Introduction

Human societies rely heavily on natural ecosystems, which are self-organizing systems that provide a range of services to humans, including water, food, wood, fiber, agricultural products, and raw materials. Additionally, ecosystems maintain a balance among the different components of the environment, preserving biodiversity and purifying the environment (Costanza et al. 1997; Daily 1997; Millennium Ecosystem Assessment 2005). Each ecosystem has a unique structure and function that is significantly impacted by human activity and climate change. In recent decades, global climate change has posed a major threat to the preservation and sustainability of ecosystem services. According to the Millennium Ecosystem Assessment, approximately 60% of identified ecosystem services are degraded. Without proper management and planning, this degradation is expected to increase over the next 50 years (Millennium Ecosystem Assessment 2005; Wang et al. 2016).

Water yield (WY) is an ecosystem service that is affected by climate and land use changes (Guo et al. 2023; Hu and Gao 2020; Pei et al. 2022; Tijjani et al. 2022; Wang et al. 2016). It refers to the availability of water resources in a watershed, which is influenced by changes in precipitation and evapotranspiration. WY is estimated based on the difference between precipitation and evapotranspiration. This estimation is influenced by various factors, including climate change, agricultural and human activities, urban development, population growth, and global warming, which can cause significant changes in precipitation and evapotranspiration, ultimately leading to a reduction in WY. Therefore, it is essential to assess and measure the effects of climate change and human activities on WY to preserve and allocate water resources rationally across various regions (Guo et al. 2023; Tijjani et al. 2022; Xiao et al. 2020; Yifru et al. 2021).

Previous studies have investigated the impacts of climate and land use changes on WY. Guo et al. (2023) used the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) tool to examine the effects of climate and land use changes on WY in 17 sub-basins across China. They confirmed that these factors had serious impacts on WY according to the climatic, geographical, and socio-economic characteristics of the sub-basins. By utilizing the InVEST model, Mirghaed and Souri (2023) simulated WY in the Shoor River basin, southwestern Iran, and evaluated its relationship with land use changes, soil characteristics, and topographic features. Their study found that the topographic features had a higher contribution on WY than soil properties. Amiri et al. (2023) assessed the impacts of land use and climate change on the hydrological regime of the Tajan River in northern Iran using the soil and water assessment tools (SWAT). They reported that climate and land use changes will lead to water stress in their study region in the coming decades. Huang et al. (2023) conducted a study to investigate the changes in ecosystem services of WY and carbon sequestration in China’s Loess Plateau by combining InVEST and Carnegie-Ames-Stanford Approach (CASA) models. They concluded that climate change and human activities have mainly influenced the increase of WY in the region since 2001. Pei et al. (2022) analyzed annual changes in WY and their relationship with revegetation in northern China, encompassing past, present, and future effects. Their research indicated that changes in climate and land use were responsible for 88% and 12% of the changes in WY, respectively, from 2000 to 2019. Tijjani et al. (2022) used the General Circulation Model (GCM)-based SWAT model for two representative concentration pathways (RCP) namely RCP45 and RCP85 to assess the effects of short-term climate change on irrigation demand, green water scarcity, and crop yield in an agricultural watershed in southern New Jersey, USA. Their study demonstrated that increases in precipitation and temperature result in higher surface runoff, groundwater recharge, lateral flow, and total stream flow. Rafiei-Sardooi et al. (2022) simulated the effects of land use and climate change on WY, water supply, and water consumption in the Halil River basin in Iran using the InVEST software. In their study, the HadGEM2-ES model under RCP26, RCP45, and RCP85 was used to assess climate change, and the land change modeler (LCM) in TerrSet v19 was applied to map land use change. Daneshi et al. (2021) investigated the yield, supply, and consumption of water in a catchment in northern Iran using the WY model of the InVEST tool. They conducted a combined modeling of the effects of climate and land use scenarios on water security to estimate the water stress and economic costs associated with reducing WY in the future. Benra et al. (2021) assessed WY using the InVEST in 224 watersheds in southern Chile. They conducted this study for three years, 1998, 2007, and 2013 to investigate the effects of land use and climate change. Hu and Gao (2020) modeled WY over Shaanxi-Gansu Loess Plateau in China under climate and land use scenarios using the InVEST tool. Their study showed that the impact of climate change on the WY is much higher than that of land use change.

