Ecological Change in Minnesota’s Carbon Sequestration and Oxygen Release Service: A Multidimensional Assessment Using Multi-Temporal Remote Sensing Data

Highlights

What are the main findings?

• Minnesota’s carbon sequestration and oxygen release services increased overall from 1998 to 2021, with 2013 as a clear turning point.
• We propose a transferable Quantity–Intensity–Structure (Q–I–S) framework to jointly capture scale, efficiency, and trends, and to identify priority areas for targeted management.

What are the implications of the main findings?

• CSOR shows a stable “higher in the north, lower in the south“ pattern: the structure is higher and improving in the north, but lower and degrading in the south.
• Degradation risk is concentrated in a narrow central–southwestern belt over a few counties; ~80% of the region shows no persistent degradation, so this belt should be prioritized for restoration and risk control.

Abstract

Carbon sequestration and oxygen release (CSOR) are core regulating functions of terrestrial ecosystems. However, regional assessments often fail to (i) separate scale-driven high supply from per-area efficiency, (ii) detect structural instability and degradation risk from long-term trajectories, and (iii) provide evidence that is comparable across units for management prioritization. Using Minnesota, USA, we integrated satellite-derived net primary productivity (NPP; 1998–2021) with a Quantity–Intensity–Structure (Q–I–S) framework to quantify CSOR, detect trends and change points (Mann–Kendall and Pettitt tests), map spatial clustering and degradation risk (Exploratory Spatial Data Analysis, ESDA), and attribute natural and human drivers (principal component regression and GeoDetector). CSOR increased overall from 1998 to 2021, with a marked shift around 2013 from a slight, variable decline to sustained recovery. Spatially, CSOR showed a persistent north–south gradient, with higher and improving services in northern Minnesota and lower, more degraded services in the south; persistent degradation was concentrated in a central high-risk belt. The Q–I–S framework also revealed inconsistencies between total supply and condition, identifying high-supply yet degrading areas and low-supply areas with recovery potential that are not evident from the totals alone. Climate variables primarily controlled CSOR quantity and structure, whereas human factors more strongly influenced intensity; the interactions of the two further shaped observed patterns. These results provide an interpretable and transferable basis for diagnosing degradation and prioritizing restoration under long-term environmental change.

1. Introduction

Against the backdrop of global change, fossil-fuel consumption, land use, and land cover change have driven sustained increases in atmospheric CO₂, intensifying warming and increasing the risks of drought and extreme events [1,2]. Consequently, ecosystem carbon sequestration and oxygen release (CSOR) has become a key regulating ecosystem service for climate change mitigation and environmental quality maintenance [3,4,5]. This service refers to the assimilation of atmospheric CO₂ through photosynthesis and its storage in biomass and soil carbon pools, forming a carbon sink, while simultaneously releasing O₂ and supporting climate regulation and atmospheric environmental quality [6,7]. Spatially explicit, temporally consistent, and cross-region comparable assessments of carbon sequestration and oxygen release support carbon peak and carbon neutrality strategies and the implementation of nature-based solutions. They also enable the identification of ecosystem degradation, the optimization of restoration investments, and the evaluation of governance performance.

Previous studies have established a relatively mature biophysical basis in both concept and accounting. Grounded in photosynthesis and the carbon cycle, this framework uses productivity and net exchange metrics such as gross primary productivity, net primary productivity, and net ecosystem productivity to characterize carbon sequestration processes [6]. It also incorporates changes in aboveground and belowground biomass carbon density and soil organic carbon stocks [8]. Carbon fixation and oxygen release are quantitatively linked through the stoichiometry of photosynthesis [9]. Methodologically, plot measurements and allometric equations [10], remote sensing retrievals based on vegetation indices and canopy structure [9,11], and ecosystem models based on light use efficiency have been combined to support mapping and accounting from regional to global scales [12,13]. These efforts provide a robust basis for quantifying carbon sequestration and oxygen release and locating carbon stocks, and they support comparisons across ecosystem types such as forests, grasslands, wetlands, and croplands [14].

However, when the objective shifts from quantifying contributions to identifying risks and management priorities, a total amount perspective has at least three limitations. First, totals cannot separate scale effects from per-area efficiency. Ecosystems with similar aggregate values may differ substantially in functional condition, and area-driven signals can be misinterpreted as functional advantages, masking latent degradation. For example, large-area grasslands in Bayannur, Inner Mongolia, China, may show high total service supply despite having suboptimal conditions and low per-area capacity. Similarly, Minnesota, USA, may experience gradual functional decline associated with persistent water pollution and forest fragmentation that is not evident in totals. Second, total indicators are relatively insensitive to long-term structural change, limiting the detection of degradation and recovery signals and the identification of thresholds where quantitative change becomes qualitative change. Third, scale-related bias can encourage management actions that prioritize expanding area while overlooking service quality and the stability and vulnerability of underlying structures, thereby weakening the targeting of restoration efforts.

At the same time, the integration of remote sensing, geographic information systems, and ecosystem models has accelerated assessments of carbon sequestration and oxygen release services [15]. Study domains have expanded from individual ecosystems to watersheds, urban agglomerations, and ecological function zones [9,16,17]. Analytical approaches have shifted from single-year snapshots to long time series evaluations of trends, abrupt changes, variability, and responses to extreme events [6]. Driver analyses increasingly incorporate climate variables including temperature, precipitation, radiation, and drought indices; topographic variables including elevation, slope, and aspect; vegetation attributes including cover, leaf area index, phenology, and functional types; and human pressures including land-use change, ecological engineering and restoration, grazing, and pollution stress [5,18]. Methods such as regression modeling, panel modeling, structural equation modeling, machine learning, and geographical detector modeling have been used to identify dominant factors and interactions [19,20,21]. These advances improve the ability to locate change and explore potential drivers, and they support scenario analysis and policy assessment. Nevertheless, two gaps remain. First, service structure is often under-specified and inconsistently defined. Resilience-relevant information such as trends, variability, and the frequency of degradation is rarely represented in a systematic and operational manner, limiting the detection of slow-onset degradation and early warning. Second, cross-regional comparison remains constrained by inconsistent indicators, parameterization, and spatial scales. Unified assessments that link totals, intensity, and change within a single framework are still limited, reducing interpretability and weakening the translation of mechanism-based results into management priorities, especially when totals appear stable while per-area efficiency declines and variability increases.

To address these gaps, this study focuses on carbon sequestration and oxygen release services and proposes an integrated biophysical assessment pathway based on total amount, intensity, and structure, covering framework design, indicator construction, and mechanism attribution. The aim is to resolve the following three practical problems: totals can mask efficiency differences and structural degradation, cross-unit comparability is often limited, and management priorities can be biased by scale. Total amount quantifies service supply at the scale of each assessment unit. Intensity represents per-unit-area service provision to distinguish scale from efficiency and reduce area-driven bias. Structure is characterized using temporal trends, interannual variability, and degradation frequency to reflect service dynamics, stability, and risk. Using statistical methods and standard deviation-based normalization, we develop an ecosystem service equivalence indicator to transform biophysical variables with different units into a unified, comparable metric. This indicator captures both the direction and magnitude of deviation from the overall distribution across units, providing a consistent basis for integrated three-dimensional evaluation. We then characterize the spatial patterns and temporal trajectories of total amount, intensity, and structure; identify anomalous areas in each dimension; and delineate areas with strong ecological management performance and areas requiring prioritized restoration. Finally, principal component regression and the geographical detector are used to quantify the effects of climate, topography, vegetation, and human activities on total amount, intensity, and structure, thereby clarifying the coupled relationships between drivers and carbon sequestration and oxygen release services and providing evidence to support differentiated and actionable ecological management and policy design.

