Forest loss and Borneo's climate

To date, most of the climate change discourse has focused on the warming effects of carbon dioxide and other greenhouse gases released into the atmosphere (IPCC 2014). However, in many regions, the climate-related impact of forest loss may be of a similar magnitude to the warming effects of greenhouse gases (McAlpine et al 2010, Pielke et al 2011, Pielke et al 2016). Forest cover influences land-surface processes, including evapotranspiration, albedo and surface roughness, which affect the magnitude and form of energy transfer to the atmosphere (Bonan 2008, Jackson et al 2008). By altering the fluxes of heat, momentum and moisture exchanges between the land surface and the lower atmosphere, forests affect the regional climate (Syktus and McAlpine 2016). In tropical regions, forests increase the ratio of latent heat to sensible, which increases moisture recycling, cloud formation and convective precipitation. In recent decades, tropical deforestation in the Amazon Basin, Congo Basin and Southeast Asia has been has been of global concern (Hansen et al 2013).

The South-east Asian island of Borneo is a deforestation hotspot (Hansen et al 2013). Over the past century, human activities have dramatically changed the landscape of Borneo. In the 1930s, an estimated 75% of Borneo was covered in tropical forests (van Steenis 1935); however, the remaining forest cover is now estimated at approximately half of the original extent (Brookfield and Byron 1990, Curran et al 2004, Langner et al 2007, Gaveau et al 2014b). These changes have been fuelled by the global demand for timber, expanding oil palm and rubber industries, growth in small-holder agriculture and repeated forest fires (Gaveau et al 2014a). Industrial oil palm and timber plantations are now estimated to cover at least 10% of Borneo (Gaveau et al 2014b, Gaveau et al 2016), with Indonesia as a whole accounting for ∼54% of total global oil palm production (Smit et al 2013). From the mid-1970s to the 1990s, at least half the world's total exports of tropical hardwoods were from Borneo (Brookfield and Byron 1990). Furthermore, the extensive drainage of Borneo's peat swamp forests for plantation development increases their susceptibility to fire (Page et al 2002, Gaveau et al 2009), with important implications for regional air quality (Gaveau et al 2014a).

Zoom In Zoom Out Reset image size

Precipitation and temperature data indicate that Borneo's climate has changed over recent decades (Kumagai et al 2013). Analyses of maximum and minimum temperatures from 1980–2002 for 33 weather stations across Indonesia found a significant rise in maximum and minimum temperatures in most stations (MoE 2007). A study in Sabah, Malaysia, found primary forest was up to 2.5 °C cooler than selectively logged production forest and up to 6.5 °C cooler than oil palm plantations (Hardwick et al 2015). Kumagai and Porporato (2012) postulated a relationship between Borneo's forest loss, temperature increase and precipitation decline. However, the spatial and temporal nature of this relationship deserves more detailed attention.

Here, we examined the relationship between changes in forest cover and observed changes in daily temperature and precipitation over Borneo. We investigated how the conversion of forest has influenced daily temperature and precipitation and their variability. We addressed this question using interpolated historical gridded (0.25° spatial resolution) daily temperature (1961–2007) and precipitation (1951–2007) data. The analysis focus was on the entire island and for nine selected watersheds (figure 1). In addition, we also analysed land surface temperature from satellite observations derived from MODIS for the period 2002–2016 for lowland areas <200 m ASL.

2.1. Study area

2.2. Land cover change data

2.3. Gridded datasets

2.4. Statistical analyses

We analysed the interpolated historical climate data at two scales: the entire island, and for the nine selected watersheds (figure 1(b)). The APHRODITE grid cell data were assigned to a watershed based on the location of the centre point of each grid cell. We accounted for the impact of El Niño events (which typically cause a reduction in precipitation in Borneo (Mason and Goddard 2001)) using the Ocean Niño Index (ONI), which measures the El Niño-Southern Oscillation (ENSO) in the east-central equatorial Pacific Ocean (Aldrian and Dwi Susanto 2003, Hendon 2003). The running three-month mean sea surface temperature (SST) anomaly for the El Niño 3.4 region (i.e. 5°N–5°S, 120°W–170°W) was used. We define El Niño events as five consecutive overlapping three-month periods at or above a positive 0.5 °C SST anomaly. A second analysis was run excluding years where a moderate (with a 1.0 °C–1.4 °C SST anomaly) or strong (1.5 °C–2 °C SST) event occurred. This exclusion aimed to reduce the influence of the ENSO phenomena on the observed climate trends for Borneo.

