→ Boosted regression trees (BRTs) were used to build predictive models of the presence of deep peat given three topographic predictors: slope; elevation; and topographic wetness index (TWI) at 50 m resolution.

→ Peat depth data (n = 24,563) were divided into a training dataset (n = 10,000), used to build the models, and a testing data set (n = 14,563), used to evaluate model predictive performance.

→ Multiple BRT models were built on the training dataset while tree complexity and learning rate systematically altered to maximise kappa (the Proportion of Correctly Classified observations [PCC] after accounting for probability of chance agreement).

→ The bag fraction was kept at a constant 75% to maintain a level of stochasticity, which is important for increasing predictive success.

→ Models were then tested against the test dataset and evaluated using percentage deviance explained (% dev), area under the receiver operation curve (AUC) and correlation with observed values (high being better).

285000

285000

290000

290000

295000

295000

1950

0020

0000

2050

00

Project:

Pen y Cymoedd Wind Farm

Title:

Verification of Deep PeatModelling with PIMP Data

Ref: GB200326_M_921_A1

Date: 29-01-15

© Crown Copyright 2015. All rights reserved. Ordnance SurveyLicence 0100031673.

¯

Drawing by:Natural Power Consultants LtdThe Green HouseForrest Estate, DalryCastle Douglas, DG7 3XS, UKTel: +44 (0)1644 430008Fax: +44 (0)845 299 1236Email: sayhello@naturalpower.comwww.naturalpower.com

Prepared by: FC Checked by: MS

Notes:a) Information on this map is directly reproduced from digital and other materialfrom different sources. Minor discrepancies may therefore occur. Where furtherclarification is considered necessary, this is noted through the use of text boxeson the map itself.b) For the avoidance of doubt and unless otherwise stated:

1. where a line recorded in the key demarcates a boundary on this plan, the boundary edge is the outside edge of the line.2. where a line or feature recorded in the key of this plan is also shown as a line or feature by the Ordnance Survey, and that line or feature is located in a different position on the ground than shown by the Ordnance Survey, then the line or feature shall be deemed to follow the position as existing on the ground.3. this plan should be used for identification purposes only, unless specifically stated above or in accompanying documentation.4. Natural Power Consultants Ltd. accepts no responsibility for the accuracy of data supplied by third parties.

Scale @ A3: 1:52,625

Key

Site boundary

Modelled Deep Peat (p > 0.59)

PIMP Data Deep Peat (>0.5 m)*

0 1 20.5 km

Coordinate System: British National Grid

Client:Vattenfall Wind Power Limited1 Caspian Point,Pierhead Street,Cardiff Bay,CF10 4DQTel +44 (0) 1736 330171www.vattenfall.com

*Data presented is that obtained up to 27/06/2014 and represents all data collectedas part of the PIMP

A modelling approach to identify peat depth and its use within habitat management plans.

KATE DAVIES, FIONA CARYL AND MIKE SHEWRING

Peat bogs are a valued feature across many upland areas of Wales. These upland areas also provide a good wind resource, which has lead to many wind farms being located in these areas. Consent conditions of any wind farm developments require restoration of peatland disturbed by wind farm construction, often through the implementation of a Habitat Management Plan (HMP).

To identify peat-dependent habitats present within the Habitat Management Area (HMA), peat depth information is required. Peat depth data are usually collected manually by probing at regular grid intervals, which is a both time and labour intensive. Peat depths are interpolated within surveyed areas but are limited in that interpolations cannot be extrapolated to new areas. Here we demonstrate how survey effort can be reduced by using machine learning models to predict the distribution of deep peat (>0.5 m) and map areas to be covered by the HMP.

→ Pen y Cymoedd Wind Farm, South Wales

→ 76 turbines

→ HMA Area of 1441 ha

→ Several broad peat covered upland plateaus, bounded by steep sided valleys used primarily for coniferous plantation

Introduction

Study area

Methods

Results

Map 1: Peat depth interpolation

Figure 1: Peat depth interpolation

Map 2: Existing mapped soil layers

naturalpower.comsayhello@naturalpower.com

No part of this document or translations of it may be reproduced or transmitted in any form or by any means, electronic or mechanical including photocopying, recording or any other information storage and retrieval system,

without prior permission in writing from Natural Power. All facts and figures correct at time of print.All rights reserved. © Copyright 2017.

→ The top model found that slope (39%) and elevation (36%) were the most influential predictors. TWI contributed 25% to the model.

→ Generally once the slope exceeds 10o at a site deep peat is much less likely to be present; deep peat is more likely to be at elevations between 300-340 m a.s.l.; and sites with high TWI (values above 14) are more likely to contain deep peat (Fig. 1).

→ When predicted over the wind farm development area, deep peat was predicted to be present in an area of 20.4 km2 (See maps below).

→ Models were validated through comparison with peat distribution resulting from peat depth interpolation (Map 1) and with existing mapped soils layers (Map 2) and showed a high degree of congruence with each.

ConclusionThis model provides a valuable resource for rapidly assessing and understanding the distribution of peat habitats within a site, which can then be used to inform management prescriptions within habitat management plans. The model requires testing for generality at other sites, but should it prove successful it would considerably reduce the effort required to map sensitive peat habitats.

Work carried out on behalf of our client Vattenfall.