Author Contributions
The authors contributed to this work in the following ways: Conceptualization, S.L.B., S.S., G.H., L.H., C.F. and D.E.O.; Data curation, P.P.; Formal analysis, S.L.B., S.S., G.H. and C.F.; Funding acquisition, G.H. and C.F.; Investigation, S.L.B.; Methodology, S.L.B., S.S., G.H. and C.F.; Project administration, G.H.; Resources, G.H. and C.F.; Software, P.P.; Supervision, G.H.; Validation, S.L.B. and S.S.; Visualization, S.L.B.; Writing—original draft, S.L.B., S.S., G.H. and C.F.; Writing—review and editing, S.L.B., S.S., G.H., L.H., C.F. and D.E.O. All authors have read and agreed to the published version of the manuscript.
Figure 1.
(
a) Isohyets across West Africa, ranging from 100 mm/year in the far north of the Sahel to over 2000 mm/year in the south of West Africa (The amount of precipitation per year was capped at 2000 mm/year). Isohyets were calculated using yearly rainfall from the Climate Hazards InfraRed Precipitation with Stations dataset from 1981–2016 [
26]. (
b) Isopleths of ET
0 across West Africa, ranging from 5000 mm/year in the far north of the Sahel to under 1500 mm/year in the far south. (The amount of ET
0 was capped at 5000 mm/year). Isopleths of ET
0 used the ET
0 dataset produced by the National Oceanic and Atmospheric Administration’s Physical Sciences Division from 1981 to 2016 [
27,
28].
Figure 1.
(
a) Isohyets across West Africa, ranging from 100 mm/year in the far north of the Sahel to over 2000 mm/year in the south of West Africa (The amount of precipitation per year was capped at 2000 mm/year). Isohyets were calculated using yearly rainfall from the Climate Hazards InfraRed Precipitation with Stations dataset from 1981–2016 [
26]. (
b) Isopleths of ET
0 across West Africa, ranging from 5000 mm/year in the far north of the Sahel to under 1500 mm/year in the far south. (The amount of ET
0 was capped at 5000 mm/year). Isopleths of ET
0 used the ET
0 dataset produced by the National Oceanic and Atmospheric Administration’s Physical Sciences Division from 1981 to 2016 [
27,
28].
Figure 2.
Precipitation regressions for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) from 1981 to 2016, in mm/month for (a) through (e), and in mm/season for (f). The data in (a–e) is capped at −1 to 1 mm, and the data in (f) is capped at −10 to 10 mm. Any insignificant trend (p > 0.05) has been masked out in gray.
Figure 2.
Precipitation regressions for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) from 1981 to 2016, in mm/month for (a) through (e), and in mm/season for (f). The data in (a–e) is capped at −1 to 1 mm, and the data in (f) is capped at −10 to 10 mm. Any insignificant trend (p > 0.05) has been masked out in gray.
Figure 3.
(a) Trend for the cessation of the season, as measured by dekads per year, calculated from 1981 to 2016. Capped at −0.1 to 0.1 dekad/year, insignificance (p > 0.05) is masked out in gray. (b) The trend of total number of dry pentads for the season, as measured in pentads/year, calculated from 1981 to 2016. Capped at −0.15 to 0.15 pentads/year, insignificance (p > 0.05) is masked out in gray.
Figure 3.
(a) Trend for the cessation of the season, as measured by dekads per year, calculated from 1981 to 2016. Capped at −0.1 to 0.1 dekad/year, insignificance (p > 0.05) is masked out in gray. (b) The trend of total number of dry pentads for the season, as measured in pentads/year, calculated from 1981 to 2016. Capped at −0.15 to 0.15 pentads/year, insignificance (p > 0.05) is masked out in gray.
Figure 4.
ET0 regressions for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) calculated from 1981 to 2016. The units in (a) through (e) have been capped to −3 to 3 mm day month, while the units in f have been capped from −4 to 4 mm day season. Any insignificant slope (p > 0.05) has been masked out in gray.
Figure 4.
