A water storage management tool was developed and improved from Zhao et al. (2021) using the SD modeling platform Stella (iSee Systems, Inc., Lebanon, NH, USA). The tool captures processes and interactions of water allocation in the YRB based on incoming natural flow to current water storage and water demand from agriculture and hydroelectricity generation (Figure 3). We aggregated five reservoirs to one system water storage for water system operation following the priority to meet Title XII instream flow target (USBR, 2008), hydropower demand, and irrigation. The water system operation algorithm was based on total water supply for agriculture (TWSA), water demands from food, energy, and environmental instream flow, constraints of minimum and maximum water storage, flood season, irrigation season, and reservoir goal curve. The historical daily average of total reservoir storage above Parker station was used as the reservoir goal curve. Soil water dynamics played a critical role in deciding how much water was needed for irrigation according to crop growth season, water loss from ET, and irrigation efficiency level from the current irrigation system. Surface water diversion was the main supply source while groundwater was used as co-supply during non-drought years or backup supply in case of insufficient surface water supply during drought years. The use of both water resources was limited by surface water entitlements (i.e., proratable and non-proratable water rights) and groundwater entitlements (i.e., primary and standby water rights), respectively. The groundwater accumulated from MAR implementation was defined as “MAR entitlement” in this study, which is a type of water right to be used when both surface water and groundwater entitlements were exhausted. In addition to managed aquifer recharge (MAR), we developed new modules based on the SD model from Zhao et al. (2021), including greenhouse systems and crop production (section greenhouse system and crop yield), to evaluate possible management strategies to improve irrigation reliability as well as economic benefits. We chose the open field high and low value crops represented by apples and alfalfa, respectively, and tomato represented the greenhouse crop.

Greenhouse system water demand included water required for evaporative cooling systems and water loss due to ET. Evaporative cooling pulls outside air through a wetted pad and uses the heat from the air to evaporate water from pads for cooling purposes. The effectiveness of cooling depended on the ventilation rate of the fan system. An average water use of 0.18 m³/(m²·s) was assumed based on a 0.037 m³/(m²·s) air exchange rate (Sabeh et al., 2006). Daily air temperature ranged between the designed maximum (32.2°C) and minimum (18.3°C) temperature for growing crops (Shamshiri et al., 2018) to represent temperature fluctuation inside GHs. We used the Hargreaves method (Hargreaves and Samani, 1985) to calculate reference ET (ET₀) inside the greenhouse. Two year-round crop rotations for tomato were scheduled with a spring crop starting in January and a fall crop starting in July for 180 days per rotation. Crop coefficients (K_c) for spring and fall seasons are shown in Supplementary Table S2. Actual ET was calculated as the product of K_c and ET₀.

The adoption of GH systems was assumed to replace low value crops (i.e., alfalfa) due to considerable crop value difference and limited crop land areas. The land use density for GHs was at 80% with plants seeded at a density of 2.5 plants/m². Greenhouses used water tanks for flexible water supply. More water was allocated to the GH storage tanks when water demand exceeded the management allowable depletion (MAD) of the tank storage at 90%. Precipitation was another source for GH water collection and 5% of the total precipitation falling on the GH area was assumed to be stored in the tanks. The GHs were assumed to be closed hydroponic systems with 100% water use efficiency.

Greenhouses had the priority to divert water from total water allocation, and the rest was diverted to crops for open fields. The total irrigation demand included water demand from apple, alfalfa, and tomato growing (Figure 4). Final total water allocation depended on water system operation rules described in section system dynamics approach. Adopted GH area was a function of an adoption ratio calculated based on the innovation diffusion model (Sterman, 2000; Repenning, 2002) with additional impact factors (Zhao et al., 2021), such as drought severity stimuli and the effectiveness of the adoption method. As farmers switched alfalfa for GHs (Figure 4), a reinforcing causal loop (Loop R) and a balancing causal loop (Loop B1) controlled the behavior of adoption rate for GHs depending on whether GHs can help mitigate the drought impact on water scarcity for irrigation.

