System Implementation and Capacity Building for Satellite Based Agricultural Monitoring and Crop Statistics in Kenya
Over recent years the availability of data for EO multi-source/platform has greatly increased thus enabling the improvement of techniques for retrieving environmental variables in the fields of ecosystem functions, natural resource management, natural hazards and Earth system modeling.
These variables, however, rarely are used in crop models to derive proxy variables (e.g. yield, consumption of nitrates, proteins content, etc.) of strong operational value (quantitative remote sensing). To date the methods based on the use of remote sensing data for monitoring crops and, in particular, of cereal crops both at the local and regional levels, are still far from being effective and operational. This is because they are not yet able to monitor the different stages of agricultural activities, i.e. to detect the crops temporal dynamics in a systematic way to be functional for defining corrective actions at the field scale during the growing crop season.
The present proposal deals with a Italian-Kenyan initiative aiming at exploiting the combined use of optical and radar satellite data to retrieve bio-physical characteristics of crops. The synergistic use of high resolution radar and multispectral optical data (considering the lack of satellite hyperspectral data until 2017) will be exploited for extracting crop plant information (e.g. LAI, biomass map, crop yield) useful to train/assimilate these data into crop growing models such as DSSAT. Crop models that can produce valuable information for site-specific agricultural management practices.
Agricultural land cover maps are critical for monitoring the current conditions and long-term changes of crop and pasture lands. The UN FAO Africover Project generated the most recent land cover map for Central-Eastern African countries in the early 2000s, but since then, population growth and major policy changes have caused land use to shift to increased agro-pastoralism and systematic expansion of cropland area, spurring the need for updated maps of cropped areas. Thanks to the possibility of exploiting ESA (European Space Agency) CAT_01 opportunities, in this project we will try to use, beside the Landsat and Sentinel-2 images (30 m and 20 m spatial resolution, respectively), high resolution RapidEye imagery (5 m spatial resolution) to map the current extent of agricultural areas.
Vegetation indices (VIs) derived from EO are strongly correlated with crop condition parameters (Bannari, 1995). Consequently, EO data have tremendous value for monitoring crop condition and production, and for providing data necessary for developing food security and early warning systems.
The remote sensing-based agricultural monitoring system developed in the framework of this projectwill, hopefully, expand and enhance the Kenyan Government's capability to monitor crop conditions and forecast food shortages and famines. In the present project proposal the GLAM (Global Agricultural Monitoring System, developed by USDA) concept developed for MODIS images will be transferred to OLI/Landsat 8 (http://landsat.usgs.gov/) images and to the Sentinel 2 data (https://sentinel.esa.int/web/sentinel/missions/sentinel-2).
Further, in cooperation with the University of Nairobi, the Web-Enabled Landsat Data (WELD) system (Roy, 2010), developed in the framework of a NASA-funded project at the Unites States Geological Survey (USGS), will be implemented, for producing a periodic, science‐quality satellite im ages mosaics at national scales at 30 m spatial resolution. The periodic satellite based analysis of the crop status, assimilated into dynamic crop growth simulation models, could help in developing a tool for the crop productivity forecast.
Data and Method
The activity will focus on the countries of the Central Africa and in particular the countries that fall within the acquisition circle of the Broglio Space Centre (BSC) ground station in Malindi (Kenya).The first year of the project involves the development of the processing chain and the classification methodology of the MODIS/Landsat images for the area of Kenya, then, after completing the validation phase and having demonstrated the operation of the system the possibility to extend the procedure to other countries (Tanzania, Uganda etc.) could be considered.
Kenya is almost divided in two by the Equatorial plane and occupies a total area of about 583000 km2. Much of the country lies in the eastern part of the Sudano-Saheliana belt, a region often hit by drought and desertification (Waweru, 2003; Nyokabi, 2004).
The country is divided into 7 agro-climatic zones (Sombroek, 1982) of which the first three, related to wetlands, comprise 12% of the territory. Zones 4-7, classified as semi-arid and very arid land, comprise the remaining 88% of the Kenyan territory. Therefore only the 12% of country have high potential for cropping (http://www.fao.org/ag/AGP/AGPC/doc/counprof/kenya.htm). The Kenya's population has increased from about 10 million in 1960 to the current 32 million.
Briefly, the scientific fallout of this project for the Kenyan institutions can be summarized as follows:
development of a methodology for updating land use maps for Kenya;
implementation of the HW/SW infrastructures for running the WELD system and a system derived by SIAM (Satellite Image Automatic MapperTM, a Maryland University registered trademark) (Baraldi, 2011);
training of students/researchers on the use of satellite images;
involvement of the local authorities in the use of satellite images;
enhancement of the potential of the BSC remote sensing station.
The extent of the area of interest and the presence of very different environments, will require the acquisition of a greater number of ground-truth data on the vegetation cover types and their spatial distribution. The contribution of local institutions will be essential for gathering this kind of information.
The cover classes will be organized according to a standard legend like the one available in Africover (Di Gregorio, 1998) (or more evolved). To manage as best as possible the large amount of satellite data that will become available, it will be necessary to define a method which is relatively simple, reproducible and applicable to all the acquired images. The image classification can proceed on two levels. The first one, based on an object oriented approach, aims at discriminating at the highest hierarchical level the class of interest (agricultural) from the other land cover classes: forested areas, bare soils, urban areas, water bodies, etc. The second one, based on the application of a supervised algorithm for each layer defined in the previous step, aims at classifying the spectral clusters in the thematic classes defined according to the adopted legend. The produced thematic maps will be made available in the most common format (raster and/or vector) in accordance with the user needs.In particular, the application of the optical images will be devoted both to produce updated maps of the agricultural areas and to estimate the crop status through the comparison of the indices commonly used for the estimation of the vegetation status (NDVI, NDII, MSAVI, GEMI, etc) obtained by processing images acquired over the same area during the same period of previous years (Laneve, 2005).
The change detection is one of the most important remote sensing applications. The detection of changes that take place in land cover allows to understand how a particular region is changing over time (Nyokabi 2004; Waweru, 2003). Several change detection techniques have been developed in the past years to facilitate the detection of changes in land cover.
The changes of the area will be analyzed through the landscape indices (landscape metrics). These indices (diversity, uniformity, intermix, etc.) summarize some spatial aspects related to the structure of the environmental mosaics. They can be useful to identify the factors and mechanisms that influence the transformation of the landscape and to develop models of landscape change (Fichera, 2013).
The additional use of SAR images (CSK, Sentinel-1 and ALOS) will provide a further tool for estimating the state of cultures. The methodology is based on the correlation that exists between the sensor measurements of backscattering and the status of the culture (or biomass) in the pixel. In the case of ALOS/Palsar images the study will be carried out through a polarimetric analysis. Both new and archived CSK and ALOS images will be acquired on the area of interest, defined according to the results of the classification performed by means of the optical images. If possible, the field campaigns will be carried out during the periods in which the acquisition of new radar images is made possible from the satellite-ground observation geometry.
In the framework of the project, the possibility of estimating the culture productivity by means of bio-physical variables assimilation techniques will be explored. The variables will be computed by using optical remote sensing data and the Acquacrop model, recently developed by FAO (Food and Agriculture Organization) (http://www.fao.org/nr/water/aquacrop.html). Currently, this model does not use any variable retrieved by remote sensing to quantify the response of crop to water shortages in terms of reduction of the crop yield. Different methods (deterministic and stochastic) for the assimilation of EO data will be developed and compared.
A field campaign has been conducted in August 2015 and a second one has been planned for 2016.