ET0 regressions for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) calculated from 1981 to 2016. The units in (a) through (e) have been capped to −3 to 3 mm day month, while the units in f have been capped from −4 to 4 mm day season. Any insignificant slope (p > 0.05) has been masked out in gray.
Figure 5.
ET0 regressed on precipitation, where precipitation is the independent variable and ET0 is the dependent variable for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) calculated from 1981 to 2016. Any insignificant slope (p > 0.05) has been masked out in gray.
Figure 5.
ET0 regressed on precipitation, where precipitation is the independent variable and ET0 is the dependent variable for the months of June (a), July (b), August (c), September (d), October (e), and full season (f) calculated from 1981 to 2016. Any insignificant slope (p > 0.05) has been masked out in gray.
Figure 6.
(
a–
e) (Left, Top to Bottom) June (
a), July (
b), August (
c), September (
d), October (
e) payouts that occurred for 1999–2016 precipitation data. (
f–
j) (Middle, Top to Bottom) June (
f), July (
g), August (
h), September (
i), October (
j) payouts that occurred for 1999–2016 ET
0 data. (
k–
o), (Right, Top to Bottom) June (
k), July (
l), August (
m), September (
n), October (
o), Heidke Skill Score for the payouts of ET
0 to the precipitation payouts. These plots show where the precipitation data and the ET
0 data have between 0 and 5 payouts occurring in the same year. To match the ET
0 and precipitation grids, the precipitation data were averaged per-pixel and resampled using nearest-neighbor. However, although some areas in the ET
0 appear more coarsely mapped than the precipitation data or the Heidke Skill Score maps, this appearance is a known consequence of the method in
https://psl.noaa.gov/eddi/globalrefet/ used to downscale the ET
0 spatial resolution from 0.5° × 0.625° resolution to 0.125° to 0.125° × 0.125° resolution and is not an artifact in our analysis in this study.
Figure 6.
(
a–
e) (Left, Top to Bottom) June (
a), July (
b), August (
c), September (
d), October (
e) payouts that occurred for 1999–2016 precipitation data. (
f–
j) (Middle, Top to Bottom) June (
f), July (
g), August (
h), September (
i), October (
j) payouts that occurred for 1999–2016 ET
0 data. (
k–
o), (Right, Top to Bottom) June (
k), July (
l), August (
m), September (
n), October (
o), Heidke Skill Score for the payouts of ET
0 to the precipitation payouts. These plots show where the precipitation data and the ET
0 data have between 0 and 5 payouts occurring in the same year. To match the ET
0 and precipitation grids, the precipitation data were averaged per-pixel and resampled using nearest-neighbor. However, although some areas in the ET
0 appear more coarsely mapped than the precipitation data or the Heidke Skill Score maps, this appearance is a known consequence of the method in
https://psl.noaa.gov/eddi/globalrefet/ used to downscale the ET
0 spatial resolution from 0.5° × 0.625° resolution to 0.125° to 0.125° × 0.125° resolution and is not an artifact in our analysis in this study.
Figure 7.
(a) Seasonal payouts that occurred during 1999–2016 in the precipitation dataset. (b) Seasonal payouts that occurred during 1999–2016 in the ET0 dataset. (c) Heidke Skill Score for the payouts of ET0 to the precipitation payouts. This plot shows where the precipitation data and the ET0 data have between 0 and 5 payouts occurring in the same year.
Figure 7.
(a) Seasonal payouts that occurred during 1999–2016 in the precipitation dataset. (b) Seasonal payouts that occurred during 1999–2016 in the ET0 dataset. (c) Heidke Skill Score for the payouts of ET0 to the precipitation payouts. This plot shows where the precipitation data and the ET0 data have between 0 and 5 payouts occurring in the same year.
Figure 8.
Using yield data, precipitation, ET
0, and SPEI for 1996–2016, (
a) this image shows the regression of precipitation predicting yield. (
b) this image shows the regression of ET
0 predicting yield. (
c) this image shows the regression of SPEI predicting yield. The data are identified by country in different colors: Burkina Faso represented in red, Mali in green, Niger in blue, and Senegal in purple. There were five regions per country, which are further represented by different regression lines in the plots—all plotted with corresponding colors to the country they are in. Further information on specifications of the regression models can be found in
Table 1.
