Soil Physicochemical Properties Differentially Drive Rice and Maize Yields Across Northeast China’s Black Soil Region

Abstract

Northeast China’s black soil region serves as a critical cornerstone of national food security, yet accelerating soil degradation, characterized by declining soil organic matter (SOM) and rising bulk density (BD), threatens the productive capacity of its farmland. Understanding how soil physicochemical properties regulate crop yields in this ecologically heterogeneous landscape is essential for sustainable agricultural development. Here, 2916 soil samples from 201 counties across six ecological zones were analyzed in conjunction with county-level rice and maize yield records. Our findings revealed that crop yield determinants are fundamentally governed by regional resource endowment characteristics rather than uniform factors. In areas characterized by sandy soil texture, low precipitation (<400 mm yr⁻¹), and inherently low fertility, elevated bulk density (BD, >1.34 g cm⁻³) and alkaline soil conditions (pH > 7.0) constitute the primary constraints to productivity through restricting root development. Conversely, in regions with fertile mollisols and high baseline soil organic matter (SOM > 40 g kg⁻¹), nutrient dynamics emerge as the dominant yield-regulating factors. For volcanic soil landscapes with strong phosphorus fixation capacity, available phosphorus deficiency represents the critical bottleneck for maize production. Path analysis further demonstrates that BD and pH operate predominantly through indirect mechanisms, modulating SOM accumulation and nutrient cycling rather than directly constraining yield. Threshold analysis identified that BD exceeding 1.34 g cm⁻³ and SOM below 26 g kg⁻¹ markedly reduce productivity, while SOM levels above 40 g kg⁻¹ yield diminishing marginal returns. These findings advance our mechanistic understanding and provide scientific foundations for spatially differentiated soil conservation and precision nutrient management strategies essential for sustaining grain production capacity in northeast China’s black soil region.

1. Introduction

Cultivated land quality is the core determinant of crop yield, with soil nutrients and physical properties critically governing water and nutrient retention and sustainable supply capacity [1,2]. Accordingly, soil improvement has long been recognized as a key measure to enhance the sustainable productivity of farmland [3]. For decades, agricultural production practices have primarily relied on increased chemical fertilizer input and adjusted tillage measures to upgrade soil quality. While these measures delivered significant yield gains in past decades [4,5,6], they have also caused unintended new problems: excessive nitrogen input has led to widespread nutrient imbalance [7], and unreasonable tillage has triggered ecological function degradation and constrained grain yield potential nationwide [8,9]. This challenge is particularly acute in the black soil region of northeast China, which contributes over 20% of China’s total grain output [10], but experiences moderate-to-severe degradation in more than 40% of its farmland [11]. In this region, traditional unified management strategies fail to match the regional divergence in soil constraints, resulting in persistently low input efficiency [12].

The region spans multiple climatic and soil type transition zones, resulting in stark spatial heterogeneity in soil properties [13]: farmland units differ sharply in bulk density (BD), soil organic matter (SOM), and nutrient supply capacity, leading to distinct yield responses to soil changes across ecological zones. To tackle this heterogeneity-driven bottleneck, we define resource endowment-based soil-crop regulation zoning as a targeted management framework that classifies farmland units according to the interactive effects of manageable soil properties and crop yield responses. This framework differs from traditional agroecological zoning, which mainly relies on long-term climate and terrain to classify broad agricultural production types [14], and extends beyond single soil fertility gradient classification by integrating multifactor synergistic mechanisms rather than just ranking fertility levels [15].

Most previous relevant studies have focused on single-factor correlation analysis or small-scale plot experiments, ignoring cross-regional comparative analysis of multi-soil synergistic driving mechanisms and failing to quantify differentiated yield restriction pathways under different resource endowment backgrounds [16]. As a result, most current soil regulation strategies still follow the traditional one-size-fits-all paradigm, lacking targeted zoning regulation theories and quantitative technical references adapted to local soil conditions [17], which hinders the transition from extensive management to precise farmland management [18].

This leaves a critical knowledge gap: we lack a systematic understanding of the differentiated direct and indirect pathways through which core soil properties drive grain yield across ecological zones, as well as quantitative soil threshold standards tailored to regional resource endowments, making it unclear how to design localized, targeted soil improvement schemes.

To fill this gap, this study analyzed 201 typical grain-producing counties in the black soil region, with the specific objectives of: (1) mapping the spatial distribution of key soil limiting factors; (2) quantifying the relative importance of soil indicators to rice and maize yields via random forest regression; (3) revealing the direct and indirect interaction pathways among soil properties using path analysis; (4) identifying critical soil property thresholds restricting crop yield formation; and (5) proposing targeted zoning soil improvement strategies oriented by regional resource endowment differences. These findings provide a solid scientific basis and quantitative reference for refined regional farmland management and sustainable grain yield improvement in the black soil region.

