In this study, PARI was constructed based on three fundamental sub-dimensions—efficiency, diversity, and stability—in order to evaluate the performance of agricultural production systems in the face of climatic and structural shocks within a multidimensional framework (Meuwissen et al., 2019). The subsequent section delineates the definition and calculation method for these sub-dimensions.

The calculated productivity, diversity, and stability sub-indices were harmonized using the equally weighted arithmetic mean method, under the assumption that each component has an equivalent theoretical weight on agricultural resilience (Nardo et al., 2005). The composite PARI scores obtained as a result of this aggregation process were determined as the final indicator representing the agricultural system resilience of provinces and formed the basis of the spatial econometric modeling stage.

Drought: to represent the effect of meteorological drought on PARI, the 3-month Standard Precipitation Index (SPI-3) for June was used as the basic climate shock variable. The SPI-3 value for June summarizes the deviation of cumulative precipitation in April, May, and June from long-term norms. This value corresponds to the critical phenological window (stem elongation and grain filling stage) that determines yield formation, especially in dry farming systems in Türkiye (Alkan and Tombul, 2022). SPI calculations were performed in the Google Earth Engine (GEE) environment using monthly cumulative precipitation series derived from daily CHIRPS precipitation data.

During the calculation process, in accordance with the method defined by McKee et al. (1993) and standardized by the World Meteorological Organization (WMO) (2012), the monthly precipitation series for the 1991–2020 reference period were modeled separately for each month using the gamma probability density function. For non-zero precipitation values, the estimation of shape ( α = μ 2 / σ 2 ) and scale ( β = σ 2 / μ ) parameters was undertaken employing the moment method on a pixel-by-pixel basis. Furthermore, cumulative probability values were obtained for the target period 2005–2024. Values calculated from the cumulative probability function H ( x ) = q + ( 1 − q ) G ( x ) with zero probability correction were converted to standard normal space via the inverse error function ( er f − 1 ) , thereby producing the SPI series. The resulting monthly SPI surfaces were then spatially aggregated within the administrative boundaries of 81 provinces and reduced to the province-year level. In conducting the econometric analyses, the drought shock was interpreted in terms of its “severity” and transformed into an indicator-SPI format for the purposes of analysis and presentation.

Growth season temperature anomaly: to distinguish between the various channels of climatic stress affecting agricultural production, temperature-induced thermal stress has been designated as a separate variable in addition to precipitation-induced shocks. The July–September period is a time of particular significance for summer crops (e.g., corn, cotton, and sunflowers) and orchards. This period is typified by elevated temperatures and evaporation pressure rather than precipitation. Therefore, a Growth Season Temperature Anomaly variable encompassing the active growth season (April–October), when agricultural operations intensify, has been formulated as an alternative to annual averages (Gürkan et al., 2024). This indicator is defined as the deviation of each province’s average growth season temperature from 2005 to 2024 from the 1991–2020 long-term climate normals. This approach endeavors to delineate the disparate impact mechanisms of precipitation- and temperature-induced climate shocks on PARI, thereby circumventing mechanical overlap.

Other variables: in addition to climate shocks, a series of variables have been integrated into the model to control for the economic and structural determinants of agricultural resilience. To reflect financial capital access and liquidity capacity, the total agricultural credit volume at the provincial level has been realized and defined as the real credit amount per hectare (million TL/ha). To represent energy intensity and technical input use in the production process, electricity consumption per unit area (MWh/ha) has been included in the model. To capture the effects of farm structure and economies of scale, the average farm size (ha/farm) variable has been calculated by dividing the total agricultural area by the number of farms. Finally, horsepower per unit area (HP/ha) was defined to measure the level of mechanization; the horsepower characteristics of two-axle tractors were used in the calculations to ensure technical consistency and standardization. The list of variables used in the spatial econometric modeling is given in Table 1.

