Australian Continental NPP Comparison
Mini-Workshop
9:00am – 12:30pm, Monday March 4, 2002
Slatyer Room, RSBS, ANU

Workshop Report

by S. Roxburgh, April 2002

Workshop Attendees

Background

Net primary productivity (NPP) is the rate at which plants accumulate carbon from the atmosphere, and is equal to the difference between carbon gained by photosynthesis, and carbon lost by plant respiration. NPP is the primary driver of the coupled carbon and nutrient cycles, and is the primary controller of the size of carbon and organic nitrogen stores in landscapes

Quantification of NPP is also critical for assessing Net Ecosystem Exchange (NEE), and ultimately Net Biome Productivity (NBP). The units of NPP are commonly carbon (or biomass) per unit land area per unit time.

Empirical estimates of NPP are typically difficult, time consuming and expensive to obtain. This is at odds with the increasingly important requirement to quantify NPP over large spatial scales, particularly in the context of global change studies. The only feasible way forward is to use mathematical models to combine what empirical estimates we do have with our current conceptual understanding of the processes contributing to NPP. As Cramer and Field (1999) noted, such models are useless if they cannot be evaluated. In the context of NPP estimation this is problematic, as there do not exist enough empirical NPP estimates across large spatial scales to simultaneously construct the models and independently validate them. One option is to perform model inter-comparison activities, such as the ‘Potsdam NPP Model Intercomparison Workshop’ (Cramer and Field, 1999). Model intercomparisons provide a method for not only comparing consistency among model outputs, but also for highlighting differences in the assumptions underlying different approaches, and for understanding how such differences contribute to the variability observed in the outputs.

This workshop represented the first step towards a comprehensive NPP intercomparison review for the Australian continent. Note that this activity was not a formal model intercomparison per se, but was rather a comparison of NPP estimates derived from a range of different models.

Scope

Although NPP can be quantified at a range of spatial and temporal scales, this project focuses on continental long-term (or equilibrium) NPP, i.e. NPP averaged over decadal or longer timeframes. Focusing on annual NPP in a ‘climatically average’ year precludes investigation at finer temporal-scales, such as within-year/seasonal variation, and also precludes investigation of between-year variability. The justification for focusing on annual average NPP is that it is a common output of a wide range of theoretical approaches, and therefore a larger number of estimates are available for comparison. Evaluation of NPP at this coarsest of temporal resolutions is also the logical first step prior to any finer scale comparison activities. Regarding spatial resolution, the eleven NPP estimates included in the comparison were all continental in scope, with a grid-cell size ranging from 0.05 degrees (approximately 25 km²), through to 0.5 degrees (approximately 2500 km²). Spatial surfaces for eight of the eleven estimates were available for more detailed within-continental comparisons.

Preliminary Results

Prior to the workshop B. McBeth and S. Roxburgh collated the NPP estimates, massaged them into a consistent format for comparison, and generated a range of summary figures and tables. Continental summaries of the eleven estimates included in the comparison are shown in the table below, and in Figure 1.

Table 1. Continental summaries of eleven NPP estimates.

Figure 1. NPP surfaces for nine estimates, arranged in order from highest to lowest. Click here to view a higher resolution image.

Overall there was a five-fold range in predicted total continental NPP, from a minimum of 0.65 Gt C/yr (VAST), to a maximum of 3.33 Gt C/yr (RFBN). On an area-averaged basis this corresponds to a range of 88 g C/m²/yr to 434 g C/m²/yr. The distribution of estimates between these two extremes was approximately continuous.

Visual inspection of the estimates in Figure 1 reveals significant within-continental differences in the spatial distribution of the predicted productivity. For example, some estimates predict relatively high NPP in the central arid regions (e.g. RFBN, Olson & Miami), whereas others predict NPP in these areas to be much lower (e.g. Miami-Oz, BiosEquil & VAST). This within-continental variation is summarised in Figures 2 and 3. These figures show pairwise comparison matrices of one surface against another, at the level of individual grid-cells. In Figure 2 the axis scaling is linear, and in figure 2 the scaling is logarithmic to emphasise the contribution of the lower rainfall, lower productivity ecosystems which dominate the continent.

Figure 2. Each scatter- plot is a cell-by-cell comparison of one NPP surface against another. For each plot the x-axis corresponds to the surface shown at the bottom of the respective column, and the y-axis is the surface at the start of each row. Axis units are gC/m²/yr. The straight line shows the 1:1 relationship. The colours correspond to the broad vegetation types defined in the Vegetation Cover-Type map. The diagonal cells show the frequency histograms for each surface.  The table is ordered from highest NPP estimate (RFBN) to the lowest (VAST). Click here to view a higher resolution image

Figure 3. This figure has the Same layout as for Figure 2, except that the axes are on a logarithmic scale to exaggerate the relationships for the lower productivity values. Click here to view a higher resolution image

The diagonal elements of the matrices show the frequency distribution of NPP for each of the nine estimates. For all estimates the frequency distributions are skewed towards the left (a predominance of low productivity values) consistent with the domination of the Australian continent by arid ecosystems. However, the distributions for the three highest estimates (RBFN, Olson and Miami) are significantly less skewed then the others, indicating a more even spread of productivity across the continent for these estimates.

