Multitemporal optical remote sensing and meteorological gridded data constitute useful data sources for crop yield forecasting over large areas, which is important for food security, in particular to predict where crop production is likely to underperform. Using linear regression, researchers and practitioners have been able to predict crop yield in various geographical settings. Machine learning and deep learning methods are increasingly used to process Earth Observation (EO) data and have the potential to allow for more automation and standardization of yield forecasting process and improve its accuracies and timeliness. However, these methods require large sample size data to train.
In this work we investigate if and to what extent machine learning and deep learning methods can improve accuracy in the context of regional yield forecasting where data availability is typically limited.
Taking Algeria as case study, we predicted provincial yields for barley, soft wheat and durum wheat over the period 2002-2018 using four explanatory variables at 10-day time interval: NDVI from MODIS, air temperature and global radiation from ERA5 ECMWF and rainfall estimates from CHIRPS.
We first developed a robust and automated machine-learning pipeline to select the best features and model for yield prediction on a monthly basis between the start and the end of the growing season. The algorithms considered included: least absolute shrinkage and selection operator, Random Forest, multi-layer perceptron, support vector regression, and gradient boosting. After that we explored the use of deep neural network applied to the same data set. In particular we used 1D and 2D convolutional neural networks (CNNs). 1D CNNs were fed with average time series extracted at the provincial level for cropland area. For 2D CNNs we compressed the pixel values of the four explanatory variables into four histograms. Each histogram is specific to a variable and represents time on the x-axis and the distribution of the variable values on the y-axis (i.e. the variable range is divided into 64 bins and the frequency of occurrence in each bin recorded at each time step).
Accuracies of the various methods and comparison with simple benchmark models will be presented in this contribution.

In central Mali, climate change and growing conflicts over land use exacerbate food insecurity; more particularly, areas of food production are affected, and need to be identified.. The region’s heavy reliance on subsistence agriculture livelihoods means that humanitarian actors must quickly assess changes in cropland to plan the distribution of food aid. Typically, in the absence of extensive field data, publicly available land cover datasets are used to identify cropland cover. While the proliferation of such datasets (e.g. ESA-CCI or GlobeLand30) has increased over the years, they are often ill-adjusted to the Sahelian context. Assessments conducted of cropland identified by the most used land cover datasets found that none were able to meet the 75% accuracy threshold in Sahelian West Africa (Samasse et al, 2019). While countries like Mali are among those most critically in need of accurate cropland mapping, the current toolkit of land use / land cover (LCLU) datasets is woefully inadequate for the needs of humanitarian actors in an emergency context.

To address this gap, the “3-Period TimeScan” (3PTS) was developed using Google Earth Engine (Gorelick et al., 2017). This product consists of a Red-Green-Blue composite of Sentinel-2 images where the red represents the maximum NDVI value during the first period of the growing season, the green the maximum NDVI value in the middle, and the blue the maximum NDVI value at the end. This technique condenses the NDVI temporal evolution along the agricultural seasons and so to single out cropland from other LCLU types. A highly localized cropland change analysis was conducted comparing the 2019 3PTS product with the one of 2017, a year prior to the start of the central Mali’s conflict. The change status was visually determined per populated site, as supervised classifications required exhaustive manual cleaning to produce a reliable product over such a large and ecologically heterogenous zone. The resulting map was compared with georeferenced data of conflict events, indicating a strong spatial correlation between violence and cropland reductions.

In June 2019, during the planting period, a peak in both the numbers of violent events and of fatalities was recorded in central Mali. Most of the significant cropland losses occurred in localities where violent events were reported for the period between April and October 2019. Cropland abandonment, but also an effect of concentration of crops in the proximity of habitations (due to access restrictions, violent threats or attacks in farther fields) and settlement damage are as many consequences of the violence operating in central Mali, as visible from space.

The World Food Programme (WFP) operationalized the analysis from this method. By offering a map and a list of localities showing significant declines in cultivation, a more precise picture of food insecurity could be drawn, highlighting vulnerable areas in need of food assistance. These outputs were quickly absorbed into the humanitarian response planning process, notably through the Cadre Harmonisé (CH), the bi-annual national food security analysis (led by the national early warning system in collaboration with line ministries and humanitarian actors such as NGOs and UN agencies). The goal of the CH is to estimate the number of food insecure people in the country and provide coordinated targeting priorities for humanitarian response. The 3PTS-derived results contributed to estimating 757,217 persons in food insecurity for the 2020 lean season (the seasonal period where hunger typically peaks during the year). Beyond the CH, the unprecedented level of spatial precision provided by these results fed into humanitarian response mechanisms and strategic decision-making, as a tool to enhance village-scale geotargeting of most vulnerable communities. WFP used these outputs to target their humanitarian assistance for the lean season of 2020 as early as March, which is 2 to 3 months ahead of the start of the lean season.