The temporal and spatial evolution of PARI was examined using a series of statistical tests and visualization techniques within the framework of exploratory spatial data analysis (ESDA). To assess the change in inter-provincial resilience differences over time in terms of convergence or divergence dynamics, σ-convergence analysis was applied using the coefficient of variation (CV) (Kındap and Dogan, 2019). Furthermore, the statistical significance of long-term trends in PARI values was examined through the application of the non-parametric Mann–Kendall trend test at the provincial level (Gumus et al., 2021). To identify the presence and strength of spatial interactions, geographical neighbor relationships between provinces were defined using the first-degree Queen’s adjacency matrix (W), which was subsequently row-standardized (Griffith, 2018). The spatial dependency structure of the dependent variable was tested annually using the Global Moran’s I statistic, and local clustering patterns were examined through the application of Local Spatial Autocorrelation Indicators (LISA). These analyses revealed that agricultural resilience forms High–High (hot spots) and Low–Low (cold spots) clusters concentrated in specific regions.

To analyze the determinants of agricultural resilience and measure the spatial spillover effects of climatic and structural shocks, the Spatial Durbin Model (SDM), which includes spatial lags of both the dependent and independent variables, was preferred. According to LeSage and Pace (2009), the SDM exhibits the greatest robustness with respect to “omitted variable bias.” Moreover, due to the generality of its structure, which encompasses spatial autoregressive (SAR) and spatial error (SEM) models, the SDM is able to capture exogenous spatial interactions with greater flexibility. The general form of the model is expressed as follows (Equation 1):

Here, y it represents the PARI for period t in province i, W represents the spatial weight matrix, ρ represents the spatial autoregressive coefficient, X represents climate shocks and structural control variables, and θ represents the spatial spillover effects of the independent variables. μ i and γ t represent the province and time fixed effects, respectively. The choice between fixed and random effects in favor of the SDM as the final model was first made using the Hausman test. Subsequently, the reducibility of the SDM to more restricted spatial models (SAR and SEM) was tested using the likelihood ratio (LR) and Wald tests. Finally, alternative spatial specifications were compared using information criteria (AIC and BIC).

The Maximum Likelihood (ML) method was used for econometric estimation, taking into account the panel data structure. Due to the potential for direct interpretations of coefficients in spatial models to be misleading owing to the presence of spatial feedback mechanisms, the marginal effects of the independent variables were decomposed into direct, indirect/spillover, and total effects in accordance with the methodology of LeSage and Pace (2009). The decomposition process was implemented using a simulation-based impacts procedure over 1,000 iterations (R = 1,000).

Additionally, to assess the role of financial access under climate shocks, interaction terms between drought (D) and real credit intensity (ln_C) variables were included in the model; and the variation in the marginal effects of drought depending on the credit level was analyzed. This approach aims to reveal the conditional effects of financial access on agricultural resilience under climatic stress conditions. Furthermore, the potential simultaneity and reverse causality between agricultural credit and agricultural resilience were considered, acknowledging the possibility of a bidirectional causal relationship. To address this issue, alternative SDM specifications were estimated using one-period lagged values of the credit variable, and robustness analyses were conducted using winsorized variables to limit the influence of outliers.

Table 2 shows the descriptive statistics of the key variables used in the analysis, as well as the differences between provinces. As shown by these statistics, the resilience level for province-year observations is concentrated around relatively high and medium values. The average PARI index is 0.61, and the narrow IQR range (0.57–0.66) indicates a similar level of resilience in most provinces. However, the minimum value of 0.20 and maximum value of 0.93 show significant differences at the extremes. Examining the sub-dimensions reveals that diversity (mean = 0.82) and stability (mean = 0.84) are consistently high, while the level of efficiency (mean = 0.16) is lower and distributed over a much wider range (0–1). This suggests that productivity is the limiting component of resilience in many provinces.

