1. Introduction
A projected world population of 9 billion by 2050 places the international farming sector under extreme pressure as this will require agricultural production to rise an additional 70% from 2005–07 to 2050 (FAO 2009). Closing the yield gap is regarded as the most promising approach to tackle this imbalance (Godfray et al. 2010; Licker et al. 2010; Mueller et al. 2012), which is likely to be aggravated in parallel with declining global per capita agricultural production under the influence of climate change (Funk and Brown 2009). While climate change can result in the alteration of severity and frequency of all kinds of hydrometeorological extreme events (IPCC 2012) this study focuses on agricultural droughts and their two risk dimensions: On the one hand, droughts can have direct agricultural impacts, such as crop failure; on the other hand, also the mere risk of having to face a drought in the coming season can affect the decision-making of vulnerable communities. Thus, they tend to avoid taking agricultural risks that could increase their production (Rosenzweig and Wolpin 1993) in average and high-yield seasons that could serve as a buffer for drought years. The goal of weather index insurance (WII) is to limit agricultural losses caused by adverse weather events while simultaneously allowing farmers to take certain risks that aim at increasing their production.
While agricultural insurance covered 40% of the total costs of weather-related disasters in high-income countries between 1980 and 2003, only 4% were covered in low-income countries. Still, the payouts from these 4% represented 3 times the amount of international aid provided for the events (Mills 2007). WII is a new generation of mechanism, which is based on the transition from loss-based to risk-based insurance. Embedded in a network of complementary tools and mechanisms, such as loans, savings, and intentional risk-taking to promote agricultural production (e.g., via drought-resistant seeds), WII pays out based on an objectively identified risk parameter. An example for such a risk parameter is the level of critical rainfall during sensitive growing periods in the case of an index that focuses on agricultural drought. Satellite-derived data have become an indispensable source of information for the development of WII. In contrast to in situ observations, which are not available on a global scale at a sufficient spatial and temporal density, satellite data provide objective spatially and temporally harmonized estimations of various environmental variables (IRI 2013). Figure 1 illustrates the concept of weather index–based insurance on the assessment of environmental and socioeconomic conditions, the active integration of smallholder farmers via feedback loops, and the possible consequences on disaster resilience.
Schematic framework of weather index-based insurance development.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
WII is usually more affordable than traditional insurance. This is mainly due to the fact that there is no need for postdisaster loss assessment (IFAD 2011), which tends to increase the insurance premium, particularly if many small-scale loss assessments are required. Since the index concentrates on anomalous weather conditions instead of crop losses at the end of the season, payouts are usually faster than for conventional crop insurance. Ideally, these early payouts allow farmers sufficient time and financial resources to employ countermeasures, for example replanting in case of early season moisture deficits.
According to Hellmuth et al. (2009), the probability of payouts, which are driven by the probability of extreme weather events, define the final price of an insurance index. Thus, WII is very sensitive to its parameterization and historical calibration. As indicated in Fig. 2, for a rainfall-based index the key parameters include the trigger (critical threshold, which triggers the beginning of payouts), the exit (threshold below which any rainfall deficits results in a maximum payout), the payout frequency, and the time window (the critical period covered in the agricultural season). One of the most critical steps in the calibration process involves identifying the length and timing of the insurance window [usually from three to five dekads (10-day periods)] as well as the definition of daily and/or dekadal caps to limit the impact of short, heavy rainfall events on payouts. Strong, isolated rainfall events would otherwise decrease the payouts without improving the growing conditions of crops during times of drought.
Schematic illustration of index insurance trigger and exit points, linking cumulative rainfall (mm) to payouts (USD).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
2. The role of satellite data
Agriculture generally depends largely on a balanced hydrologic cycle, because vegetation activity relies on moisture supply and the atmosphere’s energy budget. In rain-fed agriculture the hydrologic cycle is driven by rainfall providing moisture for the root zone of crops. Evapotranspiration is one of the main factors that determine how much precipitation remains in the soil. Sufficient water supply and limited atmospheric water demand allow vegetation to flourish. Since satellite products that independently estimate different components of the hydrologic cycle are available it is clearly valuable to consider them as interrelated, cascading sources of information that strongly influence vegetation vigor. However, flourishing, green vegetation does not necessarily mean high agricultural yields. The normalized difference vegetation index (NDVI; Kogan 1995), which is used as a proxy for the vegetative response to plant available water in this study, represents the landscape’s biophysical response to soil moisture (Ahmed et al. 2017). One significant limitation of the NDVI is its focus on vegetation greenness. In the case of maize, for instance, it is possible that the plant’s leaves appear green while the ears are underdeveloped. As a consequence, different empirical regression models (Balaghi et al. 2008) and machine learning models (e.g., Panda et al. 2010) have been developed to use vegetation indices with and without atmospheric forcing to predict near-future yields.
Different satellite datasets have strengths and limitations based on physical retrieval mechanisms, sensor technology, or algorithmic development. To understand the strengths and weaknesses of each of the data sources across different landscapes and climate types we are interested in how closely they relate to the NDVI on the African continent. We utilize satellite rainfall estimates driven by observations of cloud-top temperatures, evapotranspiration (ET) estimates driven by a land surface temperature (LST)-focused energy balance model and direct soil moisture estimates from radars/radiometers to characterize anomalies in the hydrologic cycle. If each of these estimates is accurate it should be strongly correlated with vegetation greenness in regions, in which the NDVI itself is an effective measure of vegetation vigor. Hence, if the independent satellite-derived water cycle estimates show high correlation coefficients with each other and the NDVI this provides confidence in their accuracy and their capability to represent agricultural drought conditions. To the extent that the satellite-derived estimations disagree there is value in understanding what drives the difference. In addition, we identify the relationship between satellite-derived anomalies of all variables and on the ground experience based on farmer reports. Since farmer reports are intrinsically noisy, a mismatch does not necessarily disprove satellite data quality. However, a high agreement can be used as additional evidence for the usefulness of satellite-derived datasets to characterize agricultural drought conditions.
Since data are the foundation of a robust weather risk index this study concentrates on the following core questions:
Do satellite observations from different independent satellite sensors confirm drought conditions to a degree that allows the development of a “conclusive story” from rainfall deficit to crop failure?
Can the consideration of multiple independent but complementary satellite datasets lower the basis risk (mismatch between index payout and actual drought conditions as reported by farmers)?
Where do the moisture-focused datasets agree or disagree with each other and the NDVI? Can the strengths of one satellite-derived dataset compensate for the physical limitations of another?
3. Regions of interest
We focus on the performance and agreement of satellite-derived rainfall, soil moisture, vegetation greenness, and evaporative stress on three spatial scales: the entire African continent, the transition zone from arid areas (Sahara) to the belt of the humid savannas, and three regional case studies in central Senegal (Tambacounda), northern Ethiopia (Tigray), and Zambia (Southern Province). With regard to the three case studies, the performance of weather insurance indices was also assessed using participatory processes (anecdotal evidence) of drought conditions collected during visits to selected villages within the target zone. All calculations are based on an extended bounding box around the villages that the financial instruments sector team at Columbia University visited between 2010 and 2016. The bounding boxes around the villages and dominating farming systems (DFS) are defined as follows (Fig. 3):
81 villages in Ethiopia (Tigray): from 38°45′ to 40°E; from 12°15′ to 14°30′N; DFS: highland temperate mixed;
44 villages in Senegal (Tambacounda): from 13°15′ to 15°15′W; from 13°30′ to 14°45′N; DFS: agro-pastoral, partly cereal–root crop mixed; and
18 villages in Zambia (Southern Province): from 27°15′ to 27°45′E; from 16°30′S to 16°45′S; DFS: agro-pastoral.
