Monitoring Desertification with Open Data
The problem of desertification
Desertification constitutes a worldwide problem that directly affects millions of people. It’s a problem created by human activities and it is growing.
But what is desertification? The United Nations Convention to Combat Desertification (UNCCD) defines desertification as the degradation of dry land, primarily due to human activities and climatic drivers. The term "degradation" specifically describes a loss of biological or economic productivity and complexity in croplands, pastures or forests. "Dry lands" in turn are those where total annual precipitation is a small fraction of the atmospheric demand of water, and are therefore under regular water deficit.
Dry lands cover approximately 40% of the Earth's land surface, and tend to have scarce and irregular precipitation, large daily temperature changes and shallow soils with low organic matter. These regions usually also exhibit some specialist adaptations to drought by the ecosystem as well as the human populations which exploit it. The UNCCD uses the aridity index to define and categorise the affected areas which it targets.
A second important point that we can derive from the UNCCD definitions above is that desertification is the cause - not the effect - of land degradation. Such a subtle distinction is extremely important. Desertification is associated with improper management of land that is not very productive and normally under water deficit. Examples of improper management include overgrazing, forced crops or aquifer exhaustion. Regular application of such practices will degrade land beyond its resilience capacity. It is at that point that land degradation symptoms appear - reduced vegetation cover, loss of productivity, runoff erosion, etc. Land does not become spontaneously degraded, pushing its human population into deserts - that idea is a widely held misconception.
What is being done?
There are several efforts worldwide to prevent desertification and mitigate its effects. The UNCCD is the sole legally binding international agreement which links the environment and development to sustainable land management.
The UNCCD works to combat desertification in many ways, including the dissemination of new Agro-ecology techniques which can reverse desertification. These efforts require constant monitoring for evaluation, thus the UNCCD record several environmental and humanitarian variables which are valuable indicators of actions taken or actions needed. The results of this monitoring are reported annually by each country's representative or National Focal Point.
The MELODIES’ “Desertification indicators” service
The purpose of the MELODIES’ desertification service is to create services and tools which facilitate the production and sharing of the UNCCD environmental indicators from open data. The service will support affected countries in their reporting obligations. To best-fit the structure of the required reporting we are developing three self-contained service chains:
These sub-services will produce Technical Auxiliary Indicators which relate to some of the requirements of the UNCCD - specified in the latest Advisory Group of Technical Experts for Impact Indicator Refinement (AGTE) recommendations: the Aridity index, Trends in land productivity or function, and Trends in land cover structure.
The Land Use/Land Cover sub-service
The Land Use/Land Cover (LULC) service will generate maps of LULC and LULC-changes data for the “Trends in Land Cover Structure” indicator. The LULC service works in four steps:
We automatically identify appropriate Landsat 5, 7 and 8 images over the target region in both the wet season and dry season. These are downloaded on to the cloud server.
Here we perform corrections to prepare the data for analysis, these include:
- radiometric and geometric corrections,
- conversion to at-Sensor Brightness Temperature,
- cloud cover detection and removal,
- values normalization
3. Building LULC maps
We produce LULC maps through a per-pixel classification of the Landsat data, after outliers from the training data have been removed. The classification algorithm (Linear Discriminant Classifier) is applied to the spectral bands and NDVI of the wet and dry season Landsat images in order to distinguish given classes with seasonal changes. The training dataset may be provided by the users.
At the post-processing stage, the classification result is subject to generalization and cleaning algorithms to guarantee a user-defined Minimum Mapping Unit. An assessment of the thematic accuracy of the maps is then automatically performed, based on a random sample of pixels - the "testing set" – which are compared with the map.
In the MELODIES project, we are aiming to produce several LULC maps. From these, we will also produce maps of LULC changes between different reference years. We detect LULC changes by comparing the independently-produced classifications for the same pixel at different points in time and recording any changes between time intervals. The individual combinations of LULC changes at each pixel and time interval are used to derive a "trend class" which represents the change in LULC over time. These classes are drawn from a user-configurable matrix of every possible combination of LULC changes. So for example, a land cover change from "forest" to "herbaceous vegetation" might have the trend class “Deforestation”. The final LULC-changes map is a map of these trend classes.