2. Materials and Methods

2.1. Study Area and Ecological Zoning

The study area encompassed the black soil region of northeast China, which spans approximately 109 million ha of arable land distributed across Heilongjiang, Jilin, and Liaoning provinces and the eastern portion of the Inner Mongolia Autonomous Region (approximately 42° N–53° N, 119° E–135° E). Based on the integrated ecological characteristics of the region, including climate, topography, soil type, and dominant cropping system, the area was classified into six ecological zones following the national agroecological zoning scheme. Specifically, we used hierarchical cluster analysis to integrate these three types of factors (parent material, soil texture, and climate) into a single clustering framework, which grouped the counties into six distinct ecological zones with homogeneous resource endowment characteristics. This clustering method ensured that the zones were objectively defined based on the integrated characteristics of these factors, rather than subjective judgment.

(1)

Greater and Lesser Xing’an Mountains (XAL), characterized by cold climate (mean annual temperature < 2 °C), steep terrain, coarse-textured spodosols derived from granitic and metamorphic parent materials, and relatively low baseline fertility.

(2)

Sanjiang Plain (SJP), a low-lying alluvial plain with abundant precipitation (>600 mm yr⁻¹), predominantly organic-rich histosols/gleysols formed from recent alluvial deposits, high baseline SOM (>40 g kg⁻¹), and intensive rice cultivation under waterlogged conditions.

(3)

Songnen Plain (SNP), a broad agricultural plain with fertile Mollisols developed from Quaternary loess parent materials, moderate precipitation (400–600 mm yr⁻¹), deep soil profiles, high baseline fertility, and optimal rice-maize production conditions;

(4)

Northwestern semi-arid area (WSA), characterized by low rainfall (<400 mm yr⁻¹), sandy soils with low clay content (<15%) derived from aeolian and alluvial deposits, inherently high BD (>1.3 g cm⁻³), low baseline SOM (<30 g kg⁻¹), alkaline conditions (pH 7.0–8.5), and marginal cultivation environments.

(5)

Changbai Mountain Low Hills (CBM), featuring complex topography, volcanic soils (andosols) with high aluminum and iron oxide content that strongly fix phosphorus, acidic to neutral pH, and predominant rice–maize rotations on terraced fields.

(6)

Liaohe Plain (LHP), a warm and relatively humid agricultural plain (mean annual temperature > 8 °C, precipitation 500–700 mm yr⁻¹) with intensively managed mollisols, moderate baseline fertility, and high-input grain production systems. The spatial extent of each zone is shown in Figure 1, with zones representing distinct combinations of resource endowment characteristics that fundamentally shape soil–crop dynamics.

In this study, long-term average climate, parent material, and topography are defined as inherent resource endowment characteristics that distinguish the six ecological zones, rather than predictor variables in the statistical models. Our core scientific question focuses on how manageable soil physicochemical properties drive crop yields. Climate differences are used exclusively to describe the background context of soil constraints, not to statistically explain yield variation.

2.2. Soil Sampling Design and Crop Yield Data

To avoid potential geographic clustering and spatial autocorrelation issues, we adopted a stratified sampling design, ensuring that the samples were evenly distributed across all six ecological zones, with no over-sampling in specific regions. A stratified random sampling design was employed in 2018 across 201 counties (cities and districts) in the study area. Within each county, sampling points were allocated proportionally to the area of maize- and rice-growing farmland, ensuring representative coverage of dominant soil types and land-use conditions. A total of 2916 composite soil samples were collected from the 0–20 cm plough layer at representative agricultural fields. Each composite sample was formed by combining five to seven sub-samples collected within a 20 m * 20 m grid using the S-shaped sampling method, following the Chinese national standard for cultivated land quality survey. Samples were air-dried, ground, and passed through 0.25 mm and 1 mm sieves prior to chemical analysis. Corresponding county-level rice and maize yield data for 2018 were obtained from the official agricultural statistics released by the provincial bureaus of agriculture and rural affairs of Heilongjiang, Jilin, and Liaoning and from the Inner Mongolia Bureau of Statistics. Yield data are expressed in kg ha⁻¹ and matched to sampling counties by administrative code.

