1. Introduction
Gully erosion refers to the vertical erosion, lateral erosion and retrogressive erosion of gully flow, resulting in the formation of gullies [
1]. The phenomenon is most obvious in areas with heavy rainfall, slopes, sparse vegetation and thick loose material coverage. Despite being a natural process, excessive human activities such as long-term cultivation and overgrazing are the primary inducing factors of gully erosion [
2,
3]. Several studies proposed that gully erosion is the main soil erosion type in the black soil region of Northeast China, causing serious soil erosion that erodes cultivated land, destroys roads and causes great soil and water loss [
4,
5]. Gully mapping is very useful for land use planning, conservation practices and soil erosion mitigation [
6,
7,
8].
With the development of remote sensing techniques, various remote sensing data have become cheap and readily available. Compared with field investigation using tools such as micro-topographic profile meters [
9], total stations [
10], erosion pins [
11] or global navigation satellite systems [
12] and aerial photo interpretation by photogrammetric techniques [
13,
14], mapping gullies using satellite imagery interpretation has been considered to be a preferable method with the characteristics of lower labor and time consumption, availability to larger regions of interest and dynamic research. Thus, satellite imagery interpretation has been widely used [
15,
16,
17].
Visual image interpretation, as the earliest way of mapping gullies by satellite imagery interpretation [
18], has been applied for years due to its high precision and stability [
8,
19]. However, inefficiency and the requirement for background knowledge of the interpreter limit its application. In view of these flaws, automatic gully extraction methods have been developed [
20,
21,
22], including pixel-based and object-oriented image analysis approaches. Compared with pixel-based extraction, object-oriented extraction takes the spatial and geometric information of an object into consideration, which typically makes it outperform pixel-based extraction [
23,
24]. However, object segmentation is not just a key to object-oriented image analysis—it is a very difficult process [
25,
26]. For any gully extraction method involving satellite imagery interpretation, selecting a suitable spatial resolution of remote sensing data is an initial and significant step [
27]. Thus, Luoman et al. [
28] compared the results of several remote sensing images with different resolutions when extracting gullies by visual interpretation. They reported that remote sensing images with a higher spatial resolution did not always provide the best information as they contained more noise. Some scholars have studied the selection of a suitable pixel size when using automatic gully extraction methods. Using five datasets with spatial resolutions ranging from 2 to 30 m, Younes et al. [
29] explored the effects of different spatial resolutions on some controlling factors extracted from remote sensing data when drawing a gully susceptibility map based on pixel-based image analysis and concluded that 10 m was the optimal spatial resolution. Shruthi et al. [
30] showed that the rule set developed for gully extraction based on object-oriented image analysis is sensitive to the spatial resolution of remote sensing images. However, these studies did not consider the impact of resolution on gully extraction under different classification systems based on different application purposes.
As one of the three major black soil regions in the world, the black soil region of Northeast China has been a major grain production area of China, producing rice, corn and soybeans. With topographical characteristics of rolling hills and some poor management practices, it has been suffering from serious soil erosion, threatening food security and agricultural sustainability [
31], of which gully erosion is one of the main parts. According to the first national water conservancy census, gully erosion is widely distributed and mainly appears in sloping farmland [
32]. Furthermore, gully erosion has been intensifying, as the number of gullies is increasing and the channels have been expanding in recent years [
3].
The present study aimed to analyze the suitability of remote sensing data at different resolutions for different application purposes on the small watershed scale of the black soil region of Northeast China. This objective is addressed by (1) classifying the types of gullies according to application purposes based on field investigation, (2) establishing methods of visual interpretation and extracting gullies, (3) analyzing the suitability of different resolution data and selecting the optimal resolution based on performance analyses and (4) mapping gullies with the optimal pixel size.
