motzer et al., 2005) and B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 2.5 m2 m−2 for the pasture (Silva et al., 2016). Formulations for rs are provided in the literature (see Eqs. (2) and (8) in Silva et al., 2016), with the single leaf resistance set to 100 s m− 1 (Allen et al., 1998). The vapor pressure deficit (e∗a − ea) was calculated based on the relative humidity and air temperature. The soil evaporation (Eg) is calculated based on the water diffusion based on the gradient of the specific humidity between the topsoil and atmosphere (Oleson et al., 2010; Sakaguchi and Zeng, 2009): Eg ¼ −ρa ea −αeg ra þ rg
ð7Þ where ea. and αeg are the specific humidity in the air and topsoil pores, respectively; ρa is the air density (kg m−3); ra is the aerodynamic resistance (s m−1); and rg is the resistance (s m−1) for the diffusion of water vapor within the topsoil pores. Formulations for αeg and rg are provided in Sakaguchi and Zeng (2009). The required input parameters are the soil texture, soil temperature, and soil moisture. The soil texture is available for the study area (Ließ et al., 2012). The soil temperature was spatially estimated by using the air temperature and a linear relationship between the air and soil temperature from the daily values that were observed at the ancillary weather station. The soil water content of the top soil layer was considered to be at field capacity, which was set to 0.5 m3 m−3 according to observations for the study site and adjacent areas (Crespo et al., 2011; Knoke et al., 2014). Averages were used for the corresponding months with the highest and lowest relative air humidity (June and November, respectively) to represent humid and less-humid periods for ETsat. The adiabatic temperature and humidity lapse rates were used with the DTM to obtain temperature and relative humidity maps (Fries et al., 2012; Fries et al., 2009). The net radiation (Q*) was simulated hourly by using astronomical equations, the surface elevation, the surface albedo, and a site-specific clearness index (Bendix et al., 2010). The resulting ETsat (mm d−1) from the sum of hourly values (Eq. (4)) was referred to as EThumid for the humid and ETless-humid for the less-humid period. For calibration, the closest precipitation-free day to each satellite overpass was selected after processing the above-canopy ETsci. In addition, Eq. (4) was applied by using the entire ETsci time series and meteorological data to analyze the calibration coefficients when estimating the ETsat throughout the year. The EVI from the satellite overpasses on November 2013 and June 2012 was assumed constant within the corresponding periods, namely, the less-humid period from September to March and the humid period from April to August. 223 3. Results The results are presented after processing the SLS measurements to consider the biomass storage and identify measurements that were influenced by precipitation. The spatial and radiometric accuracy results are followed by evapotranspiration maps for the months of the satellite overpasses, namely, June and November. Then, the calibration results are shown with hourly data after selecting a precipitation-free daily course of SLS measurements closest to the day of the satellite overpass. Afterwards, the calibration results after testing the measurements throughout the year are presented to explain the influence of precipitation on the ETsat estimates. Finally, the area-wide ETsat results are summarized for individual tree crowns and the potential of indicator trees is provided. 3.1. Above-forest evapotranspiration The coefficients to calculate the biomass storage in Eq. (1) were α = 1.1 and β = −437, which resulted in an average of 12.4 (maximum of 51.7) W m−2 of stored energy in the forest biomass. After subtracting the biomass storage, the ETsci values dropped by around 32%, with a median, average, and maxima from 0.24, 0.32, and 2.4 mm h− 1 to 0.20, 0.27, and 2.0 mm h−1, respectively. The evapotranspiration measurements (ETsci) that were directly influenced by precipitation were characterized by a high positive deviation between ETPM and ETsci. Such conditions were mostly found under low radiation conditions during or after precipitation events. A direct influence from precipitation was found in 54% of the data. For data that were not directly influenced by precipitation, a good relationship was observed between ETsci and the net radiation (r2 = 0.75, p-value b 0.0001), with a ETPM-to-ETsci ratio that was larger than 1 in 88% of the cases. 3.2. Area-wide evapotranspiration at the crown level The accuracy of the area-wide evapotranspiration (ETsat) depended on the satellite data processing. The geometric correction resulted in horizontal position accuracies with a root mean square error (RMSE) of 1.2 m. Furthermore, the radiometric accuracy within the image's extent (n = 100) revealed a variation in the reflectance between 2% (median) and 11% (maximum) in the visible spectrum and between 1% (median) and 17% (maximum) in the near-infrared spectrum. When using the roof of the ECSF (n = 27), the maximum variation in the reflectance was 6% and 3% for the visible and near-infrared spectra, respectively. When using samples from road features (n = 88), the Fig. 2. Evapotranspiration maps of the average meteorological input for the (a) humid and (b) less-humid months (June and November, respectively). The main image features and vegetation cover are labeled. The flux towers are marked by plus signs and the scintillometer path by a red line. The coordinates are in UTM/WGS 84. 