FAO farming systems in Africa (Dixon et al. 2001); red boxes in inset maps indicate the regions of interest in (a) Tigray, Ethiopia, (b) Tambacounda, Senegal, and (c) Southern Province, Zambia; the dotted rectangle indicates the transition zone from the arid to the humid zone.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
a. General characteristics
Virtually every climate zone can be found on the African continent. Ninety-five percent of the sub-Saharan agricultural land is rain-fed (Wani et al. 2009). In many regions, such as the greater Horn of Africa, the dependence on precipitation with large-scale intraseasonal and interannual variability results in a strong coupling of climatic shocks, agricultural production, food prices, and food insecurity (Clover 2003; FAO 2015). Another factor that adds complexity to the investigation of drought is the continuous energy and humidity feedback mechanism between the atmosphere and the land surface. This feedback is particularly characteristic for the Sahel, where Koster et al. (2004) found a strong influence of soil moisture levels on precipitation. Miralles et al. (2012) highlighted that, during boreal summer (June–August), in particular in intermediate regions between dry and wet climates, soil moisture limits evaporation.
b. Case study regions
Since a variety of satellite-based datasets are biased over “complex” terrain it is vital to know the topography in the case study regions. According to Wagner et al. (2013) the variability of the surface topography directly influences, for instance, the backscatter of radars that are used to detect soil moisture and thereby the soil moisture estimation. While Senegal is mostly characterized by more densely populated low-lying areas (Fig. 4, top), the population in Ethiopia tends to be concentrated in the country’s higher-lying areas, such as the Northern Highlands (elevation from 1500 to over 2600 m above sea level), in which the study area Tigray is located. The topographic complexity plays an important role in the estimation of rainfall amounts (Dinku et al. 2008) and the detection of light rainfall events (Bitew and Gebremichael 2010). In Zambia (Fig. 4, bottom) the topography only plays a minor role with regard to the distribution of rainfall. The rainy season is fairly homogenous (Black et al. 2016) and driven primarily by the intertropical convergence zone (ITCZ) and what is known as the Zaire Air Boundary (ZAB).
Topography (m; based on GTOPO30) and estimated population distribution (people per grid square) at a spatial resolution of 0.000 83 decimal degrees (approximately 100 m at the equator) for (top) Senegal, (middle) Ethiopia, and (bottom) Zambia; note that the legend values are unique for each country.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
The rainy seasons in Senegal range from June to September/October and in Zambia from around November to April. Because of a seasonal shift of the ITCZ Ethiopia has a long and a short rainy season. The long “kiremt” rains last from June to September; the short “belg” rains from April to May. A global monthly rainfall climatology can be accessed via an online map room of the International Research Institute for Climate and Society (IRI) at Columbia University (http://iridl.ldeo.columbia.edu/maproom/Global/Climatologies/Precip_Loop.html).
The land cover in all three regions of interest is characterized by savanna as the dominant land cover class. The Tigray region (Fig. 5, top) in Ethiopia is the only one with mentionable grassland. Most other areas are a cropland/woodland mosaic and dryland, cropland, and pasture. The Tambacounda region in central Senegal (Fig. 5, middle) is characterized by large cropland/woodland mosaic areas. The Southern Province in Zambia (Fig. 5, bottom) shows large areas with dryland, cropland, and pasture. According to the United Nations Food and Agriculture Organization (UN FAO) country briefs (http://www.fao.org/giews/countrybrief) maize and sorghum are major food crops in all three regions.
Land cover for (top) Tigray, Ethiopia, (middle) Tambacounda, Senegal, and (bottom) Southern Province, Zambia based on USGS EROS GLCCDB version 2.0. Brownish classes represent classes related to cropland; green classes represent grassland, savanna, or shrubland.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
4. Datasets and methods
a. Datasets
1) Rainfall
Since rainfall is traditionally the most used input dataset for index insurance we analyze the performance of two different datasets. The first rainfall dataset is the African Rainfall Climatology, version 2 (ARC2), which is consistent with the Rainfall Estimation version 2 (RFE2) (Novella and Thiaw 2013). However, in contrast to RFE2, ARC2 only includes data from infrared sensors and weather stations (no microwave sensors), resulting in longer time series. ARC2 is distributed from the National Oceanic and Atmospheric Administration (NOAA) and is available for the African continent from 1983 to the present on a 0.1° grid. It is based on two input data sources: 3-hourly infrared data from a geostationary satellite (Meteosat) and quality-controlled in situ observations from the Global Telecommunication System (GTS) network. The gauge observations are used as daily accumulated rainfall. In contrast to Climate Hazards Group Infrared Precipitation with Stations (CHIRPS), which is the second rainfall dataset used in this study, ARC2 is available without a monthly delay. With regard to data quality, a summer dry bias over West and East Africa has been detected. A validation study, which compared ARC2 to Tropical Rainfall Measuring Mission, version 7 (TRMM), and RFE2 over western Uganda (Diem et al. 2014), found that ARC2 tended to overestimate rainfall at all in situ stations but underestimated boreal-summer rainfall at the station in Uganda’s north. In general, ARC2 performed much better with regard to seasonal totals than for daily, dekadal, or monthly averages.
The CHIRPS precipitation dataset merges satellite measurements with station observations and a long-term mean field to produce pentadal (5 day) estimates of precipitation (Funk et al. 2015). These data run from 1981 to the present, and extend from 50°S to 50°N over all longitudes at a 0.05° spatial resolution. While a preliminary product is generated with short latency (3 days) using only a subset of available station data, the production-level product is available in the middle of the following month. CHIRPS has been shown to capture well the observed spatial and temporal variability in rainfall, as indicated by independent station data (Hessels 2015; Katsanos et al. 2016). The dataset is designed primarily for drought monitoring applications (Funk et al. 2015) but has also been successfully used in running crop or land surface models (Maidment et al. 2017) and as an independent variable in human health studies (Diouf et al. 2017).
2) Soil moisture
Within the Climate Change Initiative (CCI) of the European Space Agency (ESA) different active (radar) and passive (radiometer) sensors are combined to generate a daily surface soil moisture product with a spatial resolution of 0.25°, corresponding to roughly 28 km at the equator (Liu et al. 2011, 2012; Wagner et al. 2012). Since radiometers perform better in regions with low vegetation cover and radars work better if the vegetation density increases, a weighted blending scheme was introduced in version 03.2 (Dorigo et al. 2017), which we used for this study. The blending weight for each input dataset is calculated daily as the reciprocal of its random error variance. Regions with very dense vegetation, such as tropical forests, need to be masked, because neither sensor type performs well. Just like version 02.2 of the ESA CCI dataset, version 03.2 shows large gaps prior to 1992 (McNally et al. 2016). An operational version of the CCI soil moisture dataset, updated every 10 days, is currently in development (Enenkel et al. 2016).