The time-series analyses were designed to test for differences in trends for daily precipitation between the 1951–1972 and 1973–2007 periods, and for daily temperature between 1961–1972 and 1973–2007. The division of the time series into these periods corresponded with the mapping of natural forest loss for the period 1973–2010 for the entire island of Borneo (Gaveau et al 2014b). That study estimated forest area had declined by 168 493 km² or 30.2% from 1973–2010 (figure 1(a)). Our analyses aimed to test if the forest loss after 1973 coincided with changes in daily precipitation and temperature records for each watershed using the gridded APHRODITE datasets.

We conducted linear regression analyses of the daily temperature and precipitation variables for the two periods (for the whole of Borneo and the watersheds) in order to obtain regression coefficients and standard errors. Paternoster et al (1998) proposed the following equation in order to test for a significant difference between the regression coefficients of two slopes:

$$\begin{equation} Z = \,\frac{{{b_1} - {b_2}}}{{\sqrt {SE{b_1}^2 + SE{b_2}^2} }} \end{equation}$$

where b₁ is the regression coefficient in the first period and b₂ is the regression coefficient in the second period, SEb₁ is the standard error of b₁ and SEb₂ is the standard error of b₂. This equation provides a Z score that can be used to determine significance of the difference between two slopes. Based on this analysis, we tested if the rate of change of the climate variable in the first time period is equal to the rate of change of the climate variable in the second time period.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Linear regression analyses were applied to identify the relationship between forest loss and changes in temperature and precipitation between the two periods at the different spatial scales of analyses. The change in climate means between the two periods (percentage of increase or decrease in values) was tested against the overall proportion of forest loss. This analysis compared the proportion of forest loss with associated changes in temperature and precipitation.

For the MODIS land surface temperature, we calculated the probability density function (PDF) for forest and modified forest areas across Borneo land areas (below 200 m ASL) and for each selected watershed (below 200 m ASL) using the R statistical analysis software (www.r-project.org/). PDF allowed us to investigate the differences in the temperature range distribution between forest and modified areas.

3.1. Daily mean temperature changes

3.2. MODIS-derived temperature changes

The PDF for monthly temperatures of the forest and modified land cover classes for lowland areas of Borneo (figure 3) and selected watersheds (figure 4) were derived. Deforested areas of Borneo had higher temperatures than forest areas (figure 3(a)), with the highest temperatures concentrated in southern Borneo (figure 3(b)). We found a marked dry season (JASO) temperature contrast averaging ∼1.7 °C higher for modified forest versus forest while the difference between the tails of the high temperatures was ∼3.0 °C (figure 3(d)). In general, the temperature in forests varied less than in modified land cover areas (figure 3(a)). During El Niño conditions, the temperature differences were greater with averages ∼2.0 °C higher for modified versus forest areas (figure 3(e)). In addition, the PDF for modified land covers had a wider distribution with a higher frequency of temperatures above 31 °C.

