Symposium: Climate Change, Agriculture and Biodiversity in South Africa 17-18 August 2011 Cape Town, South Africa http://www.princeton.edu/~lestes/Symposium/Information.html Contents Introduction and Research Rationale .................................................................................................................. 2 Discussion outcomes......................................................................................................................................... 2 Key Messages .................................................................................................................................................. 4 Theme 1: Agriculture/Livelihoods Impacts .......................................................................................................... 4 Giving weight to multiple objectives for sustainable agricultural decisions under changing conditions – Oliver Crespo (Uni. Cape Town Climate Systems Analysis Group) ............................................................................... 4 Investigating and mapping yield impacts on wheat production zones in the Western Cape under future climate change – Mike Wallace (Western Cape Dept. of Agriculture) ............................................................................. 5 Preliminary assessment of economic impacts of climate change on SA agriculture – Guy Midgley (SANBI) ............ 5 Climate change impacts to wheat and maize in South Africa: a combined modeling approach – Bethany Bradley (University of Massachusetts) ........................................................................................................................ 6 Climate Change and connections between agriculture and biodiversity: what questions should we be asking? – Roland Schultz (University of Kwazulu-Natal) .................................................................................................. 8 Theme 2: Data for Climate Impact Assessments ................................................................................................ 10 Agroclimatology: Climate Network and Databank – Chrisna Henningse (Agricultural Research Council, Institute for Soil, Climate and Water) ............................................................................................................................. 10 The South African Natural Resource Information System – Garry Paterson (Agricultural Research Council, Institute for Soil, Climate and Water) ........................................................................................................................ 10 Additional derived datasets arising from research – Lyndon Estes (Princeton Uni.) ............................................ 10 UCT’s downscaled climate data - Mark Tadross (Uni. Cape Town Climate Systems Analysis Group) .................... 11 Impacts of heat stress on dairy cattle productivity under projected human induced climate change – Rantuku (Gabriel) Lekalakala (Limpopo Dept. of Agriculture) ....................................................................................... 12 Theme 3: Biodiversity Impacts ......................................................................................................................... 13 Implications of climate scenario uncertainties and approaches to dealing with these - Guy Midgley (SANBI) ....... 13 Revised biome change predictions for South Africa – some provisional outputs with discussion of potential impacts on biodiversity and protected areas – Stephen Holness (SANParks) ................................................................. 14 Modeling relationships between species’ spatial abundance patterns and climate – David Hole (Conservation International) ............................................................................................................................................. 16 Climate risk and vulnerability mapping for Southern Africa: Status Quo (2008) and future (2050) – Stephanie Midgley (OneWorld) .................................................................................................................................... 17 Global Change in rural rangelands: implications for biodiversity and livelihoods – Barend Erasmus (Uni. Witwatersrand) .......................................................................................................................................... 19 Overlap between protected areas and agricultural change – Lyndon Estes (Princeton Uni.) ............................... 20 Acknowledgements ......................................................................................................................................... 23 1 Introduction and Research Rationale On 17th -18th August 2011 a two day symposium on Climate Change, Agriculture and Biodiversity in South Africa (CCAB) was held at the Center for Biodiversity Conservation in Cape Town, hosted by Princeton University and facilitated by Lyndon Estes (Princeton Uni.) and Sarshen Marais (Conservation South Africa). The symposium brought together researchers, students, NGOs, consultants, local government officials and conservation agencies to discuss questions and share research findings around the overarching theme of climate change adaptation in the context of human livelihoods and biodiversity maintenance. The role of the symposium as a knowledge sharing and discussion exercise is part of three years of research by Lyndon Estes and Bethany Bradley (Uni. Massachusetts); in collaboration with Conservation International; to examine the use of various models to predict how both biodiversity and human livelihoods (in the form of agriculture) will be affected with ongoing climate change. The symposium was a platform to showcase their research to collaborators and interested parties within South Africa and to place it in the context of existing climate change work. There has been little focus on the interaction between projected climate change adaptation on terms of human livelihoods, and that of biodiversity and natural ecosystems. The CCAB research focuses around merging these parallel trends in climate change impacts research. This work is of particular value because human adaptation to climate change may have a larger impact on biodiversity than the direct effects of changing temperature and precipitation regimes. The research focuses on one particular human livelihood activity within South Africa: agriculture. South Africa was specifically chosen as the country of study due to its large and diverse agricultural sector, its globally important biodiversity, and its projected vulnerability to climate change. Information shared and discussed on Day 1 of the symposium was presented under the theme of Agriculture/Livelihood Impacts or Data for Climate Impact Assessments. These themes were carried through on Day 2 with presentations around Biodiversity Impacts, with final presentations considering potential interactions between biodiversity and livelihood impacts. To this end, during discussion sessions symposium members were guided by five central themes: Patterns emerging from spatial comparison of presented results (interactive GIS exercise). Methodological issues relating to strength/weaknesses and comparability/compatibility of various input datasets and modeling approaches. Possible improvements and alternative methodologies. Identifying future research and potential collaborations. What are the policy implications of results? Discussion outcomes Methodological issues relating to strength/weaknesses of various modeling approaches The value of empirical vs. mechanistic models was discussed. Bradley explained that in their experience both types of models are suitable for present day outputs, but for future projections they vary in direction and in magnitude. Mechanistic models may provide better future projections because mechanisms are changing (such as sequence of daily rainfall). However, mechanistic models require substantial detail, and the increased data requirements results in a higher risk of error. Holness suggested that there was no reason why empirical models couldn’t include more complex data, and that perhaps the issue was not around complexity of input variables. Tadross suggested that identifying the actual reasons for output variation would be very interesting. Estes also emphasized that his results are preliminary and analyses will be further refined. 