The impacts of land use and climate changes on the WY have been studied in various regions worldwide. Nevertheless, there have been limited investigations into their interactive effect on the WY in Iran. In previous studies, other critical factors such as topography and vegetation, which could impact the WY, have not been adequately evaluated. The Tajan watershed is a crucial component of the Hyrcanian forest ecosystem in northern Iran, serving as a vital habitat and ecological resource, as well as playing a significant role in providing regional water resources. In recent decades, environmental tensions and crises have arisen due to climate and land use changes, which have affected water resources and productivity in this region. Consequently, habitat conditions for humans and other creatures have been impacted. Therefore, this study investigated how land use and climate changes impact WY in the Tajan watershed, located in the north of Iran. Additionally, it sought to determine the relationship of WY to topography (slope and elevation) and vegetation (NDVI: Normalized Difference Vegetation Index). The study objectives were: (i) the simulation of climate and land use changes in the Tajan watershed over a 30-year period (2022 to 2052), (ii) the modeling of WY in the study area using climate and land use scenarios through the InVEST 3.13 software, and (iii) the assessment of the relationships of WY to elevation, slope, and NDVI at the sub-watershed level of the region in 2022.

Materials and methods

Study area

The Tajan River watershed, covering an area of about 4720 km2, is situated between 52° 18′ 17″ E to 54° 9′ 8″ E and 35° 55′ 57″ N to 36° 49′ 26″ N in Mazandaran province, northern Iran (Fig. 1). It boasted an average elevation and annual temperature of 1179 m and 25 °C, respectively. The region is predominantly composed of mountains, plains, and coastal plains, which account for around 85% of its terrain. The remaining 15% is occupied by foothills, hills, and river terraces. Agriculture and construction dominate the plains in the north, while forests and rangelands are found in the mountainous regions of the south and central watershed. The tributaries of the Tajan River, originating from the southern region, flow through the lowlands and highlands of the watershed, ultimately leading to the Caspian Sea. Most of the area is covered by Hyrcanian forests, demonstrating the ecological significance of its habitats for conserving biodiversity.

Fig. 1
figure 1

Geographical location of the study area

Full size image

Data collection

The dataset utilized in this study is presented in Table 1. The digital elevation model (DEM) was extracted from the shuttle radar topography mission (SRTM) images through the google earth engine (GEE). The GEE was also used to download Landsat 5, 7 and 8 images to create land use maps for 1992, 2012 and 2022. The meteorological statistics from 14 stations in and around the region spanning 30 years (1992–2022) have been used to gather data on precipitation, temperature, and radiation. The plant available water content was estimated based on soil’s clay, sand, silt, and organic matter percentages, obtained from global soil grid data downloaded from the GEE.

Table 1 Dataset used in this study
Full size table

Land use change simulation

Land use maps for the study area in 1992, 2012, and 2022 were created by analyzing surface reflectance images from Landsat 5, 7 and 8, respectively, downloaded from the GEE platform and classified using a support vector machine (SVM) algorithm in the ENVI 5.3 software. The classification accuracy was evaluated through visual interpretation and a confusion matrix. The matrix was based on ground reference points taken from Google Earth images and used to calculate statistics including user’s accuracy (UA), producer’s accuracy (PA), commission error (CE), omission error (OE), overall accuracy (OA), and kappa coefficient. The statistics were computed using the following equations (Eastman 2012):

$$UA = \frac{1}{n}\left( {\sum\limits_{i = 1}^{n} {\frac{{P_{ir} }}{{P_{tr} }}} } \right) \times 100$$
(1)
$$PA = \frac{1}{m}\left( {\sum\limits_{j = 1}^{m} {\frac{{P_{jc} }}{{P_{tc} }}} } \right) \times 100$$
(2)
$$CE = 1 - UA$$
(3)
$$OE = 1 - PA$$
(4)
$$OA = \frac{{P_{c} }}{{P_{t} }} \times 100$$
(5)
$$Kappa = \frac{{P_{t} \times \mathop \sum \limits_{i = 1}^{k} P_{c} - \mathop \sum \limits_{i = 1}^{k} P_{ir} \times P_{jc} }}{{P_{t}^{2} - \mathop \sum \limits_{i = 1}^{k} P_{ir} \times P_{jc} }}$$
(6)