Crop yield, defined as crop production per unit area, was mainly affected by crop water use through actual ET. Generally, there is a positive relationship between yield reduction and reduced soil moisture content as a result of water supply shortage (Doorenbos and Kassam, 1979). We calculated crop yield for apple and alfalfa with Equations (1) and (2) that describe the yield response to actual ET (Steduto et al., 2012). Greenhouse crop yield was assumed constant [28.2 kg/m² (OSU Extension Service, 2002)] representing the average yield for two growing seasons together.

where Y_a is actual crop yield; Y_x is the maximum crop yield; K_y is a yield response factor indicating the sensitivity of yield reduction to ET reduction; ET_a and ET_x are the actual and maximum ET. Based on data from the United States Department of Agriculture (USDA, 2017), maximum crop yields for apple and alfalfa were fixed at 5.4 and 1.64 kg/m², respectively. K_y was 1.1 for alfalfa and 1 for apple (Steduto et al., 2012). ET_a was calculated based on soil water dynamics in the irrigation supply module of the model.

We used the total reservoir storage and observed daily streamflow at Parker station during 1979–1999 to calibrate the model after adding new modules and improving the algorithm of water system operations based on the SD model version in Zhao et al. (2021). Then, we validated the model with the same variables for the next 15-year time period (2000–2015). Calibration used the embedded Powell optimization tool (Powell, 2009) in Stella. Model performance for validation used the Nash-Sutcliff Efficiency (NSE) coefficient (Nash and Sutcliffe, 1970). In addition, we evaluated the performance using RMSE and R² (see Supplementary material).

where Q_obs and Q_sim are the observed and simulated values, respectively; Qobs¯ is the average of the observed value; n is the number of data points.

Future climate showed a wet-to-dry trend with different levels of wetness and dryness at different time periods. We divided the 70-year climate projection (2020–2090) into seven decades to represent future climate conditions (CCs) similar to Malek et al. (2018a). Historical climate from 1979 to 2015 served as a baseline, while relative changes of precipitation and temperature for each CC relative to the baseline mostly increased from 2020 to 2090 (Table 2). Severe drought conditions are likely to occur with increased average temperature and decreased average precipitation. RCP 8.5 had more intense CCs with greater relative increases in temperature, compared to RCP 4.5.

Each CC could occur in the future instead of following the specific timeline simulated from GCMs. Managers and stakeholders can optimize a ten-year water conservation plan facing each possible CC for their various combinations and levels of droughts. Long-term plans can also be evaluated by using the 70-year climate projection (see section impact of adoption in long-term management).

Agriculture in the YRB is relatively vulnerable facing water scarcity because irrigation heavily depends on climate conditions. From historical records during 1979–2015, droughts occurred seven times with proration rate lower than 70%. Failure of sufficient water supply during droughts resulted from mismatch between demand and supply, such as the lack of reservoir storage (insufficient supply) and inefficient irrigation systems (high demand). Drip and sprinkler irrigation methods are favored by orchard owners in the YRB, and drip irrigation only covers a small portion (13%) compared to sprinkler systems (49%). Inefficient methods, such as rill and flood irrigation (36% of total irrigated land) are still common especially in the Kittitas valley and lower valley of the YRB (Johnson, 2000). In the model, the average irrigation efficiency (IE) was assumed to be 0.75 based on all irrigation methods in the YRB.