2. Materials and Methods

2.1. Study Area and Data Resources

2.1.1. Study Area

Minnesota is located in the north-central United States and spans the intersection of four major ecoregions: aspen parkland, prairie, deciduous forest, and coniferous forest. This convergence underpins high biological and ecosystem diversity [22]. Abundant lakes, rivers, and wetlands create distinctive hydrological and climatic conditions that further support biodiversity [23]. Minnesota is administratively divided into 87 counties (Figure 1). The state hosts more than 630 vertebrate species, approximately 20,000 invertebrate species, and over 5000 plant species [24].

Rapid urban expansion, infrastructure development, and land conversion have reduced and fragmented natural habitats. Forests, wetlands, and water bodies have experienced widespread degradation and pollution, placing substantial pressure on biodiversity, and more than 130 native species are listed as endangered [25,26,27]. Minnesota has a continental climate with four distinct seasons. Precipitation decreases from the southeast to the northwest, whereas temperature decreases from south to north. A pronounced urban heat island is evident in the Minneapolis–Saint Paul metropolitan area [28,29]. Over recent decades, both precipitation and air temperature have increased, and extreme rainfall events have become more frequent, posing persistent stress to ecosystem structure, habitat quality, and wildlife persistence [30,31].

Minnesota covers approximately 219,000 km², with water bodies accounting for 8.40% of the area and giving it its nickname, the “Land of 10,000 Lakes” [32]. The landscape, which is dominated by plains, includes prairie and intensive agriculture in the west, partly cleared deciduous forests in the southeast, and extensive forests with low population density in the north [33]. As of 2025, the population is about 5.8 million, with more than half residing in the Minneapolis–Saint Paul metropolitan area [34]. The economy has shifted from resource-based sectors such as agriculture, mining, and timber toward services and higher value-added industries, including healthcare and life sciences, high-tech manufacturing, and food processing; agriculture remains a foundational sector [35,36].

Minnesota has a pronounced seasonal cycle with warm summers and cold winters and is generally classified as having a continental climate. Under the Köppen system, most of southern Minnesota is characterized by a hot-summer humid continental climate (Dfa), whereas the northern two-thirds of the state are classified as warm-summer humid continental (Dfb). Long-term normals indicate a clear south–north temperature gradient, with mean annual temperature about 3–8 °C higher in the south than in the north. Mean annual precipitation is approximately 500–900 mm and decreases from southeast to northwest. Atmospheric moisture, reflected by variables such as dew point temperature and specific humidity, is typically higher during the warm season and lower during winter, consistent with enhanced moisture transport and stronger evapotranspiration in summer.

2.1.2. Data Resources

The statistical data used in this study include population, economic indicators, grain yield, air quality, and motor vehicle counts for Minnesota. Population data were obtained from the U.S. Census Bureau at https://www.census.gov/data.html (accessed on 1 August 2023). Economic data were retrieved from the Minnesota Department of Employment and Economic Development at https://mn.gov/deed/ (accessed on 1 August 2023). County-level administrative boundaries were obtained from the Minnesota Geospatial Commons at https://gisdata.mn.gov/ (accessed on 1 August 2023). All remote sensing data and derived products were accessed through the Google Earth Engine public data catalog, and the datasets used are summarized in

Table 1. Datasets from the public GEE data catalog used in this study.

No.	Dataset	Spatial Resolution	Notes
1	Daymet V4: Daily Surface Weather and Climatological Summaries	1 km	Precipitation
2	MOD11A1.006 Terra Land Surface Temperature and Emissivity Daily Global	1 km	Daytime land surface temperature; Diurnal temperature range
3	MOD13Q1.006 Terra Vegetation Indices	250 m	NDVI
4	Landsat Gross Primary Production CONUS	30 m	Gross primary production (GPP)

.

2.2. Biophysical Assessment of Carbon Sequestration and Oxygen Release Ecosystem Services

In this study, ecosystem oxygen production is used as the biophysical indicator of CSOR. Direct long-term monitoring of atmospheric oxygen at large spatial scales is not feasible because of measurement limitations and the lack of consistent historical records. Therefore, the study estimates oxygen production indirectly using gross primary productivity (GPP) and the stoichiometry of plant photosynthesis, which is widely adopted for quantifying these services [37,38,39].

As shown in Figure 2, the oxygen released by green plants is produced during photosynthesis associated with carbon fixation. Based on the photosynthetic reaction, the mass ratio of fixed organic carbon to released oxygen is approximately 44:32. Net primary productivity, defined as the net accumulation of organic matter per unit area and time, is a key indicator of ecosystem carbon sequestration capacity and is therefore used here to represent the biophysical magnitude of carbon sequestration and oxygen release services.

In the cumulative mass model, ${\mathrm{GPP}}_{\mathrm{t}}$ denotes the gross primary productivity at time step t, expressed as the carbon fixation flux per unit area and per unit time (kgCm⁻² t⁻¹). The time resolution of the GPP data used is 16 days, and $\Delta \mathrm{t}$ is 16 days. Summing ${\mathrm{GPP}}_{\mathrm{t}}$ over all time steps within the study period and multiplying by the study area A (m²) and the time step duration Δt yields the cumulative carbon fixed over the period and area, ${\mathrm{C}}_{\mathrm{fixed}}$ (kgC). Using molar mass conversion factors, the fixed carbon is then converted to the cumulative mass of ${\mathrm{CO}}_2$, ${\mathrm{m}}_{{\mathrm{CO}}_2}$(kgCO₂), and the cumulative mass of O₂ released ${\mathrm{m}}_{{\mathrm{O}}_2}$ (kgO₂), shown in Equation (1).

$$\left\{\begin{array}{c}{\mathrm{M}}_{\mathrm{C}}=\mathrm{A}{\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\mathrm{GPP}}_{\mathrm{t}}\Delta \mathrm{t} \\ {\mathrm{m}}_{{\mathrm{CO}}_2}={\displaystyle \frac{44}{12}}{\mathrm{M}}_{\mathrm{C}}\\ {\mathrm{m}}_{{\mathrm{O}}_2}={\displaystyle \frac{32}{12}}{\mathrm{M}}_{\mathrm{C}}\end{array}\right.$$

(1)

2.3. Statistical Trend Detection Methods

2.3.1. Mann–Kendall Trend Test

The Mann–Kendall (M–K) test is a non-parametric method for detecting a monotonic trend in a time series. It does not require normality or linearity and is widely applied to environmental and ecological datasets [39]. The null hypothesis states that the series shows no monotonic trend, whereas the alternative hypothesis indicates a monotonic trend. For a series with $n$ observations ${(x}_1,{ x}_2,\dots ,x_n$), the Mann–Kendall statistic $S$ is defined in Equation (2) [40]:

$$S={\sum }_{i=1}^{n-1}\left[{\sum }_{j=i+1}^{n}sign(x_j-x_i)\right]$$

(2)