One major advantage of soil moisture retrieval via radars or radiometers is that they are largely independent from weather conditions (e.g., cloud cover). However, there are physical limitations, such as the penetration depth, that limit the ESA CCI dataset to soil moisture in the top layer (the top few centimeters) of the soil (Qiu et al. 2014; Wagner et al. 1999). Also, in cases of complex topography (e.g., mountainous terrain) and frozen/snow-covered soils higher errors are observed (Dorigo et al. 2012). While soil moisture retrieval algorithms had originally been applied to sensors designed for other purposes they are now used to develop fully operational datasets and applied to dedicated soil moisture sensors (Dorigo and de Jeu 2016).
3) Evaporative stress index
The evaporative stress index (ESI) (Anderson et al. 2007, 2013) represents standardized anomalies in the ratio of actual evapotranspiration to potential evapotranspiration (PET) (fPET = ET/PET), where ET and PET are instantaneous or daily estimates retrieved using the Atmosphere Land Exchange Inverse (ALEXI) two-source energy balance algorithm (Anderson et al. 1997). Normalization by PET serves to minimize variability in ET due to seasonal variations in available energy and vegetation cover, further refining focus on the soil moisture signal. In the agricultural water management community, fPET is often referred to as the crop coefficient (Kc)—a fixed value or seasonal curve that can be used to estimate seasonal crop water use (Allen et al. 1998). In the case of ESI, however, fPET is retrieved from satellite-based estimates of actual ET rather than prescribed a priori.
To highlight differences in moisture conditions between years, standardized anomalies in fPET are expressed as a pseudo z score, normalized to a mean of zero and a standard deviation of one with respect to baseline fields describing “normal” (mean) conditions over the period of record. Extensive assessments of fPET and ESI in comparison with soil moisture observations, standard drought indicators, and crop yield datasets have shown the ability of land surface temperature to act as a proxy for surface soil moisture conditions (Anderson et al. 2011, 2013, 2015, 2016, Hain et al. 2009, 2011, 2012; Otkin et al. 2013, 2015).
Here, the ESI within ALEXI is generated over the analysis period of record using two different primary land surface temperature inputs: 1) a 3-hourly, geostationary-based LST data generated from the Gridded Satellite (GridSat-B1) Climate Data Record (CDR), which provides data from 1979 to current over Africa at a spatial resolution of 0.07°, and 2) twice-daily estimates of LST from MODIS Terra from 2000 to current over Africa at a spatial resolution of 0.05° (Wan et al. 2015). Satellite-based vegetation amount information needed by ALEXI for characterization of the vegetative canopy, here the leaf area index (LAI), is taken from the NOAA LAI Climate Data Record (Claverie et al. 2014). Meteorological inputs including atmospheric profiles of potential temperature, specific humidity, and geopotential height and surface variables such as air temperature, surface pressure, incoming solar radiation, and wind speed are taken from the Climate Forecast System Reanalysis (CFSR; Saha et al. 2010) product. CFSR provides data at 6-h analysis time steps with from one to six forecasts in between analyses, effectively providing full three-dimensional atmospheric data at an hourly temporal resolution.
4) Vegetation greenness
Vegetation greenness is provided via the NDVI, which uses spectral information from the visible and near-infrared bands to indicate the relative greenness of the vegetation. For this project, data were acquired from two sources: 1) the Global Inventory Modeling and Mapping Studies (GIMMS), provided at 0.08° for the period from 1981 to 2015 from Advanced Very High Resolution Radiometer (AVHRR), and 2) the MODIS Terra NDVI monthly climate modeling grid (MOD13C2) provided at 0.05° for the period 2000–15 (Huete et al. 2010). The GIMMS dataset represents the maximum NDVI value for the month, which allows the user to see the greenest possible pixel. The MODIS dataset represents the maximum NDVI value for the month after filtering for atmospheric contamination, clouds, and cloud shadows. All data were spatially averaged after filtering for the contaminants listed above, yielding the highest-quality value available for a given 0.05° pixel for that month.
5) Drought assessments based on participative processes
The International Institute for Climate and Society at Columbia University is continuously carrying out interactive exercises with farmers in all three study areas. These exercises (cf. Ouni and Dinh 2017) aim at the assessment of rainfall deficits during sensitive periods of the agricultural season, potential agricultural losses and related coping mechanisms, the performance of weather insurance indices, and the evaluation of the worst drought years in history. For this study, the latter is particularly crucial as reference information for comparison with satellite-detected weather and land surface anomalies. Farmers were specifically asked to remember rainfall deficits instead of crop failure (which could have been caused by other factors such as floods or diseases) between 1983 and 2015 (for Ethiopia and Zambia) or between 1990 and 2015 (for Senegal). Since the assessments are usually noisy due to disagreements among farmers or differences in assessments over multiple visits, appendix A (Table A1) provides an aggregated dataset of drought years.
b. Methods
According to Maidment et al. (2013) the aggregation of satellite data over space as well as over time generally improves their performance. Also with regard to predicting agricultural losses cumulative rainfall shows a better performance than instantaneous precipitation at one location (Black et al. 2016). Hence, all data used in this study (except soil moisture, which has a native resolution of 0.25°) were aggregated to a spatial resolution of 0.25° (roughly 28 km at the equator). The aggregation excluded pixels that were labeled as “missing” in the input products, primarily due to clouds or other atmospheric factors. The average of all candidate pixels within a 0.25° grid cell was calculated and used to populate the output file for a given month.
All spatial analyses were carried out exclusively for pixels that provide information for all datasets. Hence, no dataset has an advantage with respect to its individual masking or spatial or temporal resolution. This is vital for the objective of this study, which concentrates on strengthening the narrative of how satellite-derived variables represent drought conditions on the ground rather than exploiting the higher spatial resolution of individual datasets. Furthermore, it is still possible to use such coarse-resolution data for the detection of large-scale droughts.
To understand the relationships of independent satellite-derived estimates and their relationship to droughts as reported by farmers in three regions of interest we rely on the following methodical steps (Fig. 6):
We calculate standardized correlations for the anomalies of the moisture-focused variables (rainfall, soil moisture, ESI) and the NDVI lagged by one month to allow for the response time of vegetation greenness to alterations in moisture. Please keep in mind that the NDVI anomalies are an indicator of vegetation vigor but cannot directly be interpreted as crop yield. All calculations are carried out for the entire African continent. Analytical efforts focus on regions with exceptionally low or high agreement as well as on possible methodical explanations for disagreements. In addition, we analyze the predictive skill of ESI, CHIRPS, and ESA CCI anomalies regarding the GIMMS and MODIS NDVI in different climates from 2001 to 2015. Since the climate zones and agricultural seasons vary strongly with latitude we distinguish three zones: the northwest (west of 25°E, 0°–15°N), northeast (east of 25°E, 0°–15°N), and south (0°–35°S).
Second, we concentrate on the transition zone from arid to humid climates (11°–15°N) and the spatial correlation of all datasets over different land covers. The standardized lagged correlations are again calculated for all datasets with GIMMS and MODIS NDVI from 2001 to 2015.
Regarding the three regional case studies we match monthly and seasonal standardized anomalies of all datasets with drought years identified via multiple field visits to see which satellite datasets are capable of tracking the worst reported drought years. In the regional case studies we rely on the GIMMS NDVI to track the agreement of vegetation greenness and reported drought years beyond the availability of the MODIS NDVI (no data before 2001).