The 3PTS method fills a gap left by existing landcover datasets in covering cropland in the West African Sahel. Likewise, it benefits from a quick turnaround time in producing annual cropland change maps and the use of free and open source data and tools (Sentinel 2 and Google Earth Engine). As a result, 3PTS is well-positioned as a methodology to respond to the needs of humanitarian actors seeking to use remote sensing to monitor food security.

Food security in West Africa is considered as one of the major development challenges of the region. Food security issues are particularly prevalent there due to strong demographic growth, household food and subsistence based mainly on rain-fed agriculture and high rainfall variability. Added to these factors are the security and health risks facing the region, making agricultural production systems particularly fragile and fluctuating. Thus, the cyclical aspects of agricultural production are combined with those structural aspects of the vulnerability of populations. Since the major droughts of the early 1970s, several crop monitoring and food security early warning systems (EWS) have been developed in the region to enable decision-makers to anticipate crises, and to help plan emergency measures emergency by targeting populations and / or areas at risk. Since 2016, the GEOGLAM Crop Monitor for Early Warning (CM4EW) conducts a deliberative evidence-building process to reach an agreement on monthly crop conditions produced by the different EWS (Becker-Reshef et al., 2020). In these systems, satellite information is mainly used to derive vegetation index anomalies from time series of low spatial resolution images (MODIS or PROBA-V type), and precipitation data. Vegetation anomalies maps are produced from the NDVI value obtained for the current decade and compared to the average NDVI value for the same decade calculated over a reference period or what is assumed to be a normal situation.

If satellite images remain the main source of information for EWS at national and regional scales, their use still raises a number of issues: (i) the observed discrepancy between vegetation anomalies produced by different crop monitoring systems (Lemettais L., 2021), while the basic satellite data are identical, and (ii) the detection of a vegetation anomaly in real time is not sufficient to establish a diagnosis on the agricultural production of a region because many factors come in consideration. To complete the information collected from the field, local newspapers and new media report certain climatic or socio-economic events which may partly explain the vegetation anomalies observed. These events are not all listed, but they provide qualitative (good development, stunted growth, etc.), diagnostic (drought, flooding, etc.) and localized information which is complementary to that of satellite images and field information. Thus, the hypothesis underlying our work is that textual data from newspapers and other media represent a still untapped resource that can be used, in combination with products derived from satellite images, to provide a more comprehensive framework on food security of a region. In this context, we aim to strengthen food security monitoring systems through the use of advanced Natural Language Processing (NLP) and data science methods. More specifically, in a context of a rapidly growing technological and scientific offer, the methodological objective of our work is to enrich the crop monitoring and early warning systems of West Africa, with information derived from the analysis. The applied objective is to provide additional information based on textual data coming from online media (e.g. newspaper).

The proposed process is based on 3 stages (Figure 1). The first one consists of implementing different machine and deep learning approaches using heterogeneous data including satellite images in order to predict food security indicators (see step 1 - Figure 1). In order to address this issue, the FSPHD (Food Security Prediction based on Heterogeneous Data) framework is implemented and evaluated. To address the structural heterogeneity of the data, four groups of data are treated with adapted machine and deep learning methods, i.e. time series (NDVI, land temperature, rainfall, but also market prices), conjunctural data, thematic maps and data, and high spatial resolution (HSR) images. Different variants of the framework are experimented to obtain the final results. We observed the modest but significant contribution of deep learning models (convolutional neural networks - CNN) to HSR data processing. This work is detailed in Deleglise et al. (2021). The second part is dedicated to applying NLP approaches (Fize et al., 2020) for highlighting some textual data extracted in newspapers in West Africa (see step 2 - Figure 1). The last step consists of mapping these keywords (e.g. locust invasion, flood, terrorism) in order to propose contextual interpretations of results obtained with machine and deep learning approaches (see step 3 - Figure 1).