When S > 0, later observations tend to be larger than the earlier ones; when S < 0, the opposite holds. For n ≥ 8, S is approximately normally distributed with mean E(S) = 0. The variance is given in Equation (3):

$$\mathrm{Var}\left(\mathrm{S}\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)-{\sum }_{i=1}^m{t}_i(i-1)(2i+5)}{18}}$$

(3)

The Mann–Kendall test statistic $Z_c$ is defined as Equation (4):

$$Z_c=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{\mathit{Var}\left(S\right)}}}, S>0\\ 0, S=0\\ {\displaystyle \frac{S+1}{\sqrt{\mathit{Var}\left(S\right)}}}, S<0\end{array}\right.$$

(4)

The standardized test statistic is defined in Equation (4). Under the null hypothesis, $Z$ follows the standard normal distribution. A positive $Z_c$ indicates an upward trend; a negative $Z_c$ indicates a downward trend. At a given significance level $\mathsf{\alpha }$, a monotonic trend is considered statistically significant if $\left|Z_c\right|>Z(1-\alpha /2)$, where $Z(1-\mathsf{\alpha }/2)$ is the corresponding critical value.

Trend magnitude is estimated using Sen’s slope, a robust non-parametric estimator. The slope is calculated as Equation (5):

$$\beta =\mathrm{Median}\left({\displaystyle \frac{x_j-x_i}{j-i}}\right), j>i$$

(5)

where the Sen’s slope statistic $\beta$ represents the rate of change. $\beta >0$ indicates an increasing trend, whereas $\beta <0$ indicates a decreasing trend.

2.3.2. Pettitt Change-Point Test

The Pettitt test is a rank-based non-parametric approach for detecting a single abrupt change in the central tendency of a continuous time series [41] and is commonly used for hydrological and climatic applications. The null hypothesis $H_0$ assumes a constant location parameter with no change point, whereas the alternative hypothesis $H_1$ assumes one change point. The test statistics are defined in Equations (6) and (7).

$$K_T=\max \left|U_{t,T}\right|$$

(6)

$$U_{t,T}={\sum }_{i=1}^t{\sum }_{j=t+1}^{T}sgn(X_i-X_j)$$

(7)

If the test is significant, the most likely change point occurs at time $t$, where the statistic reaches its maximum absolute value.

The significance level is evaluated using the associated p-value. For a given $\alpha$, $H_0$ is rejected when $p < \alpha$, indicating a statistically significant change point at confidence level $1-\alpha$. The $p$-value can be approximated using Equation (8).

$$p\cong 2\exp \left({\displaystyle \frac{-6K_T^2}{T^3+T^2}}\right)$$

(8)

Applying the Pettitt test to the annual ecosystem service series from 1998 to 2021 allows the identification of statistically significant shift years and provides evidence on service stability. A significant change point indicates an unstable trajectory, whereas the absence of a change point suggests relative stability over the study period.

2.4. Construction of a Multidimensional Assessment Framework

To characterize ecosystem service levels comprehensively, this study constructs a multidimensional evaluation framework with Quantity ($Q_n$), Intensity ($I_n$), and Structure ($S_t$) as coordinate axes. These dimensions correspond to total output, per-area performance, and temporal evolution, respectively.

Quantity (${\mathrm{Q}}_{\mathrm{n}}^{\prime }$) represents the total annual amount of carbon sequestration and oxygen release provided by an ecosystem unit, reflecting the overall service scale, as defined in Equation (9). Intensity (${\mathrm{I}}_{\mathrm{n}}^{\prime }$) represents the annual service amount per unit area, reflecting per-area efficiency, as defined in Equation (10). Structure (${\mathrm{S}}_{\mathrm{t}}^{\prime }$) represents the interannual change in intensity between adjacent years, reflecting the direction and magnitude of temporal evolution in per-area service provision, as defined in Equation (11). To ensure cross-regional comparability, the three indicators are standardized using $z$-$\mathrm{score}$ normalization to obtain variables with mean 0 and standard deviation 1.

$${Q_n}_{ij}={\displaystyle \frac{{Q_n}_{ij}^{\prime }-\overline{{Q_n}_j^{\prime }}}{{\mathsf{\sigma }}_{{{\mathrm{Q}}_{\mathrm{n}}}_{\mathrm{j}}^{\prime }}}}$$

(9)

$${I_n}_{ij}={\displaystyle \frac{{I_n}_{ij}^{\prime }-\overline{{I_n}_j^{\prime }}}{{\mathsf{\sigma }}_{{{\mathrm{I}}_{\mathrm{n}}}_{\mathrm{j}}^{\prime }}}}$$

(10)

$${S_t}_{ij}={\displaystyle \frac{{S_t}_{ij}^{\prime }-\overline{{S_t}_j^{\prime }}}{{\mathsf{\sigma }}_{{{\mathrm{S}}_{\mathrm{t}}}_{\mathrm{j}}^{\prime }}}}$$

(11)

Positive values indicate above-average performance for the study area in the corresponding dimension, whereas negative values indicate below-average performance. Larger absolute values indicate stronger deviation from the mean. For example, values exceeding +3 fall approximately within the top 0.1% of the distribution for that indicator (

Table 2. Classification and interpretation of ecosystem service indicators.

Ecosystem Value	Distribution Rank	Category
I ≥ +3	Top 0.1%	High
+3 > I ≥ +2	Top 2.3~0.1%	Secondary high
+2 > I ≥ +1	Top 15.9~2.3%	Tertiary high
+1 > I ≥ 0	Top 50~15.9%	Lowest high
0 > I ≥ −1	Bottom 50~15.9%	Lowest low
−1 > I ≥ −2	Bottom 15.9~2.3%	Tertiary low
−2 > I > −3	Bottom 2.3~0.1%	Secondary low
I ≤ −3	Bottom 0.1%	Low

).

In the Q–I–S space, the domain is partitioned into eight states (

Table 3. Spatial classification of the Q–I–S assessment framework.

Space	Quantity	Intensity	Structure	Typical Characteristics	State
I	+	+	+	Quantity, intensity, and structural evolution are all above average	Progression
II	+	+	−	High quantity and intensity, but a weakening structural trend	Degradation
III	+	−	+	High quantity obtained by relying on larger area; low intensity but improving	Progression
IV	+	−	−	High quantity but low intensity with unfavorable evolution	Degradation
V	−	+	+	Small area/quantity but high and continuously improving intensity	Progression
VI	−	+	−	High intensity but insufficient total quantity and unfavorable evolution	Degradation
VII	−	−	+	Low quantity and intensity, but with signs of recovery	Progression
VIII	−	−	−	Low in all three dimensions with continuous degradation	Degradation

) based on the sign combinations of the three indices. Each state represents a distinct service condition and trajectory, as summarized in Table 3.

Within this framework, each ecosystem unit in a can be represented as a point in the three-dimensional space. Its position enables a rapid diagnosis of service scale, per-area performance, and whether the service trajectory is improving or degrading.