Schematic illustration of the methods used to identify the agreement between rainfall, soil moisture, evaporative stress index, and vegetation greenness indicators (top) on the pan-African scale, (middle) over different land covers in the transition zone, and (bottom) for three specific regions of interest.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
5. Results
a. Pan-African scale
The correlation analysis is based on the following assumption: If the correlation results of ESI and soil moisture with the lagged NDVI are comparable to or higher than for CHIRPS and ARC2 with the NDVI, this indicates both their predictive skill with regard to near-future vegetation greenness and their potential added-value to cross-validate satellite rainfall. Appendix B (Figs. B1–B8) illustrates the monthly lagged correlation of standardized anomalies for Africa between 1992 and 2015. For CHIRPS rainfall (Fig. B1) the highest correlation coefficient with the NDVI occurs for eastern and southern Africa between November and April, the months that cover the main agricultural season in southern Africa or the short rains for countries such as Kenya. During the same period, the correlations in central and western Africa are comparably lower or negative. We find very similar patterns in ARC2 (Fig. B2) with an on average slightly lower correlation coefficient. Also with regard to surface soil moisture (Fig. B3) we observe a similar pattern, whereas in May the correlation coefficient for the eastern regions of southern Africa is comparably higher than for CHIRPS and ARC2. In August, satellite soil moisture shows a slightly positive correlation coefficient with GIMMS NDVI in southern Africa compared to a negative one for CHIRPS rainfall. The lagged correlations of ARC2 with soil moisture (Fig. B5) and of CHIRPS with soil moisture (Fig. B7) look almost identical, again with slightly higher R values for CHIRPS. Also the correlation coefficients of ARC2 (Fig. B6)/CHIRPS (Fig. B8) and the ESI are comparable, with the exception of a lower correlation coefficient in western Africa during December and January for ARC2.
The ESI provides the strongest correlations with lagged NDVI anomalies in southern and eastern Africa (Fig. B4). However, the correlation coefficient during January and April is lower and often negative in large parts of Mauretania, Burkina Faso, Mali, and Niger due to incomplete cloud clearing of the thermal inputs. As in the case of soil moisture, the correlation coefficient is still very high at the end of the agricultural season in countries like Namibia or Botswana in which maize and millet are usually harvested around June. On average, the performance of satellite soil moisture and the ESI over the entire sub-Saharan region is very similar.
b. Results over different climate zones and land cover classes
The correlation analysis for different climate zones (based on the Köppen Geiger climate classification) results in a consistent picture for the northwest, northeast, and southern regions (Figs. C1–C3; see appendix C). With a few exceptions, the correlation coefficient for shifted monthly anomalies of all datasets with the MODIS NDVI (and EVI; not shown) is higher than for the GIMMS NDVI between the years 2001 and 2015. Naturally, we find a vegetation-dependent seasonality in the lagged correlation, for instance in the temperate regions of the northeastern zone (Fig. C2). With regard to individual datasets the ESI ranks among the highest correlations for the northeastern and southern zones. However, in the northwestern area (Fig. C1) it reveals systematic weaknesses for most climate zones (discussed in greater detail in section 6).
Both rainfall datasets (CHIRPS and ARC2) show a distinct temporal pattern in their individual correlation with the GIMMS NDVI and the MODIS NDVI in the southern zone (Fig. C3). In all major climate zones the correlation coefficient of both datasets ranks far lower compared to the other datasets. After the end of the agricultural season (around June) the correlation coefficient drops drastically with a maximum difference in R up to 0.4 (subtropical highland) to all other datasets.
Soil moisture performs best with regard to the lagged correlation results (around R > 0.8) in the humid subtropical zone (northwest) but shows a sharp drop in R at the end of the season (around September/October). Throughout all climate classes in the northwestern region soil moisture shows the highest correlation coefficient with the MODIS NDVI (slightly lower for GIMMS NDVI) averaged over all climate zones. Together with the ESI, the soil moisture dataset shows the highest R over all climate zones in the southern zone and the highest R in the dry semiarid climate zone with the MODIS NDVI (slightly lower for GIMMS NDVI).
The lagged standardized correlations for different [USGS Earth Resources Observation and Science (EROS) version 2.0] land cover classes (Fig. C4) concentrate on the transition zone from dry to wet climates (11°–15°N). In general, they result in a high variability for the correlation coefficient over the “cropland/woodland mosaic” and “dryland, cropland, and pasture” land cover classes. In the case of the dryland, cropland, and pasture class the correlation coefficients are very low or negative until June for both GIMMS and MODIS NDVI. In the cropland/woodland mosaic class only soil moisture reveals comparatively high R values throughout the year for GIMMS, whereas the R values for the ESI are very high (>0.9) from September to October (MODIS NDVI) but far lower (<0.5) before August. The correlation coefficient over all land cover classes exposes a strong seasonality with increasing R values around June.
c. Regional case studies
Figures 7–9 illustrate the anomalies for all satellite-derived variables compared to drought years as reported by farmers. With regard to the detection of drought conditions and the development of WII these figures show two main characteristics. First, the seasonal anomalies in individual datasets tend to have a high “hit rate” for the worst reported drought events. Whenever the seasonal anomaly does not show a negative anomaly for particular drought years we look at monthly anomalies early in the season (e.g., ESI in Tigray, 1984 and 1987). Second, the GIMMS NDVI anomalies miss major drought years (e.g., 2004 in Ethiopia; Fig. 7) or indicate droughts in years that were not identified as drought years by farmers [e.g., 2008 in Ethiopia (Fig. 7) and 2007 in Senegal (Fig. 8)]. These circumstances highlight the need to identify and analyze different independent data sources for the development of robust insurance contracts. Also the additional consideration of indices that estimate vegetation health independently from vegetation greenness, such as the normalized difference water index (Gao 1996), should be considered.
Comparison of standardized seasonal (June–September) anomalies for ARC2 (turquoise), CHIRPS (blue), ESA CCI (orange), and ESI (red) for Tigray (Ethiopia). Anomalies at the beginning of the season (June) are displayed in gray. The seasonal GIMMS NDVI anomalies (green) cover August–December (dark green), September–December, October–December, and November–December (light green); the black rectangles highlight the worst years as reported by farmers; areas shaded in gray highlight drought years; the ESI during 1998/99 is blank due to known retrieval errors in the region.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
As in Fig. 7, but for the Tambacounda region (Senegal).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Comparison of standardized seasonal (November–February) anomalies for ARC2 (turquoise), CHIRPS (blue), ESA CCI (orange), and ESI (red) for the Southern Province (Zambia). The seasonal GIMMS NDVI anomalies (green) cover December–April (dark green), January–April, February–April, and March–April (light green); the black rectangles highlight the worst years as reported by farmers.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
The 2014 drought in Tambacounda serves as a case study to track moisture deficits and the response of vegetation through one particular season. The event had been identified as one of the most severe recent droughts during multiple field visits. Also official sources, such as Famine Early Warning Systems Network (FEWS NET) reports (http://www.fews.net/west-africa/senegal/alert/december-3-2014) confirmed a 45% reduction in cereal and cash crop production compared to the 5-yr average. Tambacounda suffered from yield deficits but was not one of the most severely affected regions. While the NDVI anomaly shows a negative anomaly early in the season (until July) and a positive anomaly at the end of the season, both rainfall and soil moisture show a consistent picture (Fig. 10). They are mostly below the 1992–2015 average throughout the season. Despite the slight moisture deficit at the start of the season, the ESI (bottom right, Fig. 10) increased in August and September. Farmers that participated in the R4 Rural Resilience Initiative (UN World Food Programme and Oxfam) in 2014 received a small WII payout from WII to compensate the early season moisture deficit.