where Pir and Pjc represent the number pixels correctly classified in row i and column j of the error matrix, respectively. Ptr and Ptc show the total number of pixels in row i and column j, respectively. Pc is the number of correctly classified pixels, while Pt shows the total number of evaluating pixels. n and m are the number of rows and columns of the error matrix. It should be noted that the kappa coefficient greater than 0.8 indicates acceptable classification accuracy (Shivakumar and Rajashekararadhya 2018).

The land change modeler (LCM) in the TerrSet v19 software was utilized to project land use changes for 2052. The LCM model integrates the Markov chain model, multi-layer perceptron neural network, logistic regression, and multi-objective land allocation, thereby rendering it highly efficient in simulating land use changes. The LCM modeling follows four crucial steps, namely assessment of land use changes, modeling of transition force, modeling of land use change, and accuracy assessment. In the transition force modeling, the LCM model utilizes auxiliary variables to evaluate the potential of each pixel of an image to change from one land use type to another. Auxiliary variables are drivers that influence land use change. The correlation between variables and land use is measured using Cramer’s V, which ranges from 0 to 1. Variables with a Cramer’s V greater than 0.15 are considered useful for predicting land use changes (Eastman 2012).

In this study, the land use maps of 1992 and 2022 served as inputs of the LCM model to project land use map of 2052. The study utilized several auxiliary variables, including elevation, slope, aspect, distance from roads, distance from residential centers, distance from agricultural lands, distance from rangelands, and distance from waterbodies. These variables were selected based on the environmental conditions of the region and Cramer’s V higher than 0.15. The model was validated by comparing the ‘reference’ map (the land use map of 2022 produced using the SVM method) with the ‘comparison’ map (the predicted land use map of 2022). The projected map of 2022 was created using the LCM model and the land use maps of 1992 and 2012. To measure the validity and accuracy of the model, the kappa values (Kno, Klocation, KlocationStrata, and Kstandard), and the statistics of DisagreementGridCell, DisagreementQuantity, AgreementQuantity, AgreementChance, AgreementGridCell, Hits, False Alarms, and Misses were calculated using the Validation module in TerrSet v19 software (Eastman 2012; Leta et al. 2021). For more information about the statistics, refer to Pontius and Suedmeyer (2004), Eastman (2012) and Leta et al. (2021).

Climate change projection

Various tools and models have been developed to project climate change. LARS-WG 6 has been frequently employed to project future climate changes through various global circulation models (GCMs). It is a user-friendly open-source tool that can simulate future climate scenarios based on different greenhouse gas emission scenarios (Semenov et al. 2002). By applying the LARS-WG 6 software, climate changes in the study area were projected for 2052 based on the CMCC-CM model under two different representative concentration pathways (RCP) scenarios, namely RCP45 and RCP85. The CMCC-CM model outperforms other GCMs for simulating climate parameters, specifically temperature and precipitation, over Iran (Abbasian et al. 2019). The model was supplied with data inputs comprising of the daily minimum temperature, daily maximum temperature, daily precipitation, and daily radiation collected from 14 different meteorological stations situated in and around the region.

Water yield modeling

InVEST models WY using average annual rainfall (Px), actual evapotranspiration (AETx), reference evapotranspiration (ET0), plant available water content (PAWC), edaphic-climatic characteristics, root-limiting soil depth and land use based on the water balance hypothesis according to Eq. 7 (Guo et al. 2023; Sharp et al. 2020).

$$WY = \left( {1 - \left( {\frac{{AET_{x} }}{{P_{x} }}} \right)} \right)*P_{x}$$
(7)

where WY, Px, and AETx are based on mm/pixel. The ratio of AETx/Px for each land use is determined by applying the Budyko curve expression (Fu 1981; Zhang et al. 2004).