2.5. Exploratory Spatial Data Analysis (ESDA)

To examine spatial clustering and heterogeneity in ecosystem services, we applied Exploratory Spatial Data Analysis (ESDA) to the multidimensional assessment results. ESDA, a core approach in spatial statistics, quantifies spatial dependence among geographic units and identifies clustering, dispersion, and spatial outliers using global and local measures. For local spatial association, we used hot spot analysis to detect statistically significant clusters of high and low values. Based on the ${Gi}^{\ast }$ statistic (Equation (12)), this method compares the spatially weighted sum of values for a unit and its neighbors with the global mean to determine whether unusually high or low values are spatially concentrated around I. Its formal expression is

$${Gi}^{\ast }={\displaystyle \frac{{\sum }_{j=1}^n{w}_{ij}{x}_j-\overline{x}{\sum }_{j=1}^n{w}_{ij}}{\sqrt{\frac{{\sum }_{j=1}^n{x}_j^2}{n}-\frac{1}{x^2}}\sqrt{\frac{[n{\sum }_{j=1}^n{w}_{ij}^w-\left({\sum }_{j=1}^n{w}_{ij})\right]}{n-1}}}}$$

(12)

where ${\mathrm{x}}_{\mathrm{j}}$ denotes the observed value of spatial unit $\mathrm{j}$, $\overline{\mathrm{x}}$ is the mean across all $\mathrm{n}$ units, and ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}$ is the spatial weight between units $\mathrm{i}$ and $\mathrm{j}$. The ${\mathrm{G}\mathrm{i}}^{\ast }$ statistic is standardized to a z-score and associated p-value to assess significance under the null hypothesis of spatial randomness. A significantly positive ${\mathrm{G}\mathrm{i}}^{\ast }$ indicates a high-value hot spot; a significantly negative ${\mathrm{G}\mathrm{i}}^{\ast }$ indicates a low-value cold spot.

The study conducted Getis–Ord ${\mathrm{G}\mathrm{i}}^{\ast }$ analysis separately for the three standardized indices, Quantity, Intensity, and Structure, to characterize spatial clustering in CSOR. The results provide a spatial basis for identifying priority conservation areas and areas requiring targeted improvement.

2.6. Analysis of Primary Influencing Factors

Geographically referenced linear regression models are commonly used to quantify relationships among spatial variables and to infer difficult-to-observe outcomes from readily available predictors. The ordinary least squares (OLS) method yields unbiased estimates under standard assumptions, but its performance deteriorates in the presence of multicollinearity among predictors, often leading to unstable coefficients and failed diagnostics. Principal component regression (PCR) mitigates multicollinearity by regressing the response on a set of orthogonal principal components rather than on the original correlated variables. Because the original candidate drivers considered in this study exhibit multicollinearity (see Appendix A.1), we employed PCR to model the relationship between ecosystem service equivalents and the key influencing factors.

2.6.1. Integration of Multiple Influencing Factors

Geographical systems are shaped by interacting natural, climatic, and human factors, which often produces a large set of correlated variables and complicates attribution analysis. To reduce dimensionality while retaining most of the information, we used principal component analysis (PCA) to transform the original indicators into a smaller number of composite, uncorrelated components [42]. PCA projects the n × p data matrix (Eq 13) onto principal components that explain most of the variance, thereby reducing computational complexity and alleviating multicollinearity [43].

$$\mathrm{X}=\left[\begin{array}{cccc}{x}_{11}& x_{12}& \cdots & x_{1p}\\ x_{21}& x_{22}& \cdots & x_{2p}\\ ⋮& ⋮& ⋮& ⋮\\ x_{n1}& x_{n2}& \cdots & x_{np}\end{array}\right]$$

(13)

Set $Z$ denotes the composite indicator that reflects the main information of the original variables $\mathrm{X}$ (Equation (14)).

$$\left\{\begin{array}{c}{z}_1=l_{11}{x}_1+l_{12}{x}_2+\cdots +l_{1p}{x}_p\\ z_2=l_{21}{x}_1+l_{22}{x}_2+\cdots +l_{2p}{x}_p\\ \cdots \cdots \\ { z}_m=l_{m1}{x}_1+l_{m2}{x}_2+\cdots +l_{mp}{x}_p\end{array}\right.$$

(14)

The principles for solving the coefficient matrix $l_{Ij}$ in the above equation are as follows:

① The components $z_i$ and $z_j(\mathrm{i}\ne \mathrm{j};\mathrm{i},\mathrm{j}=1,2,\dots ,\mathrm{m})$ are mutually uncorrelated.

② $z_1{,z}_2$, …, ${ z}_m$ are linear combinations of the uncorrelated variables ${\mathrm{x}}_1,{\mathrm{x}}_2,\dots ,{\mathrm{x}}_{\mathrm{p}}$ ordered by decreasing variance.

③ To retain key information and reduce dimensionality, we select the first few principal components with a cumulative contribution rate above 85% as new variables representing the core structure of the original data.

2.6.2. Principal Component Regression Model

Assume the ecosystem service index $\mathrm{y}$ is determined by $\mathrm{k}$ factors $x_1, x_2, \dots , x_k$, with $\mathrm{n}$ observations $y_{\alpha }, { x}_{1\alpha },{ x}_{2\alpha }, \dots , x_{k\alpha }, \alpha =1, 2, \dots , n$. After standardization, the independent variables $\mathrm{X}$ are transformed by PCA into mutually uncorrelated principal components $\mathrm{Z}$. Given that $\mathrm{Z}$ removes multicollinearity, we first regress $\mathrm{Y}$ on $\mathrm{Z}$, then convert the coefficients back to the original variables $\mathrm{X}$ using the loading matrix $\mathrm{P}$.

We then perform PCA on X to obtain principal components Z; use ordinary least squares to regress Y on Z, obtaining coefficient matrix A; and convert to coefficients for the original variables via $\mathrm{B}=\mathrm{P}\mathrm{A}$.

The multiple linear regression model can be written as Equation (15):

$$y_{\alpha }={\beta }_0+{\beta }_1{x}_{1\alpha }+{\beta }_2{ x}_{2\alpha }+\dots +{\beta }_k{x}_{k\alpha }+{\epsilon }_{\alpha }$$

(15)

The least squares method finds parameters by minimizing the sum of squared residuals. To assess model reliability, an F-test is used.

2.7. Geographical Detector

The geographical detector is a spatial statistical approach for quantifying spatially stratified heterogeneity and identifying its potential drivers. The underlying premise is that if an explanatory factor substantially affects a dependent variable, their spatial patterns should be broadly consistent [44]. The method partitions the study area into strata defined by an explanatory variable and compares the variance of the dependent variable for the entire region with the within-stratum variances. This comparison tests whether spatial stratified heterogeneity is significant and quantifies the explanatory power of each factor. In addition to assessing individual drivers, the geographical detector can evaluate interaction effects between factors, thereby supporting attribution in complex coupled human–natural systems. It has been widely used in studies of construction land expansion, ecosystem service value patterns, food security, and disease surveillance [45].