2 Data sharing It was agreed that although there would not be another seminar of this nature as Estes’ CCAB postdoc research is coming to an end; there was a need for continuous information sharing which would be the responsibility of all individuals involved as the need arises (through websites, databases, partnerships etc. such as the adaptation network and ACCESS). It was decided that SANBI’s BGIS website could be an appropriate platform for data, provided it had a biodiversity conservation angle. Until such time as BGIS can upload this information, the data will be available by request through Estes’ CCAB website http://www.princeton.edu/~lestes/Symposium/Information.html. It was suggested that conditions of use accompany all data as well as any project report/publication info, and that data is loaded in the format used for the workshop (processed by Genevieve Pence) in order to avoid unnecessary re-formatting. Phoebe Barnard expressed interest in being able to share David Hole’s bird data and Stephen Holness’ ecosystembased adaptation (EbA) maps with her students. It was also suggested that the CCAB model outputs be shared at the upcoming Grasslands Partners Forum. Future Research It was suggested that information from Guy Midgley’s presentation regarding uncertainty around rate and intensity of change would be very valuable for framing future impact studies, as well as for guiding researchers in terms of knowing what levels of importance one can attach to impact studies, based on how confident one is of the accuracy the input data. Guy Midgley highlighted the importance of gathering data now with which to test species models results in the 20202050 time frame. If data is not continuously collected as the climate changes, then we will have nothing to test models against. Emphasis should be on the value of real data, rather than investing so much in current modeling. Guy Midgley also suggested that if one could find out what had triggered agricultural practices to change in the past (for example, crop changes), these historical shifts in farm activity could be combined with modeling techniques. From a biodiversity conservation perspective, areas of gained suitability for crops in the future were thought to be of more use than areas of decreased suitability. This is in contrast to the value of lost suitability predictions for economic/social planning. It was suggested that there would be value in looking at economic/social issues (Stephanie Midgley’s work) and integrating these through modeling to guide the types of human adaptation that might happen in relation to protected areas (Holness’ work). This could also link into policy as suggested below. Combining CCAB’s modeling work with Olivier Crespo’s work on optimization of decision making was also suggested as an interesting idea for future research. Regarding the CCAB wheat and maize suitability work, it could be useful to explore suitability changes for different crops (for example, olives in the Overberg if wheat becomes unviable). Rather than further refinement on the wheat and maize results (which is Estes’ intention as the results are preliminary), one could produce similar quality models for other crops, as the combined picture is most important. There were musings around the possibility of modeling livestock suitability in areas such as the Swartland. Policy Implications There is a need for an integrated discussion between sectors of mutual interest and conflict, for example rangeland expansion into areas predicted for EbA doesn’t have to result in conflict, if properly managed we could have positive outcomes for all. It was suggested that data layers be made available to inform decision makers during processes such as wind farm scoping – could ‘future EbA’ or ‘future agricultural value’ be incorporated into the Environmental Impact Assessment process? 3 Similarly, it was suggested that these layers be made available for use as part of South Africa’s NBSAP (National Biodiversity and Spatial Action Plan), and that they would be of great value if included in National Agricultural Plans. It was suggested that once the data have been lodged on Estes’ CCAB site/BGIS, that the National and Provincial departments of Agriculture be made aware that they were available for download. Estes suggested that by overlaying Stephanie Midgley’s vulnerability or future problem areas map with their map demonstrating future yield loss and unviable agricultural areas, they could produce a useful guideline for where agricultural/social support would be most required. Promotion of these areas in terms of governmental policy and finance could help communities to maintain productivity or at least build their adaptive capacity through new livelihood options. Key Messages Messaging around climate change findings and implications must be unified and consistent. Researchers focusing on South African climate change impacts should strive to use the same downscaled GCMs for comparative purposes. We should be monitoring change over time, using the ‘natural experiment’ to gather data with which to test models. For example, by collecting data now we can test model results in the 2020-250 time frame. It is of value to design studies and view outputs holistically in SA context: political, social, environmental implications. Presentation summaries Day 1 Theme 1: Agriculture/Livelihoods Impacts Giving weight to multiple objectives for sustainable agricultural decisions under changing conditions – Oliver Crespo (Uni. Cape Town Climate Systems Analysis Group) As Crespo explained, agriculture is significantly affected by climate variables, and as we know the climate is changing. With colleagues at CSAG he has been exploring ways for agriculture to benefit from or at least adapt to climate change using crop models to simulate crop responses and then exploring multiple objectives under uncertainty (e.g. maximizing yield and minimizing N loss in uncertain climate/weather conditions), with the simulations being subject to historical, seasonal, forecast and climate change scenario data. They do this by evaluating how a decision, e.g. irrigation, affects a criterion, e.g. yield. They then simulate the criterion range produced (in this case, range of yields) by a decision range (in this case, no irrigation to intense irrigation), and simulate this subject to multiple weather alternatives. This is then performed for multiple decisions and multiple criteria under multiple weather alternatives, resulting in a multi-criteria group evaluation based on a group of decisions, which is then optimized. This does not result in a single answer, but rather a range of options that one can present to farmers based on which decisions and criteria they consider of highest value. Using the example of rain-fed wheat in the Swartland, Crespo demonstrated that they had determined a rising band of efficient decisions – using present day climate data it was evident that one had the option to sow earlier and use less fertilizer, or to sew later and have a higher yield but more fertilizer loss, thereby requiring more fertilizer. By applying current and future climate models the same rising band was evident in present day, but in the future higher fertilizer at a later sowing date would be the optimum option for highest yield. In summary, this work provides a multi-objective perspective of how farming patterns may change over time and provides a limited number of efficient outcomes with pros and cons of all to aid the final decision making. 4 Investigating and mapping yield impacts on wheat production zones in the Western Cape under future climate change – Mike Wallace (Western Cape Dept. of Agriculture) Wallace began by explaining that wheat farmers have not only recently been struggling because of climate change – The deregulation of the Wheat Board in 1997 changed the industry completely and numerous farmers went out of business. This was followed by many dry years in the Western Cape, with farmers starting to ask the Dept. of Agriculture ‘is this climate change? Is this what we are going to continue to experience in the future?’ As Wallace explains, it is impossible to provide a blanket outlook of what farmers can expect, particularly since climate is so variable across the Western Cape, but they needed to provide some kind of answer for them. The Dept. of Agriculture set out to identify the most appropriate datasets, models and methods, and to quantify predicted climatic changes at a local scale. Climate data were obtained from CSAG, and regional production data were extracted from written records dated to 1989. Wheat areas per zone were mapped and verified, and allowed for identification of areas with high, medium or low productivity. No crop-model specific soil profile data were available, synthesized data from various other sources was used. As Wallace points out, it was a huge effort to parametize the model, and he found very few reliable long term yield data. He also assessed wheat production history per zone, which is challenging as it is not always evident if wheat is a primary or rotation crop, or if it is in fact kept purely for speculative value. The APSIM crop model was used, which was developed in Western Australia in conditions very similar to the Western Cape. It models biological growth processes in response to moisture, heat units, solar radiation and plant nutrition, and runs on daily data. He used two crop modeling approaches: perturbing historical data, and using downscaled data. With the perturbed data yield was very influenced by a change of 2 degrees, and certain areas such as the sandy Cape west coast were not well represented as leaching was not a factor in the model. Interestingly, some areas were found to have decreased yield with increased rainfall due to waterlogging impacts, highlighting the importance of looking at climate impacts on a fine scale across the Western Cape. With the modeled data (using 8 GCMs), he measured yield change between observed and future averages per zone and mapped them in either direction – loss or gain. From this it was evident that the north western half of the Western Cape was likely to experience yield loss, whereas areas further south and south east maintained yield levels. Preliminary conclusions from this study show a loss of yield due to temperature increase largely counteracted by CO2 fertilization in better wheat areas, with current marginal zones and western wheat zones predicted to be most negatively impacted. Preliminary assessment of economic impacts of climate change on SA agriculture – Guy Midgley (SANBI) We know that climate change will influence agricultural demand and supply, and Midgley used the diagram below to demonstrate the variables which farmers can control (e.g. input costs) and how this affects profit, which will ultimately affect land use decisions. 5 Figure: Diagram demonstrating how climate change might factor into commercial farming profits and land use decisions Blignaut et al. (2009) in South African Journal of Science examined how agricultural profits are likely to be affected by climate change, rather than just looking at yield change. They found that almost all regions warmed (by 0.5 deg. av.) and dried (by 6mm av.) over 1970-2006. They were then able to derive algorithms to predict what profitability would do under climate change, keeping all other variables constant. They then looked at average rainfall change and temperature change across all 10 GCM models, and for all the provinces (90 th percentile best and 10th percentile worst case scenario, and median). They combined best case temperature and best case rainfall to get a robust representation of full future range of scenarios. Given these different scenarios, cereals are predicted to be hardest hit. Horticulture is least affected on average, but most sensitive. Using this info, we can start to say something about what kind of an income signal will be sent to farmers – based on best case scenario there is not a very strong signal being sent to farmers to change their land use practices, but based on worst case they should have a strong signal. This is a problem for what message we want to send – we don’t know if it’s a red flag yet, and there is a huge risk for the potential knock on effects on biodiversity. We need to build a deeper understanding at the farming level to see if it will trigger land use change. He suggests we look at the following areas, and that these outlined issues would be a very useful way forward for CCBA research: - Spatial scale; sequences of years may force farmers to change land use practices (in the 20s people abandoned their land – climate variability); how climate variability will affect input costs (reduced labour and input costs, become mechanized, this has greater impact on land and negative consequences for biodiversity); and how do profits relate to farming decisions? (Presumably there must be data available? This can help to influence decisions and related biodiversity impact). Climate change impacts to wheat and maize in South Africa: a combined modeling approach – Bethany Bradley (University of Massachusetts) The goal of their collaborative project is to forecast future shifts in crop suitability and yield in South Africa. As Bradley explained, for indigenous (non-crop) species, one normally uses empirical models to predict how species will change over time because detailed physiological information is always not always available. But can such empirical models realistically be used to predict future changes? Ensemble approaches to modeling have become increasingly popular in the forecasting literature to try to counteract some of the uncertainty associated with any single model. The idea is that if you run multiple scenarios together, you have greater confidence in areas where most of the models overlap (or don’t overlap). This approach is often taken with multiple GCMs or climate scenarios (SRES). This approach is also frequently applied in ecological forecasting using multiple different statistical or empirical 6 models that each treats data slightly differently. Recently, there has been increasing focus on the need for model inter-comparison between different types of modeling approaches; in this case, empirical and mechanistic models. In this study, two different modeling approaches were used: an empirical habitat suitability modeling using MAXENT, which classifies areas as suitable/non suitable; and a physiological/mechanistic model that simulates crop growth and yield, DSSAT. MAXENT has become the dominant habitat modeling programme in ecology. MAXENT takes a species across the landscape, looks at the biogeography of that species, compares that to any variable (soil, land use) and plots that species distribution according to those environmental variables. It essentially takes environmental space and maps it back onto geographic space, and whatever falls into that bubble becomes suitable for that species. The physiological model DSSAT contains much more detailed information than MAXENT, but the tradeoff is that it can be really difficult to take a crop model designed for one crop field and apply it to large landscapes. Input variables were matched as much as possible for the two models. In a comparison of both models’ abilities to correctly identify current maize suitability (with current maize crop field distribution used as a ‘truth’), each model proved to be 87% accurate for maize. In contrast, DSSAT over-predicted suitability for wheat, while MAXENT underpredicted it. Figure: Comparing models of current maize yield The models’ abilities to simulate maize productivity were also assessed. DSSAT yield values and MAXENT suitability values (range 0-1) were compared against observed maize productivity, which was assessed using a satellite-derived yield proxy (integrated NDVI from the MODIS sensor). As shown above, DSSAT showed a linear relationship between modeled and observed yield (albeit with relatively low