Monitoring individual rice field flooding dynamics over large scales to improve mosquito surveillance and control

doi:10.21203/rs.3.rs-5181799/v1

Download PDF

Research Article

Monitoring individual rice field flooding dynamics over large scales to improve mosquito surveillance and control

https://doi.org/10.21203/rs.3.rs-5181799/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Progress in malaria elimination has been hindered by recent changes in mosquito behaviour and increased insecticide resistance in response to traditional vector control measures, such as indoor residual spraying and long-lasting insecticide-treated nets. There is therefore increasing interest in the use of larval source management (LSM) to supplement current insecticide-based interventions. However, LSM implementation requires the characterization of larval habitats at fine spatial and temporal scales to ensure interventions are well-placed and well-timed. Remotely sensed optical imagery captured via drones or satellites offers one way to monitor larval habitats remotely, but its use at large spatio-temporal scales has important limitations. Here, we propose a method using radar imagery to monitor flooding dynamics in individual rice fields, a primary larval habitat, over very large geographic areas relevant to national malaria control programs aiming to implement LSM at scale. We demonstrate this for a 3,971 km² malaria-endemic district in Madagascar with over 17,000 rice fields. We combined rice field mapping on OpenStreetMap with Sentinel-1 satellite imagery (radar, 10m) from 2016–2022 to train a classification model of radar backscatter to identify rice fields with vegetated and open water, resulting in a time-series of weekly flooding dynamics for thousands of rice fields. From these time-series, we obtained over a dozen indicators for each rice field useful for LSM implementation, such as the timing and frequency of flooding seasons. These monitoring tools were integrated into an interactive GIS dashboard for operational use by vector control programs, with results available at multiple scales (district, sub-district, rice field) relevant for different phases of LSM intervention (e.g. prioritization of sites, implementation, follow-up). Scale-up of these methods could enable wider implementation of evidence-based LSM interventions and reduce malaria burdens in contexts where irrigated agriculture is a major transmission driver.

health surveillance system

OpenStreetMap mapping

synthetic aperture radar

larval source management

decision-making tools

malaria

Sustained progress in the global fight against malaria has been achieved in recent decades. A keystone of this effort has been the scale-up of insecticide-based vector control measures, such as indoor residual spraying and long-lasting insecticide-treated nets in malaria-endemic countries. However, global progress in malaria elimination has been hindered in recent years, with the annual number of cases reaching 249 million in 2022, an increase of 18 million from 2015 [1]. This slowing of progress has been attributed to changes in mosquito behaviour and the spread of insecticide resistance, both of which could further threaten future progress by reducing the effectiveness of vector control measures. For instance, diurnal and outdoor biting can contribute to a substantial fraction of overall transmission [2, 3], and a majority of endemic countries has already reported mosquito resistance to commonly used insecticides such as pyrethroids, organochlorines, carbamates and organophosphates [1]. As a result, there is increasing interest in the use of larval source management (LSM) to supplement current vector control interventions, especially in settings of moderate to low transmission and in high transmission settings with high rates of insecticide resistance or outdoor biting [4].

LSM interventions could be especially well-suited for sub-Saharan Africa, a region that accounts for over 90% of malaria cases globally [1]. LSM interventions are particularly feasible and cost-effective in settings where habitats are ‘few, fixed and findable’ according to the WHO, which recommends environmental management (i.e. habitat modification and manipulation) as the primary LSM intervention whenever possible, and the regular application of biological or chemical larvicides in habitats that cannot be removed or modified [4]. In particular, there has been an important focus on LSM interventions targeting rice fields, where the use of larvicides, the introduction of fish and intermittent irrigation practices have all proven to be effective in reducing larval abundance [5, 6]. Demand for rice has surged in Africa in recent decades, and the area of cultivated rice has more than doubled in the last 20 years [7]. Aquatic conditions in rice fields are particularly favourable to the breeding of some of the most important malaria vectors in Africa, such as Anopheles funestus and the species complex Anopheles gambiae sensu lato [8], and the abundance of adult mosquitoes is often higher near rice-growing areas [5]. A recent meta-analysis focused on the past two decades found that rice-growing communities in Africa have experienced significantly higher malaria prevalence levels than communities that do not grow rice [9], further evidence for the importance of these water bodies as larval habitat. However, for LSM interventions to target rice fields and other larval habitats, their implementation requires an accurate characterization of these breeding sites to ensure interventions are well-placed and well-timed [4, 10].

The use of remote sensing for the environmental monitoring of rice fields and water surfaces is increasingly common in agriculture and vector surveillance thanks to the growing availability of data from earth observation satellites and drones [11]. One technique consists of generating spectral indices from optical satellites with relatively high spatio-temporal resolution (e.g. Sentinel-2, LandSat, MODIS), but these indices require field validation to be sufficiently accurate [12, 13]. In addition, the quality and availability of optical images depends on cloud cover, which can have a major impact on the signal and therefore on the number of usable images, especially in humid tropical areas during rainy seasons, where malaria transmission is often highest. Drones can generate spatial information at high resolutions (under 10cm) and at temporal intervals determined by the user, and can be of particular interest in settings already using drones for LSM interventions (e.g. application of larvicides) [14–17]. Unfortunately, in addition to implementation barriers such as community acceptance, national regulations and costs, drones can only cover small areas, which prevents their routine use at scale [16].

Synthetic aperture radar (SAR) data are better suited to detect changes in surface water over large geographic areas regardless of weather and light conditions [18, 19]. However, the resolution of freely available SAR data from Sentinel-1 satellites can be insufficient to detect small breeding sites and to correctly classify particular land cover features of interest, such as rice fields [16, 19]. To build decision making tools that can enable and guide the widespread implementation of LSM interventions targeting larval habitats in rice fields over larger areas (such as districts or provinces), complementary approaches are needed for larval habitat monitoring that leverage imagery at very high spatial resolution for breeding site identification and SAR data for temporal monitoring of flooding dynamics.

The goal of this study was to develop and validate a method that allows for the longitudinal monitoring of flooding dynamics in individual rice fields over very large geographic areas relevant to national malaria control programs aiming to implement LSM at scale. We demonstrate this for a 3,971 km² malaria-endemic district in Madagascar, a major rice producing country where malaria incidence is increasing and where LSM interventions targeting rice fields are being piloted [20]. We applied a field-validated classification model to radar imagery to characterize rice field dynamics from 2016–2022 for over 17,000 rice fields. These dynamics were summarized into indicators representing seasonality and made available via an online application. Scale-up of these methods could enable wider implementation of evidence-based LSM interventions and reduce malaria burdens in contexts where irrigated agriculture is a major transmission driver.

