Relationships between weather attributes
Solar radiation and temperature attributes were positively correlated with each other but negatively related to precipitation (rain fall and relative humidity) (Table 2 and Supplementary Table 1). There were very strong relationships between some of the weather attributes. For this reason, not all weather attributes are presented. Average temperature at 2 meters is presented for maximum temperature at 2 meters, minimum temperature at 2 meters, temperature range at 2 meters, and earth skin temperature. Their correlation coefficients with average temperature at 2 meters were 0.977, 0.864, 0.816, and 0.984 (P<0.001) respectively. Precipitation attributes were also positively correlated with each other. RH at 3 pm is not presented except for a few peculiar cases.
Weather effect on seed set before pollination
Weather effects on seed set were significant right from 105 DBP when plants were still in the vegetative growth stage; these effects were few and mostly weak. High temperature recorded at 9 am was correlated with seed set (r=0.359, P<0.001) in ‘Mshale’ when the customary pollination technique was used (Supplementary Table 2). High solar radiation was required by ‘Nshonowa’ when pollination was done with PGM (r=0.214, P<0.05) while rainfall was the limiting factor for ‘Enzirabahima’ pollinated without PGM (r=0.275, P<0.001). At 90 DBP, high rainfall (r=0.332, P<0.01) was required particularly for ‘Mshale’ that was pollinated without PGM (Supplementary Table 3).
At 75 DBP, weather mainly influenced ‘Nshonowa’ with minimal effect on ‘Mshale,’ but not on ‘Enzirabahima.’ High average temperature and high solar radiation were positively significantly correlated to seed set in ‘Nshonowa’ (Supplementary Table 4). Low rainfall and RH were also required for seed set in ‘Nshonowa.’ At 60 DBP, high average temperatures and high solar radiation had weak effects on both Mchare cultivars when pollinated with PGM (Table 3 and Supplementary Table 5). Low rainfall only affected ‘Mshale’ while low RH and low temperature at 9 am increase seed set in both Mchare cultivars.
At 45 DBP, weather effects on seed set in Mchare cultivars were somewhat similar to effects 60 DBP though there was a reduced magnitude of association. On the other hand, high solar radiation (r=0.188, P<0.05) and low RH (r=-0.247, P<0.01) had an effect on seed set in ‘Enzirabahima’ when bunches were pollinated with PGM (Supplementary Table 6). At 30 DBP, high average temperature significantly increased seed set in Mchare cultivars when pollinated with PGM (Supplementary Table 7). Low RF in Nshonowa and low RH in both Mchare cultivars increased seed set. At 15 DBP, high average temperatures, high solar radiation, and low RH had a significant effect on seed set in ‘Enzirabahima’ especially when pollinated with PGM. A similar pattern was observed for Mchare cultivars and ‘Enzirabahima’ 15 DBP. However, the correlation coefficients were generally higher for Mchare compared to ‘Enzirabahima’ (Table 4 and Supplementary Table 8).
Weather effect on seed set at the time of pollination and after pollination
At the time of pollination, only ‘Enzirabahima’ and ‘Nshonowa’ were positively significantly associated with high average temperatures especially when bunches were pollinated with PGM. ‘Mshale’ required high RH at 9 am especially for bunches pollinated without PGM (Table 5 and Supplementary Table 9). At 3 pm, low RH was required in ‘Enzirabahima’ especially when pollinations were made with PGM (r=-0.260, P<0.001). ‘Mshale’ pollinated without PGM needed high RH (r=0.252, P<0.05) while ‘Nshonowa’ pollinated without PGM needed low RH (r=-0.237, P<0.05). Enhancing stigma receptivity in ‘Mshale’ resulted in a reduced level of association of seed set with high RH at 9 am from r=0.299, P<0.01 to r=0.265, P<0.05 (Table 5 and Supplementary Table 9).
At the stage of 15 DAP, high morning temperatures were necessary for seed set in ‘Nshonowa’ if pollinations were made without PGM (r=0.350, P<0.001). Interestingly, high RH at 3 pm was required for seed set in Mchare when ‘Mshale’ was pollinated without PGM (r=0.328, P<0.01) and in ‘Nshonowa’ when pollinated with PGM (r=0.269, P<0.01). On the contrary, ‘Enzirabahima’ needed low RH at 3 pm in bunches pollinated with PGM (r=-0.165, P<0.05). Also noted was the low solar radiation necessary for seed set in ‘Nshonowa’ pollinated with PGM (r=-0.330, P<0.001). A somewhat similar pattern was observed 30 DAP with all cultivars pollinated with PGM requiring high morning temperatures for seed set (r=0.150, P<0.05 for ‘Enzirabahima,’ r=0.360, P<0.001 for ‘Mshale’ and r=0.337, P<0.001 for ‘Nshonowa’) (Table 6 and Supplementary Table 11). Interestingly, low RH at 3 pm was somehow still associated with seed set in ‘Enzirabahima’ pollinated with PGM (r=-0.170, P<0.05). On the other hand, high RH at 3 pm was needed for seed set in ‘Nshonowa’ pollinated with PGM (r=0.260, P<0.01) and in ‘Mshale’ pollinated without PGM (r=0.328, P<0.01).