224 B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 Fig. 3. Satellite evapotranspiration (ETsat) (a) estimated at 12:00 for the satellite image on November 05, 2013; with meteorological forcing (b) at the pasture site during the measurements on November 11, 2013; and with meteorological forcing (c) at the flux towers in the forest site on November 19, 2014. The area-wide estimates are overlaid by the footprint analysis (gray to red colors), and the footprint area is delineated at 80% of the flux (black contours) of each propagation path (red line). maximum variation in the reflectance was 5% for the visible spectrum and 2% for the near-infrared spectrum. Hence, the spatial and radiometric levels of the accuracy were considered adequate for analysis at 2.5-m horizontal resolution with a radiometric deviation of b 6% based on the known image features. The ETsat maps that were produced after adding the meteorological forcing to the area-wide ETsat estimations (radiation, temperature, and humidity) are presented for the months of June (humid) and November (less humid) (Fig. 2). The ETsat during the humid period (Fig. 2a) showed lower average values and less contrast within the slopes that were covered by pasture (northern section) or within the forest (southern section) compared to those during the less-humid period. High spatial frequency was observed for the ETsat values in the forest during both the humid and less-humid months. Markedly lower ETsat values were observed in the pasture during the humid month compared to the forest. The pasture patches also showed values N1.5 mm d−1 during the less-humid month, although these values were predominantly lower than 1 mm d−1 (Fig. 2b). Higher spatial frequency with both very low (b 0.5 mm d−1) and very high (N1.5 mm d− 1) ETsat values were observed in the forest, especially during the less-humid month. Generally, a larger difference was observed between the pasture and forest during the humid month, while high spatial frequency was observed within both the pasture and forest sites during the less-humid month. Fig. 4 shows the validation and calibration results when comparing ETsat to ETsci. Fig. 4a and b show the day course and the scatterplot between the average ETsat within the footprint and ETsci along the day (November 19, 2013) at the pasture site, respectively. Fig. 4c and d show the corresponding daily course and scatterplot for the day (November 19, 2014) at the forest site. In terms of the daily sums, ETsat was underestimated compared to ETsci (as shown by the linear regression compared to the 1:1 relationship in Fig. 4b and d) and should be leveled by a factor of 1.77 and 1.43 for the pasture and forest sites, respectively. A very good agreement was observed in both cases for the daily course, with a coefficient of determination of r2 = 0.99 and r2 = 3.3. Comparison of the area-wide evapotranspiration with a scintillometer The SLS observation closest to the day of the satellite overpass was considered for validation (first at the pasture site). Then, additional SLS measurements were used to calibrate the area-wide evapotranspiration (ETsat) by assuming a constant EVI within the humid or lesshumid period. Fig. 3 shows the estimated area-wide evapotranspiration at noon during the satellite overpass (November 5, 2013) and the corresponding footprints (% of contribution to the measured signal per m2 ground) for the pasture (November 19, 2013) and forest (November 19, 2014) sites. The ETsat values at noon within the footprint of the pasture site were slightly higher, with an average of 0.41 (σ = 0.06) mm h−1 compared to an average of 0.39 (σ = 0.05) mm h−1 in the forest site. At the pasture site, the footprint of the day covered the eastern side up to around 35 m from the propagation path. The annual predominant wind direction in the valley (easterlies) was also reflected in the footprint at the forest site, where the footprint area stretched up to around 125 m from the propagation path at its easternmost point. Fig. 4. Calibration of the satellite evapotranspiration with hourly scintillometer data (ETsci) (a, c, in black) and satellite estimates (ETsat) (a, c, in blue). The latter includes the maximum (blue line), mean and minimum values (traced lines) within the corresponding footprint. Data from clear sky days were used for (a, b) the pasture (November 19, 2013) and (c, d) the forest site (November 19, 2014), and the linear correlation (blue line in b, d) was measured for those days. The one-to-one relationship is shown in red (traced line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 0.98 for the pasture and forest sites, respectively. Interestingly, the ETsci values in the forest site (Fig. 4c and d) dropped at around 17:00 to values close to zero (Fig. 4c), but ETsat did not follow this drop, which produced overestimated ETsat compared to ETsci during the late afternoon on clear-sky and relatively dry days (15% of cases, when the daily ETsci was higher than 4 mm d−1). In contrast to the values at noon (Fig. 3), the daily sum at the pasture site (Fig. 4a) was lower for both ETsci (5 mm d− 1) and ETsat (2.8 mm d− 1) compared to that at the forest site (Fig. 4c; 5.3 and 3.8 mm d−1, respectively). Considerably