The Grass Is Not Always Greener on the Other Side: Seasonal Reversal of Vegetation Greenness in Aspect‐Driven Semiarid Ecosystems

Our current understanding of semiarid ecosystems is that they tend to display higher vegetation greenness on polar‐facing slopes (PFS) than on equatorial‐facing slopes (EFS). However, recent studies have argued that higher vegetation greenness can occur on EFS during part of the year. To assess whether this seasonal reversal of aspect‐driven vegetation is a common occurrence, we conducted a global‐scale analysis of vegetation greenness on a monthly time scale over an 18‐year period (2000–2017). We examined the influence of climate seasonality on the normalized difference vegetation index (NDVI) values of PFS and EFS at 60 different catchments with aspect‐controlled vegetation located across all continents except Antarctica. Our results show that an overwhelming majority of sites (70%) display seasonal reversal, associated with transitions from water‐limited to energy‐limited conditions during wet winters. These findings highlight the need to consider seasonal variations of aspect‐driven vegetation patterns in ecohydrology, geomorphology, and Earth system models.


Introduction
Variations in insolation between opposing polar-facing slopes (PFS) and equatorial-facing slopes (EFS) can be up to~30% for moderate hillslope gradients at midlatitudes (Del-Toro-Guerrero et al., 2019;Zou et al., 2007). These differences generate contrasting micrometeorological conditions, mainly in air temperature and potential evapotranspiration (PET), which drives differences in available soil moisture (Branson & Shown, 1989;Metzen et al., 2019). These microclimatic variations result in a relatively cooler and mesic environment on PFS and a relatively warmer and xeric environment on EFS. Hence, aspect-driven ecosystems, characterized by differences in vegetation density and type, emerge as a result of variations in insolation (e.g., Flores-Cervantes et al., 2014;Pelletier & Swetnam, 2017;Sternberg & Shoshany, 2001). These distinct variations for opposing hillslope aspects are more pronounced in semiarid ecosystems as they tend to exhibit strong feedbacks between soil moisture and vegetation.
Much of the previous research on aspect-controlled ecosystems is based on field observations, which provide important details but are obtained over short time periods (mostly the growing season). In addition, results from these studies are based on a limited number of field plots on opposing hillslope aspects, which leads to lack of uniformity and homogeneity in the monitoring process (Barbosa et al., 2006;Lawley et al., 2016). These limitations can be partially addressed by using remote-sensed vegetation indices (VIs), as they provide temporally regular and spatially distributed data with consistent coverage for the assessment of vegetation dynamics (Becerril-Piña et al., 2015;Hoek van Dijke et al., 2019;Lawley et al., 2016;Wallace et al., 2006). VIs combine two or more spectral bands, and their use for assessing ecological properties of vegetation has dramatically increased over the last two decades (Asner & Martin, 2009;Moreno-De Las Heras et al., 2012;Nagendra, 2001). The normalized difference vegetation index (NDVI), which compares the intensity of reflectance in the visible red and near-infrared bands, is the most commonly used VI to quantify the presence of live green vegetation (Rouse et al., 1973). By relying on a ratio of band intensities, NDVI removes a large proportion of noise caused by cloud shadows, topographic and solar angle variations, and atmospheric attenuations existing in visible red and infrared bands, which makes NDVI less susceptible to illumination conditions (Huete et al., 2002;Justice et al., 1981;Martín-Ortega et al., 2020). NDVI values have been used as a proxy for vegetation greenness in hillslope-scale studies looking at differences in vegetation between hillslope aspects (e.g., Del Toro-Guerrero et al., 2016;Jin et al., 2009). These studies found that, in those particular catchments, PFS have higher values of NDVI than their corresponding EFS throughout the year due to differences in solar radiation.
Recent research, however, challenges the general understanding that vegetation greenness is higher on PFS than on EFS, particularly for areas in which transitions between water-limited and energy-limited conditions for plant growth occur (Fan et al., 2019;Johnstone et al., 2017;Yetemen et al., 2015). Modeling results for aspect-controlled semiarid ecosystems in monsoonal semiarid climates (Yetemen et al., 2015) suggest that EFS can briefly display higher vegetation greenness than PFS during the growing season because they benefit more from summer rainfall; however, this advantage may not last long due to the fast depletion of available water. In individual catchments in Gabilan Mesa, California, USA, Johnstone et al. (2017) also found higher values of NDVI on EFS during winter, coinciding with the rainy season, and a return to higher NDVI on PFS during summer.
The complex relations between water, energy, and vegetation across PFS and EFS constitute some of the lesser understood issues in global change research and Earth systems modeling (Fan et al., 2019). Although there is some evidence that transitions between water-limited and energy-limited phases of semiarid ecosystems may challenge the current paradigm of higher vegetation greenness on PFS, there has been no global study to verify this assertion. Here, we present vegetation, topographic, and climatic data from 60 catchments located across all continents, except Antarctica, that allow us to analyze global patterns of vegetation in aspect-controlled ecosystems. We investigate if reversals of higher vegetation greenness patterns from PFS to EFS are common features of aspect-controlled semiarid ecosystems and the conditions under which such reversals occur.