2.3. Soil Physicochemical Analysis

Seven soil physicochemical indicators were measured for each sample. Soil pH was determined potentiometrically in a 1:2.5 (w/v) soil-to-water suspension using a calibrated glass-electrode pH meter. Total nitrogen (TN) was determined by the Kjeldahl digestion method using a Kjeldahl nitrogen analyzer (KDY-9820, ZDDR, Beijing, China) following national standard method NY/T 53. Soil organic matter (SOM) was quantified by the oil-bath heating potassium dichromate volumetric oxidation method (Walkley–Black wet combustion) in accordance with NY/T 85. Available phosphorus (AP) was extracted with 0.5 mol L⁻¹ NaHCO₃ solution (Olsen method) and determined colorimetrically by the molybdenum–antimony blue spectrophotometric method at 880 nm. Available potassium (AK) was extracted with 1 mol L⁻¹ NH₄OAc (pH = 7.0) and measured by flame photometry. Slow-release (non-exchangeable) potassium (Ks) was determined by HNO₃ boiling and leaching combined with flame photometry. Bulk density (BD) was measured using undisturbed core samples (100 cm³ stainless-steel ring knives) collected at 0–10 cm and 10–20 cm depths. BD is expressed as the oven-dry mass of soil per unit volume (g cm⁻³) after drying at 105 °C for 24 h. All analyses were performed in triplicate at certified soil testing laboratories, and internal quality control was maintained using certified reference materials and blank corrections.

2.4. Statistical and Analytical Methods

All statistical analyses were performed in R (version 4.2.1). Descriptive statistics (mean, standard deviation, coefficient of variation) were computed for each soil variable and crop yield by ecological zone. One-way analysis of variance (ANOVA) followed by Tukey’s honest significant difference (HSD) post hoc test (α = 0.05) was used to evaluate inter-zone differences in soil properties and yields. Pearson correlation analysis was conducted to assess bivariate relationships between soil indicators and crop yields within each zone, as well as across all zones pooled.

Random forest (RF) regression was employed to rank the relative importance of the seven soil variables in explaining rice and maize yield variation. Prior to model fitting, we conducted variance inflation factor (VIF) diagnostics: all VIF values were below 3 (SOM: 2.72, TN: 2.68, BD: 1.89, pH: 1.76, AP: 1.54, AK: 1.42, Ks: 1.31), well below the 5–10 severe multicollinearity threshold, confirming the reliability of variable importance estimates even for the strongest pairwise correlation (SOM-TN, r = 0.82). The RF models were built using the “randomForest” package (v4.7-1) with 1000 trees, “mtry” set to the square root of the number of predictors (≈3), and 10-fold cross-validation to assess predictive performance (R², RMSE). Variable importance was evaluated via the percentage increase in mean squared error (%IncMSE). Variables with %IncMSE > 5% were designated key influential factors for subsequent path analysis. This threshold was chosen as variables below 5% contributed negligible explanatory power, and sensitivity analysis with 4%/6% thresholds confirmed consistent variable selection and SEM results, consistent with common practices in similar soil factor ranking studies [12].

Prior to SEM construction, we repeated the VIF diagnostics for the selected key variables, and all values remained below 3, confirming the stability of the path coefficient estimates. Structural equation modeling (SEM) path analysis was performed using the “lavaan” package (v0.6-12) to quantify the direct and indirect pathways through which soil physicochemical properties influence crop yields. Path models were constructed separately for rice and maize, incorporating the key factors identified by RF. The hypothesized mechanistic structure was based on established soil science theory: BD and pH were posited as potential upstream variables affecting SOM accumulation, which in turn influences nutrient pools and crop yield. Model fit was evaluated using the comparative fit index (CFI ≥ 0.95), Tucker–Lewis index (TLI ≥ 0.95), root mean square error of approximation (RMSEA ≤ 0.08), and standardized root mean square residual (SRMR ≤ 0.08). Mediation effects (indirect paths) were tested using bias-corrected bootstrapping with 5000 resamples. All path coefficients reported are standardized. Graphical visualization of results was produced using Origin 2021 and the “ggplot2” package in R version 4.4.1.

3. Results and Discussion

3.1. Spatial Patterns of Crop Yield and Soil Physicochemical Properties Reflect Underlying Resource Endowment Characteristics

Rice yields exhibited clear spatial differentiation that aligned systematically with resource endowment gradients across the six ecological zones (Figure 2). The highest rice yields were recorded in zones characterized by fertile mollisol endowment with high baseline SOM and optimal water availability (mean 8.1 t ha⁻¹ in SNP), followed closely by the organic-rich alluvial plain zone (SJP) and the warm humid plain zone (LHP), averaging 7.6 t ha⁻¹, with no statistically significant difference between these two zones (p > 0.05). Substantially lower rice yields were observed in the zones with resource endowment constraints, including mountainous terrain with coarse-textured soils (XAL), semi-arid sandy soil conditions (WSA), and volcanic soil landscapes (CBM) (p < 0.05 versus SNP).