Zoom In Zoom Out Reset image size

The role of the forest in moderating temperature extremes was further illustrated by comparing watersheds WS40 and WS143 with contrasting levels of forest loss (figure 4). The PDF of the dry-season temperatures over the still largely forested WS40 was similar for forested and modified forested areas, whereas the PDF for the largely deforested WS143 showed a marked temperature contrast (figures 4(a) and (b)). For WS143, the temperature was warmer on average by ∼1.2 °C and with a much higher frequency of monthly temperatures above 30 °C. For the El Niño years (figures 4(c) and (d)), WS40 showed a very small difference in the PDFs for forested and modified areas. In contrast, the highly deforested WS143 with large areas of oil palm plantations had a very broad temperature PDF for modified areas with half the values above 31 °C. Forested areas in this watershed were cooler on average by ∼1.4 °C with more than half the PDF values below 29 °C. The impact of forest cover on monthly temperatures was further illustrated by the dry season PDFs for the other watersheds (appendix 5) and for El Niño years (appendix 6). In general, the more heavily deforested watersheds showed broader PDFs with a greater frequency of extreme temperatures in modified areas. These values were particularly marked during El Niño events (appendix 6).

3.3. Mean daily precipitation changes

In this study, we have demonstrated a relationship between forest cover change and local climate change in Borneo. Our analyses of the historical interpolated climate data demonstrates that the loss of Borneo's forests is associated with increased daily temperatures and reduced daily precipitation. Highly deforested watersheds with >15% forest loss had a significant (>15%) reduction in rainfall. The impact of deforestation on climate was most pronounced in the southeast of Borneo which has experienced extensive deforestation since 1973 (Hansen et al 2013, Gaveau et al 2014b, Miettinen et al 2016). Northern watersheds which have retained substantial forest cover had a more stable climate.

A separate analysis using MODIS-derived land surface temperature confirms that the loss of Borneo's lowlands forests is resulting in significant temperature differences between forested and human-modified landscapes. These differences were most pronounced for watersheds with a high degree of clearing and extensive oil palm plantations. This analysis showed a strong relationship between the extent of deforestation and mean monthly temperature. For the lowland areas of Borneo, deforested areas were warmer by 1.7 °C than forested areas. At the watershed-scale, modified forests are hotter and have a higher frequency of extreme temperatures (>31 °C mean monthly) than forested areas. The temperature contrast between forest and modified forest areas was most pronounced during the dry season and El Niño years. These results show that forest loss results in hotter temperatures with a significant increase in the frequency of hot temperature extremes. These extremes are most pronounced in watersheds in southeast Borneo (East Kalimantan and Central Kalimantan), whereas watersheds in north and northwest Borneo (Sarawak and Brunei) have a significantly lower extent of forest loss and have a more stable climate with similar differences in the distribution of mean and extreme warm temperatures.

A number of recent studies demonstrate that deforestation has a significant impact on local and regional temperatures (Alkama and Cescatti 2016, Schultz et al 2017). Schultz et al (2017) showed that the impact of deforestation on temperature was strongest in the tropics, where the daytime warming is on average 4.4 K. The loss of tropical forest cover results in decreased evaporative cooling resulting in temperature increases. Numerous studies describing the climate impacts of large-scale deforestation of the Amazon show reduced evapotranspiration and evapotranspiration and moisture convergence (Pires and Costa 2013), decreasing rainfall, increasing surface temperature and even affecting meso-scale circulation patterns (Lejeune et al 2015, Spracklen and Garcia-Carreras 2015, Wu et al 2017). There is recent evidence showing that these changes can also impact the length of the dry season and forest resilience, which can lead to permanent forest loss (Zemp et al 2017). Theory and evidence indicate that tropical forests may also contribute to the low atmospheric pressures that draw winds into these regions (Sheil and Murdiyarso 2009, Makarieva et al 2013, Makarieva et al 2014). However, the impact is likely to be more pronounced over large continental regions compared to tropical maritime region of Southeast Asia (Ellison et al 2017, Lawrence and Vandecar 2015).

Tropical rainforests play a critical role in regulating regional temperatures by maintaining the optimal range of temperatures. This optimal range is maintained by the evaporative cooling from the extensive, intact forest canopy (Bonan 2008). The clearing of tropical forest cover disrupts the climate regulation function of forests and results in a different frequency distribution of daily temperatures, characterised by hotter and more extreme temperatures (Thirumalai et al 2017). This effect is especially pronounced in the South-East Asian region (including Borneo) during El Niño events which are associated with abnormally hot and dry conditions compared to the long-term average (Bowman 2017, Taufik et al 2017).