Study area

This study was conducted in Ifanadiana, a rural district located in the Vatovavy region of southeastern Madagascar (See Supplementary Fig. 1, Additional File 1). The district covers an area of approximately 3,970 km², receives an annual precipitation of 1,200 mm, has a humid tropical climate and a population of approximately 200,000 inhabitants. The district consists of 15 communes (administrative units with an average of 13,000 inhabitants) and 195 fokontany (local administrative units consisting of several villages with an average of about 1,000 inhabitants). Ifanadiana is characterized by its geographical diversity, including mountains, valleys with water sources, rivers and dense, very humid vegetation, ideal for agriculture. The district is predominantly rural, with an economy dominated by agriculture, primarily rice cultivation. Rice cultivation practices in the district vary according to the geographical context, resulting in a diversity of cultivation seasons. Some farmers plant rice once a year, while others plant rice twice a year. This variation is influenced by irrigation availability and access to water, critical factors for understanding how seasonal differences contribute to the diversity in the spatio-temporal dynamics of rice field flooding across the district.

Ifanadiana district is malaria endemic, with an estimated annual incidence of approximately 500 cases per 1,000 inhabitants after accounting for underreporting and prevalence rates as high as 80% during the malaria season [21]. The district exhibits seasonal transmission dynamics linked to climatic patterns, with a low transmission season between June and November and a high transmission season between December and May. Previous studies have shown that rice fields favor malaria transmission, in combination with other climatic and socioeconomic factors in the district [22]. Therefore, we aimed to understand the flooding dynamics of all rice fields in the district in order to contribute to precise and targeted malaria control efforts.

Identification of rice field flooding thresholds using field data and Sentinel-1 images

Monitoring the spatio-temporal dynamics of rice field flooding is based on three fundamental steps: field data collection, exploitation of SAR satellite data, and determination of optimal thresholds to classify different water classes (Fig. 1). First, field observations were conducted to obtain temporal and spatial representativity of rice field flooding dynamics (See Supplementary Fig. 1, Additional File 1). A temporal sampling campaign was conducted monthly on 7 rice fields between February 2021 and February 2022 (Table 1). A spatial sampling campaign was conducted in the dry season (October 17–21, 2022) and the rainy season (February 27-March 3, 2023), covering 54 rice fields in the district (Table 1). These visits, conducted by a team of 5 people, aimed to collect representative samples of three categories of flooding dynamics ("open water", "vegetated water", and "no water"), and to collect information on rice plant height, water depth, rice field characteristics, observation landmarks and their orientation, and observed water class types. We used the NextGIS Mobile (version 2.6.49) to geolocate each rice field and store information and photos directly in the attributes of the geographic features [23]. These data allowed us to manually classify pixels in each rice field that corresponded to the three water class categories to use to train the classification model.

Second, SAR satellite data were obtained on these field-monitored rice fields. This technology is particularly suitable for monitoring the spatio-temporal dynamics of flooding in rice fields due to its ability to penetrate clouds and operate day and night [24, 25]. Among the different polarizations used by SARs, the VH (Vertical transmit and Horizontal receive) polarization of Sentinel-1 imagery is particularly relevant for detecting the presence of water [26]. We used Sentinel-1 SAR data at the ARD (Analysis Ready Data) level available on the Google Earth Engine platform [27]. These data have already undergone the major preprocessing steps: orbital corrections for precise localization, radiometric corrections to convert raw values into usable backscatter coefficients and remove noise effects, and terrain geocoding to project images onto a cartographic grid that accounts for topography and accurately represents ground geometry.

Third, using data collected in the previous two phases, optimal thresholds of VH polarization values were determined to distinguish between the three categories of flooding: "open water", "vegetated water", and "no water". Our approach began with data preparation, which included georeferencing and vectorizing observed plots as shapefiles, where each plot corresponded to an observation point containing a water class based on field observations (Fig. 1A). We then processed Sentinel-1 imagery acquired on the date closest to the field visits (range of 0–14 days difference), identifying pixels corresponding to each plot and extracting VH polarization values for each field-collected water class. To identify the best discriminative thresholds, we used two different ROC curves: one to discriminate "open water" from "vegetated water", and another to discriminate "vegetated water" from "no water". To ensure robustness and generalizability of our thresholds, we spatially stratified the data, using data around the EAST-WEST axis as training data (77%) and those from the NORTH-SOUTH axis (23%) as validation data, allowing us to calculate the sensitivity and specificity for each threshold (See Supplementary Fig. 2, Additional File 1).

Table 1

Number of field observations used to train the classification model.
	Number of rice fields	No. Observation points (photos)	No. Polygons
Temporal sample
Monthly (February 2021- February 2022)	7	63	89
Spatial sample
Dry season (October 17 2022- October 21 2022)	54	203	222
Wet season (27 February 2023–02 March 2023)	54	181	204

Estimation of individual rice field flooding dynamics in Ifanadiana district

We applied the classification models and thresholds previously obtained to SAR imagery corresponding to all rice fields within the district from January 2016 to December 2022 (Fig. 1B). To do so, we first identified all of the rice fields, buildings, and administrative boundaries in the district via a participatory mapping exercise on the OpenStreetMap (OSM) platform in 2019. Full details of this mapping exercise are reported in [28]. We downloaded this data using the QGIS QuickOSM extension [29] [August 23 2023], obtaining the boundaries of 17,321 rice fields and over 100,000 buildings. We also obtained the administrative boundaries of the district's 15 communes and 195 fokontany [30], as well as precipitation data for the duration of the study from Africa Rainfall Climate 2 [31], which was used to calculate the daily precipitation average for the district.

We used Sentinel-1 images collected every 12 days from January 01 2016 - December 23 2022 at the same ARD processing as described above. For each Sentinel-1 image, we filtered the pixels located fully-within the rice fields mapped on OSM and classified them into the same three water categories by applying the optimal thresholds determined by the classification model. Finally, we calculated the areas and proportions of coverage of each water class in each rice field for each acquisition date of the available Sentinel-1 images.

