Evaluation of spatial prediction methods for selected soil properties to support land suitability mapping for rice in Zambia.

Thumbnail Image
Makungwe, Mirriam
Journal Title
Journal ISSN
Volume Title
The University of Zambia
Rice is one of the staple food crops and is a profitable smallholder cash crop in Zambia. It has the potential to contribute to increased incomes and employment among rural producers. However, rice is the only staple crop for which domestic production does not meet domestic demand due to low productivity, among other factors. One step towards addressing this problem is the identification of land with greatest potential for production. This can be done through a land suitability evaluation. This study focuses on how to map the spatial variation of selected soil properties across Zambia to support evaluation of land suitability for rice production. When mapping the spatial variation of selected soil properties, legacy data on the target variable were available along with additional environmental covariates as predictor variables. The options were to undertake spatial prediction by geostatistical or machine learning methods. Also addressed was how to robustly validate models from legacy data when these have, as is often the case, a strongly clustered spatial distribution. The validation statistics results showed that the empirical best linear unbiased predictor (EBLUP) with the only fixed effect a constant mean performed better than the other methods used for predicting soil pH and the EBLUP with fixed effect performed better than other methods used for predicting soil organic carbon and soil Phosphorus. Random forests had the largest model-based estimates of the expected squared errors in all predictions. It was observed that the random forest algorithm was prone to select as “important” spatially correlated simulated random variables. The maps produced using the best performing methods were used as factors in land suitability assessment of paddy and upland rice under rainfed and irrigation conditions. Land suitability was evaluated while accounting for important multiple factors, and which considers their joint effect of a hierarchical model of constraints. The suitability classes were ranked according to the FAO land suitability classification. Four land suitability maps where produced. Results showed that potential land classified as highly and moderately suitable was 27 percent for rainfed paddy rice, 29 percent for rainfed upland rice, 25 percent for irrigated paddy rice and 54 percent for irrigated upland rice. The results show limited potential for production of rainfed paddy, rainfed upland as well as irrigated paddy rice production but great potential for irrigated upland rice production. Therefore, irrigated upland rice production in Zambia would help expand the potential production area of rice.
Land suitability--Rice production. , Rice--Production.