To reduce the vulnerability to droughts, we propose four climate adaptation methods for water management that either reduce irrigation demand or increase irrigation supply. The selected adaptation methods include GHs, shifting crop planting time to earlier in the season, irrigation technology for IE improvement, and managed aquifer recharge (MAR). For MAR, this study only considered surface recharge by delivering water through existing canals and spreading water on agricultural land for infiltration and percolation. We listed the measurement for each method to represent their adoption levels (Table 3). Greenhouse adoption was measured by the fraction of land switched from alfalfa to GHs, named GH area fraction (GHAF). Management for crop planting time shift (PTS) was measured with days where positive values of shifted days indicated planting crops earlier relative to the default planting date (at day 92 from the beginning of a year for alfalfa). The crop growing season length was unchanged (208 days). Improvement in irrigation technology (IT) was measured by the average IE, which was defined as the ratio of actual water stored in the root zone of open field crops (apple and alfalfa) vs. water delivered for irrigation. MAR was measured by the infiltration area (IA) with a constant infiltration rate (0.43 m/day) and hydraulic conductivity (1.52 m/day) (Zhao et al., 2021). The initial available MAR entitlement was set at 0.1 × 10⁹ m³ for RCP 4.5 and 0.5 × 10⁹ m³ for RCP 8.5 to spin up the adoption of MAR (Zhao et al., 2021). This was also to overcome the fact that MAR application was a time-dependent method.

In this CC analysis, we used the model to find the minimum requirement for each of the four adaptation methods that can fully mitigate drought effects on irrigation for each CC under RCP 4.5 and 8.5, while keeping other methods unchanged from the baseline (Table 3). The minimum requirement was defined as an application level of an adaptation method that improved irrigation reliability to 1 (irrigation water supply met irrigation demand) for every year within the selected CC. For example, when 15% of the total alfalfa land switched to GHs (GHAF = 15%), the water shortage for agriculture resulting from droughts was fully mitigated for CC1 under RCP 4.5. Thus, a GHAF at 15% was the minimum requirement for CC1. It was assumed that each CC represented an independent, ten-year planning window not affected by other CCs. For all CCs, we calculated the precent change from baseline in agricultural water demand and supply resulting from each adaptation method. Greater percent change indicated greater sensitivity of demand or supply dynamics to the adaptation method.

Long-term planning required consideration of the outcomes from adopting proposed adaptation methods. Innovation adoption describes a predictable process of innovative methods being adopted within a given community. We assume adoption behaviors are affected by advertisement, word of mouth, effectiveness of adaptation methods, cost-effective driven adoption, and drought stimuli, explained in Zhao et al. (2021). Effectiveness of adaptation methods in mitigating drought impact on irrigation reliability was evaluated differently for each method: (1) for GH, the ratio of reduction in ET loss to total ET loss; (2) for MAR, the ratio of MAR supply to water demand deficit (i.e., demand that is not met by water supplied without MAR); (3) for IT, the ratio of water saved from improving irrigation efficiency to total water supply. Adoption behaviors vary over time as the drought tolerance changed as a result of implementing adaptation methods. Cost-effective driven adoption was to reflect the fact that cheaper infrastructure tends to be diffused faster, which was assumed as MAR > IT > GH. We analyzed the role of adoption behaviors in drought mitigation from 2020 to 2090 with the initial adoption ratio at zero in 2020.

Adoption ratio is the ratio of the number of adopters over the total amount of potential adopters. Thus, adoption ratio approaching 1 indicates that the adopters approach the total population (e.g., irrigation units). We defined maximum threshold for the measurement of each adaptation methods, including GHAF, MAR IA, and IE, to represent the limitations of financial and natural resources. For example, infiltration area at full adoption (IAFA), defined as the maximum infiltration area at a 100% adoption ratio, is an upper threshold for MAR measurement. The upper thresholds for measurements were 70%, 3.0 km², and 0.95 for GHAF, MAR IA, and IE, respectively. The lower thresholds were baseline values (Table 4), and the measurement value at time t for each adaptation method was calculated by Equation (4). Five scenarios were designed to evaluate the impact of individual methods and combined methods on drought mitigation (Table 4) under RCP 4.5 and 8.5.

where Mit is the measurement value at time t for method i, i = GH, MAR, or IT; TH_{min, i} and TH_{max, i} are minimum and maximum threshold for the measurement of method i, respectively; rit is the adoption ratio for method i at time t, which is the ratio of irrigation units that adopt method i over total irrigation units. rit=0 means no irrigation unit adopts the method i at time t; rit=1 means all irrigation units adopt the method i at time t.

