Integrate water constraints in optimized land spatial allocation: Part Two: what’s the LUCC pattern using hotspot, Hexagon aggregation and space-time cube skills
Land, vital but fragile, and diversified developed
As Sir William Petty noticed hundreds of years ago, that hands were the father as lands were the mother of wealth, agricultural development integrated labour and land wisely for countries. The land, covering 29% of the Earth’s surface, is so vital for human beings, but also with features that it can easily be cultivated, converted and transformed (i.e. LUCC, which means land use and cover change) towards diversified directions by different fields, in which urbanization has been competing with the agricultural industry for the past decades.
In this period, China has been boosting agricultural production for poverty-fighting, which heavily rely on high land reclamation rate and production quality. But the gradual acceleration of urbanization accompanied by some policies such as returning farmland to forests, farmland has shown a declining trend, affecting the sustainability of agricultural production and ecosystem. Many agriculture-dominant countries and regions share the same situation.
For land analysis and planning project, my team focused on a clear solution to handle interrelated land and water for a city in North China. After we set the strategy, the priority was to analyze the status-quo of the development and utilization of farmland and water resources for agriculture. The following questions needed to be answered: (1) In which pattern did farmland change historically? (2) Did any criteria affect the changing farmland? (3) Did changing have a negative effect on food production?
Data for ArcGIS desktop
Land use and cover change need complex raster and vector manipulation. With supports from Institute of Geographical Sciences and Natural Resources Research in China, we got land use data of research area in 1992 and 2000 after the interpretation of remote sensing images, using coverage (Arc/info) format with scale 1:100,000 and Krasovsky geographic coordinate system. We used ArcGIS Map 9.X Clip tool for spatial calculation and Eviews for impact analysis.
Classic but old-fashioned presenting
LUCC is a traditional field of GIS application, mostly focusing on the historical data and its comparison with recent land use. At that time GIS applications for land use showed great help for visualization of where and how changes happened and were so cool for officials who were in charge of land use management. Here was what we did.
(1) Applied the analytical tool Union to integrate the two comparable coverage data files. Be sure to select the optional choice of JointAttributes for all. Then we converted .shp file to raster file for next round of analysis. The figure below is a comparison of farmland and build-up areas in case study area.
(2) Add new field of attribute table after union, and calculated using Python to check changing. We defined isSame function for the newly-added field ChangingType.
ChangingType = field.isSame( !Class! , !Class_1! )
When attributes did not change, defined the new field to be 1, otherwise 0.
def isSame(x, y):
fieldA = str(x)
fieldB = str(y)
if fieldA == fieldB:
return 1
else:
return 0
(3) Analyzed changing pattern using Excel to show which was the most increased or decreased land use in the period. We can see 3.62% farmland transferred to build-up areas in our case study area. Then we also analyzed the impact indicator using Eview software and statistical data from yearbooks, which does not include in this article. The changing maps and
Morden and vivid show with today’s geospatial skills
Maps are excellent storytellers, especially comparison of historical maps with recent ones. In the past, local officials paid more attention to numbers especially the baseline for farmland and potential land for build-up area in the future. Land use maps, usually presented by ArcGIS, SuperMap, MapGIS, AutoCAD, Photoshop and others, were focusing on development strategy and land use allocation other than smart management, and not one of the central jobs, thus didn’t tell any deeper “stories” with less smart & precise management intention.
So what can maps and geospatial skills do as the techniques have been evolving for decades?
After years of evolving, GIS analysing and presentation skills are growing greatly, from traditional and static data to stream and real-time database. With great help from ArcGIS Pro on which I have been closely concentrating for months, some ideas for renovating my former project occur to me: (1)to apply advanced statistical tool, e.g. hotspot, grid aggregation for more trends behind the land use changing, (2) to explore space-time cube with nearly 20 years of land use image in ArcGIS Pro Living Atlas, (3)and to apply 3D presentation to better show allocation of where changing happened.
(1) Data preparation. I created a personal geodatabase, transferred two .shp files of land use cover of 1995 and 2000 to raster files (called Raster_A and Raster_B) with cell size 200m*200m, then combined and selected only where farmland decreased and build-up area increased from attribute table (called Raster_AB). This was the main changing pattern for the two periods. So, I omitted other non-significant changing patterns. After that, I converted Raster_AB to points file (called RasterToPts_AB) for hotspot analysis and defined attribute table with a new field “grid_code” showing whether symbolized farmland increased or decreased. All data and analysis were made by ArcGIS Pro 2.5. We can find the central city had the most changing for farmland. But where are the concentrating features?
