(1. Shandong Institute of Geology and Mineral Resources, Jinan 250014; 2. School of Civil Engineering, Shandong Jianzhu University, Jinan 250 10 1)
About the author: Xu Qiuxiao (1979—), female, assistant engineer, mainly engaged in remote sensing, hydrogeology and environmental geological exploration.
The development of remote sensing technology and geographic information system technology provides information sources and technical means for studying global change and sustainable development. As the most prominent landscape symbol of the earth's surface system, land use change is an important field of global change research in recent years. In this paper, Longkou City, Shandong Province is taken as the research area, and the land use/land cover classification method of Longkou City based on the comprehensive understanding model of remote sensing images is applied to establish a knowledge base of geoscience laws and extract land use types in different periods. Finally, the spatial and temporal changes of land use are monitored and analyzed by using geographic information map.
Keywords: land use; Remote sensing image; Geoscience auxiliary information; Information atlas
The development of remote sensing technology and geographic information system technology provides information sources and technical means for studying global change and sustainable development. As the most prominent landscape symbol of the earth's surface system, land use change is an important field of global change research in recent years. Only by monitoring and analyzing the temporal and spatial changes of land use, better understanding the process and mechanism of land use change, and adjusting human social and economic activities can land use be more reasonable, ensure the pertinence and effectiveness of national macro-strategic decision-making, and achieve the goal of sustainable land use. Longkou is an early coastal city in China, and its land use change is representative.
1 Overview of the study area
Longkou City is located in the northwest of Jiaodong Peninsula, with Penglai City in the east, Qixia City and Zhaoyuan City in the south and Bohai Sea in the west and north. The maximum horizontal distance from east to west is 46.08km, and the maximum vertical distance from north to south is 37.43km. The total land area (including Sang Island and Yi Island) is 893.32 square kilometers.
This study area is obtained by cutting the remote sensing image of the boundary of the study area generated by vectorization of 1∶ 10000 topographic map. As the remote sensing data information, it covers an area of ***892 17.45hm2 (excluding Sang Island and Yi Island).
2. Land use/land cover classification in Longkou City based on comprehensive understanding model of remote sensing images.
Image understanding is a subject that studies the interpretation of images through computer system, so as to realize the understanding of the external world similar to human visual system. In the image understanding system, there are two basic tasks: extracting the image structure or clues suitable for the model from the input image, and then completing the correct mapping between the image structure in the input image and the target in the model (Zhou Chenghu, 1999). Image understanding is different from pattern recognition. Pattern recognition usually simply classifies objects according to predefined measurement sets, while image understanding needs to describe and explain images, which requires interaction between entities at different processing levels (Wang Runsheng, 1994).
According to the classification characteristics of land use/land cover, a comprehensive understanding model of remote sensing images based on GIS information is constructed. The model is divided into two processes, namely, remote sensing image understanding process and geographic information system processing process. The specific contents of these two processes are as follows:
(1) The remote sensing image understanding process mainly completes the pre-understanding process of remote sensing images, including the following processes:
1) Remote sensing image preprocessing: including image format conversion, image correction and image enhancement transformation. Image format conversion completes the conversion from remote sensing image to processing format of REDAS software system. img)； Image correction completes atmospheric correction, geometric correction, radiation enhancement, matching and mosaic of remote sensing images; Remote sensing image information enhancement and transformation processing can highlight relevant thematic information. The main methods used this time are linear stretching, K-T space transformation and boundary enhancement.
2) Registration with geoscience auxiliary information: it mainly completes the coordinate and projection system transformation between remote sensing images and geoscience auxiliary information (all kinds of thematic GIS data are mainly slope and elevation), so that remote sensing images and adopted geoscience auxiliary information can be integrated into a unified coordinate and projection system.
3) Selection and calculation of "training area": Through the preliminary understanding of the information characteristics of remote sensing images, combined with geoscience auxiliary information and field investigation, the sample "training area" is determined. "Training area" should be typical and separable. After determining the "training area", calculate the data of the "training area" and determine the statistical information of the sample (mean, maximum and minimum, variance matrix, covariance matrix, etc.). ).