Four indicators were selected to show the outcomes of adopting various management strategies from different perspectives, including irrigation reliability (IR), groundwater supply fraction, crop production, and economic water productivity (EWP, see Supplementary material). We used these indicators to demonstrate the critical factors that may affect system resilience and sustainability in terms of environmental resources and economic values.

IR is an index for the total agricultural water supply (AWS) and agricultural water demand (AWD) ratio in terms of open fields and greenhouses consuming surface water and groundwater. The IR was calculated at an annual time step based on the concept of volumetric reliability (Hashimoto et al., 1982; Kundzewicz and Kindler, 1995).

Groundwater supply fraction (referred to as F_GW) was defined as the proportion of groundwater supply (GWS) pumped for agriculture in AWS Equation (5), indicating the irrigation dependence on groundwater as well as the level of available groundwater entitlements and surface water availability. For example, more groundwater would be pumped for irrigation purposes when surface water supply cannot keep up with crop water demand due to drought. The groundwater supply fraction was calculated on an annual basis.

Climate conditions included various drought features due to the timing and level of precipitation and temperature, which represent possible situations water resource managers may face in the future. The annual IR indicated drought severity, with the lower IR meaning a more severe drought (Figure 6). At baseline for RCP 4.5, the occurrence of droughts started to increase in the late 2040s and the possibility of more frequent and extreme droughts was high in 2060s and 2070s. The baseline for RCP 8.5 showed a clear trend of increasing drought severity after the late 2040s, and the average IR value was < 0.25 after the 2070s. Among seven climate conditions under each RCP scenario, the lowest average IR was during CC6 (2070s) under RCP 4.5 and CC7 (2080s) under RCP 8.5. However, the worst annual IR was during CC5 (2060s) under RCP 4.5. CC1 (2020s) and CC2 (2030s) for both RCP 4.5 and 8.5 appeared to be water sufficient and no action was needed for improving system water reliability. Overall, climate patterns within the CCs determined the water scarcity situations, where consecutive droughts would lead to a worse situation than if the same numbers of droughts spread within a decade.

The minimum requirement for each adaptation method varied significantly under different CCs (Table 5). Higher values for the minimum requirement of adaptation methods indicated more resources (i.e., financial, land, and permits) were needed to alleviate drought impacts on irrigation. There were no minimum requirements for all four methods under CC2 for both RCPs due to the wet climate. For CC5 under RCP 4.5, the minimum requirements for GH, crop PTS, and IT were all the highest, compared to requirements in other CCs. However, infiltration area for MAR was higher at 4.1 km² for CC6 than 2.7 km² for CC5 because temporal distribution of severe droughts was mostly at the first half of CC6 whereas droughts in CC5 occurred in the second half of the decade. This indicates that MAR depended on the long-term efforts to accumulate MAR entitlements and were more valuable at longer planning times. GHAF needed to take over more than 50% of the low value croplands to improve drought conditions for CC5–CC7. The minimum requirements of PTS ranged from day 0 to 73, which illustrated that moving crop planting time to late January can significantly drop the water demand level due to lower ET in the winter season. Improving IE appeared to be an effective method as minimum requirements for IE for most CCs were under 0.9, which can be achieved easily by switching from traditional flood irrigation to sprinkler and drip irrigation.

The minimum requirements for all four adaptation methods were much higher under RCP 8.5 than RCP 4.5 as a result of the higher frequency of severe droughts (Table 5). The most extreme CC7 required the highest application level of GHAF or IE reach for drought mitigation. However, even the highest level of IE of 0.95 was not able to fully secure agricultural water supply. Crop PTS needed to move more than 2 months earlier, in order to reduce water demand; however, crop yield may drop significantly as other agronomic factors other than water affect crop growth when planted in winter. MAR IA under CC7 was more than three times that under CC4, with 10.6 km² of IA needed to accumulate ~ 5.1 × 10⁹ m³ of MAR entitlement in a decade. Overall, RCP 8.5 created challenging CCs and each CC required well-defined infrastructures for adaptation methods to improve water security.

