Revealing spatial patterns in geo-social data of small household farmers

Organised by Laboratory of Geo-information Science and Remote Sensing

Thu 30 August 2018 10:45 to 11:15

Venue Lumen, gebouwnummer 100
Room 1

By Thijs Oosterhuis (the Netherlands)

The majority of people living in absolute poverty are smallholder farmers, thus in order to increase food security it is important to strengthen their agricultural practices. However little is known about the local behaviour of these practices. Therefore, this thesis aims to identify patterns of local be-haviour in small household farms in Uganda. In order to identify these patterns, data collected in December 2015 from the Community Knowledge Worker initiative by the Grameen Foundation was used. The initiative was launched with the aim of spreading information to farmers in remote communities through a network of peer advisors. The initiative uses mobile technology as well as human networks to help smallholder farmers in getting information to improve their practices and livelihoods. This initiative provides a dataset consisting of 87.835 separate geotagged points of questionnaire farmer data, which may give a fairly good impres-sion of practices. This is combined with science-based geo-physical data of the area, such as a elevation and a landcover of respectively 30 and 300-meter resolution. This to address another aim to identify relationships be-tween patterns of local behaviour of SHF with currently used scientific data ( land cover in relation to elevation) to detect smallholder farmer practices. The location of the geotags are spatially clustered with the DBscan method, to include a spatial clustering value. After this step, clus-ters are found in the combined dataset using the ORclus clustering method for high dimensional data sets. They are visualized using geo-vis-ualisations and charts to explore data be able to draw conclusions from these found clusters. A validation of the method is performed by setting apart one third of the data set before performing the clustering method, and after that performing the same clustering methods on this validation set. Found clusters resembled the height and landcover class per agricul-ture crop. In-depth research is needed for more specific clusters and pos-sible information gaps. Despite several imperfections of the Community Knowledge Worker dataset, we conclude that our method give meaningful insight into the patterns of questions of the SHF and where it resembles height and landcover per crop The CKW network provides a valuable addi-tion for the collection and spread of farmer information, which can be used to spread information more efficiently and eventually strengthen agricul-tural practices of small household farmers and thereby increase food secu-rity.

Keywords: Grameen Foundation; Small Household Farmer; high di-mensional point data; DBscan; ORclus