The scatter-plots compare each estimate on a pairwise basis, by plotting the NPP values of all grid-cells for two NPP surfaces against one-another. On this basis the agreement between any two surfaces is poor, as shown by the scatter around the 1:1 lines within the plots. Overall, in comparing any two estimates, the chance that the NPP at a given point will differ by more than a factor of two is c. 60%. The agreement between estimates for the higher productivity forest ecosystems only (coded red in the scatter-plots) is somewhat better, with the chance of two estimates differing by more than a factor of two reducing to c. 30%.

Summary of Discussions

S. Roxburgh opened the workshop with an overview presentation of the above results. This was followed by brief (5 minute) statements from five speakers, representing authors of five of the estimates under comparison (D. Barrett (VAST); M. Raupach (BiosEquil); M. Roderick (RFBN); J. Carter (GRASP); M. Kirschbaum (CenW)).

These brief statements provided the group with an overview of the respective modelling approaches adopted by these authors. The speakers also addressed the question of why they thought their particular method yielded the NPP predictions it did, particularly in relation to the other estimates under consideration. For example, M. Roderick noted that the RFBN estimate was most likely the highest of the group as it does not specifically discount productivity due to water limitation. M. Raupach noted that the BiosEquil model showed lower productivity in Northern Australia than many (but not all) of the others, due to BiosEquil specifically including effects of saturation deficit in these regions. J. Carter noted that the GRASP model includes only grass growth across the continent, and hence may provide an underestimate due to the exclusion of the woody component. Similarly, M. Kirschbaum noted than CenW also was likely to be an underestimate, as it includes only tree growth, and hence has productivities close to or equal to zero in most of the arid regions. D. Barrett noted that the VAST model is parameterised from the VAST empirical database [http://www.daac.ornl.gov/], and that a closer examination of this database might provide further insight into interpreting the VAST model predictions, and also the predictions generated by the other methods.

The balance of time remaining in the workshop was spent discussing the estimates, and in developing the agenda for taking the project to the next stage.

Two main activities were identified:

1 Publication of the NPP inter-comparison results: Coordinator. S. Roxburgh.

It was agreed that a review publication in a peer-reviewed journal of the NPP intercomparison activities is a priority. Three steps leading to this outcome were identified.

• Publication on the CRC website of a summary of the workshop activities, the initial results as presented above, and the facility to download the software to allow users to examine the pairwise comparisons summarised in Figures 2 and 3 in greater detail (i.e. this document)
• The authors of NPP estimates present at the workshop to prepare brief reports (1-2 pages), expanding upon the comments made during the meeting on reasons why their particular approach may have yielded results different (or similar) to others included in the comparison, with a particular emphasis on outlining any reasons why their particular approach could be greater/less than the actual but unknown value of long-term NPP for the continent. These reports should also briefly summarise the overall features of the model/theory adopted.
• Augmentation of the NPP datasets already collected with those generated from international modelling efforts. A number of spatially-explicit global NPP estimates have been published (reviewed in Cramer and Field 1999 as part of the Potsdam NPP Intercomparison Project), and these need to be incorporated into the current review.

2 Further analysis of the VAST empirical database: Coordinator. D. Barrett.

The VAST empirical database of abovegroud fine-tissue NPP in minimally disturbed sites provides the best collection of empirically-based NPP estimates available for the Australian continent. Further analysis of this dataset, with a particular emphasis on documenting the years over which the estimates were made, and the climatic conditions during those periods, was suggested to be a way to further interpret the variability within this database, and to increase the utility of this resource for future work.

The aims of the above tasks are to (1) communicate these results to a wider audience, and (2) gain a deeper understanding of the observed variation among the estimates. It is envisioned that this will lead to the identification of particular aspects of the analyses that can be further tested and validated, leading to refined NPP estimates for the Australian continent.

Additional materials

Click here to download high-resolution versions of the figures included in this document (MS Excel 2000 file, 7.1 Mb)

Click here to download a copy of the poster presented at the CRC Annual Science Meeting, April 2002 (MS Powerpoint 2000 file, 2.9 Mb).

Click here to download a copy of S.Roxburgh’s powerpoint presentation delivered during the workshop introduction. (6.1 Mb).

References for NPP estimates

General Reference

Cramer, W. & Field, C.B. 1999. Comparing global models of terrestrial net primary productivity (NPP): Introduction. Global Change Biology 5: iii-iv.