Predicting malaria outbreaks from sea surface temperature variability up to 9 months ahead in Limpopo, South Africa, using machine learning

Abstract

Malaria is the cause of nearly half a million deaths worldwide each year, posing a great socioeconomic burden. Despite recent progress in understanding the influence of climate on malaria infection rates, climatic sources of predictability remain poorly understood and underexploited. Local weather variability alone provides predictive power at short lead times of 1-2 months, too short to adequately plan intervention measures. Here, we show that tropical climatic variability and associated sea surface temperature over the Pacific and Indian Oceans are valuable for predicting malaria in Limpopo, South Africa, up to three seasons ahead. Climatic precursors of malaria outbreaks are first identified via lag-regression analysis of climate data obtained from reanalysis and observational datasets with respect to the monthly malaria case count data provided from 1998-2020 by the Malaria Institute in Tzaneen, South Africa. Out of 11 sea surface temperature sectors analyzed, two regions, the Indian Ocean and western Pacific Ocean regions, emerge as the most robust precursors. The predictive value of these precursors is demonstrated by training a suite of machine-learning classification models to predict whether malaria case counts are above or below the median historical levels and assessing their skills in providing early warning predictions of malaria incidence with lead times ranging from 1 month to a year. Through the development of this prediction system, we find that past information about SST over the western Pacific Ocean offers impressive prediction skills (~80% accuracy) for up to three seasons (9 months) ahead. SST variability over the tropical Indian Ocean is also found to provide good skills up to two seasons (6 months) ahead. This outcome represents an extension of the effective prediction lead time by about one to two seasons compared to previous prediction systems that were more computationally costly compared to the machine learning techniques used in the current study. It also demonstrates the value of climatic information and the prediction framework developed herein for the early planning of interventions against malaria outbreaks.

Keywords: South Africa; climate; machine learning; malaria; prediction; weather.

Figure 1

Seasonal cycle of malaria incidence in South Africa. The seasonal cycle of monthly malaria incidence (median) is illustrated for the districts of Vhembe and Mopani, where the highest malaria incidences are reported in the province of Limpopo, South Africa. The highest incidences are observed in austral summer when temperature and precipitation are favorable for the spread of malaria. Interannual variations as measured with the interquartile range (25th percentile to 75th percentile) are shaded.

Figure 2

Persistence of malaria incidence. A cross-correlation of malaria incidence for various reference months. Correlation is indicated in color. For a specific month (y-axis) the correlation of malaria incidence with respect to other months are shown along the x-axis.

Figure 3

Sea surface temperature precursors of malaria in South Africa. Lag-regression maps of SST on malaria incidence in South Africa (Vhembe district in Limpopo province) are shown in function of target malaria month (rows) and lag (columns). The month when the precursors are observed is indicated in the bottom-right corner of each panel. For example, lag −3 precursors of observed malaria incidence in January are observed in October. Correlation is shaded at an interval of 0.2 with blue and red shadings for negative and positive correlations, respectively. Correlations that are significant at the 5% significance level are hatched in magenta. The conditions illustrated are associated with high malaria incidence. See Supplementary Figure 1 for the location of Limpopo. The sectors used later to construct SST-based climate indices are illustrated in the bottom-right panel according to the legend.

Figure 4

The lagged relationship between climate indices and malaria incidence in South Africa. Correlations between climate indices and malaria incidences throughout the malaria season (y-axes) and lags (x-axes) are illustrated with color shadings. Correlations that are significant at the 5% significance level are hatched in magenta. They are reported for various climate indices indicated above each panel (listed in Supplementary Table 1). The sectors used for SST indices (IODwest, IODeast, IOSDwest, IOSDeast, Niño3, Niño3.4, Niño4, SPnorth, SPsouth, WPnorth, and WPsouth) and local indices (precip, pres, tmax, tmin, uwnd, and vwnd; with the subscripts LP and MZ if averaged over Limpopo or Mozambique, respectively) are illustrated in Supplementary Figure 1. Domain boundaries are indicated in Table 2.

Figure 5

Summary of malaria prediction skills for various experimental setups. Malaria prediction accuracy is averaged over the months of December-April when malaria incidence is high (Figure 1). It is reported for a multi-model voting ensemble composed of the three best classifiers for each experimental setting (y-axis) and prediction lead (x-axis). Accuracy is indicated with numbers as the percentage of successful predictions and illustrated with colors with warmer colors indicate higher accuracies. Accuracies below 50 % are shown in purple.

Figure 6

Seasonal cycle of malaria prediction skill. Prediction accuracy is shown in function of target malaria prediction month (y-axis) and prediction lead (x-axis) for (left) exp1 which makes use of SST-based climate indices over the tropical Indian Ocean (IOD) and (right) SST over the western Pacific (WP; exp10). The skill is assessed with a multi-model voting ensemble composed of the three best classifiers selected for each lead/month combination. Accuracy is illustrated and indicated with numbers and colors as in Figure 5.

Figure 7

Predictions for Mopani: Similar to Figure 6 but for the district of Mopani instead of Vhembe. Results from additional experiments, labeled as discard, in which ambiguous categories and 2017 malaria data are discarded, are also shown.

Figure 8

Atmospheric teleconnections: 500-hPa (black contours) and sea level pressure (green contours) precursor to December (left) and April (right) malaria outbreaks are shown for −5 months and −7 months lags, respectively. Precursors are illustrated as correlations with solid and dashed contours (intervals of 0.2) for positive and negative correlations respectively. They are overlaid on SST precursors shown with red/blue shading. Correlations above 0.4 and below −0.4 are significant at the 95% significance level.