Welcome, to the 6th week of Spatial Data Science and Applications. For the last five weeks, my lectures have covered introduction to spatial data science, four disciplines related to spatial data science, five layers of GIS, spatial DBMS and big data systems, and spatial data analytics. This week, you will learn five practical applications of spatial data science. Do you remember the categories of dpatial data science problems, which you studied in the second week? Dependent on required tools and data characteristics, they are desktop GIS, server GIS, spatial web, spatial data analytics, spatial data management and analytics, spatial big data management and analytics and so on. I will present practical applications which could be categorized into one of them. The first problem is that you have a client, a financial investment group, who are interested in purchasing commercial timber land in the U.S., for the purpose of diversifying investment portfolio in commodity market. Assume that the client gave you a project to find the top 5 counties for timber land investment. This is the target area, South Eastern states of the U.S., which cover areas from state of Virginia to eastern part of Texas and Oklahoma. This area, is one of the world's major wood basins, where 15 percent of world timber production is made. For the problem, softwood and hardwood acreage per county is given, which can represent timber supply, and it can be retrieved from two different sources. One is from Forest Inventory Analysis report, also known as FIA report and the other is from National Land Cover Data from USGS. For the demand side, Timber Product Output (TPO) information from U.S. Forest Service is given. The figure is NLCD data of State of Georgia, which is basically land-use and land-cover map. It contains a variety land covers, and it has three forest related land covers. deciduous forest, evergreen forest and mixed forest. The acreage of deciduous forest and 50% of acreage of mixed forest could be an estimate of total acreage of hardwood forest. Likewise, the acreage of evergreen forest and 50% of mixed wood acreage would be total acreage of a softwood forest. Another set of data source is FIA report from USFS, in which softwood and hardwood acreage and TPO tables are provided. Now, lets think about solution for the project. Where do you think are good counties worth of timber land investment? One solution would to be to find counties in which large demand exist in comparison with the supply. So, how about the counties which has large difference between demand and supply? or how about counties which has large demand to the supply ratio? For this problem, I chose the method of the differencing demand and supply, because ratioing would exaggerate the actual demand in case of small amount of supply. For the differencing method, we have two options. The first method is to use FIA report for softwood supply and hardwood supply. Demand can be retrieved only from TPO table. The final issue is the unit problem, because supply is based on acreage and demand is based on CF, which stands for cubic foot, with the assumption that the conversion ratio is 1 acre to 2,000 cubic foot for softwood and 1 acre to 1,000 cubic foot for hardwood. Based on the flowchart, we can produce difference from supply to demand map in cubic foot for softwood and hardwood respectively. Likewise, the second method is to use National Land Cover Data, NLCD, and to extract softwood acreage and hardwood acreage by means of the methods as described before. By the way, fedaral land and conserved land should be removed from available acreage, because they cannot be harvested for timber supply. The same conversion rate is applied to acreage and the final map unit is cubic foot again. For the given problem, we have the data set and proposed solutions. The next question is what would be an appropriate solution structure to deal with the datasets and to implement the proposed solutions. The data size is rather small, multiple user access is not required, data analysis is simple and visualization of the outcome is important. For the characteristics, desktop GIS, using only GIS software, such as QGIS, would be the best choice. Now let's take a look at the outcomes of each step. For the method using NLCD, we have to extract acreage of softwood and hardwood for timber supply of each county. The first map is softwood acreage of counties in the target area. Then, federal land and conserved land has to be removed from available acreage. The figure shows the softwood acreage after the adjustment of federal and conserved forest land. The softwood acreage can be also given by FIA report. The figure shows softwood supply of each county in target area, using FIA report. The information source of the softwood demand is TPO table in FIA report. The softwood demand is mapped to each county and the figure shows the softwood demand map. Now, we apply differencing from softwood supply from FIA report, to softwood demand, and the outcome of the analysis is illustrated. In the figure, red color represents counties of large deficits of softwood supply in comparison with softwood demand. In other words, counties in red color would have more potential to sell softwood timber at a higher price. Now we have a list of top 5 counties which have the largest deficit of the supply to the demand. They are Tensas, Sabine, Conecuh, Bienville and Coffee counties. By the way, it should be noted that Tensas county has zero softwood supply, so that it is not appropriate for timber land investment, though it shows the largest deficit. The figure illustrates the same differencing between demand and supply from NLCD. This time, the top 5 counties with largest deficit of the supply to the demand are Tensas, Angelina, Sabine, McCurtain and Bienville county. There are two different counties from the previous analysis, which would occur due to different mechanisms of estimating softwood supply in FIA report and NLCD data. Now we are going to solve the problem of the hardwood case, and the same procedure as softwood case will be applied. The figure illustrates hardwood acreage estimated from NLCD for hardwood supply. As mentioned, the hardwood acreage from NLCD the should be adjusted with consideration of federal and conserved lands, which are prohibited from timber harvest. You are looking at the hardwood acreage after the adjustment. Another source of hardwood supply information is FIA report, and the figure illustrates the hardwood supply from FIA report. Hardwood demand is also retrieved from TPO tables, and you are looking at the hardwood demand map of the target area. Now we are ready to get the result. The differencing between the demand and the supply is applied to the hardwood problem, and the figure illustrates the outcome. The counties in red color, again have the largest deficit of hardwood supply in comparison to the demand, which would be good candidates for timber land investment of hardwood. The top 5 counties are Bledsoe, Warren, Greenwood, Union and Marengo counties. The analysis outcome with different hardwood supply from NLCD is illustrated in the figure. The top 5 counties from the second method are Bledsoe, Union, Warren, Greenwood and Marengo counties, which are identical to the outcomes of the first method, though the order is a little bit different. How do you think about the simple approaches to solve the problem raised by a financial investment group? What would be the limitations? First of all, the problem is too much simplified, so that we did not consider spatial distribution of forestland, timber volumes and demand location and timber quality and so on. Second, activities over the county boundaries was not assumed, however, the typical range of timber procurement is around 75 miles, and the assumption, no activity over the boundary, is far from realistic. All in all, 2 Step FCA with 75 miles radius would be a good alternative solution to the problem. In this lecture, you studied the first practical application of spatial data science. The problem was to find good counties for timber land investment. And it was implemented in the solution structure of desktop GIS, the simplest framework. The solution was somewhat simple, but it could present an insight in such a high-level business problem. This is the value of Spatial Data Science. Can you realize the power of spatial data science which can simply deal with investment problem? More to come in the next lectures. Please stay tuned, and see you next time.