We applied smoothing to the resulting time series to reduce the influence of stochastic noise and erroneous data points. For each rice field, we compared four smoothing techniques: cubic-spline, polynomial, basis-spline, and natural spline. For cubic-spline, basis-spline, and natural-spline techniques, we limited the degrees of freedom to 4 per year, to allow the smooths to fit the seasonality of the data without overfitting to short-term trends. For the polynomial technique, we reduced the degrees of freedom to 2 per year due to convergence issues at higher values. For each rice field, we chose the smoothing technique that resulted in the lowest residual sum of squares via leave-one-out cross validation, and used the resulting model to create a weekly time series for that rice field (See Supplementary Table 1, Additional File 1). We applied the smoothing method to the vegetated water class and the sum of both flooded water classes (vegetated + open water), referred to as total flooded area, separately.

Individual rice field characterization and estimation of aggregate indicators

We developed an algorithm to create summary statistics regarding the agricultural season from each rice field’s smoothed time series of flooded areas (Fig. 1C). First, we performed numerical derivation on each time series to identify local minima and maxima values of the area of each class using a central derivation across a time difference of one day. We then assessed whether each local minima or maxima represented the beginning, peak, or end of an agricultural season. The beginning of a season corresponded to a local minimum that was followed by a maximum that occurred at a flooding value at least 10% of the maximum flooding value of that rice field higher than the focal minimum. The peak of a season corresponded to a local maximum that occurred at a flooding value of at least 10% of the maximum flooding value of that rice field higher than the nearest minimum on either side of the focal maximum. Maxima that did not meet this threshold were defined as intra-season maxima. The end of a season corresponded to a minimum that had a maximum between it and the following minimum that occurred at a flooding value of at least 10% of the maximum flooding value. If the threshold was not met, the period between the two minima was classified as inactive. From these identified beginnings, peaks, and ends of seasons, we then estimated the following statistics for each agricultural season and rice field: start date, end date, minimum flooding value, maximum flooding value, date of maximum flooding, and whether the season was inactive (no flooding). These statistics were then summarized to rice-field level statistics over the seven year time period.

In order to provide useful information to operational programs that implement LSM interventions, individual rice field data need to be aggregated in a way that can help with design, site prioritization, and follow-up. First, to visualize spatial trends in these indicators, we obtained the average of each statistic per fokontany (smallest administrative unit) to be displayed in corresponding maps. Second, we applied a semi-supervised time-series classification algorithm to categorize rice field flooding dynamics into a limited set of homogeneous groups. For this, we manually classified as inactive those rice fields that remained less than 20% flooded during the study period or remained inactive for more than 50% the study period. We then classified the remaining rice fields (N = 14,693) by applying an algorithmic methodology based on dynamic time warping (DTW) and partitioning around the median (PAM) to the time series [32, 33]. The approach consists of two distinct steps: calculating a metric of distance between all pairs of smoothed rice field time series (in Section 3) using DTW, and clustering of the smoothed time series using the PAM algorithm applied to the DTW distance matrix. The determination of the optimal number of clusters was based on the ratio between the inter- and intra-group variance of the rice fields (See Supplementary Fig. 3, Additional File 1), using the index quality measure of Calinski-Harabasz. This resulted in a seven-category classification for the rice field dataset, to which we gave descriptive names based on their average time-series: inactive before mid-year 2018 and low flood coverage, low flood coverage, low and moderate flood coverage, moderate flood coverage, moderate flood coverage with two seasons per year, high flood coverage, and very high flood coverage. For this classification, we used the R software, version 4.3, and the packages "parallelDist", version 0.2.6 [34] and "parallelpam", version 1.4 [35]. Finally, determining the population at risk from a particular larval habitat is an important factor to consider for site prioritization of LSM interventions. Here, we estimated the number of buildings within a 500m radius of each rice field as a proxy for population at risk from that rice field, using OSM spatial data on rice fields and households. From this, rice fields were classified into two categories: those with less than 50 surrounding buildings and with more than 50 surrounding buildings.

Integration of rice field monitoring indicators into an e-health tool

To demonstrate how rice field monitoring can be used for operational purposes during LSM implementation, we integrated all the data and results obtained, including the location and temporal dynamics of rice fields, into a web application developed with the “R Shiny” platform (Fig. 1D). The application allows for the visualization of key information at the district, fokontany, and individual rice field levels. This approach aims to facilitate the use and interpretation of results, while ensuring interoperability of information to support programmatic teams involved in mosquito larvae surveillance and control. We performed spatial junction techniques with administrative boundaries to create complete geographic profiles for each fokontany and rice field. To implement the application, we used PostgreSQL for the spatial database management system (DBMS) with the “PostGIS” extension, as well as the R packages “leaflet”, version 2.2.2 [36] for cartographic visualization and “Rpostgres”, version 1.4.7 [37] and “rpostgis”, version 1.5.1 [37] for connection to the DBMS and manipulation of geographic data.

Analyses of the optimal threshold in Sentinel-1 SAR images for each water class of interest using field data resulted in a classification where VH polarization values below − 17.98 dB were considered open water, those between − 17.98 dB and − 15.93 dB as vegetated water, and those above − 15.93 dB as no water (See Supplementary Fig. 2, Additional File 1). This classification performed well on the testing dataset. The separation of open water and vegetated water had a sensitivity (i.e. identification of vegetated water) of 0.95 and a specificity (i.e. identification of open water) of 0.93, while the separation of vegetated water and no water classes had a sensitivity (i.e. identification of vegetated water) of 0.83 and a specificity (i.e. identification of no water) of 0.63 (See Supplementary Fig. 2, Additional File 1).

After applying these thresholds to the 16,982 rice fields in Ifanadiana District and analyzing the resulting time-series of total flood coverage (open water and vegetated water combined), we were able to identify rice fields that had either one or two distinct flooding seasons during each agricultural year (Table 2), the timing of which varied considerably within the district (Fig. 2). The median percentage of water coverage in rice fields at the peak of the flooding season was 43% (IQR 29–58%) (Table 2). All rice fields had at least one “primary” flooding season, which began in July and attained its peak flooding for the most part in September-October (58.53%), before the peak in precipitation (Fig. 2B, Fig. 3B). The median duration of the primary flooding season was 336 days (IQR 294–363 days) and the flood coverage increased by 22% of the total field area (IQR 16–31%) between the beginning of a season and the moment of peak flooding (Table 2). The spatial distribution showed that rice fields in low-lying fokontany (on the eastern side of the district) were flooded later than those at higher elevation (Fig. 2A). One third of the district's rice fields (N = 5,644) remained flooded during a secondary rice growing season at least once during the study period, which began in March and attained its peak between March and April (66.03%). The secondary flooding season had a shorter duration, with a median of 177 days (IQR 160–202 days) (Table 2). The spatial distribution shows that rice fields in the southern and northern parts of the district flood earlier in the secondary season than in the rest of the district (Fig. 2C). During the 6-year study period, 37.90% of rice fields had at least one inactive season, where the rice field was not flooded at all. Similar results are available when considering only the vegetated water data class (See Supplementary Tables 2 and 4, Fig. 3, Additional File 1).