(top left) ARC2, (top right) CHIRPS rainfall, and (bottom left) ESA CCI soil moisture with MODIS NDVI for 2014 (shaded area represents the climatology) and (bottom right) MODIS NDVI vs ESI anomaly for the Tambacounda region.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
6. Discussion
It is common for WII approaches that use satellite data to rely on estimates of rainfall or vegetation health (e.g., Black et al. 2016; Turvey and Mclaurin 2012). However, there is a high potential in the joint analysis of independent satellite-derived variables to characterize drought conditions in different sensitive periods during the agricultural season. Our findings are based on the correlation of lagged standardized anomalies of datasets that allow the tracking of moisture deficits as well as the response of vegetation throughout the agricultural season. Based on the correlation results and a comparison with historical drought years as reported by farmers we highlight the importance of exploiting the convergence of evidence in independent satellite-derived estimations. In addition to rainfall and vegetation greenness estimations we consider satellite-derived soil moisture and evapotranspiration to cover key elements of the hydrologic cycle and therefore better reflect drought conditions on the ground. With regard to a WII-related diagnostic, a better characterization of actual drought conditions means lower basis risk.
Considering a time lag of one month related to the GIMMS NDVI, all satellite datasets show a high agreement with the NDVI from 1992 to 2015 in large parts of Africa. However, there are strong seasonal and spatial variations, partly caused by the growth cycle of green vegetation. In particular from November to April/May all datasets show high correlations with the NDVI in southern and eastern Africa. The correlation results for ARC2 tend to be lower than for CHIRPS. With regard to the ESI and soil moisture they are comparable or higher. The most problematic region for all datasets is western Africa during November and April/May. In the case of the ESI the most likely explanation for the low correlation results is an increased uncertainty in retrieval caused by high cloud cover and low sampling. In the case of the rainfall datasets an explanation could be the low density of GTS in situ stations, which are used for calibration (ARC2) or operational assimilation (CHIRPS).
Soil moisture performs comparably well all year round in southern Africa. Possible reasons for lower correlations in parts of West Africa can be traced back to the individual radar and radiometer datasets that have been merged into the global ESA CCI dataset. Issues related to volume scattering of microwaves in very dry soils (Wagner et al. 2013) are currently under investigation. While the correlation coefficient remains positive in large parts of East Africa despite complex topography, we observe low R values in some East African regions. In most of these regions, which are characterized by generally low vegetation cover, soil moisture retrievals from radiometers are used (Liu et al. 2012). Therefore, the correlation of soil moisture and vegetation greenness is even more susceptible to measurement errors in either dataset.
In a second step we calculate lagged correlations for all datasets over different land covers and climate zones and for two different NDVI datasets (MODIS and GIMMS). The results are consistent for 2001–2015 for virtually all classes with a higher correlation coefficient for the MODIS NDVI than for the GIMMS NDVI. Again, we observe a strong seasonality caused by the natural progression of vegetation. Both ESI and soil moisture are more closely related to near-future vegetation greenness than CHIRPS and ARC2 during large parts of the agricultural season in southern Africa. This does not only indicate a high predictive skill related to vegetation health, but also their added value as additional sources of information for WII.
The case studies for Senegal, Ethiopia, and Zambia demonstrate a generally high agreement of all satellite-derived anomalies with regard to the detection of severe drought events as mentioned by farmers. We manage to track all severe droughts via seasonal or early season (monthly) anomalies. Moderate drought events, however, are not consistently detected by all satellite variables, which highlights the importance of understanding and considering multiple independent data sources. With regard to one particular drought event that struck Senegal in 2014 (Fig. 10), we show how the severity of anomalous conditions can be tracked by analyzing independently observed seasonal time series, climatologies, and corresponding anomalies to better reflect actual drought conditions on the ground.
7. Summary and conclusions
Weather insurance indices that rely exclusively on rainfall ignore the degree to which atmospherically provided moisture is available to and transpired by plants. Gaps or errors in the representation of drought conditions via satellite data potentially mean a higher basis risk for WII. Thus, the main objective of this study is to investigate the agreement of independent satellite-derived key variables of the hydrologic cycle with vegetation greenness (NDVI) as a proxy for the vegetative response to plant available water. Based on an extensive correlation analysis for the entire African continent and regional case studies we highlight the added value of considering multiple independent satellite observations to design, calibrate, and validate weather insurance indices. The findings will have a direct impact on global weather index insurance activities, carried out for instance at Columbia University’s IRI.
Datasets used in this study include satellite-derived rainfall (CHIRPS and ARC2), soil moisture (ESA CCI), evaporative stress (ESI), and vegetation greenness (GIMMS/MODIS NDVI). We minimize comparative advantages with regard to differences in spatial and temporal resolution by resampling all datasets to the same spatial resolution (0.25°) and interval (monthly averages). Some datasets might outperform others if used in their native spatial resolutions, but this study concentrates on medium- to large-scale spatiotemporal trends and agreements rather than their performance on farm level. In addition, we only calculate the correlation coefficient of lagged standardized anomalies for pixels that provide information for all datasets. This way, datasets that are extensively masked for regions with high retrieval uncertainty (e.g., soil moisture) do not automatically result in higher correlation coefficients.
Our findings can be summarized as follows. First, there is a high agreement of satellite-derived soil moisture and the evaporative stress index (ESI) with the NDVI, which is higher than for both rainfall datasets over large geographic regions. We suggest that the design and calibration of weather insurance indices could benefit from information about soil moisture conditions, evaporative stress or a combination of both. With regard to a combination of both datasets, however, more research is needed to methodically use each source of information where it performs best. Exploiting the physical interrelationship between independent satellite datasets to track agricultural drought conditions, which are often not fully explained by rainfall anomalies, also supports the historical calibration of indices via cross-validation. In addition, this approach facilitates the development of a “conclusive seasonal drought narrative,” limiting basis risk, and the identification of problematic regions based on the datasets’ disagreement.
If the strengths and limitations of each dataset are known, the additional consideration of evapotranspiration and soil moisture, whose potential added value for WII has already been described in a regional study for Senegal (Enenkel et al. 2017), allows a better characterization of the conditions on the ground. In particular in southern Africa, the correlation coefficients for the lagged standardized anomalies of all datasets with the NDVI are comparatively high. We find the lowest correlation results in large parts of western Africa, which are likely caused by physical limitations such as cloud cover (ESI) or volume scattering (soil moisture) in very dry soils. The correlation results for ARC2, which is often used operationally in WII, tend to be lower than for CHIRPS.
