Morphological variation in well-facilitated farmland parcels after construction using multi-source remote sensing and dual branch spatiotemporal fusion network

A cropland parcel is one of the most basic units in agricultural production. Its spatial structure has also changed significantly in China in recent years, particularly with the advancement of well-facilitated farmland construction. However, it remains unclear on the spatial morphological variations in the cropland parcels after well-facilitated farmland construction. It is still lacking in the evaluation of construction and long-term benefits. In this study, a systematic and efficient monitoring was proposed to timely detect the cropland changes. Several typical regions of the well-facilitated farmland construction were also selected in the middle reaches of the Yangtze River. Sentinel-2 and GF-2 imagery data were integrated with deep learning (Dual branch SpatioTemporal Fusion Network, DSTFNet). The vectorized boundaries of cropland parcels were extracted after construction. Six landscape metrics—mean parcel size, shape index, fractal dimension index, edge density, parcel density, and standard distance index—were employed to characterize the spatial morphological features of the cropland parcels after construction. Furthermore, the Cropland Parcel Spatial Morphology Index (CPSMI) was developed to evaluate the interannual variations in the cropland use intensification. There was a trend of “small parcels larger” after construction, according to the high-accuracy cropland parcel boundary data (with over half of the accuracy metrics exceeding 90%). The mean parcel size increased from 0.80 to 0.84 hm². Contiguous cropland facilitated faster land transfer and then enhanced labor efficiency throughout agricultural production. In addition, the geometric shape of the parcels was more regular: the area-weighted mean shape index decreased from 1.53 to 1.51, the mean fractal dimension declined from 1.48 to 1.46. These indicators collectively confirmed that the shape regularity and simplified boundaries were improved after the cropland consolidation. Mean parcel density also decreased from 0.82 to 0.72, indicating a more concentrated spatial distribution. While the mean standard distance index remained unchanged, its dispersion was reduced for intensive agricultural production. Overall, the spatial morphology of well-facilitated farmland parcels in the study area tended toward greater mechanization and intensification. The CPSMI of well-facilitated farmland parcels also decreased from 0.378 to 0.357. Moreover, several challenges were presented in the well-facilitated farmland construction in agricultural modernization. There was an urgent need to optimize the site selection of cropland parcels and supporting facilities within the construction region. Rational planning was essential to consider the natural and production conditions in the local regions. The supporting facilities were expanded to encroach on the cropland resources, leading to a reduction in cropland. It was crucial to strike a balance between the production efficiency of cropland parcels and supporting facilities, particularly for the stable and high grain production. In conclusion, the cropland parcel assessment—using multi-source remote sensing imagery, deep learning, and landscape metrics—can provide an effective tool to track the spatial variations in cropland morphology. The valuable data support and theoretical foundations can also be offered to monitor, regulate, and optimize the well-facilitated farmland construction. The finding can greatly contribute to more scientifically grounded and data-driven decision-making on modern agriculture.