At 45 DAP, ‘Enzirabahima’ pollinated with PGM and ‘Nshonowa’ pollinated without PGM needed high morning temperatures for seed set (r=0.209, P<0.01 and r=0.428, P<0.001 respectively). But high average temperature reduced seed set in ‘Nshonowa' pollinated with PGM (r=-0.292, P<0.01) but not in ‘Enzirabahima’ pollinated without (r=0.188, P<0.05) (Supplementary Table 12). Mchare additionally needed rainfall for seed set when pollinated with PGM (r=0.298, P<0.01 and r=0.265, P<0.01 for ‘Mshale’ and ‘Nshonowa’ respectively). A similar pattern was observed for Mchare 60 DAP with the exception rain (Supplementary Table 13). ‘Enzirabahima’ did not have the same weather attributes significant at 45 and 60 DAP. Only high temperatures at 9 am (r=0.183, P<0.05) were required for ‘Enzirabahima’ pollinated with PGM 60 DAP as at 45 DAP. At this point, there were only very weak but significant associations of seed set and high morning temperatures at 9 am in ‘Enzirabahima’ (r=0.153, P<0.05). For Mchare cultivars, the difference was that ‘Mshale’ pollinated with PGM needed low average temperatures at 2 meters (r=-0.213, P<0.05) and high RH at 9 am (r=-0.250, P<0.05).
A relatively similar pattern was also observed 75 DAP as observed 60 DAP with a few differences (Table 7 and Supplementary Table 14). At this point, ‘Enzirabahima’ pollinated with PGM also had weak but positive association (r=0.165, P<0.5) of high RH at 2 meters with seed set just like in Mchare. For ‘Mshale’ pollinated without PGM, high RH was associated with seed set (r=0.220, P<0.05 at 9 am. Low solar radiation (r=-0.215, P<0.05) and high RH (r=0.237, P<0.05 at 2 meters) were still significantly associated with increased seed set for ‘Mshale’ pollinated with PGM. For ‘Nshonowa’ pollinated with PGM, temperature at 9 am, low average temperature at 2 meters, low solar radiation and high RH at 3 pm were required for seed set (Table 7).
There was no weather association with seed set in ‘Enzirabahima’ 90 DAP as this was about its maturity time. On the other hand, high temperatures at 9 am (r=0.300, P<0.01) and high RH at 3 pm (r=0.300, P<0.01) were necessary for increased seed set in ‘Mshale’ pollinated without PGM. There were generally the same to slightly weaker associations between weather and seed set in ‘Nshonowa’ pollinated with PGM from 75 to 90 DAP. Weaker associations were observed for average temperature at 2 meters (r=-0.307, P<0.01), solar radiation (r=0.300, P<0.01), RH at 3 pm (r=0.264, P<0.05), and average RH at 2 meters (r=0.291, P<0.01).
At 105 DAP, only high RH at 3 pm (r=0.385, P<0.001) was associated with high seed set in ‘Mshale’ pollinated without PGM (Supplementary Table 16). On the other hand, ‘Nshonowa’ pollinated with PGM required low average temperature at 2 m (r=-0.304, P<0.01) for seed set. Close to full maturity at 120 DAP, weather patterns for seed set were nearly identical to those at 105 DAP (Supplementary Table 17). High RH at 9 am positively correlated to seed set in ‘Mshale’ pollinated without PGM (r=0.403, P<0.001). For ‘Nshonowa’ pollinated with PGM at 120 DAP, low average temperature at 2 m (r=-0.277, P<0.01), and low solar radiation (r=-0.347, P<0.001) were required for seed set.
Principal component analysis (PCA)
The first four principle components (PCs) could explain over 80% variability of data in ‘Enzirabahima,’ 84% in ‘Mshale’ and 84% in ‘Nshonowa’ (Table 8, Supplementary Tables 18 and 19). The first PC coefficients are positive for temperature related measurements and negative for precipitation related measurements in the three EACBs. But maximum and average temperature related measurements had the highest absolute coefficients in the first PC. Only RH at 9 am, RF, and temperature at 9 am seemed not to explain variability using the first PC as they had low coefficients. The first PC could explain about 51% of variability in ‘Enzirabahima’ and between 53 and 59% in the Mchare cultivars whether pollinations were made with or without PGM. Maximum temperature at 2 meters, average temperature at 2 meters, earth skin temperature, and temperature at 3 pm ranked as the top four for coefficients of the first PC in all three EACB. The coefficients of these first four weather attributes for the first PC were in the range of 0.341 to 0.375.