Table 2

**Summary statistics of temporal rice field flooding dynamics for 16,982 rice fields.** Statistics were calculated for each rice field across the 6 year agricultural period (July 01 2016 - June 30 2022). Summary statistics represent the summary across the full dataset of rice fields.
Description	Mode	Median	IQR (25%-75%)	Min-Max
Full time period (6 agricultural years, N = 16,982)
Number of seasons	6	6	6–7	1–11
Number of agricultural years with more than one season	0	0	0–1	0–5
Number of inactive seasons	0	1	0–1	0–5
Maximum percentage of flooded area	35	43	29–58	0-100
Primary season (N = 16,982)
Beginning of season (month of year)¹	6	5	4–6	1–12
End of season (month of year)¹	6	5	2–6	1–12
Month of highest percentage of flooded area¹	10	10	9–10	1–12
Median length of season (days)	370	336	294–363	145–1546
Average change in percentage of flooded area during the season (maximum flooded percentage - minimum flooded percentage)	19	22	16–31	0–86
Secondary season (N = 5,644)
Beginning of season (month of year)¹	12	3	1–12	1–12
End of season (month of year)¹	6	6	6–7	1–12
Month of highest percentage of flooded area¹	3	3	3–4	1–12
Median length of season (days)	160	177	160–202	104–588
Average change in percentage of flooded area during the season (maximum flooded percentage - minimum flooded percentage)	7	15	9–24	0–99
¹Each rice field was assigned the mode of the beginning month of each season across the 6 year time period

Classification of rice fields according to their flood dynamics

Using the DTW classification algorithm, we obtained an optimum of eight groups of rice fields with similar flooding dynamics (Table 3, Fig. 3A). The classification of these rice fields shows that 28.79% have high flood coverage (high flood coverage and very high flood coverage classes), 38.84% have moderate flood coverage (low to moderate flood coverage, moderate flood coverage, and moderate flood coverage with two seasons per year classes), 18.89% have low flood coverage (low flood coverage, and inactive before mid-year 2018 with low flood coverage classes), and the rest are always inactive (i.e., no water) (Table 4). These classes show similar seasonal patterns, but differ in the overall amount of flooded area of each rice field, with the exception of the always inactive rice fields, which are very flat with little flooded area, and the class of moderately flooded rice fields with two flood seasons per year (Fig. 3A and Table 4). Rainfall temporal patterns also follow these trends, but with a substantial lag in relation to the flooding dynamics, whereby the onset and peak of flooding typically occur several weeks before those of precipitation (Fig. 3). When the different rice field classes were combined with the number of households surrounding each rice field, results showed that the vast majority of rice fields (82.30%) were surrounded by fewer than 50 households, including nearly 25% of those that had high or very high flood coverage (Table 3). Only 17.7% of rice fields were surrounded by 50 or more households, including less than 5% of those with high or very high flood coverage (Table 3).

Table 3

Distribution of 16,982 supervised rice fields by the number of nearby households (within a 500m radius) and time series classification group.
Time series	Number of household < 50		Number of household > = 50
	Number	Total area (Ha)	Number	Total area (Ha)
Inactive	1,939 (11.42%)	872	350 (2.06%)	163
Inactive before mid-year 2018 with low flood coverage	804 (4.73%)	188	207 (1.22%)	66
Low flood coverage	1,748 (10.30%)	884	449 (2.64%)	325
Low to moderate flood coverage	2,095 (12.34%)	1,817	542 (3.20%)	794
Moderate flood coverage	1,715 (10.10%)	1,165	337 (1.98%)	276
Moderate flood coverage with two seasons per year	1,490 (8.77%)	2,609	416 (2.45%)	997
High flood coverage	2,776 (16.35%)	2,107	474 (2.79%)	507
Very high flood coverage	1,410 (8.30%)	436	230 (1.35%)	58

Table 4

**Descriptive statistics of rice fields by flood season and time series class.** Results show the median and interquartile values (25% and 75%), for each indicator in the eight flooding classes over a period from July 01 2016 to June 30 2022 (6 agricultural years). Similar information is available for the vegetated water flooding category alone (See Supplementary Table 3, Additional File 1).
Flood class (IQR: 25% − 75%)
Description	Inactive (N = 2,289)	Inactive before mid-year 2018 with low flood coverage (N = 1,011)	Low flood coverage (N = 2,197)	Low to moderate flood coverage (N = 2,637)	Moderate flood coverage (N = 2,052)	Moderate flood coverage with two seasons per year (N = 1,906)	High flood coverage (N = 3,250)	Very high flood coverage (N = 1,640)
Full time period (6 agricultural years)
Number of seasons	6 (5–7)	7 (6–8)	6 (6–7)	6 (6–8)	6 (6–7)	6 (5–8)	6 (6–7)	6 (6–7)
Number of agricultural years with more than one season	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)
Number of inactive seasons	1 (1–2)	1 (1–2)	1 (0–2)	1 (0–1)	1 (0–1)	0 (0–1)	1 (0–1)	1 (0–1)
Maximum percentage of flooded area	14 (10–17)	30 (25–39)	29 (24–37)	35 (30–42)	45 (38–53)	49 (42–57)	58 (51–66)	79 (71–87)
Primary season (N = 16,982)
Beginning of season (month of year)¹	5 (4–6)	6 (4–6)	5 (4–6)	5 (4–6)	5 (4–6)	6 (5–6)	5 (4–6)	4 (3–6)
End of season (month of year)¹	5 (3–6)	4 (2–6)	5 (3–6)	5 (2–6)	4 (2–6)	6 (5–6)	5 (2–6)	3 (2–5)
Month of highest percentage of flooded area¹	9 (8–10)	10 (9–10)	10 (9–10)	10 (9–10)	10 (9–10)	10 (3–10)	10 (9–10)	9 (9–10)
Median length of season (days)	322 (282–356)	307 (258–349)	336 (296–363)	339 (294–363)	342 (307–363)	349 (289–366)	342 (301–364)	341 (308–363)
Average change in percentage of flooded area during the season (maximum flooded percentage - minimum flooded percentage)	8 (6–11)	19 (15–24)	17 (14–22)	21 (17–26)	24 (19–30)	31 (25–39)	29 (22–37)	30 (24–37)
Secondary season (N = 5,644)
Beginning of season (month of year)¹	3 (1–12)	2 (1–12)	2.5 (1–12)	3 (1–12)	2 (1–12)	11 (1–12)	10 (1–12)	10 (1–12)
End of season (month of year)¹	7 (6–8)	6 (6–7)	6 (6–7)	6 (6–7)	6 (6–7)	6 (6–6)	6 (6–7)	6 (6–7)
Month of highest percentage of flooded area¹	4 (3–5)	4 (3–4)	3 (3–4)	3 (3–4)	4 (3–4)	3 (3–4)	3 (3–4)	4 (3–5)
Median length of season (days)	188 (167–223)	168 (153–189)	175 (160–203)	175 (160–201)	174 (159–201)	175 (161–188)	175 (161–202)	189 (168–230)
Average change in percentage of flooded area during the season (maximum flooded percentage - minimum flooded percentage)	45 (3–8)	11 (7–16)	12 (8–17)	14 (9–20)	15 (10–22)	27 (18–36)	20 (13–30)	18 (12–28)
¹Each rice field was assigned the mode of the month of each season across the 6 year time period

