| Summary: | Frequent frost occurrences in the Kenyan highlands have had devastating effects on agricultural productivity. With inadequate management systems to mitigate the impacts, farmers have often had to bear the burden of losses resulting from frost damage. While agriculture in Kenya remains dependent on weather and climate, the agricultural economy of Kenya continues to suffer, underscoring the need for building local knowledge as basis for development of early warning systems. The current paper attempts to delineate frost zones by statistically characterizing them based on known risk factors related to topography (elevation, convexity, aspect, upslope flow length) and Land Surface Temperature (LST) derived from Moderate Resolution Imaging Spectroradiometer (MODIS). Through binary logistic regression, a logistic regression model was developed utilizing observation data (frost occurrence and non-occurrence) as a binary dependent variable to estimate the probability of frost occurrence. Assuming a 0.5 probability cut-off threshold between frost occurrence and non-occurrence, an overall accuracy of 81% with area under Receiver Operating Characteristics (ROC) Curve of 0.88 was obtained. No evidence of lack of model fit was detected. This model outperforms the currently operational model that utilizes MODIS LST alone to detect frost zones in the Kenyan tea plantations. It provides an improved method for effective delineation of frost zones by incorporating local topographic characteristics.
|
|---|