Gully Recognition and Spatial-temporal Evolution Law of Gully Based on
SegNet Model
Abstract
Gully erosion is one of the main modes of slope erosion on the Loess
Plateau, which plays a connecting role in the slope gully erosion
system. The Loess Plateau has wide and densely distributed gullies. The
study selected a typical small watershed in the hilly and gully region
of the Loess Plateau to measure the morphological characteristics and
spatial-temporal distribution of gullies. A deep learning image semantic
segmentation model was used to identify and extract the morphological
features of gullies at the watershed scale from 2009 to 2021 based on
remote sensing images (0.5 m resolution) and then analyze their temporal
and spatial distribution characteristics. The results revealed that: (1)
most gullies occurred in the hilly southern parts of the watershed,
which has complex landforms and large slope gradients; (2) gully number
increased from 1,159 in 2009 to 2,312 in 2021 (average 97 per year),
with a frequency development rate of 2.87 km –2 y
–1; (3) gully length generally ranged from 25–40 m,
with an average growth rate is 1.66 m y –1 and
density development rate of 0.12 km km –2 y
–1; (4) gully width ranged from 0.5–1.5 m, with an
average growth rate of 0.04 m y –1. (5) the total
gully area increased from 0.0566 km² in 2009 to 0.1072 km² in 2021, with
a development rate of 4,213.39 m² y –1 and dissection
degree development rate of 0.0125% y –1. This study
provides a theoretical and scientific basis for gully erosion control
and eco-environmental protection at the watershed scale on the Loess
Plateau.