Second, we find generally higher R values for lagged correlations with the MODIS NDVI than with the GIMMS NDVI for different land cover classes and climate zones. This is likely due to the higher native spatial resolution and better-quality mapping of MODIS (250 m) compared to AVHRR (8 km). Nevertheless, the GIMMS NDVI is a useful temporal reference for periods prior to the availability of datasets from MODIS (available since 2001). All correlation results presented in this study are based on the understanding that the NDVI is a proxy for vegetation health but cannot be directly translated into yield estimates. This would require the development of a regression model (e.g., Lopresti et al. 2015) or a machine learning approach (Panda et al. 2010).
Third, comparing satellite-derived anomalies to drought years assessed via field visits is not a straightforward task. It requires an in-depth understanding of possible cognitive bias as well as a careful subseasonal analysis to identify how and when the drought affected vegetation health. This could be a rainfall deficit, but also high rates of evapotranspiration that result in soil moisture depletion and subsequent wilting. In addition to an overall dry season, it could be a deficit early in the season that is compensated late in the season or a strong season onset resulting in an end of season deficit that affects the reproductive stage of staple crops. Again, evapotranspiration and soil moisture can fill sensitive gaps by providing vital additional evidence for moisture deficits in different parts of the hydrologic cycle. Our results indicate that seasonal anomalies show a good agreement with severe drought years as reported by farmers, but less consistent results for moderate drought years. We recommend running logistic regressions for reported bad years and subseasonal satellite-derived variables to identify their skill for improved index calibration as well as for the selection of insurance windows.
Fourth, it is vital that developers of weather insurance indices build up capacities to access, modify, analyze, and use satellite data and collaborate with data providers to stay up to date with algorithmic developments. Satellite-derived datasets are often based on a series of sensors that need to be intercalibrated and therefore reprocessed. New studies might reveal errors related to physical or algorithmic limitations. Also a “better” product might become available. New satellite instruments are continuously coming online and older systems are retired. This necessitates an agile environment capable of transitioning from one system to another to continue receiving up-to-date information.
Finally, the operational availability of complementary satellite-derived datasets might become an additional motivation to rely on satellite data for WII, which could promote its upscaling for covariate risks. In addition, the opportunity to track an index during the season can support the operational planning and preparedness of humanitarian aid organizations, whose emergency appeal system is often too slow to prevent distress sales of assets, which can lead to a downward spiral with regard to chronic poverty (Hellmuth et al. 2009). In both cases it will be necessary to support the validation efforts via new technologies (Enenkel et al. 2014), such as low-cost in situ sensors or incentivized/automated reporting via mobile phones.
Acknowledgments
This study was funded within the NASA Interdisciplinary Science Program (Award NNX14AD63G).
APPENDIX A
Drought Years
Table A1 lists drought years identified via multiple field visits.
Drought years identified (indicated with a boldface “Yes”) via multiple field visits (in the case of the Southern Province in Zambia, where drought years last from November to April, the 1983/84 season is listed as 1983, etc.). The worst drought years are indicated with asterisks.
APPENDIX B
Monthly Lagged Correlation Coefficients
Figures B1–B8 illustrate the spatial agreement of all variables via monthly lagged Pearson’s correlation coefficients across Africa for 1992–2015.
Monthly lagged Pearson’s correlation coefficient for CHIRPS rainfall and GIMMS NDVI, covering the entire African continent (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for ARC2 rainfall and GIMMS NDVI, covering the entire African continent (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for ESA CCI soil moisture and GIMMS NDVI over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for ESI and the GIMMS NDVI over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for ARC2 and the ESA CCI soil moisture over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for ARC2 and the ESI over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for CHIRPS and the ESA CCI soil moisture over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Monthly lagged Pearson’s correlation coefficient for CHIRPS and the ESI over Africa (1992–2015).
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
APPENDIX C
Correlation Coefficients for Shifted Monthly Anomalies
Figures C1–C4 illustrate the shifted monthly correlations of standardized anomalies from 2001–15 for different climate zones and land cover classes.
Shifted monthly correlation of standardized anomalies of CHIRPS and ARC2 rainfall, ESA CCI soil moisture, and the ESI with GIMMS/MODIS vegetation greenness from 2001 to 2015, masked for the dominant climate zones in the northwestern zone and based on a 1-month lag for vegetation greenness.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Shifted monthly correlation of standardized anomalies of CHIRPS and ARC2 rainfall, ESA CCI soil moisture, and the ESI with GIMMS/MODIS vegetation greenness from 2001 to 2015, masked for the dominant climate zones in the northeastern zone and based on a one-month lag for vegetation greenness.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Shifted monthly correlation of standardized anomalies of CHIRPS and ARC2 rainfall, ESA CCI soil moisture, and the ESI with GIMMS/MODIS vegetation greenness from 2001 to 2015, masked for the dominant climate zones in the southern zone.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
Shifted monthly correlation of standardized anomalies of CHIRPS and ARC2 rainfall, ESA CCI soil moisture, and the ESI with (left) GIMMS and (right) MODIS vegetation greenness from 2001 to 2015, masked for the dominant land cover classes and based on a 1-month lag for vegetation greenness.
Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0111.1
REFERENCES
Ahmed, M., B. Else, L. Eklundh, J. Ardö, and J. Seaquist, 2017: Dynamic response of NDVI to soil moisture variations during different hydrological regimes in the Sahel region. Int. J. Remote Sens., 38, 5408–5429, https://doi.org/10.1080/01431161.2017.1339920.
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration—Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 15 pp., www.fao.org/docrep/X0490E/X0490E00.htm.
Anderson, M. C., J. M. Norman, G. R. Diak, W. P. Kustas, and J. R. Micikalski, 1997: A two-source time-integrated model for estimating surface fluxes using thermal infrared remote sensing. Remote Sens. Environ., 60, 195–216, https://doi.org/10.1016/S0034-4257(96)00215-5.
Anderson, M. C., J. M. Norman, J. R. Mecikalski, J. A. Otkin, and W. P. Kustas, 2007: A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation. J. Geophys. Res., 112, D10117, https://doi.org/10.1029/2006JD007506.
Anderson, M. C., C. Hain, B. Wardlow, A. Pimstein, J. Mecikalski, and W. Kustas, 2011: Evaluation of drought indices based on thermal remote sensing of evapotranspiration over the continental United States. J. Climate, 24, 2025–2044, https://doi.org/10.1175/2010JCLI3812.1.
Anderson, M. C., C. Hain, J. Otkin, X. Zhan, K. Mo, M. Svoboda, B. Wardlow, and A. Pimstein, 2013: An intercomparison of drought indicators based on thermal remote sensing and NLDAS-2 simulations with U.S. Drought Monitor classifications. J. Hydrometeor., 14, 1035–1056, https://doi.org/10.1175/JHM-D-12-0140.1.
Anderson, M. C., C. A. Zolin, C. R. Hain, K. Semmens, M. Tugrul Yilmaz, and F. Gao, 2015: Comparison of satellite-derived LAI and precipitation anomalies over Brazil with a thermal infrared-based Evaporative Stress Index for 2003–2013. J. Hydrol., 526, 287–302, https://doi.org/10.1016/j.jhydrol.2015.01.005.
Anderson, M. C., and Coauthors, 2016: The Evaporative Stress Index as an indicator of agricultural drought in Brazil: An assessment based on crop yield impacts. Remote Sens. Environ., 174, 82–99, https://doi.org/10.1016/j.rse.2015.11.034.