E-health decision support tool for LSM interventions

The e-health tool was developed to help operational teams better prioritize, design, implement and monitor LSM interventions. The tool enables detailed monitoring of rice fields, with information on flooding dynamics to identify areas and periods that could present a higher risk to nearby residents. The application displays two measures of flooding (vegetated water and open water; vegetated water only) for spatio-temporal monitoring of rice field flooding via a filter bar (Fig. 4), with three spatial levels of information that can be useful for decision making. First, a visualization of the spatial distribution by fokontany to target areas of interest according to different indicators: the period of the flooding season (the beginning and end of the season, as well as the month of peak flooding), areas where rice fields are close to larger populations, and the type of rice field activity. Next, a user can select a specific fokontany to visualize rice fields by different colors according to their temporal flooding dynamics class, and view aggregated information for key indicators. Finally, the user can select an individual rice field to view the historical time series of its flooding dynamics and the location of nearby households (within 500m).

Screenshot of the application showing the spatial distribution by fokontany of the month of peak flooding for the primary flooding season. The map in the center shows rice fields within a selected fokontany (limits in yellow) with colors according to flooding class. On the right, the time series of the different rice field classes are shown, aggregated within the fokontany (median and interquartile range). On the left, the time series per year of a selected rice field within the fokontany, with a 500 m radius in blue to delineate and visualize nearby households. The application is publicly available at http://44.218.51.103:3838/lsm/ (temporary link).

Current challenges in malaria control and elimination underscore the need to invest in new control strategies and corresponding decision support tools to guide their planning and implementation [1]. Mapping and monitoring of larval breeding sites, such as irrigated rice fields, is an essential prerequisite for the targeted deployment of larval source management (LSM) strategies. In this study, we combined the precise identification of rice fields from OpenStreetMap (OSM) data with Synthetic Aperture Radar (SAR) image analysis to track flooding dynamics in over 17,000 rice fields in rural Madagascar. From these time series, we estimated indicators relevant to the monitoring of rice field flooding dynamics for LSM interventions, such as the number and timing of agricultural seasons and a classification of rice fields according to the flooding level and temporal patterns and their proximity to human populations. We demonstrated how this information can be easily integrated into decision support tools to more effectively guide the planning and targeted implementation of LSM strategies via an accurate identification and monitoring of at-risk breeding sites.

Our approach enables the classification and monitoring of rice fields at different levels of spatial aggregation (individual rice field, village, district) that are complementary in informing different phases and types of LSM interventions. For example, habitat modification strategies such as intermittent irrigation, drainage, or improved canals can help reduce mosquito breeding in land constantly flooded or with small fluctuations in water levels [4]. Our method could allow identifying rice fields with such characteristics based on historical flooding trends (e.g. less than one third of rice fields in Ifanadiana had high levels of flood coverage throughout the year), and evaluating prospectively whether the habitat modifications implemented have an impact on flooding patterns [38, 39]. Moreover, larvicide application requires frequent applications when the rice field is flooded in order to be effective [4]. In the planning phase, our results can help prioritize intervention sites based on historical flood dynamics, their size and proximity to populations, and identify key administration periods. For instance, less than 5% of rice fields in Ifanadiana had high levels of flood coverage while being within 500m of populations with 50 households or more. By monitoring flooding in selected rice fields in real time, programs can adjust the frequency, duration, and stopping criteria for larvicide application to maximize their effectiveness while avoiding non-essential sites, thus optimizing the use of resources. Beyond its usefulness against malaria through LSM strategies, our approach could be useful in guiding surveillance and control interventions against other vector-borne diseases where rice fields play an important role in maintaining mosquito populations or could even contribute to improving food security and local agricultural development programs, helping optimize rice cultivation practices to the flooding dynamics of individual rice fields [40].

Our results show significant local heterogeneity in the number of flooding seasons, the timing of season onset, and the average level of flooding of rice fields. The average results are consistent with those from other global databases, such as RiceAtlas (thematic maps at a coarse spatial resolution and district scale by country), which show similar patterns of rice growing seasonality (number of seasons, timing, etc.) as those estimated for Madagascar [41]. On the other hand, they provide an unprecedented level of detail on the spatio-temporal heterogeneity at the local level, such as the duration of flooding and rice cultivation, which varies according to the topography (high or low elevation) and the timing of their activities, making it possible to identify, for example, which villages within a district plant their rice later (Fig. 2) or which ones tend to have more than one cropping season. Taking this local heterogeneity into account may be key to the success of LSM strategies. Indeed, larval development varies considerably depending on conditions within individual rice fields, in combination with environmental and climatic conditions on a larger scale. For example, malaria vectors in Africa, such as Anopheles funestus and Anopheles gambiae, are particularly sensitive to the presence of stagnant surface water, especially in aquatic agriculture, man-made ponds, drainage ditches, and natural ponds or rain pools [42, 43]. In addition, the proximity of rice fields to dwellings may increase the risk to the population, depending on the flight range and host preferences of Anopheles mosquitoes [44, 45]. About 17% of rice fields in Ifanadiana had more than 50 households within 500m, and these parameters (i.e., population and distance thresholds) can be easily customized in order to guide site prioritization according to the local mosquito species, LSM program budget and goals.