Panel B reveals strong heterogeneity in terms of both climate shocks and adaptation capacity indicators. Although the drought indicator is close to zero on average (−0.07), the wide range between −3.50 and 3.10 indicates that extremely wet and extremely dry years coexisted. The positive average temperature anomaly (0.48) and its upper limit of up to 3.40 point to a clear warning signal throughout the period. The wide spread observed in the variables of credit (ln_C), machine power (ln_M), electricity usage (ln_E), and farm size (ln_S) indicates significant differences between provinces in terms of access to finance, mechanization, energy, and land. This suggests that spatial resilience differences may largely stem from differences in adaptation capacity.

The 0 values observed in the index sub-dimensions represent the lowest relative performance within the sample as a result of min-max normalization and do not imply absolute inefficiency/instability. Similarly, negative values in the ln_C, ln_M, ln_E, and ln_S variables are a natural logarithmic result of levels per hectare remaining below 1.

The correlation matrix reveals that PARI displays moderate positive correlations with capacity indicators that reflect mechanization and energy utilization (ln_M, ln_E). Conversely, PARI exhibits correlations approaching zero with drought (D) and temperature anomaly (T). This structure indicates two distinct points. Firstly, there is an absence of strong multiple linear connections between the determinants. Secondly, the effect of climatic pressures on resilience may become evident primarily through interaction terms and the panel/spatial model structure, as outlined in Table 3.

To reveal the spatial pattern of agricultural resilience in Türkiye during the 2005–2024 period, PARI values calculated at the provincial level were mapped for two extreme years (2005 and 2024) (Figure 2). The maps show that the mean resilience level throughout the country has experienced a marginal increase (2005: 0.600, 2024: 0.610). However, this increase does not demonstrate spatial homogeneity. It is evident that while the lower boundary of the PARI distribution exhibited an increase over the specified period (0.273 → 0.336), the accelerated rise in the upper limit (0.740 → 0.928) signifies an expansion in the disparities in resilience among the provinces. At the provincial tier, high resilience values are notably concentrated within the Western and Mediterranean regions. In the year 2024, Antalya (0.928) and Mersin (0.880) emerged as the provinces with the highest PARI values. Meanwhile, provinces such as Bursa (0.783), Izmir (0.774), and Çanakkale (0.723) were identified as constituting the upper resilience group. A factor analysis, supported by the sub-dimension change maps, reveals that the increase in these provinces is primarily attributable to productivity gains, accompanied by enhancements in the stability component (Figure 3). In contrast, it is noteworthy that resilience has declined or remained negligible in certain provinces in Eastern and Northeastern Anatolia. The declines observed in provinces such as Bayburt (ΔPARI = −0.142) and Kars (ΔPARI = −0.132) are related to deterioration in the stability and diversity sub-dimensions, rather than productivity.

As illustrated in Table 4, PARI exhibited a predominantly stable trajectory in relation to the national average during the 2005–2024 interval, with mean values ranging from 0.60 to 0.62. A notable exception was the limited increase observed in 2022. However, the increase in standard deviation and, in particular, the coefficient of variation (CV) from 0.136 to 0.17 indicates that the differences in resilience between provinces have not decreased. Rather, a dynamic of sigma-divergence than sigma-convergence prevails. Concurrently, the systematic increase in Moran’s I coefficient from the 0.26–0.33 band to above 0.40, along with its statistical significance at the 1% level in all years, indicates that resilience is not randomly distributed and that increasingly distinct spatial clusters are forming. Indeed, the arithmetic mean of the Global Moran’s I statistics, which were calculated on an annual basis for all provinces during the 2005–2024 period, yielded a value of 0.343. This finding substantiates the presence of persistent spatial clusters comprising provinces with both high and low resilience throughout the aforementioned period. Moreover, this result substantiates the necessity of incorporating models that consider spatial interactions in the context of the examined phenomenon. It is imperative to acknowledge the methodological necessity of SDM, among other similar methodologies.