Balaghi, R., B. Tychon, H. Eerens, and M. Jlibene, 2008: Empirical regression models using NDVI, rainfall and temperature data for the early prediction of wheat grain yields in Morocco. Int. J. Appl. Earth Obs. Geoinf., 10, 438–452, https://doi.org/10.1016/j.jag.2006.12.001.
Bitew, M. M., and M. Gebremichael, 2010: Evaluation through independent measurements: Complex terrain and humid tropical region in Ethiopia. Satellite Rainfall Applications for Surface Hydrology, M. Gebremichael and F. Hossain, Eds., Springer, 205–214, https://doi.org/10.1007/978-90-481-2915-7_12.
Black, E., E. Tarnavsky, R. Maidment, H. Greatrex, A. Mookerjee, T. Quaife, and M. Brown, 2016: The use of remotely sensed rainfall for managing drought risk: A case study of weather index insurance in Zambia. Remote Sens., 8, 342, https://doi.org/10.3390/rs8040342.
Claverie, M., and Coauthors, 2014: NOAA climate data record (CDR) of leaf area index (LAI) and fraction of absorbed photosynthetically active radiation (FAPAR), version 4, NOAA/NCEI, accessed 12 January 2018, https://doi.org/10.7289/V5M043BX.
Clover, J., 2003: Food security in sub-Saharan Africa. Afr. Secur. Rev., 12, 5–15, https://doi.org/10.1080/10246029.2003.9627566.
Diem, J. E., J. Hartter, S. J. Ryan, and M. W. Palace, 2014: Validation of satellite rainfall products for western Uganda. J. Hydrometeor., 15, 2030–2038, https://doi.org/10.1175/JHM-D-13-0193.1.
Dinku, T., S. Chidzambwa, P. Ceccato, S. J. Connor, and C. F. Ropelewski, 2008: Validation of high-resolution satellite rainfall products over complex terrain. Int. J. Remote Sens., 29, 4097–4110, https://doi.org/10.1080/01431160701772526.
Diouf, I., and Coauthors, 2017: Comparison of malaria simulations driven by meteorological observations and reanalysis products in Senegal. Int. J. Environ. Res. Public Health, 14, 1119, https://doi.org/10.3390/ijerph14101119.
Dixon, J., A. Gulliver, and D. Gibbon, 2001: Farming Systems and Poverty: Improving Farmers’ Livelihoods in a Changing World. FAO?and World Bank, 412 pp.
Dorigo, W., and R. de Jeu, 2016: Satellite soil moisture for advancing our understanding of Earth system processes and climate change. Int. J. Appl. Earth Obs. Geoinf., 48, 1–4, https://doi.org/10.1016/j.jag.2016.02.007.
Dorigo, W., R. de Jeu, D. Chung, R. Parinussa, Y. Liu, W. Wagner, and D. Fernández-Prieto, 2012: Evaluating global trends (1988–2010) in harmonized multi-satellite surface soil moisture. Geophys. Res. Lett., 39, L18405, https://doi.org/10.1029/2012GL052988.
Dorigo, W., and Coauthors, 2017: ESA CCI soil moisture for improved Earth system understanding: State-of-the art and future directions. Remote Sens. Environ., 203, 185–215, https://doi.org/10.1016/j.rse.2017.07.001.
Enenkel, M., and Coauthors, 2014: Drought and food security—Improving decision-support via new technologies and innovative collaboration. Global Food Secur., 4, 51–55.
Enenkel, M., C. Reimer, W. Dorigo, W. Wagner, I. Pfeil, R. Parinussa, and R. de Jeu, 2016: Combining satellite observations to develop a daily global soil moisture product for near-real-time applications. Hydrol. Earth Syst. Sci., 20, 4191–4208, https://doi.org/10.5194/hess-20-4191-2016.
Enenkel, M., D. Osgood, and B. Powell, 2017: The added value of satellite soil moisture for agricultural index insurance. Remote Sensing of Hydrometeorological Hazards, G. P. Petropoulos and T. Islam, Eds., CRC Press, 69–81.
FAO, 2009: How to Feed the World in 2050: Report from the High-Level Expert Forum. Food and Agriculture Organization of the United Nations, 35 pp.
FAO, 2015: The state of food insecurity in the world—Meeting the 2015 international hunger targets: Taking stock of uneven progress, UN Food and Agriculture Organization, 58 pp., www.fao.org/3/a-i4646e.pdf.
Funk, C., and M. E. Brown, 2009: Declining global per capita agricultural production and warming oceans threaten food security. Food Secur., 1, 271–289, https://doi.org/10.1007/s12571-009-0026-y.
Funk, C., and Coauthors, 2015: The Climate Hazards Infrared Precipitation with Stations—A new environmental record for monitoring extremes. Sci. Data, 2, 150066, https://doi.org/10.1038/sdata.2015.66.
Gao, B., 1996: NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ., 58, 257–266, https://doi.org/10.1016/S0034-4257(96)00067-3.
Godfray, H. C. J., and Coauthors, 2010: Food security: The challenge of feeding 9 billion people. Science, 327, 812–818, https://doi.org/10.1126/science.1185383.
Hain, C. R., J. R. Mecikalski, and M. C. Anderson, 2009: Retrieval of an available water-based soil moisture proxy from thermal infrared remote sensing. Part I: Methodology and validation. J. Hydrometeor., 10, 665–683, https://doi.org/10.1175/2008JHM1024.1.
Hain, C. R., W. T. Crow, J. R. Mecikalski, M. C. Anderson, and T. Holmes, 2011: An intercomparison of available soil moisture estimates from thermal infrared and passive microwave remote sensing and land surface modeling. J. Geophys. Res., 116, D15107, https://doi.org/10.1029/2011JD015633.
Hain, C. R., W. T. Crow, M. C. Anderson, and J. R. Mecikalski, 2012: An ensemble Kalman filter dual assimilation of thermal infrared and microwave satellite observations of soil moisture into the Noah land surface model. Water Resour. Res., 48, W11517, https://doi.org/10.1029/2011WR011268.
Hellmuth, M. E., D. E. Osgood, U. Hess, A. Moorhead, and H. Bhojwani, Eds., 2009: Index insurance and climate risk: Prospects for development and disaster management. International Research Institute for Climate and Society, 112 pp., https://cgspace.cgiar.org/handle/10568/932.
Hessels, T. M., 2015: Comparison and validation of several open access remotely sensed rainfall products for the Nile Basin. Master’s thesis, Geoscience & Engineering, Delft University of Technology, 233 pp.
Huete, A., K. Didan, W. van Leeuwen, T. Miura, and E. Glenn, 2010: MODIS vegetation indices. Land Remote Sensing and Global Environmental Change, B. Ramachandran, C. O. Justice, and M. J. Abrams, Eds., Springer, 579–602, https://doi.org/10.1007/978-1-4419-6749-7_26.
IFAD, 2011: Weather index-based insurance in agricultural development: A technical guide. International Fund for Agricultural Development, 66 pp, https://www.wfp.org/content/weather-index-based-insurance-agricultural-development-technical-guide.
IPCC, 2012: Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. Cambridge University Press, 582 pp.
IRI, 2013: Using Satellites to Make Index Insurance Scalable: Final IRI Report to the International Labour Organisation–Microinsurance Innovation Facility, International Research Institute for Climate and Society, Columbia University, 60 pp.