The Land Condition sub-service
This sub-service is targeted at the new UNCCD indicator “Trends in land productivity or functioning of the Land” (LCS). LCS aims to assess the extent and intensity of the loss of ecological function and of soil and land degradation. Because this represents an extreme on the spectrum of ecological functionality, the focus of our sub-service is on the wider concept of land condition. To do this, we work with two of the UNCCD indicators: Net Primary Productivity (NPP) and Rain Use Efficiency (RUE)
NPP is defined as the amount of biomass produced by vegetation in a given time span. It is a direct measurement of ecosystem vigour, but its value as an indicator is relatively low because biomass production depends heavily on other limiting factors. As a result, NPP tends to oscillate according to these factors, which makes it harder to measure the result of a target process. In dry lands, water availability is the main controlling factor for NPP; in these areas RUE can correct the dependency problem because it is defined as the NPP per unit rainfall.
The sub-service of land condition is built using NPP and RUE as its foundations; it has two separate functions: assessment and monitoring. Assessment makes a synchronic comparison of every location to all other locations in a similar level of aridity across space, in order to detect states of land condition and their corresponding indices. Monitoring makes a diachronic comparison of every location with itself over time, to detect gain or loss of biomass in order to detect land condition trends. Finally, assessment and monitoring are merged in to a final land condition map following some explicit rules.
The indicator “Trends in land productivity or functioning of the Land” is explicitly addressed by the monitoring component of the Land Condition service. It separates intrinsic trends of land productivity (which are truly associated with land degradation or land progression under ecological succession) from extrinsic trends (i.e. those related to climate drift).
In addition to the land condition data, the aridity index is also computed at several points in time and aggregation levels. This is also offered by the sub-service as an indicator.
The Susceptibility to Desertification sub-service
The Susceptibility to Desertification Indicator (ISD) provides us with a picture of desertification change at local and national levels and allows us to identify regions that are more predisposed to change. ISD provides - for each land cover patch - the susceptibility to desertification due to vegetation, soil and climate. ISD’s are relevant to UNCCD indicators VI - Degree of land degradation, VII- Drought index and IX – Land cover status.
ISD is an integration between biophysical and climatic parameters. The biophysical parameters include the health of the vegetation at the end of the growing season in specific land cover, derived from the Normalized Difference Vegetation Index (NDVI), and the quality of the soils (in turn derived from the soil brightness in specific land cover). Both these parameters are assessed with Earth Observation data.
The climatic parameters are computed using precipitation dynamics, which are derived from the ECWMF climate reanalysis datasets. These parameters are then integrated with geostatistical methodologies in order to determine ISD.
How will the work in MELODIES benefit the monitoring of desertification?
The developments that we make within MELODIES will improve the monitoring and reporting of environmental indicators of desertification, both by improving of the services (and consequently the quality of the generated products), and by improving the technical infrastructure on which the service is hosted, allowing retrieving new results in a timely fashion.
Cloud computing and parallel processing
The integration of the Desertification Indicators service in the MELODIES technical platform will allow us to remotely generate our products using cloud computing and parallel processing. This will enable faster delivery to the users and easier refining of the results, e.g. if the quality assessment of the Land Cover products is not satisfactory, the user inputs new training data and will retrieve the new results quickly.
Linked Open Data
Converting the Desertification Indicators service data products to a Linked Open Data format will allow users more flexibility in exploring our environmental data, providing further information to help the combat and mitigation of Desertification. As well as this, opening the doors on our data either with a new licensing solution and/or with easy links to other datasets via Linked Open Data, will pave the way for new products, new users and new solutions to forthcoming problems in the future.
This video demonstrates how users can assess LULC changes by querying Linked Open Data (please turn subtitles on):
The User Interface
Finally, conveying these services through a common and dedicated web user interface will ease the reporting processes, either by allowing users to refine available parameters of the services themselves or by enabling online visualization and reporting tools.