R2), while MAXENT suitability values quickly saturated with increasing values of integrated NDVI. From this it was determined that although both models are successful in determining suitable areas in current climate conditions, MAXENT will not may not be useful for detecting future changes in maize productivity. Turning to future projections (2046-2055 time frame) of maize suitability based on nine GCMs run under the A2 emissions scenario, DSSAT showed little change relevant to correct maize suitability. In contrast, MAXENT projected substantial lost suitability: see below. a) 7 b) Figure: Future maize suitability predictions by a) DSSAT and b) MAXENT. In conclusion, DSSAT and MAXENT produce comparable present day models for maize suitability, but not wheat suitability. Under future climate conditions DSSAT projections are more optimistic, possibly because 1) MAXENT can only model within the range of current conditions, so may underestimate what is actually possible; and 2) DSSAT may be overly optimistic in terms of the water use efficiency gains, particularly in drier areas. It is also important to note that model inter-comparison is challenging based on temporal resolution, as DSSAT uses daily time sets and MAXENT monthly averages. Climate Change and connections between agriculture and biodiversity: what questions should we be asking? – Roland Schultz (University of Kwazulu-Natal) In this talk Schulz suggested that we need to move from researching outputs around means and magnitude, and focus climate change research in the following areas: Variability, extremes and thresholds. Location, yields, timing and sensitivity. Memory, amplification, critical drivers, and confidence. To give context, Schultz explained how we already live in a high risk climatic environment in South Africa, with high year to year rainfall variability and a very high aridity index. In terms of quinary catchments, natural vegetation is no longer dominant in South Africa and has been replaced by other land uses, such as agriculture. Large areas of the country are degraded, partially as a legacy of subsistence farming during apartheid. In short, we are dealing with a complex system of forward flows and feedbacks (see below) and climate change will have an overarching affect in the Climate-Land Cover-Population-Natural Resources cycle. 8 Figure: Climate- Land Cover- Population-Natural Resources cycle In terms of adding the climate dimension, Schultz explains that South Africa is divided into 5838 Quinary Catchments, each with a 50 year daily weather dataset. From CSAG he used 5 GCMs (A2 scenario), using 20 year time slices to represent ‘present’, ‘near future’ and ‘distant future’. Derived from empirical downscaling to 2600 rainfall stations and 400 temperature stations daily, then overlaid this with a daily hydrological model (inclusive in this were crop yield components). Then analyzed monthly changes in various measurements such as frost days, monthly rainfall variability, number of days when threshold rainfall was exceeded, etc to project how these measurements would change over time. Schultz puts these projections in context by asking: what are the tipping points, how are they projected to change with climate change, and what are the consequences? What are the implications of changes in variability, and changes in extremes? For example, from their work they project the duration of frost days to get shorter, and the date of first frost to come later in the season. What are the implications of this in terms of pest breeding and success, and crop planting time and growing season? Similarly, heat waves three days above 30˚C) and extreme heat waves (three days above 35˚C) are predicted to increase in the future – what are the agricultural implications? Short duration local extreme rainfalls (thunderstorms) are projected to increase – what are the consequences? Will this result in more erosion, and more fires from lightning strikes? More regular large phenomenon such as tropical cyclones are predicted, as well as more regional flooding. What are the agriculture and biodiversity consequences? Changes in chill units, changes in planting dates, changes in harvest time (shorter/changing seasons) all have agricultural and biodiversity implications which we do not yet understand. His models identify changes in sensitivity of areas, such as Lesotho being highlighted as an area sensitive to temperature change – what does this mean in terms of international water agreements between these two countries? In South Africa, the CFR and the center of the country come out as major sensitive climate zones. Critical driver dynamics may change, such as the relationship between biomass and climate drivers in grasslands. Schultz highlighted an essential theme: how confident are we with our results? His work attempted to overcome this challenge by developing an ‘index of consistency of change’, demonstrating where numerous GCM’s agreed on the same outcome. He also covered the idea of ‘memory of change’ – if plants have a memory of previous season conditions. On the question of amplification, many issues are heightened because of other changes, for example, changes in rainfall are amplified by changes to the hydrological system, and hydrological system is amplified by meteorological drought. Changes in mean annual temperature are predicted to be significant, with significant amplification over time – the rate of change per decade is predicted to be about a ¼ to ½ degree up until 2050, then becoming twice as rapid. Similarly, evaporation is predicted to increase at a rate of 5 – 10% up until 2050, jumping to a rate of 15 – 25% to 2090. This has severe consequences for our dams. 9 He concluded by re-emphasizing the need to look at what the predicted implications are of the models that the climate community have developed, highlighting that we have many more questions than answers. Theme 2: Data for Climate Impact Assessments Agroclimatology: Climate Network and Databank – Chrisna Henningse (Agricultural Research Council, Institute for Soil, Climate and Water) Henningse outlined the role of the Climate Network and Databank held by the ARC. The purpose of their project is to capture, archive and process climate data, and make it available for varied purposes such as disease warning and model inputs. Their databank started in 1990 on a DOS platform and moved to MS Access in the late 1990’s. However, due to difficulties with MS Access the data is now held on an SQL database. The Institute for Soil, Climate and Water have 571 weather stations around South Africa, specifically erected for climate change model verification in collaboration with research institutes locally and worldwide. The central collection point for all weather data is in Pretoria, where data are collected daily (every 10 seconds, stored as an hourly value). Reports are produced for farmers whose land is used for weather stations, and clients can obtain daily data and long term reports.. The South African Natural Resource Information System – Garry Paterson (Agricultural Research Council, Institute for Soil, Climate and Water) Patterson presented information around the available soil databases at ARC which could be of significant value to input into climate models. As he explained, ARC has a land type database, a soil profile database, and a Western Cape and Gauteng soil database. The land type data was collected over 30 years (completed in 2002), with over 400 000 soil observations being taken. This has resulted in full GIS coverage of soils for the whole country, which is a resource that is unique in Africa. The land type database has spatial (actual boundaries) and non-spatial components (info and list of soil types on crests, mid slopes, valley slopes and valley bottoms). Soil properties such as clay content, depth and texture are reported, and each form is divided into one of 17 soil classes. The soil profiles database consists of more than 15 000 profiles. The minimum requirement for soils in the database is that they have to have some description of form or family, co-ordinates and some analyses. Patterson explained that