Study Sites
A careful review of previous studies reporting the existence of pronounced aspect-induced differences in vegetation was undertaken in order to select study sites with minimal anthropogenic impacts. In addition, each site was also visually inspected using Google Earth™ to further assess the lack of anthropogenic activities so that they can be assumed to be in near pristine conditions.
In order to minimize topographic-illumination bias in NDVI, we did not include sites with rugged topographies or at high latitude locations (Holben & Justice, 1980;Moreira & Valeriano, 2014;Teillet et al., 1982). While varying illumination conditions caused by topography have been observed to have negligible impacts on NDVI applications due to its band-ratio format (Matsushita et al., 2007), they still can produce some bias especially on rugged, steep terrain with low solar angles. We only included sites with mean catchment steepness lower than~25°and located at latitudes lower than 45°. We also verified that the topographic-illumination effect was negligible at our selected sites ( Figure S1 in the supporting information).
Based on these considerations, a total of 60 different catchments across 23 sites worldwide were selected for this study (Figure 1). Most of these study sites are located in water-limited ecosystems at midlatitudes. Around 52% of these sites are situated in the United States, within the states of Arizona, California, Colorado, and New Mexico. Sites outside the United States were selected in Australia, Chile, Spain, and Sudan (Table S1). According to the descriptions given in previous studies (Table S1), vegetation at these sites is characterized by mesic and relatively dense ground cover on the PFS and xeric and sparse vegetation on the EFS.
The size of the study catchments ranges from~0.5 to 200 km 2 . The mean elevations and slopes of these sites vary from 300 to 2,600 m and~5°to 25°, respectively. These regions have a variety of semiarid climate conditions, including monsoon-influenced continental and Mediterranean climates. Mean annual temperatures range from 5°C to 24°C, and mean annual precipitation (MAP) ranges from 130 to 760 mm. Descriptions of the site locations and the corresponding references to the published studies are given in Table S1.

Climatic and Topographic Data Sets
Climate data for this study include monthly mean values of temperature and total precipitation at~1 km 2 spatial resolution (Fick & Hijmans, 2017) obtained from WorldClim2 (http://worldclim.org).