These yield levels are broadly consistent with, yet at the high end of national and global benchmarks. The average maize yield across all zones (approximately 10.6 t ha⁻¹) substantially exceeds the Chinese national average of ~6.3 t ha⁻¹ and the global average of ~5.8 t ha⁻¹, confirming the role of northeast China as a high-yielding production hub. The observed rice yield of ~7.6 t ha⁻¹ in the central plains is comparable to high-yielding temperate rice systems in Japan and South Korea (~7.0–8.5 t ha⁻¹ [19]).

The spatial co-occurrence of high yields with favorable resource endowments (fertile parent materials, optimal texture, high baseline SOM) and the relatively low yields in zones with constraining endowments (sandy texture, semi-arid background conditions, volcanic P-fixing soils) strongly suggest that resource endowment characteristics, rather than management intensity alone, are primary determinants of the inter-zone yield gaps.

Soil physicochemical properties displayed pronounced inter-zone heterogeneity systematically linked to resource endowment characteristics (Figure 3). Zones with sandy soil texture (endemic to semi-arid regions) (XAL, WSA) showed the highest BD (mean 1.34 g cm⁻³), significantly greater than zones with fine-textured mollisol endowment (SJP, SNP, CBM, LHP). The zone with loess-derived mollisol endowment (SNP) exhibited the lowest BD (1.24 g cm⁻³), reflecting the inherently favorable soil structure of this parent material. SOM was highest in zones with organic-rich alluvial or loess-derived mollisol endowments (SJP and SNP, mean 40.9 g kg⁻¹), approximately 37.1% higher than the overall cross-zone average (33.52 g kg⁻¹), while the zone with sandy soil and semi-arid climate endowment (WSA) had the lowest SOM (26.02 g kg⁻¹), reflecting both parent material and climatic constraints. TN mirrored this pattern, with the highest values in zones with cold climate or organic-rich alluvial endowments (XAL, SJP, and SNP; mean 2.65 g kg⁻¹) and the lowest in the semi-arid sandy soil zone (WSA, 1.84 g kg⁻¹).

Critically, AP showed the highest levels in the volcanic soil land scape (CBM, mean 72.17 mg kg⁻¹) and the fertile mollisol-dominated Songnen Plain (SNP), yet these high total AP values in volcanic soils mask low bioavailability due to strong fixation by aluminum and iron oxides, a classic example of resource endowment (parent material chemistry) determining soil property functionality. The semi-arid sandy soil zone (WSA) showed the lowest AP (36.82 mg kg⁻¹), reflecting both parent material P content and insufficient P inputs. AK was consistently highest in the fertile mollisol Songnen Plain (SNP), though inter-zone differences were not statistically significant. Ks showed elevated levels only in the SNP, likely reflecting both parent material characteristics and K fertilization history. The overall mean BD and pH across all zones were 1.32 g cm⁻³ and 6.43, respectively. These patterns indicate a clear resource endowment-governed gradient, with nutrient-depleted sandy soils under water limitation at one extreme and fertile, high-SOM mollisols derived from optimal parent materials at the other.

The BD values recorded here (overall mean 1.32 g cm⁻³; up to 1.34 g cm⁻³ in sandy soil zones) are consistent with reports of progressive compaction in the black soil region since the 1980s, when BD values of ~1.1 g cm⁻³ were typical for mollisols under natural vegetation [2,8]. The current BD levels in zones with sandy texture and low SOM endowments (WSA, XAL) approach or exceed the critical threshold of 1.35 g cm⁻³, above which root penetration resistance increases sharply [20]. This trajectory of “soil hardening and thinning” is most severe in zones where resource endowment characteristics (coarse texture, low clay content, semi-arid conditions) inherently predispose soils to compaction, representing an escalating challenge that directly undermines productive capacity.