The minimum requirement identified the potential implementation scale for each method. Most of the extreme requirements were to evaluate the degree of effort needed using a single adaptation method without knowing if these would be entirely realistic in future settings. For example, switching 100% of low value cropland to GHs exemplifies the severity of droughts in terms of GH implementation scale. However, all the minimum requirements may not be met due to limited natural and financial resources, environmental feasibility, and applicability. For example, considering damages from first frost/freeze, over-moist soil in the early growing season, and limited growing degree-days for proper crop development when spring alfalfa is sown in winter, the higher value of crop PTS may not be valid. Alternatives can include planting cool season crops in the fall for winter/spring harvest to achieve the goal of ET reduction. However, evaluation of cropping patterns and practices were beyond the scope of this paper.

All four adaptation methods reduced irrigation demand by 5–80% under most CCs (Figure 7). A positive relationship exists between drought severity and agricultural water demand reduction under adaptation management. For example, percent change in water demand reduction for CC2 under both RCP 4.5 and 8.5 was 0 because water demand was already satisfied by available supply (Figure 6) without any adaptation methods. In contrast, the highest percent change in water demand reduction for RCP 4.5 was in CC6 as a result of its extreme water scarcity status. Similarly, under RCP 8.5, water demand from CC3 to CC7 showed a decreasing trend relative to the baseline in all adaptation methods except for IE management in CC7. This was because IE reached its maximum threshold (0.95) in CC7, but the method was still unable to erase all the drought impacts.

The reduction level of agricultural water demand was dependent on the adaptation method being used. In fact, drought created a situation where the soil water content kept decreasing with continuous water loss from ET and insufficient irrigation, leading to increasing water demand over time. All methods directly or indirectly reduced water demand in three ways: (1) reduced biological water demand of crops from reducing ET (e.g., PTS); (2) reduced irrigation water wasted (e.g., IT, GH); and (3) reduced soil water deficit by increasing supply (e.g., MAR). Adaptation methods, such as PTS, whose purpose was to decrease crop water demand, had the most reduction compared to others (Figure 7). Shifting planting time earlier reduced ET due to increased growing days during winter season, where ET remained low compared to ET in summer. In addition, earlier harvest avoided high risks of having limited water during summer.

Agricultural water supply mostly decreased with application of adaptation methods except for GHs and MAR (Figure 7). The most reduction was from crop PTS in CC5 at about 14% under RCP 4.5 and 13% under RCP 8.5. Agricultural water supply and demand depended on each other via soil water dynamics, for example, reduced demand led to less supply. However, MAR, as an exception, increased water supply because MAR created extra water storage for emergency use. GHs resulted in mixed positive and negative changes of relative difference in water supply under RCP 8.5 (CC5–CC7). This is because increased GHAF led to contradictory consequences on water supply according to feedback relationship shown in Figure 4. On the one hand, reduced alfalfa acreage greatly reduced ET losses, and thus reduced water supply. On the other hand, GHs required more consistent water supply throughout the growing season with water storage infrastructure, which allowed more water to be stored for use during droughts.

The most effective individual methods for reducing irrigation deficit were MAR for RCP 4.5 and GHs for RCP 8.5, respectively, whereas a combination of all the adaptation methods (S-ALL) achieved the best results (Figure 8). The reduction of irrigation deficit can be the result of decreased irrigation demand or increased irrigation supply. In the baseline of RCP 4.5, the agricultural water demand reached a peak value of around 12 × 10⁹ m³ in the 2060s and remained high in the next two decades, compared to a value around 2.5 × 10⁹ m³ before 2050. In the baseline of RCP 8.5, the median annual water demand by decade started to increase from 2050s at an average rate of 3 × 10⁹ m³ per decade. Most water demand ranged from 13 to 18 × 10⁹ m³ in the 2080s, which is about seven times the median water demand of 2.5 × 10⁹ m³ in the 2020s. With intense water shortage in the second half of the century, both GHs and IE were effective and lowered annual water demand for RCP 8.5 to around 5 × 10⁹ m³, compared to the MAR. With the combination of GH and MAR or all three methods, the water demand slightly decreased with time for both RCPs, which can be used as an effective and flexible strategy to mitigate drought when a certain method has limitations.