Methods
The combined information from NDVI and catchment topography was used to identify the spatio-temporal variations of vegetation on different hillslope aspects. Slope, aspect, and elevation for each catchment were first determined from the DEM data. A sink-filling algorithm in ArcGIS (ESRI, 2015) was used to fill the internal depressions and enable the identification of a continuous drainage network for accurate catchment delineation. Hillslope aspect data were segregated into four cardinal directions measured clockwise (N = 315-360°and 0-45°; E = 45-135°; S = 135-225°; W = 225-315°). The north (south) aspects were considered as PFS in the NH (SH) and the south (north) aspects as EFS in the SH (NH).
The spatio-temporal variation of NDVI on different aspects can reveal differences in vegetation phenology and cover (Del Toro-Guerrero et al., 2016). In semiarid ecosystems, which are generally characterized by low to moderate values of NDVI (rarely reaching saturation levels), the dependence of NDVI value on vegetation phenology and cover is strong and linear (Fensholt et al., 2004;Huete et al., 2002). Hence, NDVI was used as a surrogate for vegetation greenness on opposing hillslopes in our analysis. NDVI values range between −1 and 1, but we ignored the negative values because they correspond to lack of vegetation (Pettorelli et al., 2005).
Locations that receive snow cover may generate uncertainty in the values of NDVI, as snow cover increases NDVI values in the same way as vegetation growth (Shabanov et al., 2002). To identify the potential interference of snow in our NDVI analysis, we used NDSI, a commonly used metric to detect and estimate snow cover. NDSI is defined as the ratio of the difference between visible green and shortwave near-infrared bands to the sum of these two bands. NDSI values range from −1 to 1, with higher values of NDSI representing greater snow cover. In this study, two different NDSI thresholds are considered for detecting snow presence: 0 and 0.1. These thresholds (0 and 0.1) represent a conservative approach and are substantially lower than the widely used threshold of 0.4 (Klein et al., 1998;Riggs et al., 2017).
For every catchment, we calculated the monthly mean NDVI and NDSI values for the study period (2000-2017) on PFS and EFS separately, based on the aspect data obtained from the topographic analysis. To quantify the relative contrast in NDVI with aspect, we calculated the difference between monthly mean values of NDVI on PFS and EFS, NDVI diff , (Johnstone et al., 2017), where NDVI PFS and NDVI EFS refer to the monthly mean NDVI values within a catchment measured on PFS and EFS, respectively. To quantify the significance of the results, we tested the hypothesis that monthly mean values of NDVI on PFS and EFS are statistically different in all catchments by using the Student's t test. Finally, we analyzed the seasonal variations of NDVI diff across the PFS and EFS in relation to NDSI, precipitation, and PET.

Results
Seasonal variations of NDVI diff for all catchments can be classified into three distinct patterns: (a) Type I pattern: NDVI diff is positive throughout the year; (b) Type II pattern: NDVI diff is positive during the summer, but a seasonal reversal leads to a negative value during the winter; and (c) Type III pattern: NDVI diff is very low or negligible for a large part or the entire year. All three NDVI diff patterns are observed at different sites in the NH; however, sites in the SH exhibit only a Type II pattern. To account for the potential interference of snow presence in the NDVI analysis (Liu et al., 2013;Shabanov et al., 2002), NDSI values were checked at all catchments. Snow presence at every catchment for each month is reported in Figure 2. The northernmost site in the NH has the largest NDSI values, followed by sites located at high elevations. Also, catchments at East Kaibab, Arizona, located at high elevations (~2,250 m above msl) have a considerable snow cover (NDSI > 0.1) from November to February (Billingsley et al., 2008). Similarly, in the SH, Mount Wilson in Australia and Chilean catchments present snow coverage in the austral winter.
As seen in Figure 2, snow-free catchments still display the three observed patterns, especially the Type II NDVI diff pattern. The monthly mean precipitation and PET of all the sites are displayed in Figures S4 and  S5, which shows that all sites from California in the NH and Chile in the SH have wet winters and dry summers. The Gabilan Mesa and Lower Kern sites in the NH receive 60-70% of its annual rainfall in the winter months, as do the Chilean sites in the SH. Rainfall in the winter months in the Mediterranean climate favors vegetation growth on the EFS as compared to the PFS.