lower values were observed for ETsat during the humid period on overcast days. Fig. 5a and b show the ETsat values at noon and the footprint when using the calibration day of July 1, 2014, respectively. Importantly, measurements that were not directly influenced by precipitation were rare during the humid season. The closest day to the satellite overpass was July 1, 2014 for 25% of the days with more than five observed hours. The validation that was applied to this day illustrates what occurs during a humid period. Fig. 5c and d show lower ETsci (3.1 mm d−1) and ETsat (2.1 mm d−1) compared to the lesshumid period. The larger footprint extent on July 1, 2014 is caused by lower wind speed, with a median and maximum value of 0.68 and 0.81 compared to 0.95 and 1.2 for November 19, 2014. The coefficient of determination was r2 = 0.91 and the calibration coefficient (a) was 1.55 (Fig. 5d). The occurrence of a short rainfall event at 14:00 in Fig. 5c led to an input of canopy water that evaporated in the following hours. The observation at the precipitation time is labeled (gray dot) in Fig. 5d. Our analysis of the daily values throughout the year revealed variations in the coefficient of determination for Eq. (4). Fig. 6 shows the monthly aggregated daily results, where Fig. 6a includes all the data and Fig. 6b omits data that were influenced by precipitation (i.e., high canopy evaporation). The coefficient of determination was higher during the less-humid periods, and the statistical significance was lower 225 during the humid period (high p-values). The statistical significance decreased because of precipitation events, which also created very high daily calibration coefficients. Consequently, the median value was used as a robust representation of the monthly calibration coefficient. Fig. 6a reveals slightly higher monthly calibration coefficients during the humid (gray background) period compared to the less-humid periods. When using all the data, the calibration coefficient was 2.12 and ranged from 1.15 to 3.18. The averages for the periods J-M, A-A, and SD were 2.11, 2.79, and 1.78, respectively. When omitting data that were directly influenced by precipitation (Fig. 6b), the calibration coefficient decreased to 1.75 and ranged between 1.45 and 2.06. In this case, the calibration coefficients for the periods J-M, A-A, and S-D were 1.83, 1.69, and 1.63 respectively. Fig. 6b shows that the lowest coefficient of determination (r2) increased from 0.72 to 0.88 when omitting data that were directly influenced by precipitation. 3.4. Change in evapotranspiration per individual tree crown Fig. 7a shows the measurement footprint for the year of SLS measurements ranked by the monthly ETsat. Within this footprint, 24% of the crowns showed a positive change in monthly ETsat; values of 4.7 (σ = 0.52) and 4.1 (σ = 0.69) mm d−1 were observed for the humid and less-humid periods, respectively. The transpiration component of ETsat was estimated to be 2.8 and 3.8 mm d−1 for the humid and lesshumid periods, respectively. The average ETsat that was assigned to the crowns was 2.5 (σ = 0.13) (for July 1, 2014) and 4.2 (σ = 0.41) mm d−1 (for November 19, 2014) when considering the selected days within the humid and less-humid periods. Fig. 7b shows the change in the spectral signature between the acquisitions during the humid (left boxes) and less-humid (right boxes on gray background) periods. A relative drop in reflectance was observed in the blue, green and red spectral ranges. The reflectance also tended to be lower in the red-edge and near-infrared (NIR1 and NIR2) ranges. When using the KS-test, a significant decay was observed in the green (KS-value = 0.7) and blue ranges (KS-value = 0.6). The Red and NIR1 ranges both presented a KS value of 0.5. Other ranges showed KS values that were lower than 0.5. For all the tests, the p-value was lower than 0.001. The main implication of these changes in the EVI could be analyzed in the ETsat products. According to the probability distributions, a slightly lower mean value was observed during the less-humid period (EVI = 0.56) compared to the humid period (EVI = 0.58). The humid period revealed a narrower probability distribution function with a minimum and maximum of 0.51 and 0.66 during the humid period compared to 0.2 and 0.79 during the less-humid period. The higher EVI variability during the less-humid period was related to diverging changes in the spectral properties and thus the ETsat of the crowns as a response to a drier atmosphere. 4. Discussion 4.1. Above-canopy measurements Fig. 5. Calibration of the satellite evapotranspiration (ETsat) for (a) the overpass on June 12, 2012 in the humid period at 12:00, with the footprint area (b) obtained by using the wind vector in the same period (July 01, 2014). Data from an overcast day were used, including (c) scintillometer measurements (ETsci) (black line) and satellite estimates (ETsat) within the footprint (blue line for the maximum and traced blue lines for the mean and minimum values) and for a short precipitation event (cyan). The linear correlation between ETsat and ETsci (d, blue line) lies above the one-to-one relationship (red traced line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) The biomass heat storage and the influence from precipitation were incorporated into the SLS evapotranspiration measurements. A considerable drop (45%) in the average values was observed, while the amount of biomass storage (~12 W m−2) closely matched the literature value (Lindroth et al., 2010). As expected, ETsci rarely outpaced ETPM. One reason was that the potential ETPM was calculated by using a single parameterization for the plant area index. The plant area index (PAI) is analogous to the leaf area index (LAI) but includes trunks, stems, and branches and can be estimated in the field by the gap-fraction method, among other methods. The green leaf area index was certainly lower than the surface area index, which made ETPM a potentially valid value for a 100% green surface area. Further investigation is required to analyze the influence of different leaf area index values or canopy models. For instance, the LAI or a specific parameterization could be 226 B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 Fig. 6. Regression results throughout one year, including the monthly “calibration coefficient” (blue connected dots), “determination coefficient” (red dots), and “p-value” (vertical bars). The satellite overpasses on June 12, 2012 and November 05, 2013 were considered for the humid period (gray background) and less-humid period, respectively. The results are presented for (a) all the data and (b) omitting direct influence from precipitation. The linear relationship for the number of days is shown for each month above each figure. The p-value of 1 × 10−4 is shown (traced line) for orientation. The meteorological data were from April to December 2014 and from January to March 2015. incorporated into the partitioning approach, where ETsci measurements were identified to strongly deviate from ETPM because of additional canopy evaporation. 4.2. Image processing The image processing results indicated good-quality multi-spectral data and resulting evapotranspiration products. For instance, literature that studied the same satellite data showed a horizontal accuracy between 0.4 and 2.2 m (Aguilar et al., 2014). A horizontal accuracy of 1.2 m could be considered very good and appropriate for analysis with a spatial sampling of 2.5 m because of the scarcity of quality ground control points in a surface that is dominated by vegetation. After radiometric rectification, the variations in the visible and near-infrared spectra indicated an uncertainty of 9% in the EVI according to the known image features. When using random samples, variation sources other than leaf pigmentation and structure, which were considered in the EVI, could have also been included in the 11% variation in the visible range and 17% variation in the near-infrared range. For instance, the lower variation in the visible range can be explained by the similar illumination geometry and shadow length within the canopy, both of which are known to lower reflectance (Koslowsky, 1993). The higher Fig. 7. Spatial distribution of (a) individual tree crowns and corresponding ETsat change ratios, where values above one (bluish colors) indicate a positive change (yellow for negative) from humid to less-humid periods (ETless-humid/EThumid). The rectified spectral reflectance of each available band is provided in (b) for the overpasses on June 12, 2012 (humid period, marked in gray) and November 5, 2013 (less-humid period, white background). B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 variation in the near-infrared range could be attributed to the vegetation phenology (i.e., senescent crowns or understories), including bright patches of dense foliage and areas of secondary vegetation in the pasture. The higher spatial frequency that was observed in the EVI for the mountain forest and thus in ETsat was mostly caused by forest gaps, canopy shadows or healthy leafy treetops. If isolated, the resulting uncertainty in the near-infrared reflectance could be explained by directional effects. Future investigations should combine a canopy model and directional canopy spectroscopy. 4.3. Area-wide evapotranspiration The high correlations for the pasture and forest sites showed optimal performance under clear-sky conditions. The ETsat estimation was easier for the pasture, which did not require calculating the biomass storage. In addition to the canopy storage, a partitioning approach was used to represent canopy evaporation in the forest site. ETsat could be calibrated for canopy transpiration and evapotranspiration after analyzing the scintillometer measurements (ETsci) in the forest. A calibration coefficient of 1.75 was suggested to represent the transpiration component of the evapotranspiration in the forest site. The calibration coefficient was 2.12 when including data after precipitation events; thus, ca. 21% of the ETsat values were associated with canopy evaporation. When using the year of measurements and the partitioning approach, the transpiration component of the evapotranspiration was estimated to be 1086 mm per year, which was around 82% of the total evapotranspiration (i.e., including canopy evaporation) of 1316 mm per year. These values match those in previous investigations for the study site, where values between 561 and 1281 mm per year (Bruijnzeel et al., 2011; Motzer et al., 2005) were found for canopy transpiration when using estimates at the tree and forest-stand levels. Higher values between 1310 and 1580 mm per year (Wilcke et al., 2008) were found by measuring the water balance at the micro-catchment