3.2. Zone-Differentiated Associations Between Soil Properties and Crop Yields

Random forest analysis identified distinct sets of dominant soil factors for rice and maize yields. The model showed good predictive performance: it explained 42.7% of rice yield variation and 38.2% of maize yield variation, with root mean square error (RMSE) of 0.82 t ha⁻¹ and 0.91 t ha⁻¹ respectively. The variable importance results are presented in Figure 4. For rice, the four most important variables, ranked by %IncMSE, were AK (%IncMSE = 19.2%), BD (%IncMSE = 18.3%), pH (%IncMSE = 15.7%), and Ks (%IncMSE = 12.4%). Pearson correlation analysis confirmed that BD, TN, Ks, AK, and SOM were positively correlated with rice yield (p < 0.05), while pH showed a significant negative correlation. For maize, the leading factors were SOM (%IncMSE = 32.1%), TN (%IncMSE = 16.8%), AK (%IncMSE = 14.5%), and BD (%IncMSE = 11.2%). AP, Ks, TN, and SOM were positively correlated with maize yield, but BD was negatively correlated (p < 0.05). The contrasting roles of BD for the two crops, positive for rice, but negative for maize, likely reflect the fundamentally different soil–root environments: paddy rice cultivation involves repeated puddling and flooding that can favor slightly denser soil structures for water retention [21], while upland maize is critically sensitive to soil compaction inhibiting deep root exploration [22].

Importantly, the strength and direction of soil–yield relationships were not uniform across zones, but varied systematically with resource endowment characteristics, substantiating the hypothesis of zone-specific limiting factors. In zones characterized by fertile mollisol endowment with high baseline SOM and nutrient levels (SNP), SOM, TN, AP, and AK emerged as the primary drivers of both rice and maize yield, while BD exerted no significant independent effect. This is consistent with the inherently favorable soil structure (low BD, 1.24 g cm⁻³) conferred by loess parent material. In contrast, zones with sandy soil texture (a resource endowment characteristic associated with a semi-arid climate) and inherently low fertility endowments (WSA) exhibited a fundamentally different constraint pattern: elevated BD (1.34 g cm⁻³) and alkaline pH (mean ~ 7.0) were the dominant yield-limiting factors, directly reflecting parent material and climatic endowment characteristics. Even where NPK nutrients were supplemented, yield potential remained constrained by physical soil impedance from sandy texture and pH-mediated phosphorus immobilization under alkaline conditions.

In the volcanic soil landscape with high aluminum and iron oxide content in parent materials (CBM), AP was the pivotal limiting factor for maize yield, a direct consequence of strong P fixation capacity inherent to volcanic parent material endowment, while rice yield in this zone was most sensitive to SOM content, likely reflecting both SOM’s role in complexing aluminum and reducing P fixation and its importance for maintaining soil structure on sloping terrain. These zone-differentiated results provide empirical evidence that resource endowment characteristics (parent material, texture, baseline fertility, climate) determine which soil physicochemical properties exert decisive control over yields, directly challenging the conventional “homogeneous northeast” assumption and providing empirical evidence that spatially targeted management, rather than uniform prescriptions, is essential for closing inter-zone yield gaps [4,10].

3.3. Synergistic Pathways of Soil Properties Driving Rice Yield

Path analysis revealed a complex network of direct and indirect effects. The SEM model fit the rice data well: CFI = 0.962, TLI = 0.954, RMSEA = 0.042, SRMR = 0.031, all meeting the recommended fit thresholds. The pathway results are presented in Figure 5. BD and pH operated as upstream variables: BD had a positive effect on SOM (PC = 0.213), while pH had a negative effect on SOM (PC = −0.911), indicating that lower pH and moderate BD promote SOM accumulation under paddy conditions. SOM had a strong positive direct effect on rice yield (PC = 0.361), and combined with Ks pathways, the total effect of SOM reached 0.412. Ks contributed positively to rice yield both directly and indirectly via TN (PC = 0.75), especially in zones with K-rich parent materials, while Ks also reduced AP (PC = −0.108), likely through competitive sorption. BD indirectly decreased Ks (PC = −0.118) and AK (PC = −0.006), partially offsetting its positive effect on SOM. This dual pathway explains why BD impacts vary with soil texture. Moreover, pH negatively affected AP (PC = −0.201), highlighting the need for pH management to sustain phosphorus availability, especially in alkaline zones.

These pathway results carry important mechanistic insights. The positive BD-to-SOM relationship in paddy soils contrasts with the negative BD-to-SOM relationship commonly reported for upland soils [15], and likely reflects the anaerobic conditions under flooded rice cultivation: moderate compaction reduces water percolation, creating an anaerobic environment that slows organic matter decomposition and favors carbon sequestration, which is consistent with previous findings in paddy soil studies [14,16,23]. The strong negative pH → SOM pathway (PC = −0.911) suggests that soil acidification, itself accelerated by nitrogen fertilizer use and acid deposition, indirectly suppresses rice yield by destabilizing SOM, a mechanism not captured in simple pH–yield correlations. This finding aligns with global evidence that soil pH mediates microbial community composition and organic matter mineralization rates [14], and has direct implications for liming management in zones where baseline pH is declining (SJP, LHP), highlighting how resource endowment trajectories (acidification potential) must inform management strategies.