Geophysical Research Letters
In order to illustrate and further analyze the different NDVI diff patterns, we show in Figure 3  Figures 3g-3i highlight how monthly PET values at three representative sites follow expected trends in solar insolation. However, seasonal differences in precipitation can produce a switch from water-limited (i.e., P < PET) to energy-limited conditions (i.e., P > PET). The Kendall site has wet summers under the influence of the North American monsoon (NAM), with monthly PET values always exceeding the corresponding monthly precipitation (Figure 3g). In contrast, Gabilan Mesa has a semiarid Mediterranean climate and monthly precipitation can substantially exceed monthly PET during winter (Figure 3h). Ramat Avisur

10.1029/2020GL088918
Geophysical Research Letters also has relatively wet winters, although much drier than those at Gabilan Mesa, and monthly PET largely exceeds monthly precipitation (Figure 3i).
To further analyze vegetation variability with respect to all possible aspects, the mean monthly NDVI values were computed at 30°intervals for two selected months, one for NH summer (June) and one for NH winter (November) (Figures 3j-3l). The Kendall site shows that NDVI on PFS (north) is always higher than NDVI on EFS (south) for both June and November (Figure 3j). The seasonal reversals of NDVI at Gabilan Mesa sites show asymmetric response in NDVI values; while higher NDVI values are observed on the northerly aspects (0-60°and 300-360°) in June, the southerly aspects (120-210°) exhibit higher values in November (Figure 3k). NDVI values are similar in both months for the Ramat Avisur site (Figure 3l).

Discussion
We found that the Type I NDVI diff pattern, with higher NDVI values on PFS than on EFS throughout the year, occurs only on a small fraction (~8%) of the studied catchments. This pattern agrees with previous field-based observations of a denser canopy on PFS and a sparser cover on EFS (e.g., Armesto & Martinez, 1978;Gutiérrez-Jurado et al., 2006;Sternberg & Shoshany, 2001). This pattern may be attributed to relatively lower insolation and less PET demand on PFS, which reduces the risk of hydraulic failure in plant vascular tissue and plant mortality (Earles et al., 2018;Fan et al., 2019). Surprisingly, the Type I NDVI diff pattern, which would be consistent with the current understanding, is not the dominant pattern in our analysis. However, it is important to note that most previous studies have analyzed data for Monthly mean values of NDVI for PFS and EFS aspects (a-c); NDVI diff values (d-f); precipitation (blue circles) and potential evapotranspiration (red circles) (g-j); and binned mean NDVI values for different aspects at 30°intervals for June and November (j-l) for three different patterns: Kendall, Arizona (Type I); Gabilan Mesa, California (Type II); and Ramat Avisur, Israel (Type III).
Our results show that many catchments in both the NH and the SH exhibit a seasonal reversal in NDVI diff (Type II NDVI diff pattern) (Figure 2). Some of the plausible explanations for these reversals in NDVI diff are the potential effect of snow on NDVI values, elevation, climate seasonality, and plant phenology. There are many sites that still show NDVI diff reversal even after excluding snow-affected sites with low/moderate NDSI values. This finding implies that, even though snow interference can affect NDVI observations (Liu et al., 2013), it does not explain the observed seasonal reversal of NDVI for many catchments. Sites located at high elevations (Socorro, Upper Rio Salado, East Kaibab, and Colorado) have low temperatures that may act as a major limiting factor for plant growth. Hence, EFS can benefit from more insolation and have longer growing seasons than PFS leading to early growth on EFS in winter but not sustainable growth in summer (e.g., Salzer et al., 2014;Zhang et al., 2015). Temperature limitation can be a reasonable explanation for NDVI diff reversal for places where severe environmental conditions prevail; however, the elevation of most of the study sites is less than 2,000 m and not prone to extremely cold conditions (Boisvenue & Running, 2006).
Though water availability is the main controlling factor on evapotranspiration in most semiarid ecosystems, some areas may also experience periods of energy limitation during the year as a result of rainfall seasonality (Garcia et al., 2014;Ivanov et al., 2008;Moran et al., 2009). This transition from water limitation to energy limitation is more pronounced when evaporative demand is exceeded by rainfall input as occurs in the Mediterranean climate (Figure 2h). For example, in the Gabilan Mesa sites in California and the Chilean sites, winter precipitation exceeds PET demand ( Figures S4 and S5); hence, the vegetation on EFS can benefit from more insolation during this period and grow faster than that on PFS.
Catchments in areas influenced by the NAM have been classified as Type I NDVI diff as PFS have higher NDVI throughout the year. However, these areas may also experience periods in which rainfall exceeds evaporative demand, but only for a short period (much shorter than Mediterranean climate). Results from modeling experiments suggest that energy-limited conditions might occur in the NAM region following intense storms during summer, inducing a short-lived energy-driven advantage for vegetation growth in EFS as compared to the opposite shady PFS areas (Yetemen et al., 2015). However, because of the short-lived and sporadic character of these events, they are unlikely to be captured by the mean monthly values over the entire study period used in our analysis.
Studies from aspect-controlled sites often report the existence of different plant functional types (PFTs) on opposing hillslopes (i.e., tree, shrub, and grass) (Gutiérrez-Jurado et al., 2013). For example, observations from an oak-grass savanna site in California indicate that grasses respond to the onset of the wet season (winter) earlier than trees, which do not respond until spring (Liu et al., 2017). Grasses tend to respond faster to winter rains greening-up before the tree leaf-out period, but trees keep their foliage for a long period after grasses have already senesced (Liu et al., 2017). During winter in the Gabilan Mesa sites, the early greening of grasses on the sunny EFS results in relatively higher NDVI values than PFS which can be covered by oaks. Later, when oaks leaf out in late spring, they can cause the NDVI values on PFS to increase, leading to NDVI values on PFS exceeding those on EFS and seasonal reversal in NDVI diff . It is important to note that the relatively faster growth response of grasses on EFS at this site occurs during a period when evapotranspiration is energy limited, which suggests that the phenological difference between PFTs can further promote NDVI diff in favor of EFS for Mediterranean climate when PET demand is exceeded by precipitation input. Chilean sites in the SH, which have similar climate, may follow the same pattern ( Figures S4 and S5) (Parsons, 1976).
Though the majority of the sites show seasonal variations in NDVI diff , a few sites (~20%) display negligible NDVI differences between opposing hillslopes (Type III pattern). This is the case for the Ramat Avisur (Israel) (Figure 2c) and Erkowit (Sudan) sites. The absence of discernible difference in NDVI diff may be ascribed to the extremely arid climates prevailing at these sites. However, some catchments in the Tabernas Desert (Spain) and Upper Rio Salado (Arizona) also display negligible values of NDVI diff , which deviate from the patterns obtained for the other catchments located in the same sites. This might be attributed to regional control of geology and other aspects of stratigraphic dips or bedrock structure and lithology (Churchill, 1981 10.1029/2020GL088918