with the climate change implications of crop suitability changing with different scenarios, that old soil records can be used to generate baseline soil conditions from historical profile records. This is of value to assist with making strategic decisions on a crop-by-crop basis. They at ARC have developed a general agricultural algorithm with parameters for each crop (soil requirement, terrain suitability, climatic requirements) and have applied this to each soil entry per land type. From this a map is created showing distribution of allocation various classes for each specific crop, using soil, climate and terrain factors. They have also looked at irrigated areas of South Africa in order to use this as a guide alongside the soils database to map future scenarios of where it would be most viable to irrigate (based on the condition of the soils, which hold water efficiently etc.) Additional derived datasets arising from research – Lyndon Estes (Princeton Uni.) Estes explained that the purpose of this presentation was to describe how they have used various datasets provided by CCAB collaborators for use in developing their combined crop modeling methodology, and was looking for feedback and comments on their methodology. For soil data, they used the ARC land type database (>7000 mapped land types). The land types were mapped to the level of terrain unit by Hein Beukes, resulting in >27000 individual “soils” that Mr. Beukes considers to be spatially accurate. It gives basic soil texture information in the form of a clay %. They used that information with a variety of pedotransfer functions to generate the parameters necessary to run the DSSAT model (e.g. drained upper limit, wilting point, runoff curves). Spatial soil organic carbon estimates were missing, which Estes modeled using the following method: he obtained % topsoil organic carbon (%OC) for 3377 ARC modal profiles, and modeled these 10 predictors using regression kriging with eight gridded spatial predictors (slope, rainfall, elevation, % tree cover, etc.). The modeled topsoil %OC surface had a root mean square error (RMSE) of 0.89. In order to get an estimate of the change of organic carbon with depth, the proportional change of organic C with depth was calculated using the subset of modal profiles that contained %OC values for the subsoil. A logistic model was used to estimate the change in %OC with depth. Estes suspects that soil organic carbon could be improved by using a variety of methods, e.g. using land cover data to map actual rather than potential OC, and using independent OC data to validate the models. Estes also mentioned that it would be useful to try model silt and clay content for SA in order to obtain improved estimates of soil water-holding properties, and asked the participants of they had any ideas on how to take this further and how it could be done, as he thinks it could improve agricultural models. He used climate related – base data from UCT CSAG and Roland Schultz. In order to get a future projection for solar radiation he employed an algorithm used by Sepo Hachigonta and Olivier Crespo from CSAG. This particular algorithm depends on an empirically fit atmospheric extinction coefficient (KRs), which was obtained by selecting the KRs value that minimized the error in predicted solar radiation values relative to historical solar radiation values provided in the records for South Africa’s 5838 Quinary Catchments. The results were partially validated using the same procedure to fit KRs values using ARC automatic weather station data. Estes also provided additional information on the methodology used to obtain crop yield proxy data from MODIS NDVI data. They used this approach to validate crop model outputs - validating was necessary because farm level yield data were not available for South Africa. To extract crop NDVI signatures that were uncontaminated by background vegetation and other features, they used the aerial crop observations collected by PICES surveys to identify pixels representing maize and wheat, and then selecting only those pixels in which 75% or more of the pixel fell within the digitized crop field boundary in which the PICES point was recorded. For each pixel, NDVI values were integrated between the minimum NDVI values preceding and following the crop-growing season, and NDVI integrals were averaged within 20X20 km neighborhoods first by year (2007-2008 for wheat, 2006-2009 for maize), then across years in order to provide values comparable in scale to the crop models, and which minimized inter-annual yield variability. Estes was looking for feedback on this approach, and asked if it seemed suitable for crop yields, or if there is any actual crop yield data that they can validate their model against? UCT’s downscaled climate data - Mark Tadross (Uni. Cape Town Climate Systems Analysis Group) GCMs are the primary forecasting tool used by climate scientists, with 21 different models around the world. Not all are independent, and some use slightly different physics for modeling, but in Tadross’ opinion there are around 10 realistically independent configurations, and climate scientists generally use as many of them as possible to see what the likelihood of change is. One can look at model consistency in various ways – either by using the median, which gives the tendency for the group of models used to move in one direction or the other; or by using model percentage, which shows where models agree there will be a change. Tadross acknowledges that downscaling is difficult, as using predictions from one grid square GCM results in all the variation being averaged out, for example from escarpment to ocean. One can downscale in two ways: either using regional models, or through statistical techniques. For each location that you require a downscaled climate model for, you need to find all the different weather systems that happen around each location, and for each weather system you build a probability function, and map those weather systems over the focal distribution with the GCMs. See image below: 11 Tadross explains that we are limited by our records of weather systems for the area, as only existing conditions can be included in the future. It is not a good way to get to extremes – but it is a limitation that modelers are aware of. Forecasts are currently ‘downscaled’ using state of art techniques for 3000 stations across South Africa with high spatial resolution ≈ 0.25° ≈ 25km. Forecasts are currently available for 7-8 global models, which is good for assessing probability of change (see below). However, not all models predict a decrease in rainfall across South Africa, therefore one needs to assess changes in terms of probability as well as risk. a) b) Winter rainfallFigure: expected climate change in 2050 for a) winter rainfall and b) summer rainfall. Summer change rainfall change Day 2 Theme 1: Agriculture/Livelihoods Impacts (continued) Impacts of heat stress on dairy cattle productivity under projected human induced climate change – Rantuku (Gabriel) Lekalakala (Limpopo Dept. of Agriculture) In South Africa, animal production contributes a significant amount to the economy. Most climate change work has focused on crops, not on animal production, and the Limpopo Department of Agriculture is addressing that by using climate models to look at the impact of heat stress on dairy cattle. They have used the Temperature-Humidity Index (THI) as an indication of the degree of heat stress experienced by dairy cows and their milk production. As the temperature-humidity index increases, respiration rates and dilation of blood vessels increases, and decreased milk production and dry matter intake decreases. This becomes heightened until they die. Even in low temperatures with high humidity, cattle experience heat stress. At a threshold THI of 72, cattle decrease their food intake by around 12 9% and milk production declines by around 21%. Naturally, profit risk linked to the frequency of exceeding stress thresholds. Using CSAGs downscale data they have mapped different stress thresholds, and have found that currently there is little stress as a result