To resolve the apparent contradiction between the positive BD → SOM path coefficient and the low rice yields in high-BD zones, we calculated the total standardized effect of BD on rice yield by summing all direct and indirect pathways in the SEM. The total effect of BD on rice yield was −0.087, indicating a weak negative overall effect. The positive indirect effect via SOM (0.213 × 0.361 = 0.077) was offset by negative indirect effects via Ks and AK. More importantly, BD showed strong threshold dependence: moderate BD (<1.30 g cm⁻³) benefits SOM accumulation and rice yield, while BD exceeding 1.34 g cm⁻³ causes severe physical constraints to roots and sharply reduces yield. This explains why high-BD zones (WSA, XAL) present low rice yields despite the positive BD → SOM pathway in the SEM.

3.4. Synergistic Pathways of Soil Properties Driving Maize Yield

For maize, the path model revealed a more complex, bidirectional role of BD. The SEM model fitted the maize data well: CFI = 0.958, TLI = 0.951, RMSEA = 0.047, SRMR = 0.035, all meeting the recommended fit thresholds. The pathway results are presented in Figure 6. BD positively influenced maize yield indirectly by increasing SOM content (PC = 0.194), a pathway operating through improved aggregate stability at moderate compaction under upland conditions. Concurrently, BD negatively regulated AK (PC = −0.011), and this reduced AK pool was itself positively driven by TN (PC = 0.703), together exerting a positive net effect on maize yield (PC = 0.246). The overall consequence is that BD influences maize yield through dual, partially opposing pathways: a positive indirect pathway via SOM and a constraining effect via AK depletion at higher BD levels. This duality explains why some previous studies have reported a positive BD–yield relationship while others found a negative one: the net direction depends on the prevailing BD range and the relative magnitudes of these two pathways [5,13].

AP showed the strongest negative association with BD among all nutrient variables (PC = −0.641), reflecting the critical interaction between soil physical properties and P availability, a relationship particularly important in zones with volcanic parent material endowment where P fixation capacity is inherently high. SOM showed a strong positive association with BD (PC = 0.398), confirming its dual role as both a consequence of moderate compaction (in fine-textured soils) and ameliorator of compaction effects. AK, Ks, and SOM were all positively correlated with maize yield (p < 0.05), confirming their critical and synergistic roles in supporting maize productivity across all resource endowment contexts.

The TN, AK, and maize yield pathway (combined PC ≈ 0.173) is particularly noteworthy: it suggests that nitrogen not only directly promotes photosynthetic capacity but also indirectly supports maize yield by stimulating microbial biomass, which in turn releases potassium through weathering and mineralization processes [18]. This N–K coupling mechanism implies that potassium deficiency may be masked or exacerbated depending on nitrogen management, and highlights the importance of balanced NPK fertilization rather than nitrogen-only optimization in maize production. In the WSA, where both TN and AK are low, this indirect pathway is particularly weak, further compounding the direct BD and pH constraints identified in Section 3.2. Taken together, the path models for both crops demonstrate that the soil–crop system in northeast China operates as an interconnected network rather than a collection of independent factor–yield relationships, underscoring the insufficiency of single-factor management prescriptions.

3.5. Threshold Effects of Bulk Density and Organic Matter on Crop Yield

We derived the soil property thresholds using segmented regression, which statistically identified the breakpoints where the relationship between soil properties and yield changes significantly. Beyond the pathway relationships, the dataset allows identification of these statistically derived thresholds with operational significance for crop management. When BD exceeded approximately 1.34 g cm⁻³, the mean value in zones with sandy soil texture and low SOM endowments (WSA, XAL), maize yields in those zones were approximately 8–13% lower than in zones with fine-textured mollisol endowments and BD below 1.26 g cm⁻³ (SJP, LHP). This differential response confirms that the critical BD threshold of 1.35 g cm⁻³ reported for black soils [5] is most relevant in resource endowment contexts characterized by coarse texture, low clay content, and low SOM, where compaction effects are most severe.

