3.1.A6. General issues of assessment of water-related ES
Terrestrial ecosystems perform between 11% and 96% of the modeled ES (Table 31A6-1). Ecosystems have the strongest impact on baseflow supply and erosion prevention, performing these functions almost entirely (93–96%). ES maps show that under the bare ground scenario, baseflow is almost absent (Section3.1.A2), meaning that the existing baseflow is almost entirely provided by terrestrial ecosystems. At the same time, under the current land cover, erosion is virtually absent (Section 3.1.A3), indicating that ecosystems almost completely prevent it. Only in the case of ES for flood mitigation under the average spring rainfall scenario (12 mm) was the effect of ecosystems negligible. Runoff retention and quick runoff values change only slightly in absolute terms between the current land cover and the bare ground scenario. However, even in this case, ecosystems reduce quick runoff by 14%.
Table 31A6-1. Results of ES modeling for the territory of Armenia.
| ES and InVEST model | Ցուցանիշ | Հողածածկ 2023 թ | Մերկ գետնին scenario | ES Provided by natural ecosystems | The share of ES provided by ecosystems % |
| Seasonal water flow regulation and baseflow provision SWY | Baseflow | 51.3 mm (BFI * = 34%) | 3.4 mm (BFI = 3%) | 47.8 mm | +93% |
| Արագ հոսք | 98.0 mm | 120.2 mm | −22.2 mm | −18% | |
| Prevention of soil water erosion and sediment transport to waterbodies SDR | Էրոզիա | 2.3 t/ha/year 6.8 Mt/year | 48.6 t/ha/year 147.2 Mt/year | Խուսափեց էրոզիայից −46.4 t/ha/year −140.4 Mt/year | −95% |
| Sediment export | 0.15 t/ha/year 0.47 Mt/year | 4.5 t/ha/year 13.5 Mt/year | Avoided sediment export −4.3 t/ha/year −13.0 Mt/year | −96% | |
| Flood risk mitigation, 50 mm rainfall scenario UFRM | Quick runoff, mm | 13.3 | 17.4 | −4.1 | −24% |
| Runoff retention, m3 | 3.7 | 3.3 | 0.4 | +11% | |
| 12 mm rainfall scenario UFRM | Quick runoff, mm | 0.19 | 0.22 | −0.03 | −14% |
| Runoff retention, m3 | 1.18 | 1.18 | 0 | 0 | |
| Cooling effect UC | Cooling capacity | 0.19 | 0.15 | 0.04 | +21% |
* BFI—baseflow index, BFI = B/(B + QF).
Consistency of the obtained results with other ES estimates for Armenia and expected ES performance across vegetation zones
Our average estimate of the erosion rate for Armenia, 2.3 t/ha/year, is very close to the values for Armenia (2.44–2.47) in the global database of modeled erosion values [65,66]. Neighboring countries (Georgia, Azerbaijan, Iran, Turkey) have similar estimates in this database—around 2–3 t/ha/year. According to Eurostat, erosion in most Mediterranean countries has a similar intensity, ranging from 2 to 5 t/ha/year [67].
The average share of baseflow in total flow, calculated based on SWY modeling results, is 34%, which corresponds to the baseflow index estimate for Armenia according to the AQUASTAT data and methodology of 35.5% (the overlap share of the internal renewable surface water resources) [68,69].
The modeling results for the prevention of erosion and sediment transport (SDR model) align most closely with the commonly accepted understanding of this ES. The SDR model identified forests as the most effective land cover class for preventing erosion, with rangelands and croplands performing worse. Among natural vegetation types, forests and woodlands provide this ES most effectively, followed by mountain meadows and then by steppes (Figure 5b,d). The model also showed that avoided erosion and avoided sediment export are the highest in areas with pronounced terrain and steep slopes, indicating that this ES is most important in those areas.While the SDR model gives plausible outputs, its accuracy depends heavily on soil, evapotranspiration, and rainfall data. The coefficients we used are based on global or European values, which should be adjusted to Armenian conditions and agricultural practices accurately.