3. Results

3.1. Ecosystem Service Quantity of Carbon Sequestration and Oxygen Release in Minnesota

The Mann–Kendall test applied to the 1998–2021 time series of ecosystem oxygen release in Minnesota yields ($Z$ = 0.94, p = 0.34), indicating an upward but statistically non-significant trend. After removing outliers and applying a 5-year weighted LOWESS smoother, the trend becomes significant ($Z$ = 2.51, p = 0.012) while remaining positive and approaching the 0.01 significance level (Figure 3). This pattern suggests a gradual long-term increase in oxygen release, but substantial interannual variability may obscure the underlying signal in the raw series. Ecologically, the results indicate an overall improvement in oxygen regulation services during the study period, although the service remains sensitive to short-term disturbances.

The Pettitt change-point test identifies a significant shift in 2013 (p = 0.003), indicating a potential turning point in ecosystem functioning. During 1998–2012, oxygen release fluctuated slightly and showed a weak decline, whereas a clear increasing trend has occurred since 2013. Although the monotonic trend for the full period does not meet the conventional (α = 0.05) threshold in the unsmoothed series, the consistent post-2013 increase and the statistically significant change point jointly support a long-term strengthening. Overall, Minnesota’s carbon sequestration and oxygen release services appear to have transitioned from a weakening phase to an enhancement phase.

3.2. Spatiotemporal Evolution of the Multidimensional Assessment Equivalent

3.2.1. Spatial Distribution and Evolution of the Total Equivalent

From 1998 to 2021, the total equivalent shows a persistent “high in the north, low in the south“ pattern. High-value areas are concentrated in northwest, north-central, and northeast Minnesota, whereas low-value areas occur mainly in the west-central, southwest, south-central, and central regions (Figure 4). Hot spot analysis indicates stable and significant spatial clustering over the study period: hot spots are primarily located in the northeast, while cold spots cluster around east-central Minnesota (Figure 5). This configuration is consistent with the spatial distribution of forest-dominated landscapes and major urbanized areas. Overall, the total equivalent exhibits strong north–south differentiation and a stable clustering structure characterized by distinct high–high and low–low agglomerations.

3.2.2. Spatial Distribution and Evolution of the Intensity Equivalent

The intensity equivalent also exhibits a persistent north–south contrast, but with an opposite regional pattern compared with the total equivalent. High-intensity areas are mainly distributed in west-central, central, southwest, south-central, and southeast Minnesota, and are dominated by secondary and tertiary high-intensity classes with relatively stable extents. Low-intensity areas occur in northwest, north-central, northeast, and east-central Minnesota, primarily in secondary and tertiary low-intensity classes; some areas shift from secondary to tertiary low intensity over time (Figure 4). Hot spot analysis reveals extensive spatial clustering: a stable belt-shaped cold spot extends from northwest to east-central Minnesota, whereas hot spots are concentrated in the southwest, centered on south-central Minnesota, and expand over time (Figure 5). In summary, the intensity equivalent is characterized by (1) pronounced regional differentiation with gradual improvement in northern low-intensity areas and relative stability in southern high-intensity areas, and (2) persistent, large-scale hot- and cold-spot clusters.

3.2.3. Spatial Distribution and Evolution of the Structure Equivalent

The structure equivalent index exhibits localized variability superimposed on a broad spatial gradient. High values persistently cluster in northern to northeastern Minnesota, forming a continuous or semi-continuous belt system that constitutes the backbone of the statewide structural pattern. In contrast, central to southern Minnesota—especially the southwest—remains dominated by low to medium values, maintaining a stable contrast with northern high-value zones. Within this large-scale pattern, high-value patches are interspersed with low- and medium-value areas, indicating substantial local heterogeneity. Over time, the cores of high-value areas remain largely stable, while their spatial extent expands or contracts, with county-level shifts among classes. Overall, the structural pattern is broadly stable but exhibits interannual variability.

3.3. Identification of Degraded Areas of Carbon Sequestration and Oxygen Release Ecosystem Services

Using equally spaced observation windows is a standard approach for temporal discretization and supports consistent comparisons across years. A 5-year interval reduces the influence of short-term fluctuations (e.g., extreme climate events, episodic disturbances, and observational noise) and facilitates identification of staged changes. Accordingly, we selected 1998, 2003, 2008, 2013, 2018, and 2021 as representative time windows to assess degradation.

Degraded counties were classified by the frequency of degradation across the six windows (

Table 4. Classification and grading of degraded areas of carbon sequestration and oxygen release ecosystem services in Minnesota.

Level	Number of Degradation Occurrences	Counties	Count
A	≥5	Todd, Cottonwood, Redwood, McLeod, Renville	5
B	4	Lac qui Parle, Watonwan, Meeker, Wadena, Fillmore, Brown, Sibley, Wright, Sherburne, Carver, Stearns, Freeborn	12
C	3	Hubbard, Jackson, Pope, Waseca, Murray, Big Stone, Chippewa, Faribault, Mower, Dodge, Olmsted, Wabasha, Blue Earth, Nicollet, Mille Lacs, Chisago, Isanti, Le Sueur, Goodhue, Benton, Stevens, Morrison, Douglas, Crow Wing, Martin, Houston, Lyon, Steele, Kandiyohi, Winona	30
D	2	Aitkin, Kanabec, Lake, Pennington, Traverse, Yellow Medicine, Swift, Red Lake, Clearwater, Cook, Grant, Pine, Rice, Dakota, Anoka, Hennepin, Mahnomen, Cass, Otter Tail, Nobles	20
E	≤1	Rock, Koochiching, Becker, Norman, Marshall, Lake of the Woods, Lincoln, Roseau, Scott, Ramsey, Washington, Clay, Wilkin, Kittson, Beltrami, St. Louis, Carlton, Polk, Itasca	19

). For each spatial unit, the number of degradation occurrences n were counted and used to assign a severity grade: Grade A, n ≥ 5, indicating the most persistent degradation; Grade B, n = 4; Grade C, n = 3, Grade D, n = 2, and Grade E, n ≤ 1, representing the least persistent degradation. This frequency-based scheme captures degradation persistence and stability, with higher n indicating more sustained deterioration.

Table 4 shows that Todd, Cottonwood, Redwood, McLeod, and Renville Counties exhibit degradation in six key years and are classified as Grade A. Twelve counties are Grade B, indicating degradation in four windows (13.79% of all counties) and reflecting relatively severe and persistent decline. Grade C–E counties account for most of Minnesota (79.31%), indicating that sustained degradation is not widespread statewide. Nevertheless, a subset of counties show persistent degradation and warrant targeted attention.

Spatially, degraded areas show a clear pattern of “severe degradation in the central region and relatively better conditions toward the periphery” (Figure 6). All Grade A and Grade B counties are concentrated in central and southwest Minnesota, forming a distinct degradation core. Outside this core, no sustained degradation trend is evident. Degradation intensity generally decreases with distance from the core, although local “enclaves” of higher degradation and embedded low-to-medium zones occur, likely associated with land-use conversion, urban expansion and transport corridors, agricultural intensification, and the distribution of water bodies and wetlands. Together, these processes produce a mixed pattern combining broad-scale attenuation with a patchy mosaic.

3.4. Analysis of the Main Influencing Factors of Carbon Sequestration and Oxygen Release Services

3.4.1. Dominant Factors and Mechanisms

Because carbon sequestration and oxygen release are primarily mediated by green vegetation, six variables were selected as key drivers: climate factors including precipitation (RF), daytime temperature (T), and diurnal temperature range (DTR); a vegetation index (NDVI); and socio-economic proxies for human activity including population (POP) and gross domestic product (GDP). PCA was performed on these six variables. The first three components explain 86.46% of total variance (

Table 5. Principal component analysis results for carbon sequestration and oxygen release ecosystem services.