level, which correspond to the total evapotranspiration at the study site. In addition, the estimated evaporation of approximately 21% of the total evapotranspiration fell within the range for rainfall interception loss in the study area, which was between 15 and 42% for the sites nearest the flux towers (Fleischbein et al., 2005). The assumption of a constant vegetation index, which was used as a crop factor in ETsat, remained valid during all hours of the day for the pasture. The same trend held for the forest site, although a divergence between ETsat and ETsci was observed during the late afternoon because of contrasting higher vapor pressure deficits when the net radiation declined. Both variables are equally important for ETsat when using the Penman-Monteith model, where the wind speed can also maximize the above-mentioned contrast between ETsat and ETsci. Another point for further investigation is the stomatal control, which could occur because of photo-inhibition under dry and sunny days and was hitherto not considered in ETsat by assuming a constant EVI. For now, the calibration in the forest was observed to increase with the water input from precipitation or dense fog. This relationship especially held for the beginning of the year, where unset conditions predominated. In particular, overcast and rainy days during the humid period produced low ETsci and ETsat, which in turn increased the uncertainty of the calibration coefficients. In addition, the uncertainty of the meteorological data could be seen in the instruments that were used to calculate ETPM. The net radiation matched 10%, the temperature 1%, and the humidity and wind 2% of the observed values. A value of 6% of the hourly values was calculated by applying this deviation to ETPM, where ETPM N 0.1 mm h−1. A 15% uncertainty could be assigned to the ETsat by adding the uncertainties in the EVI. Our method is intended to support the operational monitoring of tropical mountain forests. This method is based on two techniques: VI-based satellite ET estimation (e.g., with the Penman-Monteith model) and tower flux measurements (e.g., with a laser scintillometer). Other techniques (e.g., eddy covariance) that are supported by the 227 literature (Hoedjes et al., 2007) might be easily incorporated for calibration/validation with existing data in a broader spatial context. Uncertainty in ETsat at the crown level could occur, for instance, because of forest canopy traits (e.g., phenology or physiology) or mechanisms that could break the assumption of a constant EVI or its direct relationship to ET, thus requiring the fine-tuning of ET estimations. Further research is recommended to explain the variability in the calibration coefficient with time and among different locations. For instance, canopy shadows should increase the calibration coefficient, which could explain the higher calibration coefficient in this study (a = 2.12) compared to the actual ET (a = 1.22) for crops and riparian vegetation (Nagler et al., 2009b). Although our results should not directly hold for different ecosystems, our work can be considered within the domain of tropical mountain forests, which cover more than one million square kilometers worldwide (Spracklen and Righelato, 2014). In addition, our method could be applied to seasonal recurring calibration within the scope of operational monitoring based on high-resolution sensors and weather stations. 4.4. Analysis at the crown level The advantage of high-resolution ETsat lies in the possibility of analysis at the crown level. Observed variations in the near-infrared range can be attributed to the canopy phenology by assessing the changes between humid and less-humid periods at the crown level, i.e., changes in the leaf structure from low moisture in the soil and air. At the same time, a general decay in the visible range (green, blue, and red) showed no impairment in the pigmentation. A gradient of the change between humid and less-humid periods (ETless-humid/EThumid) could be identified within the footprint area after crown-based analysis. A trend towards higher evapotranspiration was observed in around 21% of the crowns, which should match the regulation of a forest ecosystem in a drier atmosphere. Extreme cases with more or less sensitive tree individuals could also be identified with our approach. This method could be easily used to find trees that react to droughts. For instance, the daily analysis per individual crown showed a bimodal distribution in the ET change ratio (ETless-humid/EThumid), where 1.67 and 1.80 were the most common values. Accordingly, 51% of the trees responded to a drier atmosphere by a factor higher than 1.80 and could be considered sensitive individuals in terms of water relationships. However, further research is required to consider the deployment of additional techniques to investigate if the link between individual tree water relationships (e.g., measured by sap flux and gas exchange sensors) is coupled with remote sensing estimates and to determine which topographic or biotic factors are relevant to water regulation mechanisms (e.g., forest composition or tree density). The species- or trait-specific parameterization of mechanistic models can also be used with assistance from ground inventories to support the selective management of individual tree species or groups of trees where ecosystem water regulation should be maintained. This suggestion especially holds for tropical forests, where the tree diversity was reported to range from 15 to 42 species per 50 stems (Poorter et al., 2015). Furthermore, crown-level ET analysis could reveal individual trees as bio-indicators of changes in ecosystem water regulation. 