The SWY model predicted the highest baseflow values—155 and 137 mm—in alpine and subalpine grasslands, while the forest zone showed a minimal baseflow of 29 mm, similar to that of the steppes (25 mm); both are lower than those in the semidesert and desert zones (Figure 5a,c). The proportion of baseflow contributed by ecosystems is also minimal in the forest and steppe zones (89%). This counterintuitive result, in our view, is explained by the combined effects of multiple factors that determine baseflow—precipitation, terrain slope, and soil permeability. Very high absolute baseflow values in mountain grasslands result from the high precipitation in the mountains. In other mountainous regions, higher baseflow values have also been found in upper elevation areas (e.g., [70]). The low baseflow values in the forest zone are most likely the result of forests occurring predominantly on the steep slopes of gorges and mountains. According to our assessment, the highest mean slope among the vegetation zones occurs in the forest and juniper zones —about 20°, whereas mountain grasslands and steppes occupy gentler slopes from 10° to 17°, and the semideserts and the single desert patch lie on plains with an average slope of about 6°. The moderate baseflow of 38 mm and the high proportion of it contributed by ecosystems (94%) in the semidesert zone are most likely due to its location in areas with the gentlest relief and a high proportion of highly permeable soils. The only small desert patch remaining in Armenia exhibits moderate baseflow of 34 mm an extremely high proportion of baseflow provided by ecosystems (98%), probably because it is entirely located on soils with the highest permeability (for detailed maps, see the project web-GIS [44]).
In this set of CC values across land-cover classes, the most surprising point is that grasslands have a lower CC than bare ground, that is grasslands are warmer than bare ground. However, in arid zones such a CC relationship is possible because evapotranspiration from grasslands is minimal or absent during the dry season, and the albedo of dry bare soils can exceed that of dried grass. Additionally, due to surface roughness, dry grass cools more slowly than bare ground. There are examples of dry vegetation being warmer than bare soil from the tropical zone (Feldman et al., 2022) and from Central Europe (Hesslerová et al., 2013). Nevertheless, this CC relationship for Armenia requires careful verification. Сhanging any of the coefficients determining СС (evapotranspiration, albedo, and tree canopy cover (shade) can alter the ratio of CC among different land cover classes. This highlights the need for model calibration.
Consistency of the tested InVEST models with Armenia’s natural conditions
The SWY model does not account for snow accumulation and melt, which is a major factor in Armenia’s highland hydrology. The approach we used in this study ignores snow sublimation and local variations in melt timing. For a more accurate assessment, it is clearly necessary to incorporate specialized models, such as SNOW-17, which can significantly improve runoff predictions [73]. Another significant limitation is the lack of accounting for geological structure, which is important for baseflow assessment.
Modeling the ES of flood risk mitigation (UFRM model) showed meaningful ecosystem effects only under an extreme rainfall scenario (50 mm). For average spring rainfalls (12 mm), the model barely registered any difference between current land cover and the bare ground scenario, which is due to low amounts of precipitation. It suggests the model may not be picking up more subtle but still important differences in landscape runoff retention under typical rainfall events. That raises questions about the model’s sensitivity under more typical weather conditions. Moreover, the UFRM model accounts only for the water retention capacity of ecosystems but does not consider water flow across the terrain, which makes it poorly suited for the mountainous conditions of Armenia. Slope has a critical impact on the rate of water runoff, which is why topography must be taken into account—as was done, for example, in [10].
These issues point to a clear need for InVEST model calibration (i.e., adjusting the model to match observed local data) before using its outputs in ecosystem accounts. According to [11], among the publications that used SWAT, 79% carried out some form of calibration, whereas for InVEST, only 13% of the studies did so. However, calibrated InVEST models can provide a sufficiently reliable ES assessment for strategic decision-making [74,75]. Our experience shows that InVEST models can be useful at the scoping stage, a necessary step before initiating ecosystem accounting.
However, more accurate ES assessment and mapping across the entire territory of Armenia—essential for informed decision-making—are hindered because some important coefficients in InVEST models are assigned single values, either for the entire area (the number of rainy days in the SWY and UFRM models) or for broad land cover classes (Kc in the SWY model), assuming that land-cover classes are uniform across the assessment area. As a result, models do not account for differences among areas at varying elevations or across climatic zones within Armenia.
Thus, at the preliminary stage, InVEST models proved useful for demonstrating general approaches to integrating ES assessments and maps into Armenia’s ecosystem accounting. However, given the aforementioned model uncertainties and simplifications, the estimates we obtained should be regarded as ES proxies rather than reliable data for management decisions or monetary valuation and should not be used directly in national accounting without proper calibration.
As ecosystem accounting and the corresponding data collection system develop, it may become reasonable to transition to the use of hydrological and climatic models that account for a greater number of processes and local data. However, this requires another milestone in Armenia, namely the open access to such data. At later stages, it is advisable to use different models for different purposes and decision-making contexts. InVEST models can be applied for rapid and simplified ecosystem service modeling to obtain a general overview. SWAT and other detailed hydrological and climate models are necessary for producing high-resolution and accurate assessments. Decision-support models (such as RIOS, AQUATOOL, and others) are useful for the practical application of ecosystem service assessments and maps in management contexts [12,76,77].