In the second PC, temperature at 9 am was the most important for ‘Enzirabahima’ with coefficients of 0.654 and 0.549 for pollination with and without PGM respectively. RH at 9 am was the highest in the second PC for Mchare cultivars with coefficients of 0.543 and 0.528 for ‘Mshale’ pollinated with and without PGM respectively. On the other hand, ‘Nshonowa’ had 0.634 and 0.528 for pollinations with and without PGM respectively. In the third PC for ‘Enzirabahima,’ RH at 9 am had the highest coefficient of 0.423 when pollinated without PGM whereas seed set per 100 fruits had the highest coefficient of 0.644 when pollinated with PGM. For Mchare cultivars, temperature at 9 am had the highest coefficients of 0.826 and 0.780 for ‘Mshale’ and 0.722 and 0.760 for ‘Nshonowa’ when pollinated with and without PGM respectively.
Multiple linear regression approach
For ‘Enzirabahima’ pollinated without PGM, regression analysis was only significant (P<0.01) 30 DAP development stage (Supplementary Table 30). On the other hand, regression for ‘Enzirabahima’ pollinated with PGM was significant at all considered time intervals except at 105, 75, 15 DBP, and 15 DAP (Supplementary Tables 20 to 34). For ‘Mshale’ pollinated without PGM, regression was significant for all considered time intervals except at 45 and 30 DBP as well as at the time of pollination. Pollination of ‘Mshale’ with PGM reduced error and regression was significant at all floral development stages considered. Regression was significant for ‘Nshonowa’ pollinated with and without PGM except for pollination with PGM at 105 DAP (Supplementary Tables 20 to 37).
Use of PGM generally reduced the error and increased the percentage of variability accounted for in the three EACBs (Figure 1). ‘Nshonowa’ generally had the highest variability accounted for across considered development stages, followed by ‘Mshale’ and lastly, ‘Enzirabahima.’ For ‘Nshonowa,’ the highest percentage accounted for was at 45 DBP. On the other hand, percentage variability accounted for by regression in ‘Mshale’ pollinated with PGM was fairly consistent at all considered time intervals. Variability accounted for by regression in ‘Mshale’ pollinated without PGM was inconsistent across considered development stages with the lowest at 45 DBP. ‘Enzirabahima’ did not exceed 30% variability account for in regression analyses across all considered floral development stages whether pollinated with or without PGM. The highest percentage variability accounted for was at the time of pollination (28.7%) when pollinations were made with PGM (Supplementary Table 28). Error variability exceeded regression variability at 60, 45, 30 DBP, at time of pollination and 60 DAP for ‘Enzirabahima’ pollinated without PGM.
Combining significant weather attributes from accumulated ANOVA tables (Supplementary Tables 38 to 55) from all considered floral development in a multiple linear regression with bunch size as groups revealed significance. Only ‘Mshale’ pollinated with and without PGM was revealed bunch size as a significant grouping factor (Table 9). The highest total variability accounted of 96.7% for came from ‘Mshale’ pollinated with PGM. For ‘Nshonowa’ pollinated with PGM, there were a total of 90 counts of significant weather attributes at various stages. But all these counts of significant weather attributes could not all be accommodated in the fitted model. Consequently, earth skin temperature and temperature range were dropped for average temperature at 2 meter and temperature range at 2meters was dropped for maximum temperature at 2 meters. Their correlations were r=0.984, P<0.001 and r=0.916, P<0.001 respectively.
Relating correlation, PCA and multiple linear regression
PCA teased out maximum and average temperature at 2 meters as the most important weather attributes for seed set. Regression also revealed that use of PGM increase percentage variability accounted for by the regression model. A plot of correlation coefficients of seed set with average temperature at various development stages for the three EACB pollinated with PGM was therefore used describe the ideal trend for seed set (Figure 2). The three EACB seemed to follow a similar pattern though Mchare cultivars were more closely related. Mchare cultivars had two peaks of positive temperature correlation with seed set before pollination. The first and lesser peak happens between 75 and 60 DBP while the second and greater peak happens at 15 DBP. ‘Enzirabahima’ had a lesser peak about 45 DBP and greater peak at 15 DBP and time of pollination. After pollination, low average temperatures are generally required for seed set especially for Mchare. The magnitude of temperature association with seed set after pollination is less than before pollination (Figure 2).