5. Conclusions The area-wide evapotranspiration of a mountain forest was successfully estimated at high spatial resolution. A concise set of techniques was required to automatically process field and satellite data. The spectral canopy's properties were reflected in the evapotranspiration product and could be applied to ET estimations either in a continuous landscape or on an individual crown basis. The latter provided an individual-based analysis that elucidated the gradient of individual tree responses to a drier atmosphere, where the potential use of tree crowns as 228 B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 indicators of changes in water relationships could be described. This study showed the calibration of area-wide ET estimates from the perspective of an operational application, which can be applied in future monitoring with high-resolution satellite data. Requirements included continuity in the meteorological data and satellite acquisitions. The latter should benefit from forthcoming technical developments, such as improvements in radiometric, spectral and spatial resolutions. Acknowledgements This work was financially supported by the German Research Foundation (DFG) [grant number BE 1780/38-1] within the Platform for Biodiversity and Ecosystem Monitoring and Research in South Ecuador [DFG PAK 823-825]. Logistic support by the foundation Naturaleza y Cultura Internaciónal, Loja/San Diego is gratefully acknowledged. We thank J. Zeillinger alongside Y. Svoiskiy and his crew for organizing and executing the laser scanning of the study site. We thank J. Schmidt and K. Vaivode, who helped with the interpretation of the tree crowns and validation of the geometric processing. We thank our colleagues and helpers during the construction and operation of the flux towers. We thank three anonymous reviewers whose comments helped improve the manuscript. Finally, we thank the Ministerio del Ambiente for the research permission No. 021-2013-DPL-MA and No. 010-IC-FAU/ FLO-DPZCH-MA. References Aguilar, M.A., del Mar Saldaña, M., Aguilar, F.J., Lorca, A.G., 2014. Comparing geometric and radiometric information from GeoEye-1 and WorldView-2 multispectral imagery. Eur. J. Remote Sens. 47, 717–738. Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop Evapotranspiration - Guidelines for Computing Crop Water Requirements - FAO Irrigation and Drainage Paper 56. Irrigation and Drainage. Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A., Holtslag, A.A.M., 1998. A remote sensing surface energy balance algorithm for land (SEBAL): 1. Formulation. J. Hydrol. 212– 213, 198–212. Beck, E., Bendix, J., Kottke, I., Makeschin, F., Mosandl, R. (Eds.), 2008. Gradients in a Tropical Mountain Ecosystem of Ecuador. Springer-Verlag, Berlin Heidelberg. Bendix, J., Silva, B., Roos, K., Göttlicher, D.O., Rollenbeck, R., Nauß, T., Beck, E., 2010. Model parameterization to simulate and compare the PAR absorption potential of two competing plant species. Int. J. Biometeorol. 54, 283–295. Bendix, J., Beck, E., Bräuning, A., Makeschin, F., Mosandl, R., Scheu, S., Wilcke, W. (Eds.), 2013. Ecosystem Services, Biodiversity and Environmental Change in a Tropical Mountain Ecosystem of South Ecuador, first ed. Springer-Verlag, Berlin Heidelberg. Bruijnzeel, L.A., Mulligan, M., Scatena, F.N., 2011. Hydrometeorology of tropical montane cloud forests: Emerging patterns. Hydrol. Process. 25, 465–498. Cataldo, J., Zeballos, M., 2009. Roughness terrain consideration in a wind interpolation numerical model. Proceedings of the 11th Am. Conf. Wind Eng. (AWCE). San Juan, Puerto Rico. Chadwick, D.K., Asner, P.G., 2016. Organismic-scale remote sensing of canopy foliar traits in lowland tropical forests. Remote Sens. 8 (2), 87. Chasmer, L., Kljun, N., Hopkinson, C., Brown, S., Milne, T., Giroux, K., Barr, A., Devito, K., Creed, I., Petrone, R., 2011. Characterizing vegetation structural and topographic characteristics sampled by eddy covariance within two mature aspen stands using lidar and a flux footprint model: scaling to MODIS. J. Geophys. Res. Biogeosci. 116, G02026. Choudhury, B.J., Ahmed, N.U., Idso, S.B., Reginato, R.J., Daughtry, C.S.T., 1994. Relations between evaporation coefficients and vegetation indices studied by model simulations. Remote Sens. Environ. 50, 1–17. Condit, R., Engelbrecht, B.M.J., Pino, D., Pérez, R., Turner, B.L., 2013. Species distributions in response to individual soil nutrients and seasonal drought across a community of tropical trees. Proc. Natl. Acad. Sci. U. S. A. 110, 5064–5068. Crespo, P.J., Feyen, J., Buytaert, W., Bücker, A., Breuer, L., Frede, H.-G., Ramírez, M., 2011. Identifying controls of the rainfall–runoff response of small catchments in the tropical Andes (Ecuador). J. Hydrol. 407, 164–174. Crockford, R.H., Richardson, D.P., 2000. Partitioning of rainfall into throughfall, stemflow and interception: effect of forest type, ground cover and climate. Hydrol. Process. 14, 2903–2920. Dalponte, M., Ørka, H.O., Ene, L.T., Gobakken, T., Næsset, E., 2014. Tree crown delineation and tree species classification in boreal forests using hyperspectral and ALS data. Remote Sens. Environ. 