Figure 8.
Using yield data, precipitation, ET
0, and SPEI for 1996–2016, (
a) this image shows the regression of precipitation predicting yield. (
b) this image shows the regression of ET
0 predicting yield. (
c) this image shows the regression of SPEI predicting yield. The data are identified by country in different colors: Burkina Faso represented in red, Mali in green, Niger in blue, and Senegal in purple. There were five regions per country, which are further represented by different regression lines in the plots—all plotted with corresponding colors to the country they are in. Further information on specifications of the regression models can be found in
Table 1.
Figure 9.
The regression tree for yield across all regions using seasonal precipitation, seasonal ET0, and seasonal SPEI to classify the yield into classes. Seasonal SPEI was ultimately not used for splitting the data in this regression tree.
Figure 9.
The regression tree for yield across all regions using seasonal precipitation, seasonal ET0, and seasonal SPEI to classify the yield into classes. Seasonal SPEI was ultimately not used for splitting the data in this regression tree.
Figure 10.
(a) The regression tree for Burkina Faso, Mali, and Niger plots ET0 against precipitation. The rules for Burkina Faso and Mali have the first rule of precipitation greater (less) than 680.2 mm and the second rule of precipitation greater (less) than 460.6. The rule for Niger has precipitation greater (less) than 314.7 mm. (b) The regression tree for Senegal plots ET0 against precipitation. The first rule is ET0 greater (less) than 749.1 mm. The second rule is ET0 greater (less) than 997.5 mm. The final rule is precipitation greater (less) than 367.4 mm.
Figure 10.
(a) The regression tree for Burkina Faso, Mali, and Niger plots ET0 against precipitation. The rules for Burkina Faso and Mali have the first rule of precipitation greater (less) than 680.2 mm and the second rule of precipitation greater (less) than 460.6. The rule for Niger has precipitation greater (less) than 314.7 mm. (b) The regression tree for Senegal plots ET0 against precipitation. The first rule is ET0 greater (less) than 749.1 mm. The second rule is ET0 greater (less) than 997.5 mm. The final rule is precipitation greater (less) than 367.4 mm.
Figure 11.
(a) The regression tree results for Burkina Faso, Mali, and Niger plotted with ET0 against yield, with different dot shapes denoting which node in the regression tree the yield falls into. The dots are separated by different colors for the precipitation rules—for Burkina Faso and Mali, the precipitation greater than 680.2 mm is represented as pink, the precipitation between 680.2 mm and 460.6 mm is represented as dark red, and precipitation less than 460.6 is represented as green. For Niger, the precipitation greater than 314.7 mm is represented with blue, and the precipitation less than 314.7 is represented as orange. (b) The regression tree results for Senegal, plotting ET0 against yield, with different dot shapes representing which node in the regression tree the yield falls into. The dots are separated by different colors for the precipitation rules, and there are lines representing the ET0 rules. The first rule is ET0 greater (less) than 749.1 mm, and the second rule is ET0 greater (less) than 997.5 mm. The next rule is precipitation greater than 367.4 mm (represented by pink) and precipitation less than 367.4 mm (represented by orange).
Figure 11.
(a) The regression tree results for Burkina Faso, Mali, and Niger plotted with ET0 against yield, with different dot shapes denoting which node in the regression tree the yield falls into. The dots are separated by different colors for the precipitation rules—for Burkina Faso and Mali, the precipitation greater than 680.2 mm is represented as pink, the precipitation between 680.2 mm and 460.6 mm is represented as dark red, and precipitation less than 460.6 is represented as green. For Niger, the precipitation greater than 314.7 mm is represented with blue, and the precipitation less than 314.7 is represented as orange. (b) The regression tree results for Senegal, plotting ET0 against yield, with different dot shapes representing which node in the regression tree the yield falls into. The dots are separated by different colors for the precipitation rules, and there are lines representing the ET0 rules. The first rule is ET0 greater (less) than 749.1 mm, and the second rule is ET0 greater (less) than 997.5 mm. The next rule is precipitation greater than 367.4 mm (represented by pink) and precipitation less than 367.4 mm (represented by orange).
