1. Introduction
Albedo is a critical attribute of the Earth’s land surface and plays an important role in shaping the Earth’s energy budget. Changes to albedo can modify local atmospheric conditions and instigate land-atmosphere feedbacks with potential continental repercussions. A canonical example of an albedo-atmosphere feedback is the classic hypothesis of Charney [
1] invoked to explain multi-decadal drought in the Sahel. Albedo-rainfall interactions may also operate on seasonal timescales [
2,
3]. Changes in albedo are also associated with major effects of deforestation on climate and vegetation-precipitation feedbacks, e.g., [
4,
5,
6]. More recently, observational and mechanistic evidence has linked fire-induced albedo changes to observed reductions in precipitation in the following wet season [
7,
8,
9]. This concept of fire-induced rainfall suppression hinges on the presence of widespread positive anomalies in albedo (i.e., brightening) following extensive wildfire. The basic premise is that in fire-prone regions dominated by convective rainfall systems, such as tropical Africa, reduced energy available for boundary layer growth under brighter-than-normal albedo conditions decreases the likelihood that the top of the atmospheric boundary layer will cross the lifting condensation level, a necessary condition for the formation of convective rainfall. Given the strong linkage between humans and fire [
10], this could represent a plausible way in which humans modify regional hydrologic cycling. Understanding the scope and magnitude of brightening is critical to assessing the viability of this mechanism on a continental scale and, more generally, to understand how wildfire affects the surface energy balance.
Fire is common in Africa, contributing to 70% of the global burned area [
10]. However, currently, there are inconsistent findings with regard to how albedo anomalies develop after a fire and recover over time in Africa, both in term of the sign and magnitude of the anomalies. Ground-based measurements of albedo change during and after a fire are lacking in Africa. Therefore, previous researchers have turned to long-term satellite reflectance datasets, such as those derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), to investigate fire-induced land surface changes. Previous studies have reported instantaneous decreases in albedo after fire by up to 50%. Gatebe et al. [
11] reported significant darkening in Northern Hemisphere sub-Saharan Africa that consistently lasted up to two years after fire for a majority of fire-affected pixels in the MODIS dataset. There was limited evidence of brightening in some cropland regions during the year following fire, as well as in most regions when fires occurred outside of the typical fire season, but the number of pixels reporting brightening was statistically dominated by instances of extended darkening. Recently, Dintwe et al. [
12] measured widespread, immediate darkening after fires over all of Africa using a single pixel-based measurement approach. This approach used albedo anomalies derived from albedo baselines at the same location as the burn, but sampled in the years directly preceding and succeeding the year of the burn.
On the other hand, Saha et al. [
8] found evidence of strong brightening in burn scars (up to ~7% above baseline albedo) in the Kalahari region of Southern Hemisphere Africa in the months following dry season fire [
8]. The measured effect was more pronounced in more arid environments. Wetter regions showed slight darkening over the same timescale. Elsewhere on the globe there are reports of the immediate halving of albedo following intense fires in Australia [
13] and significant brightening in the years following intense wildfires in Greece [
14]. Notably, all of these studies relied on MODIS reflectance data, which suggests that data source is not the driving factor between seemingly contradictory observations of fire-induced albedo change.
The extent geographic or methodological differences affected the outcomes of these studies is unclear. Each of the aforementioned studies computed changes in albedo, due to fire differently. Perhaps the main issue is in determining a reference, the baseline value for albedo so that an anomaly can be calculated after a fire occurs. For example, if the anomaly were purely temporal (i.e., comparing the albedo in a burn scar to the albedo in that same region before or well after it burned) then it is possible that the computed anomaly would capture factors like a drought that temporarily modifies both the albedo and the likelihood of fire. On the other hand, if only a spatial reference were used (i.e., comparing albedo of a burn scar to a different, unburned region at the same time), this would not ensure that the surrounding pixels actually had a similar long-term baseline albedo. Indeed, the reason why some areas burn while others do not burn can be related to differences in vegetation composition, which could result in differences in baseline albedo.
Part of this difficulty stems from the fact there is no standard way to calculate albedo anomalies using a spatial or temporal reference. Given the natural spatial covariance of a spreading process like fire, there is some difficulty in establishing a nearby, representative reference pixel because of many potential reference pixels have also burned. Indeed, Gatebe et al. [
11] used a spatial window ranging from 2.5 km to 30 km in their reference pixel matching scheme to associate pixels that experienced fire with similar reference pixels. There is a question of how truly representative a reference 500 m pixel that is as much as ~30 km away from the burned pixel actually is. Dintwe et al. [
12] point out another potential issue with spatial references. The fact that a potential reference region in proximity to a burn scar did not burn could be indicative of underlying dissimilarity of vegetation between the two pixels, thereby making it an unsuitable reference to measure albedo anomalies. Without correcting for underlying biases between the two pixels, this comparison would be invalid.