Figure 3 shows that the long-term net change in PARI during the 2005–2024 period exhibits a distinct spatial divergence across Türkiye. The high positive ΔPARI values observed in the Southwest and Mediterranean regions (particularly Antalya, Mersin, and surrounding provinces) indicate that agricultural resilience has strengthened in these regions over time. These increases are consistent with the relatively developed irrigation infrastructure in these provinces, the high value-added and diversified structure of the product mix, and the more widespread availability of financial access. In contrast, the concentrated negative ΔPARI values in Eastern and Northeastern Anatolia (e.g., the Erzurum–Kars plateau and surrounding areas) indicate that agricultural resilience in these regions has weakened over the long term. These provinces struggle to sustainably increase resilience gains due to limited production diversity, high exposure to climatic variability, and relatively low adaptive capacity.

In Central Anatolia and transitional regions, the clustering of ΔPARI values around zero suggests that agricultural resilience in these areas has remained stagnant. This situation can be explained by the production structure relying heavily on dry farming and limited technological–institutional alignment. Notably, the spatial clustering of positive and negative changes indicates that the long-term evolution of PARI is not random; rather, it is shaped by the combined effects of regional production systems, climatic conditions, and policy environments.

Figure 4 shows the temporal evolution of the annual Global Moran’s I coefficients calculated for PARI during the 2005–2024 period. The findings reveal that spatial autocorrelation was consistently positive and statistically significant throughout the period, and also showed a marked upward trend, particularly after 2015. This trend indicates that agricultural resilience levels are progressively diverging into stronger spatial clusters across provinces, with resilience gains becoming concentrated in geographically limited areas.

As illustrated in Figure 5, the long-term trends of PARI at the provincial level during the 2005–2024 period manifest a spatially heterogeneous structure. The application of the Mann-Kendall test to the data reveals that statistically significant increasing trends prevail in most of Western and Central Anatolia, while decreasing trends are spatially concentrated in Eastern and Northeastern Anatolia. Furthermore, provinces where no significant trend was detected indicate that resilience dynamics followed a relatively stable or fluctuating course over time.

As Figure 6 illustrates, PARI demonstrated notable and statistically significant spatial clusters at the provincial level in 2024. The High–High clusters in the western and southwestern regions show that provinces with high agricultural resilience are contiguous with neighbors displaying comparable characteristics, while the Low–Low clusters in Eastern and Northeastern Anatolia suggest that low resilience levels are geographically persistent and regionally endemic. The limited number of High–Low and Low–High observations confirms the existence of spatial heterogeneity, pointing to provinces that stand out from their surroundings. Overall, the results obtained from LISA confirm that agricultural resilience is not distributed randomly but rather exhibits a strong local spatial dependency structure.

Figure 7 shows that high drought intensity is predominantly concentrated in the southeastern region of Anatolia and the southern part of central Anatolia; in contrast, lower levels of drought pressure are observed in the western and coastal regions.

Table 5 shows the diagnostic tests that were performed to determine the suitable model specification for the analysis. The Hausman test, which was applied in the first stage to determine the panel data structure, was found to be statistically significant at the 1% level (161.607, p < 0.01). This result suggests that the fixed effects estimator is preferable to the random effects estimator for controlling for unobservable, region-specific heterogeneity. The “general-to-specific” approach proposed by LeSage and Pace (2009) was used to select the spatial model. Specifically, it was tested whether SDM, could be reduced to the restricted models: SAR and SEM. The results of the Likelihood Ratio (LR) test prove the superiority of the SDM model over the restricted models and reject model simplification (p < 0.001). Likewise, Wald tests confirmed that the spatial lags of the independent variables (θ = 0) differ significantly from zero and that the “common factor hypothesis” is invalid. In addition to the findings from the diagnostic tests, the SDM had the highest log-likelihood value (3855.79) and the lowest Akaike information criterion (AIC: −7685.57) compared to other models. Based on this statistical evidence, it was concluded that the Spatial Durbin Model with Fixed Effects best explains the spatial dependency structure in the data set.