Figure 12.
(a) The regression tree results for Burkina Faso, Mali, and Niger with yield plotted against precipitation; the different dot shapes represent the nodes the observation is under and the different dot colors represent the country the observation is from, with lines representing the regression tree rules for precipitation. The first rule for Burkina Faso and Mali data is precipitation greater (less) than 680.6 mm, and the second rule is precipitation greater (less) than 460.2 mm. The rule for Niger observations is precipitation greater (less) than 314.7 mm. (b) The regression tree results for Senegal with yield plotted against precipitation, the different dot shapes representing the node the observation is under and the different dot color representing the ET0 rules, with a line representing the rule for precipitation. The first rule splits data into greater than 749.1 mm and less than 749.1 mm (represented in orange). The second rule splits data into greater (less) than 997.5 mm, with data that is greater than 997.5 represented in pink. The data that falls between 997.5 mm of ET0 and 749.1 mm are represented in blue, and are subject to the precipitation rule of greater (less) than 367.4 mm.
Figure 12.
(a) The regression tree results for Burkina Faso, Mali, and Niger with yield plotted against precipitation; the different dot shapes represent the nodes the observation is under and the different dot colors represent the country the observation is from, with lines representing the regression tree rules for precipitation. The first rule for Burkina Faso and Mali data is precipitation greater (less) than 680.6 mm, and the second rule is precipitation greater (less) than 460.2 mm. The rule for Niger observations is precipitation greater (less) than 314.7 mm. (b) The regression tree results for Senegal with yield plotted against precipitation, the different dot shapes representing the node the observation is under and the different dot color representing the ET0 rules, with a line representing the rule for precipitation. The first rule splits data into greater than 749.1 mm and less than 749.1 mm (represented in orange). The second rule splits data into greater (less) than 997.5 mm, with data that is greater than 997.5 represented in pink. The data that falls between 997.5 mm of ET0 and 749.1 mm are represented in blue, and are subject to the precipitation rule of greater (less) than 367.4 mm.
Figure 13.
(a) The precipitation plotted against the ET0, with the rules of the classification tree shown as lines. The first rule is precipitation is greater (less) than 317.7 mm, the next split is ET0 is greater (less) than 892.5 mm. Crop failures are shown in red, and crop successes as defined by the 20th quantile are shown in blue. (b) the classification tree is shown, with 416 observations, 83 of which are crop failures. In Node 2, the precipitation is greater than 317.7 mm, with 317 of the 359 observations being crop successes (88%) and 42 of the observations being crop failures (12%). In Node 3, the observations have less than 317.7 mm of precipitation, resulting in 16 crop successes of the 57 observations (28%) and 41 crop failures (72%). Node 4 has ET0 less than 892.5 mm, and of the 24 observations, 16 are crop successes (66.7%) and 8 are crop failures (33.3%). Finally, Node 5 has ET0 less than 892.5 mm and has 33 observations, all of which are crop failures (100%)
Figure 13.
(a) The precipitation plotted against the ET0, with the rules of the classification tree shown as lines. The first rule is precipitation is greater (less) than 317.7 mm, the next split is ET0 is greater (less) than 892.5 mm. Crop failures are shown in red, and crop successes as defined by the 20th quantile are shown in blue. (b) the classification tree is shown, with 416 observations, 83 of which are crop failures. In Node 2, the precipitation is greater than 317.7 mm, with 317 of the 359 observations being crop successes (88%) and 42 of the observations being crop failures (12%). In Node 3, the observations have less than 317.7 mm of precipitation, resulting in 16 crop successes of the 57 observations (28%) and 41 crop failures (72%). Node 4 has ET0 less than 892.5 mm, and of the 24 observations, 16 are crop successes (66.7%) and 8 are crop failures (33.3%). Finally, Node 5 has ET0 less than 892.5 mm and has 33 observations, all of which are crop failures (100%)
Table 1.