The innovation of our approach lies in the use of freely available SAR satellite imagery combined with OSM geographic data to track individual rice field flooding dynamics at very large scales. Although this approach is flexible and easily adaptable to other malaria endemic settings (Sentinel-1 data covers the whole globe), its main limitation is the availability of complete, high quality data on public platforms such as OpenStreetMap needed to locate the individual rice fields and residential areas [46]. Photo-interpretation mapping performed by humans, although resource intensive, can achieve remarkable accuracy, limit classification bias that is commonplace in AI-generated datasets, and help enrich public databases for other uses [47]. For example, through various initiatives, our team has mapped over 109,000 rice fields and 1,170,000 buildings in 7 districts of southeastern Madagascar, which may allow our approach to be scaled up in several regions of Madagascar where malaria is highly endemic. Beyond Madagascar, parts of Africa and Asia are already relatively well mapped on OSM [46], demonstrating the potential of this approach for larger scales and diverse contexts. To circumvent the limitations of public data on rice fields, some studies have focused on using neural networks to identify favorable habitats for mosquitoes, such as rice fields, with benefits in terms of reduced financial and labor costs, although these methods are prone to error and rarely use very high-resolution data [48, 49]. An interesting alternative is the use of drone imagery, which provides much better resolution and can improve vector control activities. However, this approach is difficult to implement on scales required by malaria national control programs due to regulatory constraints, high costs, safety challenges, community acceptance, and privacy concerns [16]. Therefore, drones are generally recommended for targeted ground surveys in small-scale pilot projects [15, 16, 50]. Our approach represents a compromise between these two alternatives, allowing for the improvement of large-scale vector control interventions at a lower cost.

Despite the innovation and strengths in our approach, several limitations were identified. First, we did not conduct concurrent field surveys to characterize larval productivity as a function of flood dynamics or to identify sites with high mosquito production [15]. However, the link between local malaria dynamics in Ifanadiana and rice fields has been demonstrated previously [22, 51], and our approach aims to guide LSM interventions that specifically target rice fields as priority breeding sites, as is the case in Madagascar [20]. Second, the spatial resolution of Sentinel-1 satellite data (10m) prevents us from robustly tracking small rice fields (< 0.04 ha), as the small number of pixels (some of which are partially outside the rice field) leads to greater stochasticity and uncertainty in the flooding results. In our context, this represented only 13.48% of all rice fields in the district, but this could be an important limitation for the application of our method in contexts where small rice fields represent an important part of the larval habitat. Unfortunately, SAR imagery does not exist at finer resolutions than Sentinel-1, while simultaneously being freely available over such a wide area and regular passing frequency. Finally, we integrated the results of this study into a web-based decision support tool to demonstrate how it can effectively guide the local implementation of LSM interventions through a high-resolution, open-access, low-cost approach, but this study did not test the impact of the approach under real-world conditions of LSM intervention implementation. Operational research is needed to assess the usefulness of this approach in the field in close collaboration with programmatic teams.

While LSM strategies that target irrigated fields offer promising solutions for improving malaria control, there are currently few decision support tools for prioritizing and monitoring rice fields that would help optimize their design and implementation. This study presents an innovative and reproducible approach to monitor individual flooding dynamics in rice fields at large scales to guide LSM interventions, providing critical information at multiple levels of spatial aggregation that can be useful for decision making, such as the number of seasons, the exact flooding period, the level of flooding and activity. Scaling up these approaches may facilitate broader adoption of evidence-based interventions using LSM, potentially decreasing malaria impact in regions where rice fields are a significant contributor to larval habitat.

ARD

Analysis Ready Data

DBMS

Spatial Database Management System

DTW

Dynamic Time Warping

IQR

Interquartile Range

LSM

Larval Source Management

MODIS

Moderate Resolution Imaging Spectroradiometer

OSM

OpenStreetMap

SAR

Synthetic Aperture Radar

Vertical transmit and Horizontal receive

WHO

World Health Organization.

Ethics approval and consent to participate

Not applicable.

Consent for publication

All authors approved the manuscript’s submission for publication.

Competing interests

The authors declare that they have no competing interests.

Authors' information

¹ ONG PIVOT, Ranomafana, Madagascar. ² Institut de Recherche pour le Développement, UMR 224 MIVEGEC (IRD, UM, CNRS), Montpellier, France. ³ Université de Montpellier, ED 168 CBS2, Montpellier, France. ⁴ Department of Global Health and Social Medicine, Harvard Medical School, Boston, USA. ⁵ Espace Dev, IRD, Univ Montpellier, Univ. Antilles, Univ Guyane, Univ Réunion, 5 Preah Monivong Blvd, Phnom Penh 12201, Cambodia. ⁶ Espace Dev, IRD, Univ Montpellier, Univ Antilles, Univ Guyane, Univ Réunion, Montpellier, France.

Funding

This work was supported by a grant from Agence Nationale de la Recherche (Project SMALLER ANR-19-CE36-0001), with additional support from a grant from AfriCam project, funded by the French Development Agency (AFD) as part of PREACTS program and internal funding from Pivot.

Author Contribution

MR, TMR and AG conceived and designed the study. TC, CR, VH, FAI and ED used their expertise to validate the methodology and prepare the raw data for analysis. MR, TMR, AG, MVE and FAI analyzed the data. MR, TMR, AG and MVE contributed to the interpretation of the data and wrote the initial draft of the manuscript. All authors read, edited, and approved the final version of the manuscript.

Acknowledgement

We would like to thank Cathy NONINIRINA, Nadia RANOROARINALISOA, and Roger RIVOMANANA for their assistance with field data collection, and the Pivot Medical team for their work to improve health care in Ifanadiana District.

Data Availability

Data are available on OpenStreetMap (https://www.openstreetmap.org), on github the analysis code (https://github.com/mvevans89/dynamique-riz.git), on Google Earth Engine the code to extract Sentinel-1 SAR data (open water: https://code.earthengine.google.com/60e2941c1ee53a4ed55c52c281e387bf, vegetated water: https://code.earthengine.google.com/832f1a7284cc9ef6e68d4b1cf39212c7), and on the Shiny app (temporary link : http://44.218.51.103:3838/lsm/).