Geophysical Research Letters
As explained in section 2, in order to avoid topographic illumination effects, we did not include in our analysis sites with large steepness or ruggedness. Compared to the catchments studied here, steeper catchments may display a more complex topography and a more pronounced aspect-driven effect (Ivanov et al., 2008), which could potentially result in larger NDVI differences. Further research is needed to assess whether transitions from water-limited to energy-limited conditions in steeper and more complex catchments can be effective in producing seasonal reversals in NDVI patterns.

Conclusion
Our analyses of climate, topography, and vegetation data from 60 catchments located across all continents, except Antarctica, show that the overwhelming majority of sites display seasonal reversal of vegetation-greenness dominance patterns from PFS to EFS, which can be explained by transitions from water-limited to energy-limited conditions. This research constitutes a first step to better understand variations of vegetation on hillslopes with opposing aspects. Improving the representation of vegetation dynamics in shady and sunny semiarid hillslopes is of great importance to satisfy the growing demand for better ecohydrology models in Earth system (Fan et al., 2019) and landscape evolution frameworks (Yetemen et al., 2019).

Data Availability Statement
Data sets used in this study are publicly available and can be accessed from the following online repositories: Shuttle Radar Topography Mission (SRTM) SRTM 1-arc second global DEM and NDVI from GEE (https://developers.google.com/earth-engine/datasets); monthly mean temperature and precipitation from WorldClim (Version2) (worldclim.org); and monthly mean potential evapotranspiration from Consultative Group on International Agricultural Research-Consortium for Spatial Information (CGIAR-CSI) (https://cgiarcsi.community/data/global-aridity-and-pet-database/).