$$\frac{{AET_{x} }}{{P_{x} }} = 1 + \frac{{PET_{x} }}{{P_{x} }} - \left[ {1 + \left( {\frac{{PET_{x} }}{{P_{x} }}} \right)^{\omega } } \right]^{1/\omega }$$
(8)

where PETx represents potential evapotranspiration and ω is an empirical parameter that represent natural edaphic-climatic characteristics (Donohue et al. 2012; Hargreaves and Samani 1982).

$$PET_{x} = k_{c} \left( {l_{x} } \right) \times ET_{0}$$
(9)

where ET0 is the reference evapotranspiration and Kc (ℓx) is the plant evapotranspiration coefficient on pixel x. Kc regulates the ET0 values based on the crop type or vegetation cover. ET0 can be calculated using the following equation (Droogers and Allen 2002):

$$ET_{0} = 0.0013 \times 0.408 \times RAD \times \left( {T_{m} + 17} \right) \times \left( {T_{d} - 0.0123P} \right)^{0.76}$$
(10)

where RAD refers to the extraterrestrial radiation (MJ m−2 d−1). Tm shows the daily average maximum and minimum temperatures (°C). Td is the difference between the daily average minimum and maximum temperatures (°C). P specifies the monthly precipitation (mm).

PAWC, or plant-available water capacity, denotes the available water supply to plants, determined as the contrast between the field capacity and the permanent wilting point. The calculation for this is as follows (Sharp et al. 2020):

$$\begin{aligned} PAWC = & 54.509 - 0.132 \times SA\% - 0.003 \times \left( {SA\% } \right)^{2} - 0.055 \times SI\% - 0.006 \times \left( {SI\% } \right)^{2} \\ & - 0.738 \times CL\% + 0.007 \times \left( {CL\% } \right)^{2} - 2.688 \times OM\% + 0.501 \times \left( {OM\% } \right)^{2} \\ \end{aligned}$$
(11)

where CL%, SA%, SI%, and OM% demonstrate the percentage of clay, sand, silt, and organic matter, respectively. The model was calibrated using the Z parameter, an empirical index that expresses the local rainfall pattern and hydrological characteristics estimated using Eq. 12.

$$Z = \frac{{\left( {\omega - 1.25} \right)P_{x} }}{AWC}$$
(12)

Available water content (AWC) is the volume (mm) of the water that can be held by plants, determined by soil texture and effective rooting depth. It was calculated in this study using global soil grid data (Sharp et al. 2020). For more information about WY-InVEST model, refer to Sharp et al. (2020).

Water yield scenarios

Six scenarios were designed based on land use and climate factors (precipitation and evapotranspiration) to project WY in the watershed for 2022 and 2052. As described in Table 2, each scenario considered the impacts of land use change, or climate change, or their combined effects on WY. The scenarios were implemented in the InVEST 3.13 software by altering the inputs of the WY model, including annual precipitation, evapotranspiration, and land use.

Table 2 Description of Land use and climate scenarios
Full size table

Statistical analysis

The study employed statistical analyses of the Getis-Ord (G*) statistic, bivariate local Moran’s I (BLMI), and geographically weighted regression (GWR) to investigate the relationships under examination. The G* statistic is used to assess the spatial autocorrelation of a variable and identify its hot and cold spots in space, with values greater than the average denoting a high-high cluster or hot spot, and those smaller than the average indicating a low-low cluster or cold spot. BLME can be used to evaluate the spatial autocorrelation of two variables and is more effective than the G* statistic. The x-axis of the Moran scatter plot shows one variable, while the y-axis displays the spatial lag of another variable. The closer Moran’s I is to + 1 or − 1, the stronger the spatial autocorrelation, and if it approaches zero, it indicates weaker spatial autocorrelation (Anselin 1995; Getis and Ord 1992; Ord and Getis 1995). The GWR is an extended form of ordinary least square (OLS) regression that evaluates the spatial correlation between two variables by considering their local spatial relationships. It examines heterogeneous spatial relationships between dependent and independent variables at every geographic location in an unstable space. This method enables the assessment of local correlation between geographic parameters (Brunsdon et al. 1998; ESRI 2016).

