Characterization of Sorghum and Green Gram for Data Estimation Using Earth Observation in Machakos and Tharaka-Nithi Counties, Kenya
Loading...
Date
2024-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Kenyatta University
Abstract
Data management in agriculture is very vital for the planning of any country,
including Kenya. A study was conducted to establish the role that remote sensing
data can play in the assessment of spatial variability in crop data and the provision
of crop farm information. The main objective of the study was to develop a remote
sensing-based crop data information system for estimating green gram and sorghum
data, respectively, under farm field conditions. The study was carried out in Ikombe Katanga area of Machakos County for the green gram crop and in Tharaka Nithi area
for the sorghum crop. Estimating crop data was based on three approaches: crop area
estimation, spectral signature library development for sorghum and green gram, and
assessment of microclimates within agroecological zones. The following parameters
were estimated and used to calculate the most important crop data for green gram
and sorghum for the October, November, and December rains. The parameters
selected for the development of this estimation model were vegetation indices,
biomass, leaf area index (LAI), enhanced vegetation index (EVI), soil moisture
index, soil organic carbon, rainfall, land surface temperature, soil pH, soil nutrients
(NPK), and evapotranspiration. The first crop data estimation involved crop area,
determined by assessing sorghum and green gram cropping pattern data. The
identified cropping pattern for the green gram study area and crop area estimates
were mixed crop (maize and green gram), mixed crop (maize and beans), green gram,
mixed crop (maize and pigeon pea), mixed crop (maize and cowpea), and maize with
2445.93 ha, 10,034 ha, 5981 ha, 4697.82 ha, 3743.82 ha, and 586.35 ha, respectively.
The sorghum crop patterns were mixed crop (sorghum and beans), mixed crop
(sorghum and cowpea), and sorghum, with crop area estimates of 1988.46 ha, 961.65
ha, and 469.62 ha, respectively. Furthermore, the development of spectral signature
libraries was done, and the spectral reflectance ranges for sorghum and green gram
were 0.230064 to 0.321126 and 0.26900 to 0.07466 across all bands, respectively.
The results further revealed the existence of microclimates within agroecological
zones IV and V and four microclimatic zones within the agroecological zones lower
midland IV and V of Tharaka Nithi and Machakos counties. Using data for October,
November, and December (OND) rains and cropping season data, it was possible to
estimate crop yield. Twelve parameters were analyzed and ran through a random
forest machine learning algorithm to generate sorghum yield estimates. The
validation of the model was carried out using root mean square error (RMSE) and
root mean absolute error (RMAE), with results showing that RMSE was 4.036 with
R2 of 0.98 and RMAE was 3.022 for the green gram crop, while for sorghum, RMSE
was 6.51 with R2 of 0.99 and RMAE of 5.5. The yield estimates were 4.5 bags/acre
for green gram and 9 bags/acre for sorghum, respectively. The data estimation under
farm field conditions was sufficiently optimized using the farm crop data estimation
(FCropDesti) tool developed from this work using ArcGIS software. The study also
confirms that employed methodologies were important in creating a homogenous
environment on farms for the identification of cropping patterns, which further
determine crop data such as crop area, crop type, and crop yield. Crop data estimation
is important for policymakers at the county and country level to implement climate smart agriculture. The research can be replicated in other important crops in the
country.
Description
A Thesis Submitted in Fulfillment of the Requirements for the Award of the Degree of Doctor of Philosophy in Agronomy, School of Agriculture and Environment Sciences, February 2024.
Supervisors
1. Joseph Gweyi Onyango
2. Shadrack Ngene