Potential bias in assessing the role of different terrestrial ecosystems in ES provisioning
According to the SEEA-EA guidance, one of the EA tasks is to evaluate how various ecosystem types contribute to ES provisioning [1]. However, using broad land cover classes as proxies for varied and complex ecosystems can lead to significant bias. InVEST models operate with broad land cover classes such as “forest” or “grassland”. Although this approach is practical, it may obscure significant ecological diversity and misrepresent the true functioning of particular ecosystem types [78].
Given high topographic and climatic variability in Armenia, these risks are exacerbated there. With elevations ranging from 375 to over 4,000 m above sea level, the area of the country includes both lowland semi-deserts and high alpine regions. Precipitation, soil properties, temperature regimes, and land use can all change quickly in this area, sometimes within a few kilometers [33]. In Armenia the category “grassland” encompasses diverse ecosystems, ranging from alpine meadows to semideserts, that differ fundamentally in their functioning and in their capacity to provide ES. Average values of ES indicators for grasslands do not reflect the diversity of ecosystem functions and services among the various types of meadows, steppes, and semideserts (Figure 5). Likewise, not all forests have the same function in regulating hydrology; their contributions are influenced by species composition, slope gradient, canopy density, and soil depth [21–23]. Thus, conducting ES accounting at the level of broad land cover classes fails to capture ecosystem-specificity, offers little for informed ecosystem-management decisions, and in some cases can lead to incorrect decisions. For example, using the average baseflow value for grasslands (59 mm) leads to underestimating the contribution of mountain grasslands with baseflow values of 137–155 mm to the total baseflow volume.
Biases in understanding the roles of different ecosystem types in delivering ES could have negative consequences for environmental policy. Globally, an example of such a bias is the underestimation of grasslands’ roles in water provision and soil protection, alongside a primary focus on the ecological value of forests. This often leads to afforestation of natural grasslands, resulting in negative impacts on water regulation and soil quality [28,79,80].
Even within vegetation zones—which partially account for the diversity of grasslands and woody vegetation—there remains a wide spread of ES values across individual polygons, indicating the high heterogeneity of environmental conditions and plant communities within them. This raises the question of whether a more detailed ecosystem classification and mapping should be used to assess ecosystems’ roles in delivering ES.
The feasibility of assessing the entire bundle of water-related ES
Water regulation is closely linked to the prevention of soil erosion, as well as the cooling effect of evapotranspiration. Tested InVEST models use the same data and coefficients (Figure 31A6-1). Therefore, it makes sense to consider water-regulating and soil-protection ecosystem services together as an integrated whole. 
Figure 31A6-1. The relationship between the coefficients used in the tested models. *Average monthly temperature values were used to adjust the average monthly precipitation, taking into account the snow season.
Our preliminary testing of ES models did not include ES of rainfall pattern regulation, which was recently added to the recommended list of ES in the SEEA-EA framework [1]. In recent years, this ES has been increasingly recognized as fundamentally important, as it completes the hydrological cycle on land. Without accounting for this ES, vegetation appears only to evaporate moisture, reducing water availability on land. However, evaporated moisture in the atmosphere contributes to precipitation recycling, which increases the overall amount of rainfall and enables its transport further inland [24,76-79].
SEEA-EA recommendations imply a subcontinental scale of this ES, exceeding the territory of Armenia; however, several studies have also highlighted its relevance at the regional level, including in arid zones [77, 80-82]. We did not include this ES in our testing, as it is not yet represented in the set of relatively simple models like InVEST ready to use without specialized research. Moreover, as noted by Wierik et al. [83], research on this ES has focused on the global level or on tropical forests, while there is a knowledge gap for other zones, including temperate forests and grasslands.
According to estimates by Tuinenburg et al. [84], Armenia lies within a zone with high evaporation recycling ratios—typical for most land areas—meaning that nearly all evaporated moisture eventually returns as precipitation. The country also exhibits medium precipitation recycling ratios: in winter, 50–60% of precipitation originates from land evaporation, and in summer, this figure rises to 70–80%. However, these are only averaged estimates, which in reality may reflect a mosaic of areas where forests either increase or decrease water availability [85].
For Armenia, as a mountainous country, another potentially important but still poorly formalized function is the capture of atmospheric moisture by vegetation in upland areas, which act as “water towers” [24,76].
Models of ESs that return vegetation-evaporated moisture to land should be developed and included alongside other water-regulating ES in national EA.