By applying the GeoDa 1.18 software, the G* statistic was used to investigate the hot and cold spots of WY at the sub-watershed level. Spatial autocorrelation of WY to elevation, slope, and NDVI was assessed through the BLMI statistic. Additionally, the spatial correlation of WY with elevation, slope, and NDVI was also evaluated using the GWR in ArcGIS 10.7. These statistical analysis were conducted with data from the region in 2022.

Results and discussion

Land use and climate changes

Figure 2 indicates the land use maps of 1992 and 2022, as well as the projected map of 2052 generated using the LCM model. Notably, the land use classification accuracy of 1992 yielded a kappa coefficient, overall accuracy, producer’s accuracy, user’s accuracy, omission error, and commission error of 0.90, 93%, 91%, 95%, 9% and 5%, respectively. For the 2022 land use classification, these metrics were estimated at 0.91, 94%, 93%, 92%, 7% and 8%, respectively. These results demonstrated the classifications’ acceptable validation and accuracy. Ambarwulan et al. (2023) calculated the kappa coefficient and overall accuracy for modeling land use changes in the Cisadane watershed in Indonesia to be over 83%. Sisay et al. (2023) estimated the kappa coefficient and overall accuracy in land use classification in the Goang watershed, Ethiopia, for different years to be higher than 84% and 86%, respectively. The model validation and accuracy were proven by the estimated kappa values (Kno = 0.97, Klocation = 0.98, KlocationStrata = 0.98, and Kstandard = 0.96) being above 0.8 (Viera and Garrett 2005). Additionally, the statistics of DisagreementGridCell, DisagreementQuantity, AgreementQuantity, AgreementChance, AgreementGridCell, Hits, False Alarms, and Misses were recorded as 0.01, 0.01, 0.20, 0.14, 0.64, 16.1%, 1.6%, and 0.9%, respectively. These findings proved the acceptability of the model’s predictions (Eastman 2012; Leta et al. 2021). The area allocated to each land use is specified in Table 3. From 1992 to 2022, the forest, rangelands, agriculture, built-up areas, and waterbodies decreased by − 78,440, − 413, + 72,663, + 6097 and + 94 ha, respectively. Predictions showed that from 2022 to 2052, the mentioned land uses will further decrease by − 56,640, + 2117, + 46,229, + 7994, and + 3 ha, respectively.

Fig. 2
figure 2

Land use maps of the study area for 1992, 2022, and 2052

Full size image
Table 3 The area allocated to each land use in 1992, 2022 and 2052 and their changes
Full size table

The study revealed that the region experienced significant land use changes from 1992 to 2022. Approximately 22% (102,960 ha) of the region was affected by such changes. Forests experienced the highest decrease (-17%), while agricultural land increased the most (+ 15%) (Fig. 3). It is predicted that 17% (equal to 81,113 ha) of the watershed will undergo land use conversion from 2022 to 2052. Forests will experience the highest decline (− 12%), while agriculture will have the highest increase (+ 10%). In addition, Fig. 4 shows the changes in precipitation and evapotranspiration in the sub-watersheds of the region from 1992 to 2052. Precipitation and evapotranspiration exhibited a rising trend from 1992 to 2022, with downward and upward trends forecasted for them from 2022 to 2052, respectively. The evaluation of uncertainty in predicting climate parameters was based on observational and historical data. The correlation coefficient and the normalized root mean square error (NRMSE) were estimated, resulting in 0.95 < R < 0.99 and 2% < NRMSE < 10%. These values demonstrate the validity and accuracy of the climate prediction.

Fig. 3
figure 3

Land use change in the study area during the periods of 1992–2022 and 2022–2055

Full size image
Fig. 4
figure 4

Changes in a precipitation (P) and b evapotranspiration (ET) of the study area in 1992, 2022, and 2052. P and ET related to 2052 are projected based on RCP45 and RCP85