The Robust LM (RS) diagnostic tests in Table 6 show no strong and consistent evidence of spatial error dependence over the 20-year period examined. The mean RSerr statistic remains low; the median p-value is above the 0.05 significance threshold, at 0.162; and the proportion of significant years is limited to 15%. These results suggest that using SEM-type error dependence to explain the data is inadequate; a more appropriate approach is to model spatial interactions through the spatial components of explanatory variables and diffusion channels.

As shown in Table 7, agricultural resilience in the face of climate shocks depends not only on the amount of inputs, but also on the economic context and purpose of their use. While the positive coefficients of agricultural credit (ln_C) and mechanization (ln_M) in the baseline pooled OLS model suggest these inputs support production capacity under average conditions, this relationship significantly reverses when fixed effects and spatial dependence are considered. The negative direct effects of credit and mechanization in the final SDM results suggest that these resources are primarily used for defensive strategies under climate stress, such as risk management, income stability, and compensating for production losses, rather than for investments that enhance yield. These results are consistent with region-specific production practices in Turkish agriculture, where credit and mechanization are predominantly employed within small-scale and fragmented farm structures, limiting their role in supporting productivity-enhancing and climate-adaptive investments, as discussed in Section 3.2.3. The consistently negative and significant drought indicator (D) confirms that climate-induced shocks directly erode agricultural resilience at the provincial level. The negative value of the credit-drought interaction term indicates that increased credit use during drought periods functions as a short-term adaptation response to maintain current production levels, rather than as a structural transformation to strengthen resilience. On the other hand, the significance of the spatial lag variables and the positive spatial autoregressive parameter (ρ) demonstrate that agricultural resilience is shaped by both intra- and interprovincial factors, including regional interactions that spread through production conditions in neighboring provinces, market integration, and shared climate risks.

Table 8 shows that the key findings regarding the agricultural credit variable are not affected by potential simultaneity or reverse causality issues. The negative and significant credit coefficient obtained in the baseline SDM results remains consistent in magnitude and statistical significance in the alternative specification with the lagged credit variable. This suggests that the impact of credit on resilience occurs via a time-lagged mechanism rather than being determined simultaneously. Similarly, the stronger and more significant interaction term between drought and credit in the lagged model indicates that the baseline results are not dependent on periodic shocks or short-term feedback effects. The stability of the credit coefficient in the winsorized SDM specification, which limits the effect of outliers, confirms that the findings are not sensitive to extreme values or the tail of the distribution. Taken together, these results show that the study’s key findings are robust to modeling preferences and data structure and that the conclusions regarding the role of credit are based on a reliable causal foundation.

As shown in Table 9, the marginal effect decomposition clarifies that the determinants of agricultural resilience operate not only locally, but also through strong spatial channels. The statistically significant and negative direct (local) and indirect (spillover) effects of the drought indicator (D) reveal that climate shocks spread through commercial networks and shared basin risks, creating regional contagion and not remaining isolated within provincial borders. This suggests that drought in neighboring provinces can erode the focus province’s resilience through commercial integration. Similarly, the negative and significant total effect for the credit (ln_C) and mechanization (ln_M) variables shows that these inputs do not automatically increase resilience under average conditions. The negative and significant indirect effect of mechanization (−0.0047) suggests that capital concentration may carry a regional component that increases the fragility of the production structure under certain conditions rather than creating positive externalities within the neighborhood network. The negative and significant direct and total effects of the drought–credit interaction (Dxln_C) indicate that credit expansion during drought periods functions as an adjustment mechanism aimed at buffering short-term shocks (“debt rollover”) rather than as a resilience-enhancing structural investment. In contrast, the marginal effects of the scale variable (ln_S) being statistically insignificant implies that local farm size alone is not a decisive factor in resilience.