(2) Getis-Ord Gi* Hotspot analysis. From mapping clusters tools of spatial statistics toolbox, hotspot analysis is focusing on comparing within the context of neighbouring features, in order to show where there are higher values jumping out of surrounded high pixels, which can be called statistically significant hot spots. I used this toolbox for illustrating where the clusters were.
After selected RasterToPts_AB as input feature class and “grid_code” as the input field in Hot spot analysis (Getis-Ord Gi*), took Fixed distance band and Euclidean distance method as default and enabled z-valued, we could find some deep statistical information. In the figure below, red spots represent Gi-Bin results with 95% as the hot spots for loss of farmland and green ones for those with 95% as the cold spots. We can find not only the central city but also the capital city for each county and in the hilly valley (west), had statistical clusters for farmland losses. However, in the southwestern part (hilly) and some area of the east part (waterways), there was conversion from other types of land use to farmland.
But what impacts can these clusters have?
(3) Grid aggregation within the hexagon shapefile. Hexagon of 1.0 square km generated summary blocks for spatial features of farmland changing in case study area. Numbers of the cluster (points) could be counted within each hexagon, using “summary within” tool. After customized symbolization, we can find the farmland changing varied greatly: 10 hexagons with more than 80% of farmland had changed to build-up area, mostly in northern and eastern direction. This was impacts of total changes from 1995 to 2000. But how can we learn changing from each gap of one year to another?
(4) Space-time cube analysis. Nowadays in ArcGIS Pro Portal living atlas, there are a lot of images and datasets for GISers. Finally, I managed to find imagery layer “Global Land Cover 1992–2018” by ESRI which had time attribute for analysis. And I checked its feasibility for my study area. That was perfect what I need most to show something deep from space-time aspects. I decided to explore the impacts, i.e. the changing ratio within the unit area, of land cover change from 2000 to 2017, on which year I found there were fewer differences in changes between 2017 and 2018. I set analysis strategies as follows:
Step_1: Created mosaic dataset with 17-year raster data after grid aggregation. After clipped to fit the region boundary, saved images for each year, combined images between the neighbouring two years, selected only pixels by adding a new field “ChangingTypes” for signals of farmland to build-up areas in attribute manipulation (Raster_C), converted to point data (RstToPts_C) and summed up within Hexagon of 10 square km (Hex_RstToPts_C), I had 17 polygon data for mosaic dataset. But unfortunately, creating mosaic dataset only required raster data. So I had to convert again for raster files (Rst_Hex_C) with cell size 2000m.
Then added to mosaic dataset. The default property for this didn’t have time attribute, I had to add time essential attribute including variable, dimensions and StdTime, and build multidimensional info to tell the dataset which could be the comprising field and the time gaps. Later I made multidimensional raster layer. Please be noticed that the red spots represent the higher number of changing area from farmland to build-up within 10 square km from 2000 to 2017.
Step_2: Created and visualize space-time cube. This is a great tool for geospatial features changing according to different time gaps. I created the cube with extended multidimensional raster layer above-mentioned and added to the 3D scene. We can find the dark blue spots are showing recent land use changes. But how about the 17 years?
Step_3: Made emerging hot spot analysis. I selected K nearest neighbours method and 8 spatial neighbours with 1 time step as the parameters. Results showed there were fewer cold spots, but new hot spots in the northwest county showing currently a great amount of farmland converted to urban areas, much more significant than the central city area where sporadic hot spots dominated.
The space-time changing patterns are much clearer in modern skills and vivid presentation, compared to classic methods, although there is still a lot to do such as the mechanism why greater changing happened somewhere and how to make proper policy to minimize the impact of farmland and /or ecosystem value loss.
I will continue according to the mainstream, on how agricultural water resources interrelate with farmland, and how to allocate the land resource for optimization with less water and stable economic profits, and prepare some posts in the near future.
Of course, some issues including information loss in the various conversion of data format, the balance of cell size and Hexagon region, etc., will occur. So all kinds of comments are welcomed for upgrading our views on data analysis, cartography, visualization and so on.
Thanks for reading.