Full size image

Water yield

The InVEST 3.13 software was utilized to model the WY in the study area across six scenarios. The resulting data is demonstrated in Fig. 5 with an average WY of 4.9, 5.3, 2.6, 1.7, 2.7, and 1.8 mm/pixel (30 × 30 m) for the LC22, LC52, CC45, CC85, LCC45, and LCC85 scenarios, correspondingly. As presented in Table 4, the total volume of WY based on the aforementioned scenarios is estimated as follows: 23.4, 25.1, 12.3, 8.3, 12.9, and 8.7 Mm3 y−1 (on average 49.5, 53.3, 26.2, 17.6, 27.3 and 18.4 m3 ha−1, respectively). Figure 6 illustrates the classified maps of WY, while Table 5 details the area allocation for each class. The WY of 0–10 mm ha−1 is dominant in most of the region, covering 35%, 28%, 49%, 58%, 44% and 55% of its surface in the LC22, LC52, CC45, CC85, LCC45, and LCC85 scenarios, respectively. The WY of more than 50 mm ha−1 occupied 20%, 23%, 10%, 8%, 11%, and 8% of the region calculated in the mentioned scenarios, respectively. Figure 7 elucidates the trend of changes in WY at the sub-watershed level in different scenarios. The sub-watersheds 4, 11, 17, 32, 31, and 46 displayed the greatest changes in WY. Table 6 provides the WY for each land use type in all scenarios. Forests had the lowest WY calculated at 8, 18, 3, 2, 3, and 2 m3 ha−1 in the LC22, LC52, CC45, CC85, LCC45, and LCC85 scenarios, respectively, while the highest WY was found in the constructed areas estimated at 654, 642, 162, 82, 163 and 81 m3 ha−1 in the mentioned scenarios, respectively.

Fig. 5
figure 5

Maps of water yield resulting from the implementation of all scenarios in InVEST

Full size image
Table 4 The volume of water yield in the study area predicted in all scenarios and its comparison with the LC22 scenario
Full size table
Fig. 6
figure 6

Classified maps of water yield for all scenarios

Full size image
Table 5 The area (ha) allocated to each water yield class in all scenarios
Full size table
Fig. 7
figure 7

Water yield changes in the sub-watersheds of the region for each scenario

Full size image
Table 6 Mean water yield (mm ha−1) for land uses in each scenario
Full size table

Statistical analysis

The G* statistic was utilized to analyze the distribution of WY, as depicted in Fig. 8. The findings indicate that sub-watersheds located in the northern parts of the region were WY hot spots (0.04 < G* < 0.14) across different scenarios. Meanwhile, sub-watersheds situated in central parts were cold spots (0.001 < G* < 0.01). Additionally, the BLMI statistic was employed to investigate the spatial autocorrelation of WY with elevation, slope, and NDVI, as illustrated in Fig. 9. The estimated Moran’s I index for the WY with elevation, slope, and NDVI were − 0.53, − 0.62, and − 0.01, respectively. Furthermore, Fig. 10 clarifies the spatial correlation of WY with elevation, slope, and NDVI analyzed using the GWR in ArcGIS 10.7. The coefficient of determination (R2) between WY with elevation, slope, and NDVI was found to be 0.94, 0.90, and 0.96, respectively.

Fig. 8
figure 8

Results of the Gettis-Ord (G*) statistic to determine hot and cold spots of water yield in all scenarios

Full size image
Fig. 9
figure 9

Results of bivariate local Moran’s I (BLMI) to determine the relationship of water yield to a elevation, b slope, and c NDVI

Full size image
Fig. 10
figure 10

Spatial correlation of water yield to a elevation, b slope, and c NDVI obtained from geographically weighted regression (GWR)

Full size image

Discussion

From 1992 to 2022, the study area experienced significant land use changes, resulting in land use conversion in 22% of the region. The most notable alterations were the reduction of forests and the increase in agriculture, which covered 17% and 15% of the watershed, respectively. The findings indicated that the Tajan watershed was affected by land use changes during the aforementioned period. These changes reduced natural habitats and altered ecological conditions of the region. Agricultural activities have been one of the most important factors of land use change in the region, which has increased due to population growth and lack of proper land management in the past. Land use change is predicted to be severe from 2022 to 2052, significantly impacting the destruction of the region’s forests and reducing habitat quality. Additionally, it can impact water resources and soil conservation in the watershed.