Figure 8 shows that the marginal effect of drought (D) on PARI changes systematically depending on the credit level. While the effect of drought is relatively limited and statistically weak in provinces with low credit intensity, the negative effect of drought on agricultural resilience becomes significantly stronger as the credit level increases. The fact that confidence intervals fall below zero at high credit levels indicates that this relationship becomes statistically significant. This finding suggests that increased credit usage during drought periods functions as an adjustment mechanism aimed at liquidity and loss compensation rather than productive investments, and therefore does not play a resilience-enhancing buffer role.

The study of the spatial distribution of agricultural resilience (PARI index) in Türkiye between 2005 and 2024 reveals that, despite a limited increase nationwide, regional inequalities have deepened. This finding largely aligns with general trends in the existing literature regarding the spatial transformation of agricultural production. Doğruer et al. (2023) employed a study utilizing CORINE data, which revealed that agricultural regions exhibited a modest nationwide increase. However, this increase displayed non-homogeneity in spatial distribution. It is emphasized that, despite the presence of extensive agricultural areas in the Central Anatolia and Southeastern Anatolia regions, disparities in productivity and resilience persist. Similarly, Ichiminami et al. (2016) demonstrate that the agricultural structure is shaped by policies such as economic transformation, modernization, and irrigation investments. However, these initiatives are distributed unevenly across regions. While mechanization and transportation investments increase productivity in the western and Mediterranean regions, structural constraints limit agricultural performance in the eastern and northeastern provinces. This observation is further corroborated by water footprint analyses conducted by Muratoglu and Avanoz (2021). The study reveals that blue and green water use efficiency is higher in regions with developed irrigation infrastructure (Mediterranean and Aegean), which in turn strengthens agricultural resilience indicators. These findings also demonstrate a strong parallel with the historical “West–East dichotomy” debates in Türkiye’s regional development literature. Aşık et al. (2023) state that regional inequalities in Türkiye have persisted since 1913, shaped by industrialization and market access dynamics. Notably, eastern regions have exhibited an inability to converge with the national average.

The study examined PARI values concentrated in western and coastal provinces, including Antalya, Mersin, and Izmir (High-High clusters), while low resilience clusters (Low-Low) persisted in eastern and northeastern Anatolia, aligning with New Economic Geography theories emphasized by Gezici and Hewings (2004). Contrary to the law of diminishing returns on capital, the study contends that developed regions demonstrate a propensity for accumulation, thereby validating the tenets of New Economic Geography. This observation signifies that agricultural productivity and stability in production exhibit cumulative enhancement in regions endowed with climatic advantages, complemented by irrigation infrastructure and market integration. As indicated in the existing literature, regions confronted with structural disadvantages become increasingly vulnerable to climate shocks, which consequently leads to their entrapment in “regional poverty traps” (Doğruel and Doğruel, 2003; Filiztekin, 2018).

The study’s most salient finding, which contradicts prevailing assumptions in the literature, is that the utilization of agricultural credit (ln_C) and mechanization (ln_M) have a statistically significant and negative effect on resilience. Contrary to the predictions of the conventional development economics approach – which posits an increase in productivity due to enhanced financial access, facilitating investment in technology and risk management for farmers (Agbodji and Johnson, 2021) – The negative coefficients in this study (−0.0050 for credit) signify that financial instruments in Türkiye’s agricultural sector have become an unsustainable “coping” mechanism rather than an “adaptation” strategy. This situation can be explained by the phenomenon defined in the literature as “maladaptation,” wherein actions that provide short-term relief increase long-term vulnerability (Schipper and Lisa, 2020).