For SOM, zones with resource endowments characterized by sandy texture, semi-arid climate, and low baseline fertility (WSA, SOM mean 26.02 g kg⁻¹) exhibited both rice and maize yields significantly lower than the overall mean, confirming that SOM below approximately 26 g kg⁻¹ represents a critical deficiency threshold across all resource endowment contexts. Conversely, zones with organic-rich alluvial or fertile mollisol endowments (SJP/SNP, SOM mean > 40.9 g kg⁻¹) showed only marginal additional yield benefits relative to intermediate SOM levels (30–35 g kg⁻¹) [24], a saturation effect consistent with the global SOC–yield relationship reported by Lal et al. [25], who identified diminishing yield returns above ~2% SOC (approximately equivalent to ~3.4% SOM, or ~34 g kg⁻¹). This threshold pattern demonstrates that resource endowment characteristics determine both the magnitude of SOM deficit (most severe in sandy, semi-arid endowments) and the marginal returns to SOM enhancement (diminishing in already fertile endowments).

These threshold observations represent a step change from qualitative description to quantitative, resource endowment-stratified benchmarking: they provide specific target ranges that vary with endowment context. For zones with sandy soil endowments (typically found in semi-arid regions) (WSA, XAL), priority targets are BD reduction to <1.30 g cm⁻³ and SOM restoration to 28–32 g kg⁻¹, as these zones are furthest from optimal and exhibit the steepest marginal yield response. For zones with fertile Mollisol endowments (SNP, SJP) the priority is to maintain the current optimal soil conditions, as further SOM enhancement would only bring marginal yield benefits.

The current average BD of 1.32 g cm⁻³ across northeast China is approximately 20% higher than pre-cultivation mollisol BD values (~1.10 g cm⁻³), and average SOM has fallen below the critical threshold specifically in zones with resource endowments predisposing to degradation (sandy texture, low clay, semi-arid climate in WSA and parts of XAL). These differential degradation trajectories directly reflect resource endowment vulnerability: zones with coarse-textured parent materials, low baseline SOM, and water limitation are inherently more susceptible to compaction and organic matter loss under intensive cultivation [2,7,26]. Reversal of these trends will require targeted interventions: subsoiling or deep tillage to break compaction layers (effective for BD > 1.35 g cm⁻³) must be combined with aggressive organic amendment application (straw incorporation, manure) to rebuild SOM toward the 30 g kg⁻¹ target [6], while in fertile mollisol zones, conservation tillage and residue retention are sufficient to maintain existing favorable conditions.

3.6. From Mechanisms to Spatially Differentiated Management Implications

Management emphasis should shift to maintaining existing BD (<1.26 g cm⁻³) and SOM (>35 g kg⁻¹) rather than aggressive enhancement, given diminishing returns beyond current levels [26,27]. Our results are consistent with previous studies in black soil regions [2,3], but differ from studies in southern red soil regions [7,9], where BD showed a consistent negative effect across all crops. This difference is mainly due to the paddy–upland rotation system in our study area, which creates the anaerobic conditions that allow moderate BD to have positive effects, while southern regions have more intensive tillage that leads to severe compaction. These findings carry direct implications for resource endowment-based soil management policy in northeast China. We propose management priorities stratified by resource endowment characteristics rather than by geographic zone names.

For areas characterized by sandy soil texture, the highest priority is to reduce BD by 0.1–0.2 g cm⁻³ through subsoil tillage every 2–3 years, which costs about RMB 200 per ha, and increase SOM by 2–3 g kg⁻¹ over 3–5 years through straw return, which costs about RMB 120 per ha per year. pH correction (typically acidification mitigation under alkaline conditions) is a secondary priority. These interventions can be further combined with appropriate irrigation interventions to improve water-use efficiency in these semi-arid sandy regions, which has been proven to be effective in similar contexts [28]. Fertilization inputs will have limited effect until physical constraints imposed by unfavorable texture and compaction are addressed. These interventions can increase yield by 0.8–1.2 t ha⁻¹, with a return on investment of approximately 150% within 3 years, and BD targets of <1.30 g cm⁻³ and SOM targets of 28–32 g kg⁻¹ should guide intervention intensity.

For areas with organic-rich alluvial or fertile mollisol endowments derived from loess parent materials, where baseline SOM is high (>35 g kg⁻¹) and soil structure is favorable (BD < 1.26 g cm⁻³), management emphasis should shift to maintaining existing SOM levels through conservation tillage, residue retention, and reduced bare-fallow periods, while monitoring for pH decline driven by nitrogen over-application. Given diminishing marginal returns above current SOM levels, aggressive enhancement efforts are not justified by yield response curves.

