Hyperspectral remote sensing for cropland assessment and modeling for agro-ecological zones: A case study of Taita Hills, Kenya

Abstract

This dissertation seeks mainly to provide an assessment of cropland characteristics in the Taita Hills. The objectives of this research are: Identification of crops using hyperspectral imagery, delineation of agro-ecological zones, modeling the impacts of climate change on agro-ecological zones and mapping suitable agricultural land in Taita Hills. The study area is composed of low to high variation in altitude within about 22km stretch. The main economic activity is mixed farming. Research in land use and land cover in the area have been done using remote sensed datasets. The data used in this research included a high resolution hyperspectral image which covered a transect area of 22 km long and 2 km wide, running from the low to high zones of the study area. It constituted 64 bands with spatial resolution of 0.6 meters. In addition, climate data of 1960-2010 (current climate), predicted climate of 2050 (future climate), soils of Taita Hills, slope, digital elevation model (DEM) and land use / land cover maps. In order to achieve the set objectives, several methodologies were applied. Initially, the hyperspectral imagery in the low, mid and high zones were classified. Training areas of crops were generated from 25 selected plots digitized from an aerial image (Nikon D3X) taken simultaneously with the hyperspectral image. Spectral angle mapper (SAM) and spectral information divergence (SID) were used as the pixel-based algorithms for classification. A combination of the Principal Component Analysis (PCA) that generated the Eigen vectors (best data) and a multivariate geostatistical clustering techniques were used to delineate the agro-ecological zones (AEZ) for Taita Hills. A difference between the future and the current zones delineated was also done. Finally, the assessment of suitable cropland areas incorporated the development of elevation models, watersheds and soil erosion mapping that applied the revised universal soil loss empirical model (RUSLE) and multi-criteria evaluation analysis. The analysis was done using the sum weighted overlay of soil erodibility, slopes, rainfall availability and land cover in the modeling. The results achieved for the research were: Classified hyperspectral imagery showing crops with an overall mean accuracy of 81% and a kappa index of 0.8. The main crops mapped were: Maize, mangoes, bananas, avocadoes and sugarcane. Agro-ecological zones (AEZ) for 1960-2010 (current) and AEZ for 2050 (future) were modeled. The difference of the two models gave an indication of the potential impacts of climate on AEZ. In the assessment of suitable cropland areas, four categories were mapped out: most suitable, more suitable, less suitable and least suitable. In conclusion, various crops can be identified using hyperspectral imagery. It is evident that agricultural activities are more intensive in the mid zones than in both low zones (warm) and high zones (cold) of Taita Hills. Furthermore, if current climate trends persist, some agro-ecological zones will increase and others will decrease to about 1 kilometer wide from the original size of a single AEZ, hampering the agricultural activities in the future (2050). In assessing suitable cropland areas, soil erosion is seen as a potential impact especially on the steep slopes rendering the land not suitable for agriculture. Moreover, agricultural activities are projected to do well in the lower zones if farmers will irrigate crops. Farmers need to be aware of the best farming practices to adapt to changing climate variations. The county government through the Kenya Forest Service (KFS), Kenya Wildlife Service (KWS) and other stakeholders need to embrace the laws governing forest areas, protected areas and the catchment areas in order to reduce soil degradation and loss of the ecosystems and biodiversity in the region.

Keywords: Hyperspectral data, spectral angle mapper; climate variation; agro-ecological zones; principal component analysis; soil erosion and revised universal soil loss empirical model.