There are six types of land use extracted this time, namely construction land, cultivated land, water area, garden land, woodland and unused land. AOI expansion method is used to select training area. When using this method to select the training area, the threshold of initial seed pixel and spectral distance is very important, and the threshold of initial seed and spectral distance of any training area can only be determined after many experiments. These two parameters are different in different land types. Generally, after selecting 2 ~ 3 training areas, watch the alarm shielding and immediately modify and adjust these two parameters. When the training area is added, the warning mask and the previous warning mask are superimposed in time to judge the quality change of the sample data and adjust it until the warning mask is more consistent with the actual ground class, that is, this kind of training area is selected. After the first-class training, it is necessary to analyze the final training area as a whole to ensure the purity of the spectrum.
(2) geographic information system processing, with the support of GIS system, completes the processing of geoscience auxiliary information, mainly including the following processes:
1) Introduction and pretreatment of thematic information: collecting geoscience auxiliary information (including various thematic information and statistical data, etc. Through ground investigation or expert knowledge and experience, the auxiliary information is imported into GIS system to complete preprocessing, including digitalization and editing of various vector data, establishment of topological relations, collation of statistical data, geo-coding and spatialization of statistical data, etc.
2) Generation of auxiliary data: using pre-processed auxiliary data to convert various data formats, such as rasterization of vector data and interpolation of point statistical data (IDW/ Kriging and other methods) to generate area data; Finally, coordinate and projection system are unified to realize registration with remote sensing data.
3) Establishing various geoscience auxiliary factor databases: with the support of GIS software, based on the various auxiliary data generated above, a thematic database is established to form a geoscience auxiliary factor database.
(3) The generation process of knowledge base, that is, the establishment of expert knowledge base, mainly the knowledge base of geoscience laws, includes the following processes:
1) Knowledge acquisition: By comparing remote sensing images, conducting on-the-spot investigation, and according to various representative image characteristics, investigating the land use/land cover, vegetation distribution and ecological environment in all geomorphic conditions in this area, and acquiring various practical knowledge.
2) Knowledge base generation: sorting out the acquired knowledge, summarizing various laws, forming knowledge rules, and finally forming a knowledge base of geoscience laws through continuous training, modification and debugging of the data in the "training area".
Land cover/land use types are obviously dependent on altitude. According to the analysis of the existing land use map in the study area, the construction land and garden are mainly distributed below 300m above sea level; Most of the cultivated land is below the altitude of 150m; Woodlands are widely distributed at different altitudes, and the woodlands below 10m are mainly coastal shelterbelts and highway greening woodlands. Above 350m above sea level, only woodland is distributed, and there are no other land types.
Slope data can be used to distinguish some land cover/land use types. According to the results of field investigation, combined with topographic map and existing land use status map analysis, construction land, irrigation land and garden land are mainly distributed in the area with slope less than 20, and irrigation land is less when the slope is greater than 10. Therefore, the spectral characteristics of remote sensing data also show green vegetation, and slope data is a valuable parameter when it is difficult to judge whether it is irrigated land or woodland. The following are the detailed relationships between several land types and altitude and slope.
In this classification knowledge base, expert rules are expressed in the following basic forms:
IF (condition) THEN (conclusion), Confidence (conclusion credibility)
The reliability range is [0, 1]. When the value is 0, the possibility that the current pixel belongs to a given category is completely ruled out. When the value is 1, the original credibility of the pixel is maintained, and the current conclusion is always valid. Credibility can be determined by geoscience experience or expert scoring.
The representation of knowledge and the construction of knowledge base should be combined with the research characteristics of geoscience problems. By constantly modifying and debugging the knowledge base, the image interpretation results basically reach the effect of manual visual interpretation. When the value of reliability is 0, the possibility that the current pixel belongs to the given category is excluded. When the value is 1, the existing reliability value should not be changed. This representation not only considers the characteristics of remote sensing image interpretation, but also significantly reduces the number of rules in the knowledge base. This is extremely important for remote sensing data processing with a large amount of data. The following are the rules in the rule base:
If the value =1DEM is less than 300 and the slope is less than 20, the construction land is CF = 1.