of high THI. Models predict that northern parts of the country will experience stress due to increased THI in the near future, and extreme stress will occur in the north western part of the country in the distant future. The department has students involved in investigating this further, including a PhD student looking at heat stress in chickens. Their intention is to develop adaptation strategies to minimize THI effects, such as: reducing overcrowding; maximizing shade; using sprinklers for cooling; improving ventilation; new efficient building design; and the installation of thermo controlled air conditioning. The department is also experimenting with the cross breeding of nguni cattle with dairy cows, as nguni are more pest resistant and tolerant to harsh conditions but are not good milk producers. Theme 3: Biodiversity Impacts Implications of climate scenario uncertainties and approaches to dealing with these - Guy Midgley (SANBI) A range of uncertainties affect our climate projections, and these occur at different periods throughout the modeling period with different implications. This range of uncertainties affects rate of change, and pattern of chance, and has more of a severe impact on pattern of change as it means responses will be different as patterns vary. Source Rate of change Emissions X GCM (structural) X X Downscaling X X 1st order impact model X X Higher order impact model X Economic model X X X Adaptation response and effectiveness X X Net impact ? ? Pattern of change Figure: Effect of uncertainties on rate and pattern of change. Size of X show severity of effect. Over the next 40 years human mitigation responses in terms of emission reductions are unlikely to affect the rate of emissions over that time period. However, GCMs still carry substantive uncertainty, and if they have failed to incorporate a particular feedback then this will have large implications on the outcome. Downscaling adds uncertainty to pattern of change more than rate of change (see figure). We know very little about the net impacts of climate change on any particular issue, hence the question marks! Midgley gave an example of how he has encountered this uncertainty in his work: for the 1 st national communication on climate change, models demonstrated significant drying across parts of the country, with heavy risk to the succulent Karoo and Fynbos biomes. When the 4th assessment came out with much more sophisticated models, similar outcomes for rainfall seemed to hold, with substantive winter drying in the Cape, but only predicted at the 13 end of the 21st century, and by 20%. In this case, the rate of change was not as high as from the first models, but the pattern of change was the same. Now with statistical downscaling scenarios, the pattern of change is not the same – the median change has switched from substantial drying to getting slightly wetter, and only the 25th percentile demonstrates an increase in temperature. As Midgley explains, this is a real challenge for when we have to communicate our findings to the public who do not like uncertainty. Using the 2007 AR4 scenarios, Fynbos and Succulent Karoo showed contraction by 2050, but much less than the 2000 scenarios, and the Nama Karoo showed a completely different response. 2090 scenarios showed a reduction in Succulent Karoo, but the rate of change had been reduced. He used an example from Hoffman et al.’s (2011) study on ‘pan evaporation and wind run decline in the CFR’, according to their paper the Cape may no longer experience wind by 2065 – but he highlights that climate models are not good at predicting wind, so one has to be cautious of how one packages this kind of information. As he explains, one way to handle uncertainty is ensemble forecasting – which smears all uncertainties into the model, and may end up with a consensus of where they overlap, but this is not necessarily representative of how nature may respond. In the case below, he looked at plausible best and worst case scenarios (10 th and 90th percentiles), which captures a large amount of variation. Figure: Southern Cape best and worst case scenarios, 2050 Midgley emphasized that that researchers should focus on is monitoring – to use the change that is underway as a natural experiment against which to measure our climate models. Looking at CO2 paleohistory, our rising CO2 levels will push us back to what the world looked like 40-50 million years ago, when CO2 levels were above 1000ppm. Grassland-savanna systems will be replaced with forest. In collaboration with Barney Kgobe and William Bond, they grew savanna trees under pre-industrial CO2 conditions and it became evident that CO2 drove increased tree and C4 grass success. We need to include CO2 in vegetation and crop modeling as it has a substantial effect. Revised biome change predictions for South Africa – some provisional outputs with discussion of potential impacts on biodiversity and protected areas – Stephen Holness (SANParks) This work focused around identifying areas for ecosystem based adaptation (EbA), and will go into the national biodiversity assessment. Two concepts that they tried to identify were: 1) areas of biome stability (particular parts of the country are at higher structural risk) and 2) areas supporting EbA (the resilience side) To identify areas of biome stability, they reran the biome predictions, used downscaled temperature and precipitation data from a range of models, and used high, medium and low risk scenarios (90th and 10th percentile, median). They used MAXENT to develop a model of current biomes and tested the ability to predict current biomes to confirm that 14 the model was accurate. It was found to be 86% accurate, mostly with problem areas around ecotones or mixtures. The current projections show grassland declining and being replaced by savanna, coastal forests declining in hot and dry areas and expanding in warm and wet areas. Interestingly, the Succulent Karoo envelope is completely maintained, Fynbos is not squeezed as much under low and medium risk, only under high. Holness was quick to add that this does not mean there will be no impact, structurally the envelope will still look like Fynbos but there will be vegetation changes. He highlights that areas which are consistently predicted not to change under all scenarios are very useful to incorporate into conservation plans. Figure: Consistently predicted areas in all scenarios One can look at areas of stability and instability in high, medium and low risk scenarios, which he did at a fine scale in order to use the data for a ‘Biome Adaptation Plan’. However, as Holness highlighted, it is still essential to consider resilience in case our models are not correct! Here, he identified areas that are important for resilience: 1. 2. 3. 4. 5. 6. Which parts of the country have high diversity? Identifying gradients (precipitation/temp and overall) Meso scale refugia (south facing slopes and gorges) National riparian corridors and coastal processes High value connected habitats (large unfragmented landscapes which meet targets) Intact centers of endemism These criteria were equally weighted at each of the intermediate stages, and then put through a transformation filter to exclude urban areas, fields etc. with a soft filter for degraded and fragmented land. From this we get a fine scale national layer of areas important for EbA, which can then be used to make the case for broad scale climate change impacts and relevance. This product has great value because it doesn’t include climate change models so it does not have that element of uncertainty, we can confidently say that these identified areas are of ecological value. 