Principal Component	Eigenvalue	Contribution Rate %	Cumulative Contribution Rate %
1	2.2097	36.8279	36.8279
2	1.9854	33.0905	69.9184
3	0.9922	16.5374	86.4558
4	0.5302	8.8364	95.2922
5	0.2571	4.2844	99.5766
6	0.0254	0.4234	100

), indicating that the six variables can be effectively summarized by three latent dimensions.

Based on the loadings (

Table 6. Principal component loadings.

Original Variable	Principal Component 1	Principal Component 2	Principal Component 3
RF	0.3907	0.2344	0.0155
T	0.5382	0.6935	−0.2366
DTR	0.2526	−0.055	0.7057
NDVI	0.2795	−0.0514	0.5757
POP	−0.611	0.636	0.3172
GDP	−0.206	0.2324	0.1167

), the first component is strongly associated with POP and can be interpreted as human activity intensity. The second component loads strongly on T and also relates to POP; given the documented linkage between population concentration and urban heat effect, this component primarily reflects the temperature environment. The third component is strongly associated with DTR and NDVI; because diurnal temperature range is closely linked to vegetation physiological processes, this component represents vegetation growth conditions. The six variables can be grouped into three aspects: human activity intensity, thermal conditions, and vegetation growth status.

To quantify relationships between the drivers and the service indicators, PCR models were fitted between the six factors and the standardized indices, $Q_n$, $I_n$, and $S_t$ (Equation (16)).

$$\left\{\begin{array}{c}\mathrm{Q}\mathrm{n}=1.6101-3.7621GDP+3.9995POP-6.7280T+5.3355DIF+0.7924RF-0.2548NDVI\\ In=-1.3029+7.3807GDP-9.8919POP+5.0027T-3.5040DIF+1.3540RF-1.0953NDVI\\ St=1.6236+4.4556GDP-4.2083POP-7.9041T+8.3144DIF+0.1677RF-0.9430NDVI\end{array}\right.$$

(16)

The relative influence of predictors differs across dimensions. For total quantity, effect sizes rank as T > DTR > POP > GDP > RF > NDVI, identifying daytime temperature as the dominant factor. For intensity, the ranking is POP > GDP > T > DTR > RF > NDVI, indicating that human activity intensity, especially population, is the primary constraint. For structure, the ranking is DTR > T > GDP > POP > NDVI > RF. Overall, climate factors exert stronger effects on total quantity and structural change, whereas socio-economic factors more strongly influence intensity.

Model reliability was evaluated using the F-test (

Table 7. Critical values for the F-test.

Test Item	$\mathit{\alpha }$	$\mathbf{F}$ Score	$\mathit{p}$-Value	Critical Value	Result
$Q_n$	0.001	58.94	<0.0001	4.2	Reliable
$I_n$	0.001	38.3614	<0.0001	4.2	Reliable
$S_t$	0.001	28.2317	<0.0001	4.2	Reliable

). All fitted models exceed the critical values at α = 0.001, indicating statistical significance at this confidence level. Residual diagnostics show residuals concentrated around zero with no evident systematic pattern, suggesting the model was adequate and fit the quantity, intensity, and structure.

3.4.2. Spatial Coupling Between Services and Key Influencing Factors

Geographical detector analysis was used to quantify spatial coupling between the drivers and service equivalents (including Quantity, Intensity, and Structure indicators) (

Table 8. Interaction detector results for carbon sequestration and oxygen release services.

	T	TR	RF	NDVI	POP	GDP
T	0.2886
DTR	0.3884	0.1621
RF	0.4844	0.568	0.1925
NDVI	0.7431	0.7478	0.7099	0.1433
POP	0.7544	0.7311	0.5037	0.6595	0.1028
GDP	0.7147	0.7164	0.581	0.6529	0.2124	0.1196

). The single-factor ranks as T (0.29) > RF (0.19) > DTR (0.16) > NDVI (0.14) > GDP (0.12) > POP (0.10). Thus, climate factors show substantially stronger spatial coupling with ecosystem services than socio-economic factors. Daytime temperature exhibits the strongest single-factor association (q = 0.29), followed by precipitation, whereas population has the weakest (q = 0.10).

Interaction detection indicates that factor pairs generally show nonlinear enhancement or mutual enhancement, with markedly higher explanatory power than single-factor effects. This implies that the spatial pattern of carbon sequestration and oxygen release services is jointly shaped by multiple interacting drivers rather than by any single factor. The strongest enhancement occurs for POP × T interaction (q ≈ 0.7544), followed closely by NDVI × DTR (q ≈ 0.7478). These results highlight temperature regime, diurnal temperature range, vegetation condition, and population distribution as key interacting determinants of spatial heterogeneity in carbon sequestration and oxygen release services.

4. Discussion

4.1. Long-Term Trends in Carbon Sequestration and Oxygen Release

The Mann–Kendall test applied to the unsmoothed series suggests an overall increase in Minnesota’s carbon sequestration and oxygen release (CSOR) services from 1998 to 2021. However, the trend is weakly significant. After applying a 5-year weighted LOWESS smoother to reduce high-frequency noise and outliers, the upward trend becomes statistically stronger. This indicates that analyses based solely on raw remote sensing time series may underestimate long-term change, whereas a combined “smoothing + trend” strategy can more robustly extract underlying ecological signals.

The Pettitt test identifies a significant change point in 2013, consistent with a transition from a fluctuating phase to a recovery phase. This shift likely reflects lagging responses to policy and environmental changes. Since the late 20th century, the U.S. Farm Bill and Minnesota’s conservation and climate policies have promoted conservation tillage, wetland and grassland restoration, riparian buffers, and tighter controls on forest and wetland conversion. Many of these measures were implemented mainly between 2000 and 2010, and their cumulative effects on vegetation productivity and CSOR would be expected to emerge after a time lag, yielding a detectable regime shift around 2013 [46].

Concurrently, forest, grassland, and wetland areas in Minnesota generally stabilized or expanded after 2000 [47], while the growth of urban land and high-intensity agriculture slowed [48]. Broader adoption of conservation agriculture likely strengthened vegetation carbon uptake. Together with warming and, in some years, improved hydrothermal conditions, net primary productivity may have undergone a post-disturbance recovery and enhancement process, amplifying the phase shift in CSOR around 2013.

Despite the statewide increase in total CSOR, persistent degradation remains evident at local scales. The total-equivalent index exhibits a stable north–south gradient, with broad improvement in northern high-value areas and continued lagging performance in the central and southern low-value regions. Degradation mapping further indicates that persistently degraded counties are strongly clustered in central and southwestern Minnesota, forming a distinct degradation core (Figure 6; Table 4). These patterns suggest that statewide gains are partly driven by contributions from northern high-supply areas, while portions of central and southern Minnesota continue to experience declines in service condition and ecosystem functioning.