5. Discussion
A high spatial resolution was helpful to interpret gullies, but an excessively high resolution increased noise. Although the shadow caused by gully erosion, which could help determine the boundaries of gullies, was clear in the high-resolution images [
8], the shadow of vegetation growing in gullies was too clear, which covered up the outline of the gullies and made it difficult to draw them [
28]. Therefore, the interpretation accuracy of the image with a spatial resolution of 1.02 m was the highest when not considering the type of gully.
When choosing the spatial resolution of the image, one should consider not only the gully interpretation accuracy but also the price of the image and the application purpose of the research. Generally speaking, the higher the spatial resolution of an optical image is, the higher the price is. Taking the company Beijing Lanyu Fangyuan Information Technology Co., Ltd. [
37], which specializes in providing satellite images and other image processing services in China, as an example, the prices of some satellite images widely used and currently in service are shown in
Table 6. It should be noted that the interpretation accuracy of gullies became significantly worse in general when the spatial resolution was 4.08 m or lower, so only the prices of images with a spatial resolution of 0.5–2.1 m are listed. Based on the performance of gully interpretation and the prices of the images, images with a resolution of 2.04 m are the best choice when not considering the types of gullies interpreted, which could meet the general requirements of relevant professionals. In fact, many studies have used SPOT images with a spatial resolution of 2.5 m to interpret gullies [
28,
38]. When finer monitoring in a large area was carried out, images with a resolution of 1.02 m could be selected. When studying soil loss caused by gullies at different development stages, the types of gullies that need to be interpreted are ephemeral gully, permanent gully and modern incised valley. To extract the information of these three types of gullies completely, images with a spatial resolution of 0.51 m are the only choice. However, research on ephemeral gullies has little practical significance, as they have narrow width and shallow depth, do not hinder general tillage—or sometimes disappear due to tillage—and often appear repeatedly at the same position. In this situation, images with a spatial resolution of 1.02 and 2.04 m could be considered, and images with a spatial resolution of 2.04 m are better in that they are cheaper and still meet the requirements. Pullman showed that ZY-3 satellite images with a spatial resolution of 2 m were an ideal choice for interpreting permanent gullies and modern incised gullies [
28]. When studying the activity of gullies to formulate relevant governance policies, the types of gullies that need to be interpreted are active gullies and stable gullies. Images with a spatial resolution of 0.51 and 1.02 m can both be used, but considering the practical significance, price factors and interpretation performance, images with a spatial resolution of 1.02 m are undoubtedly better. When it is necessary to monitor the treatment of gullies, gullies first need to be identified as untreated gullies or treated gullies. Similar to the selection of the spatial resolution of images used to interpret active gullies and stable gullies, images with a spatial resolution of 0.51 and 1.02 m could be listed as candidates, but the images with a spatial resolution of 1.02 m are more suitable.
It should be noted that the abundance of auxiliary data might reduce the requirement for spatial resolution. Auxiliary data include high-precision DEMs, fine road network data and vegetation fractions. Studies showed that SPOT images with a spatial resolution of 2.5 m and NDVI data could be used to interpret active gullies, semiactive gullies and stable gullies, and the result was ideal [
39]. Moreover, the spectral and temporal resolutions of images are important factors affecting the characteristics of surface features in remote sensing images, besides spatial resolution [
40,
41]. The improvement of spectral resolution could provide a more precise surface spectrum, which could be used as auxiliary data to better interpret gullies. With the improvement of temporal resolution, more dynamic information could be obtained. If images before and after a rainstorm could be obtained, the rainwater accumulation shown in the image after the rainstorm could be used as auxiliary information to interpret gullies. The interpretation of gullies was only studied from the perspective of spatial resolution with limited auxiliary data in this paper. Future research needs to take spectral resolution, temporal resolution and enrichment of auxiliary data into account.