15 1.) Dark areas are important for adaptation (corridors, gradients etc) 2.) Green areas are likely to remain within their current biome 3.) Orange areas are likely to be under structural pressure Figure: EbA map for eastern part of SA, demonstrating how it can be useful to identify potential conservation areas (outlined in blue) Holness showed the immediate value of this product by applying it to a selection process for new Grasslands National park – it can aid to select areas that are likely to remain grassland in the future as well as being important adaptation sites. It is also useful for identifying CBAs to inform current land use planning, to feed into zoning and EIA information about which corridors should be maintained as much as possible into the future. Modeling relationships between species’ spatial abundance patterns and climate – David Hole (Conservation International) Under climate change, species can/will exhibit a number of different responses, ranging from behavior change to population decline and extinction. It appears that as a medium term response, species will shift their geographical distribution, and this is already being observed in birds. They are interested in how rapidly the bird turnover happens, and the present and future irreplaceability of endemic birds. By modeling this, they can get an idea of the optimum set of corridors for species. Hole explains that a good first step is to look at changes in abundance, as birds will normally change in abundance as climate becomes less suitable, before they alter their geographic distribution. This can be used as an early warning system before range shifts actually happen. Although, Hole points out that for many birds, land use change has been more of a driver than climate change. They used reporting rate as it is very difficult to measure abundance – using the SA Bird Atlas dataset, reporting rates are highly correlated with observations of local abundance. However, it is limiting in that one can’t tell abundant from very abundant, and it doesn’t account for species specific detectability (systematically under-reports less detectable species). For their study they used 78 bird species (68 endemic), four bioclimatic variables and the average of three GCMs to see how bird abundance and range distribution would shift over time. 16 Abundance -83% Range extent -74% Figure: result showing endemic species with contrasting projections For each species, an output such as the figure above is produced, demonstrating how both abundance and range extent is projected to change. This information is very valuable for conservation planning, as it allows us to gauge where species are most abundant and where they are most likely to have highly viable population. It could also be useful to determine how range reductions/changes of birds overlap with predicted agricultural suitability areas? Climate risk and vulnerability mapping for Southern Africa: Status Quo (2008) and future (2050) – Stephanie Midgley (OneWorld) The work of the OneWorld Group takes a vulnerability approach, rather than looking at impact studies. They are focused on working within the SADC region and by looking at vulnerability they can align with policy and the priorities of people and the country, so they are more likely to affect change. Midgley highlights that vulnerability in the SADC region is not as a result of climate change, it already exists, and consequently people in this area are likely to be most hard hit by climate change. With the change in variability of climate change, one gets greatly different outcomes between vulnerable and less vulnerable people. The objectives of their research were to 1) distil key, clear messages around highly complex biophysical and socioeconomic drivers of vulnerability; 2) identify vulnerability hotspots in the SADC region; and 3) investigate mapping as a science based tool for policy and adaptation development. So firstly, how to quantify vulnerability? Using indicator-based geographically disaggregated vulnerability mapping. Exposure + sensitivity = impact; impact + adaptive capacity = vulnerability, with various data inputs for each factor. Midgley explained that choice of indicator is not easy. They searched for suitable high resolution datasets to represent the three categories (exposure, sensitivity, adaptive capacity). From this, the final choice of indicators was based on the following: Data availability for Africa at the required resolution; quality of the data (statistical reliability) and availability of good metadata; permission to use a database; and avoidance of auto-correlation (independence of data sources). There was a strong focus on indicators representing agriculture (including water resources), food security and human health, but biodiversity and forestry were also included. An assumption of relationship between indicator and vulnerability was made for each. For projections of future vulnerability, they assumed that current sensitivity and adaptive capacity would remain the same into the future, although Midgley acknowledges that this as a limitation. They simply added the summary layers together, so that impact overlay = sensitivity + exposure, and vulnerability overlay (for present and 2050) = impact + exposure + adaptive capacity. They examined four output scenarios: current vulnerability remains (e.g. adaptive capacity prevents worsening); vulnerability is intensified; new vulnerability emerges (e.g. low adaptive capacity and exposure); and current vulnerability reduced (e.g. potential centers of resilience?) 17 Zimbabwe, Malawi and Madagascar have high impact risk, When you add adaptive capacity (= vulnerability map), large problem areas are Mozambique, DRC, Namibia, Zambia, Malawi and Madagascar). See maps below for present day adaptive capacity, as well as sensitivity analysis and transboundary vulnerability hotspot projections for current day and 2050. Note that many other maps of value were shown, presentation is available on request. In conclusion: Growing populations that remain highly vulnerable or become even more so under climate change are more likely to turn increasingly to natural resources for food/income and/or to forfeit sustainable land use practices. Growth of settlements will result in a growing need for land, water, problems of waste disposal, water pollution, wood collection. Peri-urban conditions could drive encroachment onto farmland and protected land. When farmers lack adaptive capacity their options are limited and long-term thinking disappears, and their response can exacerbate impact (maladaptation risk). Increasing vulnerability of neighboring countries will drive further migration; often to South Africa; thus increasing pressure. Human response to climate change could have greater impact on biodiversity than climate change itself. Figure: Adaptive capacity across sub-Saharan Africa (red =lowest adaptive capacity). 18 Sensitivity analysis (current) Sensitivity 2050 Hotspot (current) Hotspot 2050 Figure: Maps of sensitivity analysis (red = most sensitive) and trans-boundary vulnerability hotspots (deep purple = most vulnerable) for present day and 2050. Global Change in rural rangelands: implications for biodiversity and livelihoods – Barend Erasmus (Uni. Witwatersrand) Erasmus provided an interesting overview of how people in rural areas around KNP are already responding to environmental variability, and now this may provide us with insight into how they may further respond with climate change. He emphasized the value of measuring fine scale processes to predict broad scale patterns. WITS have 19 collected data on deaths, migration, cellphone access, income etc. for settlements fringing on KNP, and with such detailed information they can start to look at drivers of change. Erasmus explains that in Bushbuckridge; an area within a former homeland; people are very heavily reliant on natural resources. One of their adaptation strategies is to have many alternative income/food source methods. For example, people may grow crops in their backyard, sell airtime, grow nuts etc., and this allows for multiple different income sources. He also highlights that in these areas, social adaptive capacity is high – communities know that they can rely on their neighbours, and they can rely on you. In such a highly variable environment these communities have devised such methods to cope, but the question is how much variable they can withstand? In these rangelands, bush encroachment is common, heavy fuelwood harvesting has resulted in re-sprouting short shrubby stands of Terminalia sericea, fruiting