If the value = 2 DEM < 50 and the slope < 10, the irrigated land CF = 1.
I value =3, then waters cf = 1
If the value = 4, DEM < 300 and slope < 20, the dry land CF = 1.
If the value = 5dem < 300 and the slope < 20, the garden cf = 1.
If the value =6, woodland cf = 1
Otherwise, if DEM > = 300 or slope > = 20, the woodland cf = 1.
Elseif50 = < DEM < = 250, 10 = < slope < 20, then garden cf = 0.5.
ELSE woodland cf = 1
The accuracy evaluation of this classification result adopts stratified random sampling method, mainly referring to the current map of land use in Longkou City, and combining with visual interpretation and field verification results, the accuracy evaluation of the two classification results is carried out. The accuracy of remote sensing classification results in each period was verified by directly consulting the current map of land use in that year. Because there are some differences between remote sensing classification system and land use status classification system, land use type classification and land use status classification of remote sensing classification images should be unified properly before random sampling. Then, 300 sample points are selected from the classification results of remote sensing in each period, and more than 10 sample points are guaranteed in each category. The classification results of Longkou City in 2003 and 1989 were evaluated by the accuracy evaluation method based on error matrix. The actual Kappa coefficient calculation shows that the overall accuracy and user accuracy of land use TM classification results in Longkou 1989 and 2003 are above 75%, and the Kappa coefficient is above 0.8, which meets the requirement of minimum allowable discrimination accuracy of 0.7. These indicate that the results of remote sensing classification of land use in Longkou city are ideal, and the classification accuracy of each category is also high.
3 Longkou City, the spatial and temporal changes of land use analysis
With the rise and development of geo-information science, the resources available to people are extremely rich, and the information processing technology has also been greatly improved, especially the dynamic visualization technology has made a new breakthrough. Under this demand and technical background, Mr. Chen advocates the discussion and research of geoscience information atlas on the basis of traditional geoscience atlas. Geographic information atlas is a natural extension of geographic information atlas, which is a group of digital maps, charts, curves or images that can reflect the laws of geospatial information according to certain index gradual change laws or classification systems. Geographic information atlas is the combination of "graph" and "spectrum", which has the dual characteristics of graph and pedigree.
The analysis model of land use map adopted in this paper includes two parts: ① Transfer matrix, from which we can see the main types of land use changes in each time series unit and the supply sources of various categories. ② Analyze the land use maps of different time series units, and investigate the spatial combination and spatial-temporal displacement of the map units.
3. 1 transfer matrix
Transfer matrix plays an important role in analyzing the flow direction between land use types, which can not only quantitatively explain the mutual transformation between land use types, but also reveal the transfer probability between different landscape types, so as to better understand the temporal and spatial evolution process of land use. Transfer matrix includes transfer area matrix and probability matrix.
Table 1 1 989 ~ 2003 land use transfer matrix (unit: hm2)
Note: R is the transfer ratio of various land use types (%).
As can be seen from the table 1, from 1989 to 2003, the transfer area of construction land was 487.6 1hm2, accounting for 3.62% of the initial construction land area, and there was no obvious flow direction. The cultivated land use transfer area is 17632.40hm2, accounting for 49.85% of the initial cultivated land area. The main flow direction is garden land, accounting for 13835.82hm2, accounting for 39. 1 1% of the initial cultivated land area, followed by construction land, accounting for 3508.77hm2. The flowing area of water area is 492.92hm2, accounting for 8.72% of the initial water area, mainly flowing to gardens and woodlands, and * * * accounts for 6.4% of the initial water area. The flowing water area of the garden is 13 16.62hm2, accounting for 7.89% of the original garden area. The main flow direction is construction land, which is 759.25hm2, accounting for 4.55% of the initial garden area. The circulation area of forest land use is 1405.45hm2, accounting for 10.45% of the initial forest land area, and the main flow direction is construction land, accounting for 4.66% of the initial forest land area. The leased area of unused land is 2 158.74hm2, accounting for 47.24% of the original unused land area. The main flow direction is garden, which is 1306.85hm2, accounting for 28.60% of the initial unused land area, followed by woodland and construction land, and the ratio of * * to the initial unused land area is 65438+.