The negative relationship uncovered in the findings merits evaluation in the context of the endogenousity debates that have permeated the existing literature. Rather than confronting the fragility inherent in their operations, farms with subpar resilience and vulnerability to external shocks may have resorted to increased credit to ensure their survival. This has potentially transformed financial instruments from mechanisms that bolster resilience to means of survival debt. Guermond (2022) found that in the cases of Cambodia and India, microfinance and agricultural loans led to a situation of over-indebtedness among farmers during periods of climate shocks. These loans were found to be utilised for household consumption and debt servicing rather than for productivity-enhancing investments, consequently undermining farmers’ adaptive capacity. Similarly, the study found that the drought and credit interaction term (Dxln_C) was negative, indicating that increased credit use during droughts did not protect farmers but instead left them more vulnerable under the burden of debt. A key finding of this study is that the robustness checks demonstrate that the negative relationship does not stem from reverse causality. The negative effect persists even when lagged values of the credit variable are used. These results support the causality that “inefficient use of credit reduces resilience” rather than “low-resilience provinces borrowing credit.” Research undertaken expressly in Türkiye likewise suggests that low interest rates constitute a pivotal element in the utilization of credit by farmers. However, it is notable that the employment of loans predominantly serves to address deficiencies in working capital, rather than to engender profound structural transformation (Adanacıoğlu et al., 2017). In this context, the findings of the study confirm the prevailing financial geography literature (Taylor, 2013) that asserts the inadequacy of financial deepening as a standalone catalyst for fostering resilience. Instead, the findings reveal that, in the absence of judicious direction, financial deepening engenders a “debt trap,” thereby exacerbating the vulnerabilities inherent within the system, as evidenced by the analysis of Türkiye’s agricultural data.

A similar “productivity paradox” is observed in the mechanization variable. The strong negative effect of mechanization on resilience (−0.0165) should be linked to the phenomenon of “over-capitalization,” a chronic problem in Türkiye’s agricultural sector. Furthermore, considering the volatility in fuel and maintenance costs during the period under review, high mechanization levels may have become a financial burden that increased farmers’ fixed cost load and eroded their buffer capacity (stability) against economic shocks. The literature frequently emphasizes that the small and fragmented farm structure resulting from the division of agricultural land through inheritance hinders the efficient use of modern machinery and leads to idle capacity (Irmaklı and Aydın, 2022). Özpınar (2020), in his study conducted in the Çanakkale region, stated that small-scale farms have tractor power far exceeding their capacity, which increases operating costs and reduces profitability.

The negative mechanization coefficient and the negative indirect effects in the SDM model indicate that machinery investments are incompatible with the scale of production and that this situation increases fixed costs, reducing the farmer’s buffer capacity (stability) against financial shocks. Furthermore, agronomic studies (Kassam et al., 2010) showing that mechanization practices such as incorrect and excessive tillage reduce the soil’s moisture retention capacity, intensify the effects of drought, and weaken ecological resilience reveal another mechanism supporting this finding. Therefore, during the analyzed period, the increase in mechanization appears to have manifested itself as a form of capital accumulation disconnected from economic rationality rather than technological progress.

The findings of this study consistently explain the structural characteristics of Turkish agriculture. Türkiye’s agricultural production system is characterized by small-scale farms and a high degree of land fragmentation. This structure limits the transformation of mechanization investments into productivity-enhancing changes. Socioeconomic assessments of land consolidation projects show that the average farm size is approximately six hectares; the majority of farms have <5 hectares of land and manage multiple parcels simultaneously (Tanrivermis and Aliefendioglu, 2019). This fragmented structure increases production costs and hinders the effective use of mechanization.

Under these conditions, mechanization creates a cycle of over-capitalization, which increases fixed costs instead of boosting productivity through technological progress. Empirical studies conducted in Türkiye reveal that, in the short term, mechanization supports production; however, in the long term, it rigidifies the cost structure, reducing farms’ flexibility in the face of climate shocks (Özkan and Miçooğulları, 2026). The limited production area per machine on small, fragmented farms increases the unit cost of mechanization investments. Furthermore, the fact that these investments are mostly financed through credit increases vulnerability to income shocks.