On the other hand, there are also potential issues with temporally-derived baselines. Regions with an active fire regime show a strong fuel buildup and burn cycle. Multiple studies have demonstrated how climatic correlates, such as previous wet season rainfall modify fire in the following dry season [
7,
15]. Therefore, albedo anomalies calculated with only a temporal baseline might be influenced by initial land surface conditions that are uniquely predisposed to subsequent burning. Dintwe et al. [
12] use an albedo anomaly calculated relative to average from the year before the fire and the year after the fire. Using a baseline that is, in part, derived from the year following fire, assumes that full albedo recovery has occurred within a year. However, there is ample evidence of fire-induced land surface effects that last for multiple years, e.g., References [
11,
14].
Despite recognition of issues with both spatial- and temporal-only anomaly definitions, to date no study has attempted to account for both kinds of pitfalls on a continental scale. We aim to fill this gap with an improved methodology. The specific temporal lags and apparent disparity in albedo anomalies previously found in Africa clearly warrant further investigation. Despite the common occurrence of fire in Africa and potential impact on the continental radiation budget, a holistic continental analysis that investigates the long-term evolution of fire-induced albedo changes using a unified framework has not yet been undertaken. This is the aim of the current study.
3. Results and Discussion
We identified 1.54 million fires, amounting to 11.2 million km
2 of the burned area over the five-year period, or approximately 11% of the continental area of Africa. Due to our strict quality requirements for fire objects classification, these fire objects represent 88% of fire pixels in the MODIS dataset over our period of record. For this reason, the total impact of fire shown in results is likely to be an underestimate. Smaller fires dominated our dataset. Of the fires we identified, 1.37 × 10
6 were less than 10 km
2 in size. Fires larger than 10 km
2 and 100 km
2 occurred 1.72 × 10
5 and 9796 times, respectively. However, larger fires represented a higher proportion of the cumulative burned area. Fires larger than 10 km
2 and 100 km
2 comprised 65.4% and 31.0% of the total burned area observed. The average albedo anomaly in the year following fire was +6.51 × 10
−4 (95% confidence interval: +6.43 × 10
−4 to +6.58 × 10
−4) for all of sub-Saharan Africa. The five-year continental average was +2.71 × 10
−4 (+2.68 × 10
−4 to +2.75 × 10
−4) Overall, the return to baseline albedo generally occurred within the first two years after fire (
Figure 2, black lines).
We identified a general temporal pattern of albedo anomaly development. The lowest observed albedo anomaly was observed immediately. In most cases this initial anomaly was negative (i.e., immediate darkening) and recovered within three months. After that there was less intense, but still significant, brightening up to about one year after fire. Depending on the region, there is some variability in this signature. For example, fires that occur in the Southern Hemisphere during the wet season show an immediate brightening signal that lasts throughout the five-year lifetime of this analysis, with a peak of substantial brightening around six months after fire. Our findings of dominant darkening in the Northern Hemisphere and dominant brightening in the southern Hemisphere reconcile seemingly conflicting reports of hemispheric differences in the physical effect of fire on the land surface [
8,
11].
The amount of brightening is dependent on when and where the fire occurs. Wet season fires result in substantially more brightening than fires during other times of the year within each hemisphere (
Figure 2, blue lines). This is especially the case in the northern hemisphere. Gatebe et al. [
11] report similar findings after Northern Hemisphere fires, but suggested a possible statistical anomaly, due to low sample size. We believe that the observed effect is real and must be considered in future studies. We hypothesize that the severity of rarer wet season fires, when they do occur, tends to be greater, as they would occur only during extremely dry times, such as intense drought. Additionally, during the wet season, the vegetated unburnt reference region may be substantially less reflective than the underlying soil, leading to an intensified brightening. Further research efforts should aim to decouple these two signals of darkening char deposition and potential brightening to gain a more complete understanding of fire-induced surface changes. Vegetation cover and species could play a large role in determining the magnitude of immediate darkening, due to char.
The Kalahari region exhibits a strong brightening that lasts for over a year, on average. We hypothesize that the soil color is important in explaining the brightening, as there is a significant contrast between the brightening observed over the Kalahari sands (arenosols) with the lack of brightening immediately to the east of this region, which is characterized by darker luvisols [
20] (
Figure 3). The brightening is subdued in the northern half of the Kalahari sands. This could be an effect of spatial trends in soil moisture. The negative relationship between soil moisture and albedo across soil types is well described [
21]. It is likely that the revelation of wetter, darker soils has less of an effect on albedo anomalies. This is supported by the fact that the highest average albedo anomalies following wet season fires (1.11 × 10
−2 and 3.72 × 10
−3 for the NH and SH, respectively) occur about six months after the burn (i.e., during the dry season).
To the south of the Kalahari, in the South African ‘veld’, there is also brightening in the year after fire. However, the relatively low frequency of fire in this region precludes us from drawing definitive conclusions about fire-induced albedo changes. The timing of brightening also shows differences from surrounding areas. The fastest brightening also occurs in the Kalahari sands, on average in 31 days after fire (
Figure 4). In the northern hemisphere brightening occurs after 107 days on average, compared to 60 days for the southern hemisphere.