We selected two different dry periods for RCP 4.5 (2070–2075) and 8.5 (2080–2085), respectively, to demonstrate the impact of adaptation methods on dynamics of soil moisture content (Figure 9). During typical dry years from 2070 to 2075 for RCP 4.5, the water loss by ET led to a soil moisture content drop below 15% at baseline. The soil was even drier during 2080–2085 under RCP 8.5 indicating considerable water stress from extreme droughts. Among three single method scenarios, GHs restored soil water the most during these periods for both RCPs, followed by IT and MAR. MAR filled up soil water to field capacity during 2070–2073 with the remaining available entitlements and became less effective even if its adoption ration was approaching 1.0. Soil moisture content was the bridge connecting agricultural water demand and supply. Agricultural water demand kept increasing when the deficit between current soil water content and field capacity was not replenished, which put more pressure on future supply. The key to mitigate drought conditions was continuously wetting the soil or slow down the water loss from the soil to avoid increased soil water deficits. Different adaptation methods can be more helpful for certain climate patterns according to their basic principles in mitigating droughts.

The results of annual irrigation reliability show a finer temporal resolution of the impact of adaptation methods on extreme future climates (Figure 10). All adaptation methods performed better in improving irrigation reliability under RCP 4.5 than under RCP 8.5. Under RCP 4.5, both IT and GHs appeared to be less effective than MAR. Under RCP 8.5, GHs and IT served as a better option in increasing irrigation reliability with frequent severe droughts. The discrepancy of effectiveness of adaptation methods under RCP 4.5 and RCP 8.5 was because of their fundamental functionalities, where GHs and IT were methods focusing on reducing water demands and MAR was focusing on increasing water supply. By switching low value cropland to GHs, higher water use efficiency in GH's irrigation system decreased average water demand overall. Improving IE directly increased the volume of water stored in the crop root zone per unit volume of water irrigated to the soil. MAR did not significantly enhance water scarcity conditions after 2050 for RCP 8.5 as a result of low adoption rate at the initial stage as well as high consumption of MAR entitlements during 2050s. With all adaption methods working together, we still see there were 4 years having water shortages in 2050s due to low adoption rate (0.1–0.3) in the middle of the century.

MAR adoption ratios increased faster than the adoption ratios for IT and GH and the more extreme climate in RCP 8.5 encouraged greater adoption in a shorter time for all adaptation methods (Figure 11). Note that the values at the full adoption for GHs, IT, and MAR were 70% of the alfalfa land, IE at 0.95, and IA at 3.0 km². The actual adoption for each method started low and increased as more and more irrigation units decided to take action to improve water security. MAR was assumed as an easier adaptation method than the other two methods because no extra construction was needed to apply MAR. However, the reason of MAR failing to provide enough groundwater under RCP 8.5 (Figure 10) was due to low available MAR entitlement with low infiltration area. With a MAR area of 3.0 km² at full adoption, the IA increased to 0.36 km² and accumulated 0.15 × 10⁹ m³ of water entitlement at around year 2048 (Figure 12), when severe droughts under RCP 8.5 started to occur. Therefore, the speed of MAR entitlement accumulation could not catchup with agricultural water demand, so that the entitlement dropped to 0 in 2050 and stayed low after that. On the other hand, the same designed IAFA under RCP 4.5 could always satisfy almost all the agricultural water demand and managed to retain a good amount of groundwater entitlement by the end of the century. The entitlement accumulated as high as 0.8 × 10⁹ m³ in 2064 and declined sharply due to a dry climate for 10 years. As a cost-effective adaptation method, our findings suggest an appropriate MAR design for IAFA based on the projected CCs, the capacity of the groundwater system, and available land for infiltration can mitigate drought impact by providing large buffers for climates with alternate dry and wet conditions.