The northern parts of the region had the highest WY, with the central parts having the lowest. This trend was consistent across all scenarios. Higher precipitation and more agriculture and construction in the northern parts were the primary factors behind the WY increase there. This study evaluated WY under various scenarios to measure the impact of climate change and land use, both separately and in combination, on WY variation in the region. The LC22 was the baseline scenario implemented based on land use and climate conditions of 2022. The LC52 scenario illustrated the effect of land use change on water yield, while the CC45 and CC85 scenarios elucidated the impact of climate change. The combined impacts of climate and land use changes on WY were demonstrated by the LCC45 and LCC85 scenarios. The comparison of scenarios indicated that the LC52 scenario yielded the highest water volume (25.1 Mm3 y−1 with an average of 53.3 m3 ha−1), while the CC85 scenario produced the lowest volume (8.3 Mm3 y−1 with an average of 17.6 m3 ha−1). Comparison of the LC22 scenario with the LC52, CC45, CC85, LCC45, and LCC85 scenarios exhibited that WY varied by 7%, − 47%, − 65%, − 45%, and − 63%, respectively. Additionally, it was demonstrated that land use change contributed 2% to 7% towards an increase in WY, while climate change contributed 47% and 65% (based on RCP45 and RCP85, respectively), towards a reduction in WY. It was also found that climate change based on RCP85 impacts WY variation by 18% compared to RCP45. According to Yifru et al. (2021), WY varied from 8 to 48% in their study area due to the impact of climate change. Pei et al. (2022) confirmed that alterations in WY were predominantly due to climate and land use changes accounting for 88% and 12% respectively. Rafiei-Sardooi et al. (2022) highlighted the significant impact of climate change compared to land use change on WY. Additionally, Hu and Gao (2020) estimated that climate change had a contribution of over 92% on WY.

The results proved that the WY experienced limited changes under the influence of land use change (based on the LC52 scenario), but severe variations under the effect of climate change (under the CC45 and CC85 scenarios). The impact of climate and land use changes on extreme changes in the WY was also observed under LCC45 and LCC85 scenarios. The study affirmed that the change in climate conditions and land use had a decreasing and increasing effect on WY in the study area, respectively. However, it should be noted that the effects of climate change could be much more severe than those resulted from land use change. Land use change results in variations in vegetation cover and soil permeability, which increases WY. On the other hand, climate change leads to a decrease in precipitation and an increase in evapotranspiration, thus causing a decrease in the total volume of the WY. Yifru et al. (2021) highlighted the significant impact of climate change scenarios on WY and hydrological processes in comparison to land use change. Pei et al. (2022) demonstrated that climate has a greater impact on WY changes than land use.

The lowest WY of 0–10 mm/ha was mainly observed in central forested areas of the region, while the highest WY (> 50 mm/ha) was estimated in the northern parts that encompassed agriculture and construction activities. The hot spot analysis also revealed that the sub-watersheds located in the northern parts were the primary hot spots of WY, while the central sub-watersheds were designated as cold spots. Built-up areas had the highest WY, while forests had the lowest. The order of WY for land uses was as follows: built-up areas > agriculture > rangelands > forest. The non-permeable built-up surfaces increase WY, while the vegetation in forests helps to preserve and supply water, and thus reduces WY (Guo et al. 2023). Mirghaed and Souri (2023) evaluated the WY of the Shoor River basin in southwest Iran using the InVEST model and found that it is higher in built-up areas than in forest, rangeland, and agriculture. This is in line with the results reported by Yang et al. (2021) and Lang et al. (2017).

Analysis of hotspots in different scenarios revealed that the spatial distribution of WY in sub-watersheds varies due to the impact of climate and land use changes. This fact demonstrates that the effects of climate and land use changes vary depending on the local ecological conditions and human activities that govern each sub-watershed. As a result, the WY also differs among them. These differences can impact the water resources management in each sub-watershed. Hence, Water resource conservation policies and programs should be customized to local conditions in each sub-watershed to mitigate the impact of climate and land use changes.