We acknowledge the patchy climate station coverage is a problem in constructing and interpreting the gridded climate data for Borneo which has a mountainous interior. For many locations, including the mountainous interior, available records are not accessible, have poor quality control and assurance including significant missing data values, and do not cover sufficient temporal periods. As a result, stations are more representative of the lowland modified forest cover and consequently the interpolated signals may be stronger there than for Borneo's interior. In a recent study, Supari et al (2017) used data from 88 weather stations from Indonesia and showed significant and spatially coherent trends in temperature indices. Their results are consistent with our findings. Our analysis of the climate data for the selected catchments show robust a impact of the level of deforestation on the climate signal (figures 2(e) and (f)). Furthermore our analysis of the MODIS land surface temperature shows impact of deforestation on the frequency of extreme warm temperatures. The agreement between the results of the interpolation and station-based analyses give us confidence the results are robust. However, there is a need to further disentangle the relative contribution of global warming and deforestation on regional climate signals over Borneo.

The high rate of deforestation in combination with projected increases in temperature and the likelihood of more extreme El Niño events (Thirumalai et al 2017) will have far reaching impacts on the environment, agricultural productivity and human livelihoods in Borneo in coming decades (Paterson et al 2015, Mora et al 2017). Further, Becek and Horwath (2017) postulate that increased frequency of high temperatures and droughts in Borneo are increasing the risk of forest degradation through reduced net primary productivity with reduced soil moisture leading to canopy thinning and reduced leaf area index. If these changes are realised, it is likely that this tropical ecosystem will collapse and result in a substantially altered bioclimatic regime (see figure 2 in McAlpine et al 2010).

The expansion of oil palm plantations is the major driver of forest loss and fragmentation in Borneo. An analysis using microclimate sensors shows temperature differences of 2.8 °C–6.5 °C for oil palm plantations compared to rainforests (Luskin and Potts 2011, Ramdani et al 2014, Hardwick et al 2015). A recent study from the neighbouring region of Sumatra assessed the impact of deforestation on local warming for seven different land cover types (forest, urban areas, clear cut land, young and mature oil palm plantations, acacia and rubber plantations) (Sabajo et al 2017). Using MODIS and Landsat LSTs, they showed that local warming resulting from the conversion of forest to clear cut land was up to 10 °C (Sabajo et al 2017). In addition, temperatures below 20 °C and above 36 °C are unsuitable for growth of oil palm (Paterson et al 2015). Yields decrease by approximately 30% with an increase of 2 °C above optimum.

Between 1973 and 2015, 186 507 km² of Borneo's forest were lost, leaving just 373 347 km² of forest (205 318 km² of intact old growth) (Gaveau et al 2016). Forest loss and degradation has continued with the subsequent fires of 2015 (Field et al 2016, Purnomo et al 2017). Aside from the larger scale consequences of greenhouse gas emissions, smoke pollution and the reduced atmospheric moisture resulting from these forest losses, there are also more localized impacts. As we have shown, these include higher temperatures, reduced rainfall and greater climatic variation. We note that the previously rare combinations of conditions that allow tropical rain forests to burn (Van Nieuwstadt and Sheil 2005) may become more common with increased frequencies and intensities of drought, along with the degradation and higher risk of ignition that comes with human activities (Siegert et al 2001).

ARC Discovery Projects grant DP160102107 and the Australian Research Council Centre of Excellence for Environmental Decisions (CE11001000104). We thank Jessie Wells, Malcolm Nunn and Simon Blomberg for their assistance with the historical daily temperature and precipitation data. We are especially grateful to David Gaveau for the forest cover data, and Srivatsan Vijayaraghavan for providing advice on the APHRODITE data. We also thank Royal Society South East Asia Rainforest Research Programme for providing the climatic data at Danum Valley Field Centre used in this paper (www.searrp.org).