GE images, a type of optical image, were the only satellite image data used for mapping gullies in this study. Some other studies have used GE images to map gullies and confirmed the availability of GE images [
42,
43,
44]. Compared with other high-resolution optical images, GE images can be used to evaluate gullies in a very large area without any cost of image acquisition. However, the dates of GE images are often not homogeneous across a region provided for some platforms, which remains a limitation of the application of GE images [
45]. In addition, gully mapping was restricted to agricultural land, as forest and other surface cover precluded interpretation of gullies beneath [
46].
Comparing the GE images of the two time phases used in the study, it was found that there were almost no changes in the types and distribution of ground objects, which shows that the states of the gullies in the study area were basically the same at the two time points. Therefore, the lag of verification data acquisition had little effect on the results in this study. However, studies have shown that gullies can be greatly developed in a short time if conditions are favorable due to rainstorms and freeze–thaw erosion [
47,
48]. Therefore, the acquisition time of validation data should be as close as possible to the acquisition time of images used to interpret gullies, and it is better not to use an interval including a rainy season or a freeze–thaw period when relevant research of gullies is carried out.
This study was carried out in a typical black soil region of Northeast China. The method and results could be extended to other black soil areas in Northeast China for the following reasons: (1) there were similarly distributed rules for gully erosion across the entire black soil region, and (2) the study area had all types of gullies in the black soil area due to a long history of gully development and management. However, some adjustments need to be made to this method to obtain results suitable for local conditions when attempting to extend the present approach to areas other than the black soil region. For example, it is necessary to reselect the time phase of images and reclassify the types of gullies according to field investigation.
In response to the problem of gullies swallowing cultivated land, land managers are looking for effective land management measures to control gully erosion, such as the terracing of hillslopes [
49]. Land use planning and land degradation are interrelated and interact with each other [
50]. Further research is needed to clarify the relationship between gully development and land use planning, which can provide suggestions for gully control. Indicators are often selected and calculated with remote sensing and GIS techniques to reflect the relationship between humans and land [
51]. Based on this study, the spatial and temporal distribution of different types of gullies could be obtained by interpreting remote sensing images with optimal spatial resolution. Then, the indicators of gully severity such as gully intensity and density can be used to quantitatively study the relationship between land use planning and gully development.
6. Conclusions
The main purpose of this study was to reveal the applicability of images of different spatial resolutions for the interpretation of different types of gullies so as to select the optimal spatial resolution under different application requirements. The study draws the following conclusions: Firstly, improvement of the spatial resolution will make the contour and internal details of gullies clearer. However, at the same time, it will increase the noise information. Based on this, the interpretation performance of images with a spatial resolution of 1.02 m is the most satisfactory, without considering the type of gully. Additionally, the interpretation result of the image with a spatial resolution of 2.04 m remained excellent, with all F-scores above 95%. Therefore, images with a spatial resolution of 2.04 m are most suitable for general research, considering acquisition cost. When the spatial resolution falls to 16.32 m, gullies cannot be interpreted completely. Secondly, images with different spatial resolutions have different interpretation abilities for different types of gullies. The interpretation of all types of gullies from images with a spatial resolution of 0.51 m is outstanding. Different from images with a 0.51 m spatial resolution, it is difficult to identify ephemeral gullies from images with a spatial resolution of 1.02 m. However, ephemeral gully research has little practical significance due to minor hazards and temporary occurrences. Therefore, images with a spatial resolution of 1.02 m are the most universally appropriate choice. Thirdly, when the spatial resolution is 2.04 m or lower, the ability to interpret some other types of gullies in addition to ephemeral gullies is weakened. Therefore, the spatial resolution should be selected according to the types of gullies needed in practical application. When the spatial resolution of an image is 8.16 m, it is impossible to identify any types of gullies, and it is also very difficult to identify gullies at all.
There are some problems in this study that require further research. Besides the spatial resolution, the spectral resolution and temporal resolution also have an impact on the interpretation of gullies. In addition, the selection and enrichment of auxiliary data is another factor that changes the interpretation performance. Furthermore, different sections of the same gully may belong to different types. In this study, a simple classification was carried out, but a finer classification method is needed.