WHO. World malaria report 2023 [Internet]. Geneva: World Health Organization. 2023. https://cdn.who.int/media/docs/default-source/malaria/world-malaria-reports/world-malaria-report–2023.pdf
Sougoufara S, Ottih EC, Tripet F. The need for new vector control approaches targeting outdoor biting anopheline malaria vector communities. Parasites Vectors. 2020;13:295.
Sangbakembi-Ngounou C, Costantini C, Longo-Pendy NM, Ngoagouni C, Akone-Ella O, Rahola N et al. Diurnal biting of malaria mosquitoes in the Central African Republic indicates residual transmission may be out of control. Proceedings of the National Academy of Sciences. 2022;119:e2104282119.
WHO. Larval source management: a supplementary malaria vector control measure: an operational manual [Internet]. Geneva: World Health Organization. 2013 [cited 2023 Oct 25]. https://iris.who.int/handle/10665/85379
Chan K, Bottomley C, Saito K, Lines J, Tusting LS. The control of malaria vectors in rice fields: a systematic review and meta-analysis. Sci Rep. 2022;12:19694.
Derua YA, Kweka EJ, Kisinza WN, Githeko AK, Mosha FW. Bacterial larvicides used for malaria vector control in sub-Saharan Africa: review of their effectiveness and operational feasibility. Parasites Vectors. 2019;12:426.
FAOSTAT. Food and Agriculture Organization of the United Nations [Internet]. 2021 [cited 2024 Jul 4]. https://www.fao.org/faostat/en/#home
Lacey LA, Lacey CM. The medical importance of riceland mosquitoes and their control using alternatives to chemical insecticides. J Am Mosq Control Assoc Suppl. 1990;2:1–93.
Chan K, Tusting LS, Bottomley C, Saito K, Djouaka R, Lines J. Malaria transmission and prevalence in rice-growing versus non-rice-growing villages in Africa: a systematic review and meta-analysis. Lancet Planet Health. 2022;6:e257–69.
Fornace KM, Diaz AV, Lines J, Drakeley CJ. Achieving global malaria eradication in changing landscapes. Malar J. 2021;20:69.
Li L, Xin X, Zhao J, Yang A, Wu S, Zhang H, et al. Remote Sensing Monitoring and Assessment of Global Vegetation Status and Changes during 2016–2020. Sensors. 2023;23:8452.
Luo C, Liu H, Fu Q, Guan H, Ye Q, Zhang X, et al. Mapping the fallowed area of paddy fields on Sanjiang Plain of Northeast China to assist water security assessments. J Integr Agric. 2020;19:1885–96.
Tornos L, Huesca M, Dominguez JA, Moyano MC, Cicuendez V, Recuero L, et al. Assessment of MODIS spectral indices for determining rice paddy agricultural practices and hydroperiod. ISPRS J Photogrammetry Remote Sens. 2015;101:110–24.
Carrasco-Escobar G, Manrique E, Ruiz-Cabrejos J, Saavedra M, Alava F, Bickersmith S et al. High-accuracy detection of malaria vector larval habitats using drone-based multispectral imagery. Costantini C, editor. PLoS Negl Trop Dis. 2019;13:e0007105.
Byrne I, Aure W, Manin BO, Vythilingam I, Ferguson HM, Drakeley CJ, et al. Environmental and spatial risk factors for the larval habitats of Plasmodium knowlesi vectors in Sabah, Malaysian Borneo. Sci Rep. 2021;11:11810.
Carrasco-Escobar G, Moreno M, Fornace K, Herrera-Varela M, Manrique E, Conn JE. The use of drones for mosquito surveillance and control. Parasit Vectors. 2022;15:473.
Trujillano F, Jimenez Garay G, Alatrista-Salas H, Byrne I, Nunez-del-Prado M, Chan K, et al. Mapping Malaria Vector Habitats in West Africa: Drone Imagery and Deep Learning Analysis for Targeted Vector Surveillance. Remote Sens. 2023;15:2775.
Bayik C, Abdikan S, Ozbulak G, Alasag T, Aydemir S, Balik Sanli F, EXPLOITING, MULTI-TEMPORAL, SENTINEL–1 SAR DATA FOR FLOOD EXTEND MAPPING. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2018;XLII–3-W4:109–13.
Catry T, Li Z, Roux E, Herbreteau V, Gurgel H, Mangeas M, et al. Wetlands and Malaria in the Amazon: Guidelines for the Use of Synthetic Aperture Radar Remote-Sensing. IJERPH. 2018;15:468.
USAID/CDC. U. S. President’s Malaria Initiative – Madagascar Malaria Operational Plan FY 2023 [Internet]. PMI. 2022 [cited 2024 Jul 4]. https://www.pmi.gov/where-we-work/madagascar/
Evans MV, Ihantamalala FA, Randriamihaja M, Aina AT, Bonds MH, Finnegan KE, et al. Applying a zero-corrected, gravity model estimator reduces bias due to heterogeneity in healthcare utilization in community-scale, passive surveillance datasets of endemic diseases. Sci Rep. 2023;13:21288.
Pourtois JD, Tallam K, Jones I, Hyde E, Chamberlin AJ, Evans MV, et al. Climatic, land-use and socio-economic factors can predict malaria dynamics at fine spatial scales relevant to local health actors: Evidence from rural Madagascar. PLOS Glob Public Health. 2023;3:e0001607.
NextGIS Mobile [Internet]. 2013 [cited 2024 Sep 5]. https://play.google.com/store/apps/details?id=com.nextgis.mobile
Fletcher K, Agency ES. Sentinel–1: ESA’s Radar Observatory Mission for GMES Operational Services [Internet]. ESA Communications; 2012. https://books.google.mg/books?id=5Yy2MwEACAAJ
Zhan P, Zhu W, Li N. An automated rice mapping method based on flooding signals in synthetic aperture radar time series. Remote Sens Environ. 2021;252:112112.
Nguyen DB, Gruber A, Wagner W. Mapping rice extent and cropping scheme in the Mekong Delta using Sentinel–1A data. Remote Sens Lett. 2016;7:1209–18.
Mullissa A, Vollrath A, Odongo-Braun C, Slagter B, Balling J, Gou Y, et al. Sentinel–1 SAR Backscatter Analysis Ready Data Preparation in Google Earth Engine. Remote Sens. 2021;13:1954.
Ihantamalala FA, Herbreteau V, Révillion C, Randriamihaja M, Commins J, Andréambeloson T, et al. Improving geographical accessibility modeling for operational use by local health actors. Int J Health Geogr. 2020;19:1–15.
Herbreteau V, Révillion C, Trimaille E. GeoHealth and QuickOSM, Two QGIS Plugins for Health Applications. In: Baghdadi N, Mallet C, Zribi M, editors. QGIS and Generic Tools [Internet]. 1st ed. Wiley; 2018 [cited 2024 Jul 19]. pp. 257–86. https://onlinelibrary.wiley.com/doi/10.1002/9781119457091.ch7
OCHA, Madagascar. - Subnational Administrative Boundaries [Internet]. 2018 [cited 2023 Aug 8]. https://data.humdata.org/dataset/cod-ab-mdg
Novella NS, Thiaw WM. African Rainfall Climatology Version 2 for Famine Early Warning Systems. J Appl Meteorol Climatology. 2013;52:588–606.
Sardá-Espinosa A. Time-Series Clustering in R Using the dtwclust Package. R J. 2019;11:22–43.
Montero P, Vilar JA. TSclust: An R Package for Time Series Clustering. J Stat Soft [Internet]. 2014 [cited 2024 Mar 11];62. http://www.jstatsoft.org/v62/i01/
Eckert A, Godoy L, parallelDist KSS. Parallel Distance Matrix Computation using Multiple Threads [Internet]. 2022 [cited 2024 Jun 1]. https://cran.r-project.org/web/packages/parallelDist/index.html
Schubert E, Rousseeuw PJ. Fast and eager k < math > < mi is=true>k-medoids clustering: O(k) < math > < mrow is=true>O < mrow is=true>true>( < mi is=true>k < mo is=true>) runtime improvement of the PAM, CLARA, and CLARANS algorithms. Information Systems. 2021;101:101804.
Cheng J, Schloerke B, Karambelkar B, Xie Y, Wickham H, Russell K et al. leaflet: Create Interactive Web Maps with the JavaScript Leaflet Library [Internet]. 2024 [cited 2024 Sep 5]. https://cran.r-project.org/web/packages/leaflet/index.html
Basille M, Bucklin D, Cidre González A. rpostgis: R Interface to a PostGIS Database [Internet]. 2016 [cited 2024 Sep 5]. p. 1.5.1. https://CRAN.R-project.org/package=rpostgis
Martello E, Yogeswaran G, Reithinger R, Leonardi-Bee J. Mosquito aquatic habitat modification and manipulation interventions to control malaria. Cochrane Database of Systematic Reviews [Internet]. 2022 [cited 2024 May 22]; https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD008923.pub3/full
Jacups S, Kurucz N, Whitters R, Whelan P. Habitat modification for mosquito control in the Ilparpa Swamp, Northern Territory, Australia. J Vector Ecol. 2011;36:292–9.
Dröge S, Poudyal M, Hockley N, Mandimbiniaina R, Rasoamanana A, Andrianantenaina NS, et al. Constraints on Rice Cultivation in Eastern Madagascar: Which Factors Matter to Smallholders, and Which Influence Food Security? Hum Ecol. 2022;50:493–513.
Laborte AG, Gutierrez MA, Balanza JG, Saito K, Zwart SJ, Boschetti M, et al. RiceAtlas, a spatial database of global rice calendars and production. Sci Data. 2017;4:170074.
Arisco NJ, Rice BL, Tantely LM, Girod R, Emile GN, Randriamady HJ, et al. Variation in Anopheles distribution and predictors of malaria infection risk across regions of Madagascar. Malar J. 2020;19:348.
Debrah I, Afrane YA, Amoah LE, Ochwedo KO, Mukabana WR, Zhong D, et al. Larval ecology and bionomics of Anopheles funestus in highland and lowland sites in western Kenya. PLoS ONE. 2021;16:e0255321.
Zohdy S, Derfus K, Headrick EG, Andrianjafy MT, Wright PC, Gillespie TR. Small-scale land-use variability affects Anopheles spp. distribution and concomitant Plasmodium infection in humans and mosquito vectors in southeastern Madagascar. Malar J. 2016;15:114.
Chan K, Konan KA-C, Doudou DT, Kouadio GB, Lines J, Aunger R, et al. Rice farmers’ knowledge, attitudes and practices towards mosquitoes in irrigation schemes in Côte d’Ivoire: a qualitative study. Malar J. 2023;22:352.
Herfort B, Lautenbach S, Porto De Albuquerque J, Anderson J, Zipf A. A spatio-temporal analysis investigating completeness and inequalities of global urban building data in OpenStreetMap. Nat Commun. 2023;14:3985.
Google Research. Mapping Africa’s Buildings with Satellite Imagery [Internet]. Google Research. 2024 [cited 2024 Sep 13]. http://sites.research.google/gr/open-buildings/
Chang Y-L, Tan T-H, Chen T-H, Chuah JH, Chang L, Wu M-C, et al. Spatial-Temporal Neural Network for Rice Field Classification from SAR Images. Remote Sens. 2022;14:1929.
Pérez-Carabaza S, Boydell O, O’Connell J. Habitat classification using convolutional neural networks and multitemporal multispectral aerial imagery. JARS. 2021;15:042406.
Hardy A, Oakes G, Hassan J, Yussuf Y. Improved Use of Drone Imagery for Malaria Vector Control through Technology-Assisted Digitizing (TAD). Remote Sens. 2022;14:317.
Evans MV, Ihantamalala FA, Randriamihaja M, Herbreteau V, Revillion C, Catry T et al. Increasing the resolution of malaria early warning systems for use by local health actors [Internet]. 2024 [cited 2024 Jul 2]. http://medrxiv.org/lookup/doi/10.1101/2024.04.17.24305948

