Common beans (Phaseolus vulgaris L.) are a major subsistence crop in sub-Saharan Africa. Due to their capacity to fix atmospheric nitrogen, they are a very important source of dietary protein for the population and contribute to soil fertility. Nonetheless, common beans are known to have a low productivity and a high variability in yields. A better understanding of what causes this variability is necessary to adopt the right practices that will enhance the productivity of common beans. With this aim in mind, we analysed bush bean (a type of common bean) yields and responses to yield-improving treatments from on-farm try-outs that were led in 2016 and 2017 in two districts of Northern Tanzania as part of the N2Africa project. Detailed information about the trials was contained in a dataset, to which we added soil information during a soil sampling campaign that took place from May to July 2018. Mid-infrared diffuse reflectance spectroscopy was used to analyse the soil samples. It is a relatively recent technique that is faster and cheaper than conventional wet chemistry analyses. It allows large-scale analysis of soils but is still controversial as its accuracy is not always optimal. We aimed to assess the accuracy of soil properties predicted from soil diffuse reflectance spectra, comparing them with soil properties obtained from conventional wet chemistry analyses. The possibility of predicting bush bean yields and responses from soil information was evaluated. Moreover, we assessed the explanatory and predictive capacity of several other variables to find out which were the most important for bush bean yields and responses. This also allowed to get an insight into the importance of soil properties relative to other variables. In addition, a household typology was constructed using multivariate statistical techniques and the available information about household characteristics. The influence of the typology on yields, responses and soil properties was evaluated to get a better understanding of the constraints faced by different farmers.
We found that the accuracy of spectrally-predicted soil properties was lower than what can be found in literature. The accuracy was assessed on an independent sample set, unlike most studies where internal cross validation techniques are used, which is part of the reason for lower accuracy. We also found that the wet chemistry measured soil properties were a better predictor of yields and response than the information provided by spectroscopy. However, when other variables were considered next to soil properties, it appeared that the latter did not have a significant role in explaining and predicting yields and responses. Instead, management variables, such as the fertiliser and bush bean variety used, were found to be the major factors influencing yields and responses, along with environmental variables (temperature, altitude and precipitation) for the control yields. The household typology revealed some patterns in yields, with a low resource endowed category having significantly lower yields than the higher resource endowed categories. Nonetheless, the absolute response and the soil properties did not differ between household types. We conclude from our results that the potential response to yield-improving treatments is the same for all farmers, regardless of their soil and household characteristics. It is therefore important to reduce inequalities between farmers in terms of access to inputs and labour. This will help all farmers to apply good management practices (such as timely planting, weeding and the use of appropriate improved varieties) and fertilisers in their fields, since we have the indication that these are efficient ways to achieve higher yields.
MSc and Bcs thesis, internship reports