Türkiye’s geographical and climatic characteristics further accentuate this vulnerability. Analyses based on expert assessments indicate that drought and climate change are primary external risks for Turkish agriculture. These analyses point to an increase in the frequency and severity of droughts, particularly in the Mediterranean Basin (Kortak and Çakır, 2025). Projections of temperature increases and precipitation decreases reveal that the risk of water stress is rising in the western and southern regions, where agricultural production is concentrated (Bas and Killi, 2024). The rapid decline in fallow land leads to more intensive and continuous production, which puts additional pressure on water resources and soil structure (Kaplan, 2026).

In this context, agricultural credit functions more as a coping mechanism for short-term income loss than as a tool for structural adaptation that increases resilience under drought conditions. Studies of dairy and livestock farms in Türkiye show that access to credit significantly affects farmers’ risk perceptions and behaviors. However, these loans are mostly used to manage current expenses and debt rather than for productive investments (Özsayın, 2022). Additionally, the significant impact of farm size on economic, social, and environmental sustainability indicates that small-scale farms are inherently less resilient (Başer and Bozoğlu, 2023).

Therefore, in the Turkish context, mechanization investments financed by credit can increase vulnerability due to fixed cost burdens and debt dynamics rather than strengthen agricultural resilience, given the fragmented land structure and mounting climate pressures. The negative effects of credit and mechanization observed in the study demonstrate that the impacts of financial and technological inputs cannot be evaluated independently of contextual structural conditions.

In the context of the impact of climate shocks, the spatial spillover effects obtained by the study using the SDM reveal that drought and adaptation failures affect neighboring regions as well as local ones. The negative direct and indirect effects of the drought indicator (D) and the high spatial dependence coefficient (ρ = 0.230) illustrate the systemic risk nature of the climate crisis. The effect of drought on resilience is particularly pernicious. In addition to immediate effects, drought exerts lagged effects in the post-shock recovery process of agricultural systems, specifically manifesting as declining seed reserves and disrupted cash flow following a dry year. This consequently compromises the production capacity of subsequent years.

Studies examining the spatial distribution of drought in Türkiye show that rainfall decreases, particularly in the Central Anatolia and Southeastern Anatolia regions, threaten agricultural production and that this effect spreads over large areas when combined with basin-based water management deficiencies (Ergene et al., 2024; Kartal et al., 2024; Ozcelik et al., 2024). In the study examined, the decline in the components of “stability” and “diversity” in the eastern provinces confirms the thesis that monoculture agriculture increases risks. Kremen and Miles (2012) state that biological diversity increases the resilience of agricultural systems to shocks, while a single crop pattern leaves the system vulnerable to pests and climate extremes. The low PARI values observed in Türkiye’s eastern regions suggest that these regions have a weak absorptive capacity to absorb climatic fluctuations due to their limited crop patterns and low technology use, while western regions are able to distribute risk (portfolio effect) through methods such as polyculture and greenhouse production.

The study’s findings indicate that Türkiye’s agricultural sector is confronted with an “adaptation deficit,” and the current support policies (i.e., credit and incentives for machinery) are contributing to a “maladaptive” cycle, failing to address this deficit. These findings align with the framework delineated by Neset et al. (2019) concerning actions undertaken to adapt to climate change that nevertheless engender heightened vulnerability in the long term. Credit functions as a liquidity tool that sustains the current inefficient system rather than serving as a catalyst to help farmers align their production patterns with climate resilience, such as adopting drought-resistant seeds or drip irrigation. This situation is consistent with the urgency identified in the World Bank Group (2022) Climate and Development Report for Türkiye regarding the transition to water efficiency and climate-smart agricultural practices. This study proves that policymakers must implement a qualitative transformation that considers the quality of resource use, its suitability to the regional product pattern, and its ecological sustainability rather than focusing on quantitative targets, such as increasing credit volume or the number of tractors, because otherwise, spatial inequalities and climate vulnerabilities will continue to increase.