4.2. Advantages of the Quantity–Intensity–Structure Multidimensional Framework

The Quantity–Intensity–Structure (Q–I–S) framework distinguishes three complementary attributes of ecosystem services, scale, per-area performance, and temporal evolution and characterizes them jointly. Quantity captures the total annual CSOR provided within a spatial unit, Intensity describes CSOR per unit area, and Structure reflects the direction and magnitude of change through interannual variation in intensity. This configuration enables simultaneous answers to “how much?,” “where?,” and “is it improving or degrading?” By partitioning the three-dimensional space into eight states based on the sign combinations of Q, I, and S, the framework moves beyond simple high–low comparison toward process-oriented diagnosis. For example, units with high Quantity and high Intensity but negative Structure represent “high-value but degrading” areas with latent risk, whereas units with low Quantity and low Intensity but positive Structure indicate “low baseline but recovering” areas with restoration potential. Leveraging the spatiotemporal continuity of remote sensing data, the framework supports fine differentiation of ecological states and trajectories and provides operational evidence for targeted management. Importantly, Q–I–S is indicator-agnostic: any ecosystem service represented as a remote sensing time series, such as food production, water regulation, or soil retention, can be assessed within the same structure, supporting transferability across services.

4.3. Ecological Mechanisms of the North-High, South-Low Spatial Pattern

Northern Minnesota, including the northeast and northwest regions, exhibits high CSOR (carbon sequestration and oxygen release) in association with substantial forest cover, particularly coniferous and mixed forests. These forest ecosystems not only support greater aboveground biomass, thereby enhancing photosynthetic capacity, but also promote soil organic carbon storage, contributing to the formation of persistent carbon pools. In addition, the region’s abundant wetlands and water bodies provide favorable moisture regimes for tree growth, and studies have demonstrated that wetlands play a crucial role in carbon sequestration [49]. Wetland vegetation absorbs atmospheric CO₂ through photosynthesis and converts it into organic matter, and adjacent terrestrial vegetation also contributes to carbon storage [50]. In the southwest and central south, agricultural intensity is higher, with intensive cropping systems such as soybean and maize cultivation leading to degradation of soil structure and soil quality and a reduction in the ecosystem’s carbon fixation potential. At the same time, urbanization and land development exacerbate the loss of natural vegetation and thus attenuate CSOR, particularly within the Minneapolis–Saint Paul metropolitan area [51].

Surrounding management practices likewise influence the spatial patterns. In northern Minnesota, effective forest management measures help sustain higher levels of ecosystem health. Data from the Minnesota Department of Natural Resources indicate that forests enrolled in the Sustainable Forests Incentive Act (SFIA) cover more than 540,000 acres, reflecting substantial participation in sustainable forest management that supports carbon storage and biodiversity. Moreover, Minnesota has implemented multiple measures to control and eradicate invasive species, including the Minnesota Invasive Species Act, which further supports ecosystem health. In contrast, portions of southern Minnesota exhibit agricultural practices that are comparatively less sustainable. These practices are characterized by extensive tillage and reliance on monoculture cropping of maize and soybeans, resulting in notable soil degradation. USDA data indicate soil erosion rates in southern Minnesota on the order of approximately 5 tons per hectare per year. Additionally, the rapid urbanization of the Minneapolis–Saint Paul area has led to substantial loss of green space; although urban greening initiatives exist, protection measures have not kept pace with urban expansion. City planning data show that over the past decade, urban green space in this region declined by roughly 15%. Collectively, unsustainable cropping practices and deficiencies in urban planning significantly undermine the region’s CSOR.

4.4. A Joint Attribution Using Principal Component Regression and Geographical Detectors

This study combined principal component regression and geographical detectors to attribute the spatial variability of CSOR to natural and anthropogenic drivers. Beyond statistical attribution, the identified drivers can be interpreted as mechanistic constraints operating through three pathways: climatic regulation of photosynthesis and respiration, vegetation condition as an integrated signal of site productivity and disturbance, and the effects of land-use intensity and landscape fragmentation.

Precipitation affects CSOR primarily by regulating soil moisture and stomatal conductance, thereby controlling gross primary production and, ultimately, NPP-derived CSOR. Temperature influences enzymatic activity and phenology, yet its net effect can differ across space depending on whether warming relieves thermal limitation or increases evaporative demand and drought stress. Diurnal temperature range serves as an indicator of atmospheric dryness and cloudiness. A larger diurnal temperature range is commonly associated with clearer skies and stronger radiative forcing, but also with lower humidity and higher vapor pressure deficit, which can suppress photosynthesis by promoting stomatal closure. These mechanisms support the observed importance of hydrothermal conditions in explaining CSOR spatial heterogeneity.

NDVI represents canopy greenness and provides a close proxy for leaf area and photosynthetic capacity. Higher NDVI generally indicates greater light interception and carbon assimilation, leading to higher CSOR intensity. NDVI also integrates lagging effects of management, disturbance history, and site conditions, including crop rotation and residue retention, harvest and drought impacts, and soil fertility. This helps explain why vegetation condition remains a strong explanatory factor after accounting for climate variability.

Population and GDP act as proxies for land-use conversion pressure, impervious-surface expansion, and the intensity of human disturbance. These pressures can reduce CSOR through habitat loss, fragmentation, and microclimate alteration, and by limiting rooting volume and infiltration in developed areas. In agricultural landscapes, economic activity may be associated with management intensification, such as drainage, fertilization, and irrigation. Such intensification can raise productivity in some locations while reducing ecosystem stability and resilience, with direct implications for Structure.

Geographical detector results show that two-factor interactions typically explain spatial heterogeneity more effectively than individual factors, indicating that CSOR patterns are shaped by coupled constraints rather than a single dominant driver. The interaction between population and temperature aligns with the combined influence of urban heat effects and land surface modification, which can amplify thermal stress and shift phenology, producing spatially concentrated impacts on CSOR. The interaction between NDVI and diurnal temperature range indicates joint control by canopy status and atmospheric moisture demand. Even in areas with high greenness, a large diurnal temperature range can limit realized photosynthesis, whereas favorable conditions allow high-NDVI landscapes to translate potential productivity into actual CSOR.

From a functional perspective, climate and NDVI primarily regulate Intensity by shaping photosynthetic efficiency and canopy capacity. Land-use proxies more strongly influence Quantity by altering the spatial extent of vegetated surfaces. Structure is particularly sensitive to the temporal stability of these constraints, and negative Structure is more likely where climatic stress and human disturbance co-occur, reflecting persistent degradation or reduced resilience.

4.5. Remote Sensing-Based Degradation Identification and Spatial Management Implications

Using the Structure indicator to represent interannual change, this study developed a frequency-based degradation grading approach and, together with hot spot analysis, identified persistent degradation cores and surrounding areas with episodic degradation. The key step is to translate multi-temporal remote sensing indicators into a binary sequence of “degradation events,” and then classify each spatial unit into Grades A–E based on the frequency of degradations. This design reduces sensitivity to anomalies in any single year and emphasizes persistence, improving the detection of long-term ecological risk areas.

Overlaying degradation grades with Q–I–S states and hot/cold spots reveals a core degradation belt in central Minnesota surrounded predominantly by areas with lower degradation frequency. As the state’s major metropolitan region, the Twin Cities area concentrates population and economic activity and includes impervious surfaces, dense transportation networks, and frequent land development and renewal. These pressures intensify vegetation fragmentation and impose sustained disturbances on local ecosystems, limiting stable recovery and contributing to high-frequency, high-severity degradation clusters.