The spatial autocorrelation analysis revealed that WY has a moderate relationship with elevation and slope (-0.62 < Moran’s I < − 0.53). No significant relationship was found with NDVI (Moran’s I = − 0.01). These findings indicated that at the sub-watershed level, the spatial distribution of WY is almost opposite to that of slope and elevation in the region. However, there is no relationship between the spatial distribution of WY and vegetation. Mirghaed and Souri (2023) have also reported a significant spatial autocorrelation of WY to elevation and slope. Other studies have also highlighted the significant relationship between topographic features and ecosystem services, particularly with regards to WY (Ma et al. 2021; Shen et al. 2021).

The results of GWR analysis indicated WY has varying local spatial correlations with elevation, slope, and NDVI at the sub-watershed level in the region. Moreover, in most sub-watersheds, WY exhibited a higher local correlation with NDVI compared to elevation and slope. In 37 sub-watersheds, the local spatial correlation of WY with elevation was determined to be low to medium (0 < R2 < 0.6), while in 9 sub-watersheds it was strong (0.6 < R2 < 0.8). Similarly, the local correlation of WY with slope was found to be low to medium in 40 sub-watersheds (0 < R2 < 0.4), and strong in 6 sub-watersheds (0.6 < R2 < 0.8). The local spatial correlation of WY with NDVI was estimated to be low to moderate in 27 sub-watersheds (0.0 < R2 < 0.6), strong in 16 sub-watersheds (0.6 < R2 < 0.8), and very strong (0.8 < R2 < 1.0) in 3 sub-watersheds. These findings suggested that the local spatial correlation of WY with elevation, slope and NDVI is influenced by the geographic location. Despite such differences at the level of sub-watershed, WY had a strong spatial correlation with elevation, slope, and NDVI at the watershed scale (0.90 < R2 < 0.96), implying that topography and vegetation have a significant impact on the WY in the study region. Earlier studies have also referenced the influence of topography and vegetation on the WY (Ahmadi Mirghaed and Souri 2022; Kusi et al. 2020; Mirghaed and Souri 2023; Yang et al. 2021). It was concluded that the impact of topography and vegetation on WY can vary depending on the geographic location. These variations can influence water resource policies, planning, and management in each sub-watershed. Therefore, it is crucial to consider them for effective local water resource management.

Limitations and future prospects

This study has provided a significant understanding of WY under the influence of climate and land use changes in the Tajan watershed, northern Iran. However, it faced certain limitations. The lack of an appropriate database for the study area is considered to be a major limitation in this regard. Some of the inputs of the WY-InVEST model, such as plant rooting depth and Kc coefficient, were prepared based on scientific sources and global databases due to time and cost constraints, which may somewhat affect the results at the local scale. Therefore, it is important to establish a strong database in this regard in Iran for proper assessment and management of water resources.

InVEST evaluates WY based on the difference between annual precipitation and evapotranspiration. However, it has limitations. First, it is not able to evaluate WY at sub-annual scales (seasonal, monthly and daily). Second, it considers the characteristics of rooting depth and Kc coefficient to be the same for a land use type in different areas, which may increase the uncertainty of the results. Third, this model does not consider soil permeability and topographic features, which are important factors affecting WY.

More research is needed to study WY at different scales for water management strategies. Studying the effects of climate and land use changes on WY and their implications for sustainable water resources management is very important. Therefore, the development of advanced simulation methods to predict and reduce the long-term effects of climate and land use changes on water supply and demand is of particular importance. In future research, it is necessary to consider appropriate approaches to evaluate various environmental, economic, social and political issues related to watersheds in the world, especially in Iran.

Conclusion

This study assessed WY due to climate and land use changes in the Tajan watershed, north of Iran, from 2022 to 2052. The findings indicated that the northern parts of the watershed had the highest WY, while the central parts had the largest WY. In addition, land use changes led to a 2–7% increase in WY, whereas climate change caused a WY reduction of 47% under RCP45 and 65% under RCP85. There was a strong correlation between WY and elevation, slope, and NDVI. Moreover, their correlations can be varied depending on geographical location. In conclusion, this study confirmed that climate change had a greater impact on WY than land-use changes. Additionally, WY can be influenced by local topography and vegetation conditions. This study suggested that managers and planners should adopt appropriate strategies to conserve water in the future regarding water conservation policies, dealing with climate and land use change, and ecological conditions of the region.