140, 306–317. Fleischbein, K., Wilcke, W., Goller, R., Boy, J., Valarezo, C., Zech, W., Knoblich, K., 2005. Rainfall interception in a lower montane forest in Ecuador: effects of canopy properties. Hydrol. Process. 19, 1355–1371. Fleischbein, K., Wilcke, W., Valarezo, C., Zech, W., Knoblich, K., 2006. Water budgets of three small catchments under montane forest in Ecuador: experimental and modelling approach. Hydrol. Process. 20, 2491–2507. Fries, A., Rollenbeck, R., Goettlicher, D., Nauss, T., Homeier, J., Peters, T., Bendix, J., 2009. Thermal structure of a megadiverse Andean mountain ecosystem in Southern Ecuador and its regionalization. Erdkunde 63, 321–335. Fries, A., Rollenbeck, R., Nauss, T., Peters, T., Bendix, J., 2012. Near surface air humidity in a megadiverse Andean mountain ecosystem of southern Ecuador and its regionalization. Agric. For. Meteorol. 152, 17–30. Glenn, E.P., Nagler, P.L., Huete, A.R., 2010. Vegetation index methods for estimating evapotranspiration by remote sensing. Surv. Geophys. 31, 531–555. Glenn, E.P., Neale, C.M.U., Hunsaker, D.J., Nagler, P.L., 2011. Vegetation index-based crop coefficients to estimate evapotranspiration by remote sensing in agricultural and natural ecosystems. Hydrol. Process. 25, 4050–4062. Gonzalez-Dugo, M.P., Neale, C.M.U., Mateos, L., Kustas, W.P., Prueger, J.H., Anderson, M.C., Li, F., 2009. A comparison of operational remote sensing-based models for estimating crop evapotranspiration. Agric. For. Meteorol. 149, 1843–1853. Hall, F.G., Strebel, D.E., Nickeson, J.E., Goetz, S.J., 1991. Radiometric rectification: Toward a common radiometric response among multidate, multisensor images. Remote Sens. Environ. 35, 11–27. Hoedjes, J.C.B., Chehbouni, A., Ezzahar, J., Escadafal, R., De Bruin, H.A.R., 2007. Comparison of large aperture Scintillometer and Eddy covariance measurements: can thermal infrared data be used to capture footprint-induced differences? J. Hydrometeorol. 8, 144–159. Homeier, J., Breckle, S., Günter, S., Rollenbeck, R., Leuschner, C., 2010. Tree diversity, forest structure and productivity along altitudinal and topographical gradients in a speciesrich Ecuadorian Montane Rain Forest. Biotropica 42, 140–148. Hsieh, C.I., Katul, G., Chi, T.W., 2000. An approximate analytical model for footprint estimation of scalar fluxes in thermally stratified atmospheric flows. Adv. Water Resour. 23, 765–772. Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83, 195–213. Hyyppä, J., Hyyppä, H., Leckie, D., Gougeon, F., Yu, X., Maltamo, M., 2008. Review of methods of small-footprint airborne laser scanning for extracting forest inventory data in boreal forests. Int. J. Remote Sens. 29, 1339–1366. Jung, M., Reichstein, M., Ciais, P., Seneviratne, S.I., Sheffield, J., Goulden, M.L., Bonan, G., Cescatti, A., Chen, J., de Jeu, R., Dolman, A.J., Eugster, W., Gerten, D., Gianelle, D., Gobron, N., Heinke, J., Kimball, J., Law, B.E., Montagnani, L., Mu, Q., Mueller, B., Oleson, K., Papale, D., Richardson, A.D., Roupsard, O., Running, S., Tomelleri, E., Viovy, N., Weber, U., Williams, C., Wood, E., Zaehle, S., Zhang, K., 2010. Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature 467, 951–954. Knoke, T., Bendix, J., Pohle, P., Hamer, U., Hildebrandt, P., Roos, K., Gerique, A., Sandoval, M.L., Breuer, L., Tischer, A., Silva, B., Calvas, B., Aguirre, N., Castro, L.M., Windhorst, D., Weber, M., Stimm, B., Guenter, S., Palomeque, X., Mora, J., Mosandl, R., Beck, E., 2014. Afforestation or intense pasturing improve the ecological and economic value of abandoned tropical farmlands. Nat. Commun. 5, 5612. Koslowsky, D., 1993. The Influence of Viewing Geometry on Annual Variations of NDVI. Proceedings of the Int. Geosci. Remote Sens. SympIGARSS '93, Tokyo, Japan, pp. 1140–1142. Lawley, V., Lewis, M., Clarke, K., Ostendorf, B., 2016. Site-based and remote sensing methods for monitoring indicators of vegetation condition: an Australian review. Ecol. Indic. 60, 1273–1283. Ließ, M., Glaser, B., Huwe, B., 2012. Uncertainty in the spatial prediction of soil texture: comparison of regression tree and random forest models. Geoderma 170, 70–79. Lindroth, A., Mölder, M., Lagergren, F., 2010. Heat storage in forest biomass improves energy balance closure. Biogeosciences 7, 301–313. Liu, W., Hong, Y., Khan, S., Huang, M., Grout, T., Adhikari, P., 2011. Evaluation of global daily reference ET using Oklahoma's environmental monitoring network—MESONET. Water Resour. Manag. 25, 1601–1613. Martins, C.A., Moraes, O.L.L., Acevedo, O.C., Degrazia, G.A., 2009. Turbulence intensity parameters over a very complex terrain. Bound.-Layer Meteorol. 133, 35–45. Masek, J.G., Vermote, E.F., Saleous, N.E., Wolfe, R., Hall, F.G., Huemmrich, K.F., Gao, F., Kutler, J., Lim, T., 2006. A Landsat surface reflectance dataset for North America, 1990–2000. IEEE Geosci. Remote Sens. Lett. 3, 68–72. McGraw, J.B., Warner, T.A., Key, T.L., Robert Lamar, W., 1998. High spatial resolution remote sensing of forest trees. Trends Ecol. Evol. 13, 300–301. McMahon, T.A., Peel, M.C., Lowe, L., Srikanthan, R., McVicar, T.R., 2013. Estimating actual, potential, reference crop and pan evaporation using standard meteorological data: a pragmatic synthesis. Hydrol. Earth Syst. Sci. 17, 1331–1363. Moraes, O.L.L., Acevedo, O.C., Degrazia, G.A., Anfossi, D., Da Silva, R., Anabor, V., 2005. Surface layer turbulence parameters over a complex terrain. Atmos. Environ. 