Abstract
Introduction
Arid and semi-arid regions worldwide face a persistent double challenge: chronic water scarcity and destructive flash floods. In these fragile landscapes, floodwater spreading has emerged as a promising nature-based solution, diverting seasonal runoff onto permeable lands to recharge groundwater, improve soil moisture, and restore vegetation. Iran, dominated by drylands, has implemented numerous such projects over recent decades, yet their long-term effectiveness has rarely been rigorously proven. Most evaluations relied on scattered field observations without any control site, making it impossible to isolate the project's true impact from natural climatic variability; any observed greening or wetness could simply reflect a sequence of wetter years. Fortunately, advances in satellite remote sensing now offer a robust alternative. Freely available, high-resolution Sentinel-2 imagery enables consistent, repeatable tracking of moisture-related indices across many years. This study directly addresses the evaluation gap by adopting a Control-Impact design and applying a differential trend analysis to NDWI and NDMI time series. Our primary objective was to quantify the lasting influence of floodwater spreading on surface moisture and vegetation water content, effectively separating the human-induced signal from background climate variability. The research not only remedies critical methodological shortcomings of previous work but also delivers a clear, reproducible monitoring and evaluation framework for similar water-harvesting systems well into the future.
Materials and Methods:
The study was conducted at the Soh Meimeh Floodwater Spreading Research Station (Isfahan Province, Iran) and an adjacent control area with similar physiographic conditions but no floodwater reception, spanning from April 2016 to March 2024 (nine years). Sentinel-2 MSI Level-2A imagery was employed, and all processing was carried out on the Google Earth Engine (GEE) platform. After cloud masking and generating monthly median composites, NDWI (using bands B8 and B11) and NDMI (using bands B8 and B12) were calculated for each month, and their mean values within the study and control polygons were extracted. Within a Control-Impact design based on BACI logic, the primary response variables were defined as the concurrent differences between the two sites (ΔNDWI and ΔNDMI). Statistical analyses were performed in R and included: 1) descriptive statistics, boxplots, and histograms to assess index distributions; 2) paired t-test (with normality of differences confirmed by Shapiro-Wilk test) and Mann-Whitney U test to examine the significance of mean differences; 3) simple linear trend fitting on individual time series of each site; and 4) linear trend analysis on the differential series using linear regression, with standard errors corrected by the Newey-West method to account for any autocorrelation. Model adequacy was verified through residual analysis and the Durbin-Watson test.
Results and Discussion:
Descriptive statistics showed that the mean NDWI at the spreading site (M = -0.220) was higher than the control (M = -0.226), and similarly, mean NDMI was higher at the spreading site (M = -0.103) compared to the control (M = -0.116). The ranges confirmed this advantage, with minimum values in the control (NDWI: -0.267, NDMI: -0.144) being substantially more negative than those in the spreading site (NDWI: -0.255, NDMI: -0.126). Boxplots confirmed a systematic upward shift in distributions towards greater moisture at the spreading site. The paired t-test indicated highly significant differences for both indices (NDWI: t=11.17, p<0.001; NDMI: t=9.94, p<0.001), and the Mann-Whitney test results were consistent (NDWI: U=4381, p<0.001; NDMI: U=4838, p<0.001). Individual linear trends of indices at each site were all non-significant (p-values ranging from 0.218 to 0.992), indicating that background factors produced no discernible directional trend over the study period. The delta trend analysis, as the core of this research, revealed a different picture. ΔNDWI exhibited a significant positive slope of +0.00073 per year (p=0.037, R²=0.054), signifying a gradual increase in the moisture gap in favor of the spreading area. Conversely, ΔNDMI showed a non-significant negative slope of -0.00063 per year (p=0.215, R²=0.019), meaning that the effectiveness of floodwater spreading on deeper moisture (NDMI) remained stable with no decreasing or increasing trend. Examination of residual autocorrelation confirmed model adequacy, and the Newey-West correction did not alter the conclusions. Temporal dynamics analysis also revealed a widening NDWI gap after 2019 and an exceptional spike in the control site in April 2020 (due to heavy rainfall), underscoring the importance of long-term trend analysis.
Conclusion:
The findings provide robust evidence for the positive, sustained effectiveness of floodwater spreading on surface and vegetation moisture. The dual pattern—increasing ΔNDWI with stable ΔNDMI—reveals layered mechanisms: floodwater spreading initially stabilized deep soil moisture (stable ΔNDMI), then through improved soil structure, greater organic matter, and perennial vegetation, generated a positive feedback that specifically boosted surface moisture and active vegetation water content (increasing ΔNDWI). For the first time, these effects were quantitatively documented using a rigorous causal Control-Impact design. The non-significance of individual site trends versus the significant differential trend strongly underscores the necessity of Control-Impact designs and delta analyses; otherwise, the true intervention signal remains masked by high climatic background noise. Several limitations warrant caution. No pre-project satellite data were available, ground calibration was lacking, and R² values were relatively low. Additionally, spectral indices are indirect proxies, susceptible to atmospheric dust, sun angle variations, soil differences, and vegetation phenology. The absence of concurrent field soil moisture and biomass measurements further restricts direct validation. Therefore, generalization to other sites or climates should be careful. Nonetheless, the presented framework—combining Sentinel-2 imagery, cloud processing, and delta trend analysis—offers a scientific, cost-effective, and reproducible approach for continuous monitoring of floodwater spreading projects and other dryland restoration interventions. It can support evidence-based decision-making for water resource management and desertification control in Iran's arid and semi-arid regions. |