This integrated evidence base supports spatially differentiated management. Areas within the Twin Cities region showing persistent degradation and simultaneously high Quantity or Intensity should be prioritized for risk control and ecological restoration. In contrast, peripheral areas with low current service levels but positive Structure values are suitable for designation as zones with recovery potential in medium- and long-term restoration planning. Methodologically, the combination of time series analysis, frequency-based grading, and spatial clustering provides an operational workflow for large-scale monitoring and early warning of ecosystem service degradation using standard satellite products.

In the degraded zoning and governance framework for central and southwestern Minnesota, Grade A core zones require a targeted reversal of long-standing degradation to enhance CSOR stability and ecological resilience in order to restore the natural ecological pattern. Management should prioritize large-scale ecological restoration and reforestation, employing locally native species to establish a mixed conifer–broadleaf forest, and implement a planned regime of thinning and renewal with rotation limits and designated monitoring plots. Concurrently, control of invasive species should be strengthened, and a regional eradication and monitoring network should be established. Grade B areas should focus on halting further degradation and promoting localized restoration to consolidate existing ecological functions; within ecological restoration zones, buffers and restoration of native understory vegetation should be advanced to increase soil cover and moisture retention. For Grade C zones of moderate degradation, the objective is to maintain and stabilize existing ecological functions and prevent further deterioration. Recommendations include the promotion of ecologically friendly agriculture by implementing reduced-tillage practices, the use of cover crops, and crop rotation to reduce erosion and enhance soil organic matter. Grade D areas should pursue sustained improvement through localized, low-intensity interventions, while Grade E regions should maintain a low-degradation trajectory and prevent escalation to more severe states in order to sustain regional ecological resilience. Across the board, practices should maintain existing nature reserves and green spaces with regular monitoring of core indicators. The governance hierarchy should follow an A > B > C > D > E priority, with resources preferentially allocated to Grades A and B; Grade C should receive continual monitoring and gradual restoration to prevent escalation to higher degradation levels. Resource allocation should center on natural ecological restoration complemented by agricultural modernization, green infrastructure, and community engagement. Additionally, cross-regional collaboration should be established in the central and southwestern areas to promote replicable restoration paradigms, improving governance efficiency and scalability.

Building on this evidence, an operational pathway should link degradation Grades A–E with Q–I–S states to define tiered interventions and measurable targets. In the persistent degradation core, near-term risk control should be prioritized by restricting new land conversion, expanding permeable surfaces, and implementing corridor-based greening along riparian zones and major transportation axes. Management performance should be evaluated annually using grade transitions and changes in Structure. In surrounding areas with episodic degradation, adaptive management should focus on buffering drought and heat stress through targeted canopy enhancement and soil and water conservation, with interventions triggered when consecutive degradation events exceed a predefined threshold. In zones with recovery potential and positive Structure, medium- to long-term restoration should prioritize connectivity and habitat quality, supported by a satellite-based monitoring–evaluation–adjustment loop that dynamically reallocates resources to spatial units that fail to improve within a specified period.

4.6. Uncertainties, Limitations, and Future Research

This study contains several uncertainties related to the data, methods, and county-level results. First, CSOR is inferred from satellite-derived NPP using a stoichiometric conversion. This reflects an NPP-driven signal and does not explicitly account for ecosystem respiration, lateral carbon transport, or disturbance pulses. Thus, CSOR here should be interpreted as a potential service indicator rather than a full net carbon–oxygen exchange. Second, uncertainties in the NPP product and preprocessing may propagate to county aggregation and the classification/grading, particularly in counties with heterogeneous land cover. Third, the five-year equal-interval sampling provides consistent temporal coverage but may miss short-term extremes, which could affect the interpretation of interannual variability. Fourth, pest and disease outbreak data are highly incomplete, so this factor was not considered in the current analysis; it should be incorporated in future work when more comprehensive and reliable outbreak records become available. Future work will validate and refine CSOR using flux observations and model outputs, incorporate additional remote sensing constraints, and test the robustness of county grading under alternative temporal sampling and uncertainty analyses.

Future work should place greater emphasis on field validation and calibration, the integration of high-resolution and multi-source remote sensing data, and the development of process-driven inversion methods. It should also implement rigorous uncertainty quantification, including Monte Carlo methods and Bayesian frameworks, as well as event-driven time series analysis and rapid response mechanisms to enhance the diagnostic capacity and predictive robustness of ecological processes. These improvements should be applied to the validation, refinement, and extension of CSOR. Additional remote sensing constraints and multi-source data should be incorporated, and the robustness of county-level classifications under varying temporal sampling schemes and uncertainty analyses should be evaluated.

5. Conclusions

CSOR services in Minnesota strengthened overall from 1998 to 2021, with a statistically significant turning point in 2013. Mann–Kendall and Pettitt analyses jointly indicate a transition from an early period dominated by fluctuations with slight weakening to a later period of sustained recovery and enhancement, with 2013 as the key change point. The higher significance after temporal smoothing underscores the need for a “smoothing + trend testing” workflow when extracting long-term signals from noisy remote sensing time series.

The Q–I–S framework effectively separates service scale, per-area performance, and temporal evolution, enabling identification of both latent risk areas and zones with recovery potential. In particular, the framework highlights (1) “high-value but declining” units, characterized by high Quantity and Intensity with negative Structure, as priorities for risk prevention and restoration, and (2) “low baseline but improving” units with low Quantity and Intensity but positive Structure as strategic targets for medium- and long-term enhancement. Because the framework is not tied to a specific indicator set, it is readily transferable to other ecosystem services represented by long-term remote sensing series.

CSOR exhibits pronounced north–south differentiation and stable spatial clustering across Minnesota. Quantity remains higher in the north and lower in the south, broadly aligning with forest belts versus cropland- and urban-dominated landscapes, and forming persistent hot and cold spots. Intensity shows large-scale clustering with stable hot spots in the south/southwest and a persistent cold spot belt from the northwest to east-central areas. Structure indicates improving trajectories primarily in the north and more persistent weakening in parts of the south, together implying a stable spatial gradient coupled with localized heterogeneity.

Persistent degradation is limited in extent but highly concentrated. Frequency-based grading identifies five Grade A and twelve Grade B counties concentrated in central and southwestern Minnesota, forming a narrow high-risk belt. Approximately 80% of counties show no persistent degradation. This concentration suggests that management resources should prioritize the central belt despite generally stable conditions statewide.

Climatic factors dominate variation in Quantity and Structure, whereas human activity factors more strongly shape Intensity, and CSOR patterns are governed by multi-factor interactions. PCR identifies daytime temperature and diurnal temperature range as the most influential drivers of service scale and trajectory, while population and GDP are the leading drivers of per-area intensity.

Geographical detector results show that interaction effects consistently exceed single-factor explanatory power; combinations such as population, temperature and NDVI, and diurnal temperature range exhibit strong nonlinear enhancement, confirming that CSOR spatial patterns arise from synergistic coupling among multiple natural and anthropogenic drivers rather than being controlled by any single factor.