Katsanos, D., A. Retalis, and S. Michaelides, 2016: Validation of a high-resolution precipitation database (CHIRPS) over Cyprus for a 30-year period. Atmos. Res., 169B, 459–464, https://doi.org/10.1016/j.atmosres.2015.05.015.
Kogan, F. N., 1995: Droughts of the late 1980s in the United States as derived from NOAA polar-orbiting satellite data. Bull. Amer. Meteor. Soc., 76, 655–668, https://doi.org/10.1175/1520-0477(1995)076<0655:DOTLIT>2.0.CO;2.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140, https://doi.org/10.1126/science.1100217.
Licker, R., M. Johnston, J. A. Foley, C. Barford, C. J. Kucharik, C. Monfreda, and N. Ramankutty, 2010: Mind the gap: How do climate and agricultural management explain the ‘yield gap’ of croplands around the world? Global Ecol. Biogeogr., 19, 769–782, https://doi.org/10.1111/j.1466-8238.2010.00563.x.
Liu, Y. Y., R. M. Parinussa, W. A. Dorigo, R. A. M. de Jeu, W. Wagner, A. I. J. M. van Dijk, M. F. McCabe, and J. P. Evans, 2011: Developing an improved soil moisture dataset by blending passive and active microwave satellite-based retrievals. Hydrol. Earth Syst. Sci., 15, 425–436, https://doi.org/10.5194/hess-15-425-2011.
Liu, Y. Y., W. A. Dorigo, R. M. Parinussa, R. A. M. de Jeu, W. Wagner, M. F. McCabe, J. P. Evans, and A. I. J. M. van Dijk, 2012: Trend-preserving blending of passive and active microwave soil moisture retrievals. Remote Sens. Environ., 123, 280–297, https://doi.org/10.1016/j.rse.2012.03.014.
Lopresti, M. F., C. M. Di Bella, and A. J. Degioanni, 2015: Relationship between MODIS-NDVI data and wheat yield: A case study in northern Buenos Aires province, Argentina. Inf. Process. Agric., 2, 73–84, https://doi.org/10.1016/j.inpa.2015.06.001.
Maidment, R. I., D. I. F. Grimes, R. P. Allan, H. Greatrex, O. Rojas, and O. Leo, 2013: Evaluation of satellite-based and model re-analysis rainfall estimates for Uganda. Meteor. Appl., 20, 308–317, https://doi.org/10.1002/met.1283.
Maidment, R. I., and Coauthors, 2017: A new, long-term daily satellite-based rainfall dataset for operational monitoring in Africa. Sci. Data, 4, 170063, https://doi.org/10.1038/sdata.2017.63.
McNally, A., S. Shukla, K. R. Arsenault, S. Wang, S., C. D. Peters-Lidard, and J. P. Verdin, 2016: Evaluating ESA CCI soil moisture in East Africa. Int. J. Appl. Earth Obs. Geoinf., 48, 96–109, https://doi.org/10.1016/j.jag.2016.01.001.
Mills, E., 2007: Synergisms between climate change mitigation and adaptation: An insurance perspective. Mitig. Adapt. Strategies Global Change, 12, 809–842, https://doi.org/10.1007/s11027-007-9101-x.
Miralles, D. G., M. J. van den Berg, A. J. Teuling, and R. M. de Jeu, 2012: Soil moisture–temperature coupling: A multiscale observational analysis. Geophys. Res. Lett., 39, L21707, https://doi.org/10.1029/2012GL053703.
Mueller, N. D., J. S. Gerber, M. Johnston, D. K. Ray, N. Ramankutty, and J. A. Foley, 2012: Closing yield gaps through nutrient and water management. Nature, 490, 254–257, https://doi.org/10.1038/nature11420.
Novella, N. S., and W. M. Thiaw, 2013: African rainfall climatology version 2 for Famine Early Warning Systems. J. Appl. Meteor. Climatol., 52, 588–606, https://doi.org/10.1175/JAMC-D-11-0238.1.
Otkin, J. A., M. C. Anderson, C. Hain, I. E. Mladenova, J. B. Basara, and M. Svoboda, 2013: Examining rapid onset drought development using the thermal infrared–based evaporative stress index. J. Hydrometeor., 14, 1057–1074, https://doi.org/10.1175/JHM-D-12-0144.1.
Otkin, J. A., M. C. Anderson, C. Hain, and M. Svoboda, 2015: Using temporal changes in drought indices to generate probabilistic drought intensification forecasts. J. Hydrometeor., 16, 88–105, https://doi.org/10.1175/JHM-D-14-0064.1.
Ouni, S., and D. Dinh, 2017: Prototype tool can help maintain farmers’ involvement in index insurance projects. Consultative Group for International Agricultural Research (CGIAR) Research Program on Climate Change, Agriculture and Food Security (CCAFS), accessed 22 February 2017, https://ccafs.cgiar.org/blog/prototype-tool-can-help-maintain-farmers%E2%80%99-involvement-index-insurance-projects.
Panda, S. S., D. P. Ames, and S. Panigrahi, 2010: Application of vegetation indices for agricultural crop yield prediction using neural network techniques. Remote Sens., 2, 673–696, https://doi.org/10.3390/rs2030673.
Qiu, J., W. T. Crow, G. S. Nearing, X. Mo, and S. Liu, 2014: The impact of vertical measurement depth on the information content of soil moisture times series data. Geophys. Res. Lett., 41, 4997–5004, https://doi.org/10.1002/2014GL060017.
Rosenzweig, M. R., and K. I. Wolpin, 1993: Credit market constraints, consumption smoothing, and the accumulation of durable production assets in low-income countries: Investments in bullocks in India. J. Polit. Econ., 101, 223–244, https://doi.org/10.1086/261874.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, https://doi.org/10.1175/2010BAMS3001.1.
Turvey, C. G., and M. K. Mclaurin, 2012: Applicability of the Normalized Difference Vegetation Index (NDVI) in index-based crop insurance design. Wea. Climate Soc., 4, 271–284, https://doi.org/10.1175/WCAS-D-11-00059.1.
Wagner, W., G. Lemoine, and H. Rott, 1999: A method for estimating soil moisture from ERS scatterometer and soil data. Remote Sens. Environ., 70, 191–207, https://doi.org/10.1016/S0034-4257(99)00036-X.
Wagner, W., W. Dorigo, R. de Jeu, D. Fernandez, J. Benveniste, E. Haas, and M. Ertl, 2012: Fusion of active and passive microwave observations to create an essential climate variable data record on soil moisture. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci., I–7, 315–321.
Wagner, W., and Coauthors, 2013: The ASCAT soil moisture product: A review of its specifications, validation results, and emerging applications. Meteor. Z., 22, 5–33, https://doi.org/10.1127/0941-2948/2013/0399.
Wan, Z., S. Hook, and G. Hulley, 2015: MYD11C1 MODIS/Aqua Land Surface Temperature/Emissivity Daily L3 Global 0.05Deg CMG V006. NASA LP DAAC, https://doi.org/10.5067/MODIS/MYD11C1.006.
Wani, S. P., J. Rockström, and T. Y. Oweis, Eds., 2009: Rainfed Agriculture: Unlocking the Potential. CABI, 328 pp., https://www.cabi.org/cabebooks/ebook/20093041389.