3.2 Land use information map
In the atlas, * * * has 36 types of atlas units, that is, land use change types, of which 30 types show that land use types have changed, accounting for 26.34% of the total area of the study area. In order to read the main direction of land use type transfer more simply and clearly, and understand the main characteristics of land use change, 30 types of change map units are sorted according to the area size, and the conversion area percentage and cumulative conversion percentage of each map unit are calculated, and the 10 map unit which accounts for 92.04% of the total change area is counted, and the area sorting table of main land use map unit types from 1989 to 2003 is obtained.
1989-2003 Sorting Table of Land Use Types and Areas of Main Map Units
As can be seen from Table 2, the most significant structural feature of land use change in Longkou City from 1989 to 2003 is the transformation from cultivated land to garden land, and the cultivated land area in this direction is *** 13835.82hm2, accounting for 58.89% of the total change area. Secondly, cultivated land was converted into construction land, with an area of 3508.77hm2, accounting for 14.93% of the total change area. The third is the transformation from unused land to garden, with an area of 1306.85 hm2, accounting for 5.56% of the total change area. It can be seen that the land use types in Longkou mainly flow to building land and garden land during the study period.
4 conclusion
(1) This study used Landsat data ETM++ to monitor and analyze the land use in Longkou city in time and space, and achieved good results. However, the resolution of ETM ++ image is low, which mainly reflects some comprehensive land use information. For some areas with broken land use maps and complex land use types, it is difficult to extract features, so there is a lack of detailed monitoring of land use changes. In the future research, we should actively explore new dynamic monitoring methods combined with high-resolution satellite images, and make full use of these high-resolution remote sensing data to obtain more reliable and accurate regional land use change information.
(2) How to make full use of the rich geographical auxiliary data provided by GIS spatial database, and then automatically discover knowledge, integrate multi-scale and multi-temporal high-resolution remote sensing data, and establish a flexible and efficient reasoning mechanism to complete the automatic extraction of thematic information from remote sensing images is the further research direction.
(3) "Geographic Information Atlas" is a new academic thought, which is still in the stage of understanding, and the understanding is not very mature. More scholars and more research work are needed to improve our understanding of it. This paper holds that it is an advanced and feasible technical way to study the information map of land use evolution and development by using remote sensing technology, invert the temporal and spatial changes, and then understand the objective world, reveal and reproduce the past.
(4) The results show that during the period of 1998 ~ 2003, the land use pattern in Longkou City changed frequently, with great intensity and prominent contradictions. Mainly manifested in the continuous decrease of cultivated land and the continuous growth of garden and building land; Its main flow direction is from cultivated land to garden and building land. These changes lead to the decline of cultivated land quality and the enhancement of utilization rate in Longkou City, which brings great pressure to cultivated land protection. Therefore, the land management department should pay attention to macro-decision, handle the relationship between social and economic development and land reserve resources, and implement sustainable development.
refer to
Chen. Exploration of 200 1. Geographic information map. Commercial press
, Peng,,. 2002. Study on information map of land use evolution-A case study of Shunyi County, Beijing. Geographic Information Science, 2: (58 ~ 63).
Qi Qingwen, Chi Tianhe.2001. Theory and method of geographic information atlas. Journal of Geography, 56 (Supplement): 8 ~ 18.
Shu, Mao Zanyou. 1997. Research on knowledge-based GIS-aided remote sensing image classification method-taking land cover land use types as an example. Journal of Surveying and Mapping, 1 1(4)
Ye Qinghua. 200 1. Study on spatio-temporal composite pattern of land use/land cover change in the Yellow River Delta. Doctoral thesis of Institute of Geographical Sciences and Resources, Chinese Academy of Sciences
, Luo, Yang Cunjian, Liu Qingsheng. Geographical understanding and analysis of remote sensing images. Science press, 180 ~ 240.
Lucas I F J, frans J M, Wel V d 1994. Accuracy evaluation of land cover data derived from satellite: a review. Photogrammetry Engineering and Remote Sensing, 60(4):4 10~432