trees are left in the landscape as a ‘common use’ trees, and bird diversity is extremely low due to lack of nest sites from fuelwood harvesting. Fuelwood harvesting is a strong driver of change (fine scale process) in this landscape. As Erasmus explains, woodland stem density and structure has changed over time as larger stems have been harvested. No large stems above 10cm remained in 1999, and for those who rely on this fuelwood, the quality of the resource has severely declined. This has social implications; for example; women may expect their children to help them harvest fuelwood as resources get scarcer, and as a consequence children no longer attend school. Erasmus highlights that we need to further our understanding of the social implications of climate change, with around 15 million people relying on savannas for fuelwood. Overlap between protected areas and agricultural change – Lyndon Estes (Princeton Uni.) The CCAB project intends to understand how human adaptation to climate change might impact biodiversity. This project focuses on South African agriculture and biodiversity. Understanding this problem is difficult. It can be thought of as “a socioeconomic play being carried out an ecological stage”. That is, human activities take place within the context of our ecosystem. Climate change modifies the ecosystem, and complex human activities and behaviors will be affected by this change to the stage in difficult-topredict ways, and those reactions will in turn modify the ecological stage. This holds for understanding how agriculture adaptation to climate change might impact conservation. However, we can start at a fairly basic level, using a simple model to understand the pressure that agriculture exerts on biodiversity. At the most fundamental level, agriculture exerts pressure on biodiversity by converting natural ecosystems into farmland. This pressure to convert land into agriculture may be expressed as a function of supply and demand for particular crops. Pressure on conservation land = f (agricultural demand, agricultural supply) Demand is complex, and involves all sorts of local, national, and international factors, but it seems safe to say that it is only going to increase in the future. We thus focus on the supply side of the equation, which, at a national level, can be further expressed as: -1 Agricultural supply = Production area (ha) X productivity (kg ha ) Climate change impacts has the potential to alter agricultural supply: ∆ Agricultural supply = Area production potential X % productivity Area production potential X % productivity Agricultural modeling helps us to identify the changes in productivity and production area, which in turn allows us to identify how and where this pressure function might change, at both local and country scales, thus giving us some insight into implications for pressure on conservation lands. 20 Estes reviewed the CCAB work presented by Bethany Bradley in order to highlight areas that both MAXENT and DSSAT agreed on suitability for maize in 2050, and where DSSAT shows yield change (productivity) over time. The multiple models and climate scenarios were used jointly to identify areas to which the greatest confidence could be assigned to projections. Areas of suitability for wheat and maize in the 2055 period were identified based on agreement among 17 out of 18 scenarios for each crop model (i.e. a 95% confidence level, based on inputs from nine GCMs run under both the B1 and A2 emissions scenarios). All areas falling outside of the 95% suitability agreement were considered unsuitable. Estes illustrated this example using the maize models, where both MAXENT and DSSAT showed very little in the way of new areas of suitability, while MAXENT projected a much larger loss of suitability than DSSAT (see below). The same pattern applied to wheat projections. Figure: lost and gained suitability for maize in 2055 using 95% agreement across scenarios Estes then unioned the 95% confidence results for both models, and intersected them with the current maize and wheat fields distributions to identify areas of potential loss. Intersecting the two models 95% confidence surfaces returned the areas to which the highest confidence regarding future suitability could be attached. Within both the loss and future suitability areas, DSSAT yield change projections were examined to get a further sense of where agricultural impacts, both positive and negative, would be the greatest. It is possible that DSSAT was over-optimistic in terms of increases to crop water use efficiency gains under elevated CO2, so the yield loss map was re-examined after subtracting 25% from the yield changes, under the assumption that the model was substantially biased upward. Even after this assumption was made, the map showed a fairly small area that exceeded a 25% yield loss (yield losses of less than 25% can be compensated by breeding and technology improvements, according to Jones and Thorton (2009)). In contrast, the area with high confidence in future suitability showed a fairly large area with yield gains of 25% or higher for both crops (the eastern portions of the current wheat and maize projection regions). The areas of >25% loss and gain were then used to highlight areas where conservation interests are most likely to be put under pressure by agriculture. Current protected areas (PAs) and National Protected Areas Expansion Strategy 21 (PAES) priority areas were mapped onto these areas of agricultural change to determine which areas might be under pressure (see below for example using maize results): Figure: current and planned protected areas vs. maize loss/gain Given the high uncertainty regarding the projections of agricultural loss, and the small number of affected conservation lands, more focus should be placed on understanding the implications for conservation in areas showing increasing agricultural productivity. To get a more refined sense of the pressure to farm conservation lands within areas of increasing productivity, a Topographic Roughness Index (TRI) was used to determine how feasible it might be to farm different areas. 90% of crop fields in South Africa have a roughness value of 8.25 or less, so PAs and PAES overlapping the high productivity areas were divided into those falling above and below this threshold (see figure below). Areas falling below the threshold can be considered to be of highest risk of conversion to agriculture. High TRI areas would be less threatened as they are more difficult to farm. PAs High TRI Low TRI PAES High TRI Low TRI = 26 =3 = 63 = 15 Figure: TRI ratings for existing and predicted protected areas, and maize gain. 22 Despite it being very challenging to understand how agricultural pressure on conservation land will play out in reality (especially given a wide range of other potential socio-political-economic factors), from this output one can make some reasonable assumptions in areas of increasing productivity: one could expect increasing land prices and more farming, with the implications on conservation that priority areas are harder to buy or convert, and priority areas become increasingly isolated. In areas of decreasing productivity, it is difficult to say much of anything with confidence, because we don’t know how it will change the ability of SA to meet its overall demand for these agricultural commodities. Estes points out that we can speculate further but economists and sociologists are needed in order to obtain a holistic picture. The crop modeling results are preliminary, and will be rerun before the end of 2011 to reassess these findings. Potential implications for conservation land will be further explored by incorporating: 1. 2. A simple land use change model; and economic scenarios involving the effect of changing commodity prices on breakeven yields for wheat and maize. Acknowledgements CCBA funding is provided by the Princeton Environmental Institute. Numerous collaborators contributed to the success of this work, and are thanked for their sharing of time, knowledge and data. Thank you to all who presented and participated in this stimulating workshop. 23