Figure 1: Heterogeneous data framework


References

Becker-Reshef I., Justice C., Barker B., et al., 2020. Strengthening agricultural decisions in countries at risk of food insecurity: The GEOGLAM Crop Monitor for Early Warning. Remote Sensing of Environment, 237 : 111553. https://doi.org/10.1016/j.rse.2019.111553
Deléglise H., Interdonato R., Bégué A., Maître d’Hôtel E., Teisseire M., Roche M. 2021. Food security prediction from heterogeneous data combining machine and deep learning methods. Expert Systems with Applications. 116189. https://doi.org/10.1016/j.eswa.2021.116189
Fize J., Roche M., Teisseire M. 2020. Could spatial features help the matching of textual data? Intelligent Data Analysis, 24(5): 1043-1064. https://doi.org/10.3233/IDA-194749
Lemettais L., 2021. Analyse comparative des produits d'anomalies de végétation dans les systèmes de sécurité alimentaires en Afrique de l'Ouest, CIRAD-Univ Toulouse, 91 pp. https://agritrop.cirad.fr/599380/

Mapping Crop Types and Cropping Systems in Nigeria with Sentinel-2 Imagery
Authors:
Esther Shupel Ibrahim 1,2,3,*, Philippe Rufin 1,4,7, Leon Nill 1, Bahareh Kamali 2,5, Claas Nendel 2,6,7 and Patrick Hostert 1,7
1 Geography Department, Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany; philippe.rufin@geo.hu-berlin.de (P.R.); leon.nill@hu-berlin.de (L.N.); patrick.hostert@geo.hu-berlin.de (P.H.)
2 Leibniz Centre for Agricultural Landscape Research, Eberswalder Straße 84, 15374 Müncheberg, Germany; bkamali@uni-bonn.de (B.K.); Claas.Nendel@zalf.de (C.N.)
3 National Centre for Remote Sensing, Jos. Rizek Village Jos Eat LGA, P.M.B. 2136, Jos, Plateau state, Nigeria
4 Earth and Life Institute, Université catholique de Louvain, Place Pasteur 3, 1348 Louvain-la-Neuve, Belgium
5 Institute of Crop Science and Resource Conservation, University of Bonn, Katzenburgweg 5, 53115 Bonn, Germany
6 Institute of Biochemistry and Biology, University of Potsdam, Am Mühlenberg 3, 14476 Potsdam, Germany
7 Integrative Research Institute on Transformations of Human-Environment Systems (IRI THESys), Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany


Abstract:
Reliable crop type maps from satellite data are an essential prerequisite for quantifying crop growth, health, and yields. However, such maps do not exist for most parts of Africa, where smallholder farming is the dominant system. Prevalent cloud cover, small farm sizes, and mixed cropping systems pose substantial challenges when creating crop type maps for sub-Saharan Africa (SSA). In another vain, Africa has been underlined as one of the most vulnerable continents to climate change. In many regions of Africa, severe cases of pest and crop diseases are linked to the negative impacts of climate change such as rising average temperatures and changes in precipitation regimes. The problems are prominent in Nigeria, where rainfed subsistence farming dominates and even only slight changes in climate regimes may largely affect cropping systems.
Using remote sensing data and modeling techniques, related risks may be mitigated in the future, but a deep understanding of the mechanisms behind crop diseases or operational early-warning systems are not in place in Nigeria and most parts of SSA. There is accordingly the need to explore remote sensing data more rigorously and to develop methodologies for detecting and mapping crops affected by critical pest and disease. We here suggest a mapping scheme based on freely available Sentinel-2A/B (S2) time series and very high-resolution SkySat data to map the main crops—maize and potato—and intercropping systems including these two crops for the main crop production region of Nigeria, Jos Plateau. We analyzed the spectral-temporal behavior of mixed crop classes to improve our understanding of inter-class spectral mixing. Building on the Framework for Operational Radiometric Correction for Environmental monitoring (FORCE), we preprocessed S2 time series and derived spectral-temporal metrics (STM) from S2 spectral bands for the main temporal cropping windows. These STMs were used as input features in a hierarchical random forest classification. Secondly, we derived a spatial regression model to assess the distribution of fungal diseases in relation to weather and environmental variables. We assessed four environmental factors to explore spatial dependence, autocorrelation and significance of fungal diseases: elevation, slope, aspect, and land cover, coupled with three weather factors: relative humidity, rainfall and temperature, as well as the concentration densities of fields in the Jos Plateau cropland densities as a proxy for management practices. Increased risk of fungal disease infestations in areas where a susceptible crop densely present, is commonly assumed to occur with daily temperatures > 21oC and is favorable throughout the critical cropping season period (June-July) and frequent episodes of leaf wetness. The latter is a result of either high relative humidity (>70%) or precipitation events, which we use both as proxies.
Our crop type mapping resulted in the first wall-to-wall crop type map for this key agricultural region of Nigeria and achieved an overall accuracy of 84% for crop/non-crop discrimination, and 72% for the five most relevant crop classes (including complex inter-cropping). Plot analyses based on a sample of 1,166 individual fields revealed largely homogeneous mapping patterns, demonstrating the effectiveness of our classification system also for intercropped classes, which are temporally and spatially heterogeneous. Moreover, we found that small field sizes (75% of fields smaller than 1 ha) were dominant in all crop types, regardless of whether or not intercropping was used. Our study offers guidance for creating crop type maps for smallholder-dominated systems with intercropping and underpins the importance of understanding critical temporal windows and related STMs for crop type differentiation and disease mapping. The study supports future smart agricultural practices related to food security, early warning systems, agricultural policies, and extension services.