¹Each rice field was assigned the mode of the beginning month of each season across the 6 year time period

No competing interests reported.

Additionalfile1.docx

Download PDF

Reviews received at journal
17 Oct, 2024
Reviews received at journal
17 Oct, 2024
Reviewers agreed at journal
17 Oct, 2024
Reviewers agreed at journal
16 Oct, 2024
Reviewers agreed at journal
16 Oct, 2024
Reviewers invited by journal
14 Oct, 2024
Editor assigned by journal
01 Oct, 2024
Submission checks completed at journal
01 Oct, 2024
First submitted to journal
30 Sep, 2024

You are reading this latest preprint version

Monitoring individual rice field flooding dynamics over large scales to improve mosquito surveillance and control

Status:

Version 1

Abstract

Figures

Background

Methods

Study area

Identification of rice field flooding thresholds using field data and Sentinel-1 images

Estimation of individual rice field flooding dynamics in Ifanadiana district

Individual rice field characterization and estimation of aggregate indicators

Integration of rice field monitoring indicators into an e-health tool

Results

Classification of rice fields according to their flood dynamics

E-health decision support tool for LSM interventions

Discussion

Conclusion

Abbreviations

Declarations

Ethics approval and consent to participate

Competing interests

Authors' information

Funding

Author Contribution

Acknowledgement

Data Availability

References

Footnotes

Additional Declarations

Supplementary Files

Status:

Version 1