Groundwater is an important and vulnerable resource in water supply. Increasing groundwater use in total water supply can cause dramatic groundwater depletion when the average pumping exceeds average recharge rate. GHs and IT indirectly reduced the groundwater supply fraction (F_GW) especially in later decades under RCP 4.5 (Figure 13). MAR as a groundwater resource kept F_GW as high as the baseline level. However, groundwater pumped from available MAR entitlements was artificially recharged from surface water thus using the aquifer as extra storage which was less likely to cause negative effects on groundwater levels.

The result for RCP 8.5 showed that a combination of all methods was needed to lower the F_GW (Figure 13). The F_GW during 2020–2060 was not affected very much even with application of all adaptation methods. However, all adaptation methods combined greatly reduced the median of F_GW in last three decades especially for 2080s, the decade with the highest F_GW in baseline. During severe drought scenarios, all annual groundwater rights were mostly consumed during the summer season and time periods when irrigation water rights were not available (outside of Apr–Oct). With the help of all adaptation methods, groundwater pumping was shifted mainly before April and after September serving as a backup for limited surface water.

RCP 8.5 had more impact on crop production than RCP 4.5 for both low value (e.g., alfalfa) and high value (e.g., apple) crops (Figure 14). Average annual production of alfalfa in each decade at baseline fluctuated around 0.62 × 10⁶ tons/year before the 2050s and reduced gradually after the 2050s, especially for RCP 8.5. MAR and IT slightly increased alfalfa production relative to the baseline. GH implementation led to a significant drop in alfalfa production to 0.25 × 10⁶ tons/year on average due to reduced land area for alfalfa for both RCPs. Apple production was also sensitive to the increasing GHs adoption. For RCP 4.5, adopting GHs increased high value crop production from an average of 3.0 × 10⁶ tons/year in 2030s to 3.1 × 10⁶ tons/year in 2080s, compared to baseline with decreased production after the 2040s to as low as 2.75 × 10⁶ tons/year in 2070s. MAR and IT slightly improved apple production by 0.2 × 10⁶ tons/year on average during droughts. For RCP 8.5, apple production at the baseline showed an obvious reduction associated with the increasing trend of frequency and severity of droughts. Similar to RCP 4.5, GHs was more effective at improving apple production as adoption increased for RCP 8.5. For example, about 62% of GHAF (adoption ratio = 0.88) enhanced apple production by 0.5 × 10⁶ tons/year in 2080s from the baseline. During 2060–2090 with frequent extreme droughts, IT increased apple production by 0.4 × 10⁶ tons/year on average, whereas MAR achieved the smallest average amount at 0.2 × 10⁶ tons/year. With all three adaptation methods, we can only see a slightly increasing trend of apple production toward the end of the century.

The critical time in the year determining the amount of annual crop yield was the end of the growing season when drought impacts had accumulated as expressed by lack of water in the root zone. Lack of soil water reduced actual ET, leading to yield reduction according to Equation (1). The reason that GH was more effective in improving high value crop production was the reduced land area in alfalfa (and thus reduced total water demand) and GH water demand. With sharing water with alfalfa and ET loss from alfalfa, the soil water in the root zone was used up faster, which exhausted the water supply system more quickly during the irrigation season. Consequently, the orchard soon reached soil water deficiency in the middle season causing yield reduction. Without alfalfa, water available per unit open field for orchard increased allowing sufficient water in the root zone for ET demand. Improving IE increased the ratio of water stored in the root zone per unit water delivered to the field. However, this method had a ceiling that constrained the maximum irrigation efficiency. During extreme droughts, scenarios with high GHAF proved to be a better strategy to decrease water loss by ET.