The Doukkala irrigation scheme, in western Morocco is one of the most important irrigated areas of the country. Investigations carried out under the MOSES (Managing crOp water Saving with Enterprise Services) _H2020 project allowed the demonstration of several remote sensing information services during the agricultural seasons 2016-2017 and 2017-2018: Seasonal probabilistic weather forecast, Early-season and In-season crop mapping, Monitoring of crop water demand and Short-term forecasts of irrigation water requirements. The aim of the MOSES project was the development of a platform capable of providing a set of useful information products to different categories of users involved in irrigation water management. We have evaluated two MOSES products: a) crop maps and b) crop water demand.

Weekly Crop Water Demand forecast CWDF were analyzed against data on actual water allocation in the irrigation season 2017-2018. The weekly CWD outputs were calculated within the MOSES platform using L8/ OLI and S2/ MSI cloud free images available within the week and weekly short-term forecasts. We estimated ETc with two different methods: FAO-56 (kc—NDVI) and an analytical approach. The In-Season Crop maps (ICM) were produced exclusively out of Sentinel-2 imagery.

This study was focused on the assessment of the adequacy of the water applied to meet the crop water demand in the irrigation seasons2017-2018.

Monitoring of crop water demand (CWD) was based on the estimation of the maximum crop evapotranspiration, obtained from remote sensing data of the monitored area. Such output is updated frequently (e.g., every week) during the irrigation season and compared to the weekly surface irrigation water volumes allocated. The assessment of adequacy of allocations against the crop water demand (CWD) showed that crop water requirement (CWR) was 10-15% larger than allocated surface water for the entire area, with this difference being small at the beginning of the growing season.

In season crop mapping (ICM), shows that the best classification performance is achieved in the first part of the season. In the December-January timeframe the classifications achieved a 60%-65% Kappa accuracy. In such crop development phase, the phenology and the corresponding spectral diversity are playing a major role.

The use of MOSES products during the irrigation management operations would help the water management authority to save water, especially during the winter season, leaving additional water available to meet requirements in spring and summer.

Keywords: MOSES, irrigation management, water requirement, Forecast.

Global biodiversity and ecosystem services are under high pressure of human impact. Although avoiding, reducing and reversing the impacts of human activities on ecosystems and especially on protected areas should be an urgent priority, the loss of biodiversity continues. One of the main drivers of biodiversity loss is land use change and land degradation. In South Africa land degradation has a long history and is of great concern. The project SALDi (South African Land Degradation Monitor) aims on developing new, adaptive, and sustainable tools for assessing land degradation by addressing the dynamics and functioning of multi-use landscapes with respect to land use change and ecosystem services.
Within SALDi ready-to-use earth observation (EO) data cubes are developed. EO data cubes are useful and effective tools utilizing earth observations to deliver decision-ready products to non-remote sensing experts. By accessing, storing, and processing of remote sensing products and time-series in data cubes, the efficient monitoring of land degradation in time and space can therefore be enabled. Within the SALDi data cube pre-processed optical (Sentinel-2) and radar (Sentinel-1) satellite data since 2015 is ingested. It is mainly designed to monitor intra- and interannual vegetation dynamics as well as change detection and land degradation in a spatial high resolution (up to 10 m pixel size) in various region of South Africa. Two areas are located in the coastal lowlands, while the other four areas are located on the central high plateau. Those six research areas are evenly distributed in various regions of South Africa and cover both, a climatological as well as an ecological gradient in different ecosystems and some of them are in protected areas. To analyze these diverse regions, not only the Sentinel satellite data, but also further ready-to-use datasets are included in the data cube: 10+ vegetation indices, soil indices, digital elevation model, water mask, 70+ land cover classes. Besides building the technical infrastructure and providing the EO data, training materials and courses are designed to enable users working on the SALDi data cube.