Task 2: Develeping a Synoptic Prediction System (SPS)

2.1. Introduction
2.2. Data sources
2.3. Classification 1

Predicting contemporary landuse
Forecasting future landuse
Translating into land degradation impacts
Comments

2.4. Classification 2

Predicting contemporary landuse
Forecasting future landuse
Translating into land degradation impacts
Comments

2.5. General Comments and ideas for improvements

2.1. Introduction

The aim is to develop a fuzzy-neuro modelling framework able to link socio-economic data with physical-climatic data in order to translate climatic and environmental land use change impacts in socio-economic terms.

It is not easy to predict the impact of climatic change on land use let alone present the results in a form that decision makers can understand. Simply mapping variables which can be used to categorise and describe the land-climate (environment) is not sufficient to discover what the variables mean as indicators of climatic or landuse change. Maps are an extremely useful way of presenting the results so that they can be understood and acted upon, but much more than mapping is required during the analysis stages. The following few sections describe the development of the Synoptic Prediction System (SPS) which is designed to integrate socio-economic data with other physical land and climate environmental data in order to predict the impact of climatic and environmental change on land-use.

The first major task in classifying the likely effects of climate change on landuse involved developing and comparing present and future land use scenarios. Neural networks like those employed in Task 1 were trained to model the relationship between climate (temperature and rainfall), soil characteristics (permeability, texture, fertility, parent material), biomass, elevation, population densities, other socio-economic variables, and agricultural landuse. Having trained the neural networks to classify contemporary landuse they were applied to classify future landuse using predicted values of the input variables. The predictions and forecasts of landuse were then translated into land degradation terms using fuzzy statements about landuse changes. Further analysing the difference between contemporary and future classifications and comparing with ESA definitions in Medalus case study areas is covered in Task 3. Various future scenarios can be simulated by directly changing forecasts of the inputs. The first run is unlikely to produce results which enable the precise location of ESA areas to be determined although this become increasingly likely as the data inputs and modelling procedure improves in subsequent models. It is hoped that the first set of results make sense at a country scale and could be used directly to help allocate EU funds to mitigate land degradation, but it would be better to wait for later model outputs as the information becomes more reliable.

The first essential step in developing the SPS involved acquiring input data for now and future forecasts.

2.2. Data sources

Socio-economic data developed from Task 1;
Physical data outputs from other MEDALUS III projects.
Climate data from the CRU
Soils geographical database of Europe version 3.2
GLOBE: Global Land One-KM Base Elevation Data version 0.1
Satellite data (none acquired yet!)

Data from the above sources was manipulated using ArcInfo and ArcView into 1DM raster grid inputs for the SPS.

2.3. Classification 1

A first attempt at modelling agricultural landuse using neural networks!

Dominant landuse was classed into 4 categories; arable, trees and orchards, waste and others. Neural networks were then trained to classify each landuse seperately based on the various physical and socio-economic inputs described below. Having trained to predict contemporary landuse the neural networks were applied to generate landuse forecasts using forecast values of the predictor variables. The difference between predicted contemporary and forecast agricultural landuse was then translated into land degradation terms using fuzzy inference. inputs y seperately, then combine the independent models in a Multi-Criteria Evaluation (MCE) framework using fuzzy logic. The first landuse category selected was arable. Dominant arable landuse occurs on a large proportion of cultivated land which is usually of the highest quality.

Predicting contemporary landuse

Future landuse

Forecasting future landuse involves applying the trained contemporary landuse neural networks using forecasts of the various predictor variables. Click the

Results and translation into land degradation impacts

Maps of observed, predicted and forecast dominant agricultural landuse:

One way in which landuse change can be translated into land degradation impact maps is through the use of fuzzy inference. Details of how this is done will appear here soon. For now click here to view the resulting land degradation map.

Comments

It would be much better to use landuse probabilities instead of 0-1 locations of dominant landuse classes.
Fuzzy inference should be used to generate the soil quality layer.
Soil type, height above sea level and the quality of the soil in terms of physical properties were assumed not to change.
Some other measures of terrain roughness could be included.
As different crops like different kinds of soil when predicting agricultural landuse the soil categorisation could be different in each case. This specification should be implemented in classification 2 below.

2.4. Classification 2

Predicting contemporary landuse

Future classification

Future classification involves applying the trained contemporary landuse neural networks using forecasts of the various predictor variable. Click the

Maps of observed, predicted and forecast agricultural landuse:

Translation into land degradation impacts

Comments

2.5. General comments and ideas for improvements

Land capability, land suitability and landuse patterns can be analysed in an attempt to identify areas of appropriate and inappropriate landuse. (It would be useful to be able to identify areas which are suitable for supporting landuse Y with only a small erosion or biomass reduction risk but which is presently farmed as landuse X posing greater risks.) It maybe possible to investigate socio-economic reasons for inappropriate landuse descisions.

EU CAP subsides may contribute to landuse decisions which encouraging short term economic reasoning and result in land degradation. It maybe that an increasingly large number of farmers have become unlikely to pass on working farms to their children. There maybe a general lifestyle change driving agricultural abandonment, where there are less people maintaining the land it is more likely to degrade? Whilst this change takes place by in funding his children through college/university which requires money in the short term. Whilst his children are away the farmer may no longer be capable of looking after the entire farm properly and may opt for a subsidised crop and hope land degredation impacts never really take effect.) It maybe that farming decisions which decrease land degredation risk can be encouraged with a form of set a side as an option to inappropriate subsidised landuse and that land help from local support groups could return the land to a more natural stable state capable of being farmed at some point in the future.

There are essentially two main types of land degradation: that which is related to physical land deterioration, reduction in biomass and loss of soil; and, that which results in less profit or revenue being generated by agriculture. Although the two are closely related they are quite different. To fit in with the rest of the Medalus work here the focus is on the dynamics of landuse change which result in a switch in land from being agriculture capable to being agriculture incapable. I would argue that in this case the focus is on the physical land deterioration type of land degradation.

This page was last modified in July 1998.