Regressions Predicting Yield Numbers within the brackets () represent the standard error of the estimate directly above themselves. Different p-values are used to determine significance—* represents estimates that p-values smaller than 0.05, ** represent estimates that have p-values smaller than 0.01, and *** represent estimates that have p-values smaller than 0.001.
Table 1.
Regressions Predicting Yield Numbers within the brackets () represent the standard error of the estimate directly above themselves. Different p-values are used to determine significance—* represents estimates that p-values smaller than 0.05, ** represent estimates that have p-values smaller than 0.01, and *** represent estimates that have p-values smaller than 0.001.
| Model 1: Precipitation | Model 2: ET0 | Model 3: SPEI |
---|
Intercept | 0.4954 *** | 1.1026 *** | 0.7521 *** |
| (−0.0759) | (−0.1039) | (−0.0379) |
X1 | 0.0005 *** | −0.0006 *** | 0.2856 *** |
| (−0.0001) | (−0.0002) | (−0.0536) |
Burkina-Kenedougou | 0.1494 * | 0.2278 *** | 0.0385 *** |
| (−0.0634) | (− 0.0564) | (−0.0535) |
Burkina-Kouritenga | −0.0085 | −0.0046 | 0.0385 |
| (−0.0551) | (0.0554) | (−0.0535) |
Burkina-Mouhoun | 0.226 *** | 0.402 *** | 0.2988 *** |
| (−0.0568) | (0.0611) | (−0.0535) |
Burkina-Silassi | 0.1684 ** | 0.3503 *** | 0.2726 *** |
| (−0.0596) | (−0.0611) | (−0.0535) |
Mali-Kayes | 0.0436 | 0.0946 | 0.1179 * |
| (−0.057) | (−0.0544) | (−0.0535) |
Mali-Koulikoro | 0.0972 | 0.2928 *** | 0.1325 * |
| (−0.0546) | (−0.0699) | (−0.0535) |
Mali-Mopti | 0.0191 | 0.0131 | −0.0396 |
| (−0.0557) | (−0.056) | (−0.0535) |
Mali-Segou | 0.1347 * | 0.3019 *** | 0.1315 * |
| (−0.0539) | (−0.0716) | (−0.0535) |
Mali-Sikasso | 0.1265 | 0.3539 *** | 0.2808 *** |
| (−0.066) | (−0.0587) | (−0.0542) |
Niger-Bouza | −0.2407 *** | −0.0557 | −0.3178 |
| (−0.0571) | (−0.0902) | (−0.0535) |
Niger-Dosso | −0.2264 *** | −0.0975 | −0.2338 *** |
| (−0.05459) | (−0.0665) | (−0.0542) |
Niger-Filingue | −0.3094 *** | −0.3415 *** | −0.4191 *** |
| (−0.0602) | (−0.0582) | (−0.0535) |
Niger-Goure | −0.3062 *** | −0.3722 *** | −0.5024 *** |
| (−0.073) | (−0.0656) | (−0.0542) |
Niger-Mayahi | −0.3162 *** | −0.1432 | −0.4006 *** |
| (−0.0582) | (−0.0891) | (−0.0542) |
Senegal-Diourbel | −0.0719 | 0.1062 | −0.1752 ** |
| (−0.0596) | (−0.0945) | (−0.0535) |
Senegal-Fatick | −0.0092 | 0.1580 | −0.09 |
| (−0.0574) | (−0.0871) | (−0.0535) |
Senegal-Foundiougne | 0.1864 ** | 0.3092 *** | 0.1071 * |
| (−0.0573) | (−0.0776) | (−0.0535) |
Senegal-Gossas | −0.1458 ** | −0.0115 | −0.1868 *** |
| (−0.0549) | (−0.0725) | (−0.0535) |
Senegal-Mbacke | −0.2022 *** | −0.2458 *** | −0.2593 *** |
| (−0.0557) | (−0.0542) | (−0.0535) |
Adjusted R-squared | 0.6649 | 0.6784 | 0.6692 |
F-statistic | 42.16 | 41.65 | 42.97 |
Degrees of Freedom | 395 | 395 | 395 |