When examining the structural mechanics of agricultural spatial analysis, few datasets match the rigorous multi-variable depth of Pampean region crop trials. The evaluation of spatial data—compiled across 2,310 records within a designated Pampean lot—offers a transparent lens into how environmental covariates dictate agronomic performance over consecutive agricultural campaigns.
At the center of this investigation lies a two-dimensional spatial framework mapping UTM coordinates, topographical elevation, water table depth at flowering (designated as NapaDicC1 and NapaDicC2 for the first and second campaigns respectively), and corresponding corn yields measured in metric tons per hectare via combine-mounted yield monitors. Through conversion from Microsoft Excel source sheets to comma-separated text formats utilizing RStudio, researchers structured continuous variables to test distributional normality, spatial autocorrelation, and regression trend models.
Descriptive Metrics and Distributional Variance
The statistical profile confirms stark differences between the two monitored campaigns. For the first campaign (C1), both the mean ($\overline{z}$) and median ($Md$) corn yield ($Rm$) outpaced the second campaign ($C2$). Specifically, the primary campaign registered a mean yield of $\overline{z}_{Rm1} = 9.13 \ \text{t}\cdot\text{ha}^{-1}$ with a median of $Md_{Rm1} = 8.86 \ \text{t}\cdot\text{ha}^{-1}$ and a mode of $Mo_{Rm1} = 7.00 \ \text{t}\cdot\text{ha}^{-1}$. Conversely, the second campaign recorded a mean of $\overline{z}_{Rm2} = 7.55 \ \text{t}\cdot\text{ha}^{-1}$, a median of $Md_{Rm2} = 7.99 \ \text{t}\cdot\text{ha}^{-1}$, and a mode of $Mo_{Rm2} = 9.40 \ \text{t}\cdot\text{ha}^{-1}$. Two missing data cells were identified within the C2 yield records.
The dispersion of values surrounding the median was wider during the initial campaign. The range ($R = x_{min}, x_{max}$) for campaign one spanned from $0.9800 \ \text{t}\cdot\text{ha}^{-1}$ to $17.0850 \ \text{t}\cdot\text{ha}^{-1}$, whereas campaign two compressed to a range between $1.9381 \ \text{t}\cdot\text{ha}^{-1}$ and $12.2521 \ \text{t}\cdot\text{ha}^{-1}$. Furthermore, Shapiro-Wilk normality tests resulted in significant p-values, leading to the rejection of the null hypothesis of normal distribution for all evaluated variables. Standard transformations—including logarithmic, square root, and Box-Cox adjustments—failed to normalize the distributions, a deviation visually confirmed by Q-Q plots.
Environmental Covariates and Correlation Architecture
To understand the physical drivers behind yield variance, researchers analyzed topographical elevation and water table depth. Elevation averaged $\overline{z}_{Ele} = 244.31 \ \text{m}$ ($Md_{Ele} = 244.26 \ \text{m}$, $Mo_{Ele} = 242.77 \ \text{m}$) across a range of $241.91 \ \text{m}$ to $27.50 \ \text{m}$. Meanwhile, water table depth ($Pn$) averaged $\overline{z}_{Pn1} = 4.07 \ \text{m}$ ($Md_{Pn1} = 3.98 \ \text{m}$) in the first campaign, compared to $\overline{z}_{Pn2} = 3.44 \ \text{m}$ ($Md_{Pn2} = 3.31 \ \text{m}$) in the second.
Scatter plots and local mean smoothing lines exposed clear negative relationships between corn yield and both environmental covariates:
- Elevation: Negative correlation coefficients were confirmed at $r_{C1} = -0.56$ for campaign one and $r_{C2} = -0.73$ for campaign two.
- Water Table Depth: Stronger negative correlations emerged against yield, registering $r_{C1} = -0.86$ and $r_{C2} = -0.84$.
- Latitude and Longitude: Exhibited negligible trends, with correlation values remaining low between $0 < r < 0.35$.
- Elevation vs. Water Table: A positive correlation linked terrain height with deeper water tables ($r_{C1} = 0.67$ and $r_{C2} = 0.74$).
Trend Modeling and Spatial Autocorrelation
Using Akaike’s Information Criterion ($AIC$) to select the optimal model configuration, multiple linear regression ($y = \beta_{1}x_{1} + eta_{2}x_{2} + eta_{0}$) proved superior for both crop cycles. The selected regression models explained $73.91\%$ of the variance for C1 and $72.96\%$ for C2. Relative importance metrics demonstrated that water table depth exerted a heavier influence on yield trends than terrain elevation, a distinction that was more pronounced during the first campaign.
Examining spatial autocorrelation via Moran’s Index ($I$) on model residuals revealed low positive spatial correlation restricted to distances between $150 \ \text{m}$ and $200 \ \text{m}$. Beyond this threshold, spatial correlation degraded into random distribution patterns, aside from an isolated high positive correlation spike at distances exceeding $1,200 \ \text{m}$ during C2—likely driven by low sample counts ($n_{14} = 96$) and localized outliers.
Sources
These sources formed the evidence pack for this article. Links open the original publisher; inclusion does not imply endorsement.
- github.com original
- bcra.gob.ar original
- tipyanabidky.cz original
- sat4all.com original