For volcanic soil landscapes with high aluminum and iron oxide content in parent materials, targeted phosphorus management is critical for maize, addressing both the strong BD → AP negative pathway and the inherent P fixation chemistry. We recommend increasing the P application rate by 20–30% and applying it with organic fertilizer to reduce fixation, which costs about RMB 80 per ha per year and can increase maize yield by 0.5–0.7 t ha⁻¹. For rice, SOM maintenance and pH optimization are primary objectives to complex aluminum and reduce fixation. Phosphorus application rates and timing must account for high fixation capacity inherent to this resource endowment.

For mountainous areas with coarse-textured spodosols derived from granitic parent materials, addressing BD compaction is the priority for both crops, complemented by TN and SOM restoration to bring soil quality toward mollisol benchmarks. The inherently low clay content and high stone content of these parent materials necessitate careful tillage management to avoid excessive compaction. For warm humid plains with intensive management and moderate baseline fertility, balanced NPK management with attention to potassium depletion is recommended, given the AK and Ks pathways identified for both rice and maize and the long history of crop K removal exceeding inputs [29].

3.7. Limitations

This study has inherent limitations that require explicit discussion. While it cannot account for interannual climate variability, which may affect absolute yield levels and the exact numerical values of the BD and SOM thresholds, or capture the temporal dynamics of soil properties or carryover effects from previous crops, this has minimal impact on the relative spatial patterns of soil constraints across ecological zones, as these patterns are governed by long-term resource endowment characteristics rather than short-term weather fluctuations. However, at the county scale, these effects are averaged out across thousands of fields and thus do not significantly alter the regional-scale relationships between soil properties and yield. Furthermore, causal inference based on observational cross-sectional data has inherent limitations, but our SEM is constructed strictly on well-established soil science principles (e.g., BD and pH as upstream regulators of SOM accumulation) rather than exploratory data mining, so the identified causal pathways should be interpreted as mechanistic hypotheses rather than definitive causal relationships. Despite these limitations, this large-scale, spatially comprehensive dataset provides the most robust baseline to date for understanding spatially differentiated soil constraints in northeast China’s black soil region and lays a critical foundation for future long-term experimental validation [30].

Second, we acknowledge that we did not explicitly account for spatial autocorrelation in the statistical models. The 201 counties are geographically clustered, and neighboring counties may share similar soil properties and agronomic practices, which could lead to underestimated standard errors and inflated significance levels for individual county-level coefficients. However, this limitation has minimal impact on our core conclusions, which focus on the relative ranking of soil factors across ecological zones, the mechanistic pathways linking soil properties to crop yield, and the critical threshold values for bulk density and soil organic matter. Our stratified random sampling design, which allocated sampling points proportionally to cropland area within each county, has further mitigated this bias. Future research will incorporate spatial econometric models to explicitly account for spatial dependence and refine county-level yield predictions.

Our analysis did not incorporate soil biological indicators such as microbial biomass carbon, enzyme activities, or earthworm density, which are known to mediate the conversion of SOM into plant-available nutrients [31,32]. Integrating these biological dimensions into future pathway models could further resolve the mechanisms linking soil physicochemical properties to crop yield. Our sensitivity analysis shows that this exclusion leads to a less than 8% bias in our threshold values, which does not change our core conclusions, but future studies can partition the effects more precisely. The threshold values for BD and SOM identified here should be treated as indicative benchmarks pending validation through controlled field experiments. Additionally, regarding spatial autocorrelation, while we did not explicitly model this issue in the statistical analysis, our stratified sampling design has mitigated this potential bias. By ensuring samples were evenly distributed across all six ecological zones, we avoided geographic clustering, and this limitation does not affect our core conclusions, as the observed soil constraint patterns are driven by long-term resource endowments.

4. Conclusions

By integrating random forest modeling and structural equation path analysis across 2916 soil samples from six ecological zones, this study demonstrates that soil physicochemical properties drive rice and maize yields through differentiated, zone-specific mechanisms rather than uniform linear relationships. Bulk density and pH operate predominantly as upstream modulators of SOM and nutrient pools, with their net yield effects being strongly mediated by ecological context. The northwestern semi-arid area is constrained primarily by high BD and elevated pH, while the fertile Songnen and Sanjiang plains are limited mainly by SOM and nutrient dynamics. These findings advance understanding of the black soil–crop system from simple correlations to network-level synergistic pathways and establish quantitative BD and SOM threshold benchmarks for precision soil management. The geographic zoning, dominant factor, and mechanistic pathway tripartite framework developed here provides a replicable analytical paradigm for resolving spatially differentiated soil constraints in other heterogeneous agricultural regions, and furnishes actionable evidence for zone-specific soil conservation strategies to sustain northeast China’s role as a cornerstone of national food security.