39, 3103–3112. Motzer, T., Munz, N., Küppers, M., Schmitt, D., Anhuf, D., 2005. Stomatal conductance, transpiration and sap flow of tropical montane rain forest trees in the southern Ecuadorian Andes. Tree Physiol. 25, 1283–1293. Mu, Q., Zhao, M., Running, S.W., 2011. Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sens. Environ. 115, 1781–1800. Müller, F., 2005. Indicating ecosystem and landscape organisation. Ecol. Indic. 5, 280–294. Nagler, P.L., Glenn, E.P., Hinojosa-Huerta, O., 2009a. Synthesis of ground and remote sensing data for monitoring ecosystem functions in the Colorado River Delta, Mexico. Remote Sens. Environ. 113, 1473–1485. Nagler, P.L., Morino, K., Murray, R.S., Osterberg, J., Glenn, E.P., 2009b. An empirical algorithm for estimating agricultural and riparian evapotranspiration using MODIS enhanced vegetation index and ground measurements of ET. I. Description of method. Remote Sens. 1, 1273–1297. B. Silva et al. / Remote Sensing of Environment 194 (2017) 219–229 Nakaya, K., Suzuki, C., Kobayashi, T., Ikeda, H., Yasuike, S., 2006. Application of a displacedbeam small aperture scintillometer to a deciduous forest under unstable atmospheric conditions. Agric. For. Meteorol. 136, 45–55. Nouri, H., Glenn, E., Beecham, S., Chavoshi Boroujeni, S., Sutton, P., Alaghmand, S., Noori, B., Nagler, P., 2016. Comparing three approaches of evapotranspiration estimation in mixed urban vegetation: field-based, remote sensing-based and observationalbased methods. Remote Sens. 8, 492. Odhiambo, G.O., Savage, M.J., 2009. Surface layer scintillometry for estimating the sensible heat flux component of the surface energy balance. S. Afr. J. Sci. 105, 208–216. Oleson, K.W., Lawrence, D.M., Gordon, B., Flanner, M.G., Kluzek, E., Peter, J., Levis, S., Swenson, S.C., Thornton, E., Dai, A., Decker, M., Dickinson, R., Feddema, J., Heald, C.L., Lamarque, J., Niu, G., Qian, T., Running, S., Sakaguchi, K., Slater, A., Stöckli, R., Wang, A., Yang, L., Zeng, X., Zeng, X., 2010. Technical Description of version 4.0 of the Community Land Model (CLM). NCAR Technical Note NCAR/TN-478+STR. National Center for Atmospheric Research, Boulder, CO, USA. Poorter, L., van der Sande, M.T., Thompson, J., Arets, E.J.M.M., Alarcón, A., Álvarez-Sánchez, J., Ascarrunz, N., Balvanera, P., Barajas-Guzmán, G., Boit, A., Bongers, F., Carvalho, F.A., Casanoves, F., Cornejo-Tenorio, G., Costa, F.R.C., de Castilho, C.V., Duivenvoorden, J.F., Dutrieux, L.P., Enquist, B.J., Fernández-Méndez, F., Finegan, B., Gormley, L.H.L., Healey, J.R., Hoosbeek, M.R., Ibarra-Manríquez, G., Junqueira, A.B., Levis, C., Licona, J.C., Lisboa, L.S., Magnusson, W.E., Martínez-Ramos, M., Martínez-Yrizar, A., Martorano, L.G., Maskell, L.C., Mazzei, L., Meave, J.A., Mora, F., Muñoz, R., Nytch, C., Pansonato, M.P., Parr, T.W., Paz, H., Pérez-García, E.A., Rentería, L.Y., Rodríguez-Velazquez, J., Rozendaal, D.M.A., Ruschel, A.R., Sakschewski, B., Salgado-Negret, B., Schietti, J., Simões, M., Sinclair, F.L., Souza, P.F., Souza, F.C., Stropp, J., ter Steege, H., Swenson, N.G., Thonicke, K., Toledo, M., Uriarte, M., van der Hout, P., Walker, P., Zamora, N., Peña-Claros, M., 2015. Diversity enhances carbon storage in tropical forests. Glob. Ecol. Biogeogr. 24, 1314–1328. Rollenbeck, R., Peters, T., 2009. Climate Station Data at ECSF Established. :p. 1998. http:// www.tropicalmountainforest.org/data_pre.do?citid=497 (accessed 23.5.16). 229 Roy, D.P., Wulder, M.A., Loveland, T.R., W, C.E., Allen, R.G., Anderson, M.C., Helder, D., Irons, J.R., Johnson, D.M., Kennedy, R., Scambos, T.A., Schaaf, C.B., Schott, J.R., Sheng, Y., Vermote, E.F., Belward, A.S., Bindschadler, R., Cohen, W.B., Gao, F., Hipple, J.D., Hostert, P., Huntington, J., Justice, C.O., Kilic, A., Kovalskyy, V., Lee, Z.P., Lymburner, L., Masek, J.G., McCorkel, J., Shuai, Y., Trezza, R., Vogelmann, J., Wynne, R.H., Zhu, Z., 2014. Landsat-8: science and product vision for terrestrial global change research. Remote Sens. Environ. 145, 154–172. Sakaguchi, K., Zeng, X., 2009. Effects of soil wetness, plant litter, and under-canopy atmospheric stability on ground evaporation in the Community Land Model (CLM3.5). J. Geophys. Res. Atmos. 114. Silva, B., Strobl, S., Beck, E., Bendix, J., 2016. Canopy evapotranspiration, leaf transpiration and water use efficiency of an Andean pasture in SE-Ecuador – a case study. Erdkunde 70, 5–18. Spracklen, D.V., Righelato, R., 2014. Tropical montane forests are a larger than expected global carbon store. Biogeosciences 11, 2741–2754. Wilcke, W., Yasin, S., Fleischbein, K., Goller, R., Boy, J., Knuth, J., Valarezo, C., Zech, W., 2008. Water relations. In: Beck, E., Bendix, J., Kottke, I., Makeschin, F., Mosandl, R. (Eds.), Gradients in a Tropical Mountain Ecosystem of Ecuador (Ecological Studies 198). Springer, Berlin Heidelberg, pp. 193–201. Wulder, M., 1998. Optical remote-sensing techniques for the assessment of forest inventory and biophysical parameters. Prog. Phys. Geogr. 22, 449–476. Yarbrough, L.D., Navulur, K., Ravi, R., 2014. Presentation of the Kauth–Thomas transform for WorldView-2 reflectance data. Remote Sens. Lett. 5, 131–138. Yebra, M., Van Dijk, A., Leuning, R., Huete, A., Guerschman, J.P., 2013. Evaluation of optical remote sensing to estimate actual evapotranspiration and canopy conductance. Remote Sens. Environ. 129, 250–261.