[Paper Review] Fire seasonality identification with multimodality tests

Jose Ameijeiras‐Alonso, Akli Benali+3 more|arXiv (Cornell University)|Oct 9, 2018

Fire effects on ecosystems46 references4 citations

TL;DR

This paper proposes a nonparametric circular statistics approach using an adapted excess mass test with bootstrap calibration to identify multimodal fire seasonality patterns across a 0.5° grid in Russia-Kazakhstan. The method detects multiple fire peaks per year, revealing anthropogenic influence where fire seasons deviate from climatological patterns, with spatial FDR correction enabling systematic large-scale analysis of human-driven fire regimes.

ABSTRACT

Understanding the role of vegetation fires in the Earth system has become an important environmental problem. Although fires time occurrence is mainly influenced by climate, human activity related with land use and management has altered fire patterns in several regions of the world. Hence, for a better insight in fires regimes, it is of special interest to analyze where human activity has influenced the fire seasonality. For doing so, multimodality tests are a useful tool for determining the number of fire peaks along the year. The periodicity of climatological and human--altered fires and their complex distributional features motivate the use of the nonparametric circular statistics. The unsatisfactory performance of previous nonparametric proposals for testing multimodality, in the circular case, justifies the introduction of a new approach, accompanied by a correction of the False Discovery Rate with spatial dependence for a systematic application of the tests in a large area between Russia and Kazakhstan.

Motivation & Objective

• To detect multiple fire seasons in regions where human activity alters natural fire patterns.
• To develop a robust statistical test for multimodality in circular data (e.g., seasonal timing) that outperforms existing nonparametric methods.
• To apply the test systematically across a large spatial grid (0.5° resolution) while correcting for spatial dependence and false discovery rate (FDR).
• To distinguish anthropogenic fire seasons from climatological ones by identifying temporal mismatches between fire peaks and dry seasons.
• To create homogeneous fire patches based on land cover to group grid cells with similar fire seasonality behavior.

Proposed method

• Adapts the excess mass statistic for circular data to test for multimodality in fire occurrence timing.
• Uses a modified kernel density estimator to nonparametrically estimate the circular density function.
• Employs a bootstrap procedure with resampling from the estimated density to calibrate the test's p-values and ensure correct Type I error control.
• Applies the test to each 0.5° grid cell in a study region spanning Russia and Kazakhstan.
• Corrects for spatial dependence in multiple testing using a False Discovery Rate (FDR) procedure to control false positives.
• Groups grid cells into fire patches based on dominant and secondary land cover types to aggregate similar fire seasonality patterns.

Experimental results

Research questions

• RQ1 How many distinct fire seasons occur annually in different regions of the Russia-Kazakhstan agricultural belt?
• RQ2 To what extent do observed fire peaks deviate from the climatological dry season (June–September), indicating human influence?
• RQ3 Can a nonparametric circular test detect multimodal fire seasonality more reliably than existing methods?
• RQ4 How can spatial dependence in large-scale multiple testing be corrected to maintain statistical validity?
• RQ5 Which land cover types are most strongly associated with multimodal fire seasonality patterns?

Key findings

• The proposed bootstrap-calibrated excess mass test outperforms existing nonparametric circular multimodality tests in terms of calibration and power, even with moderate sample sizes.
• The method successfully identifies multiple fire peaks in regions where climatological conditions do not support them, indicating human-driven burning.
• In the study area, a significant number of grid cells exhibit two fire seasons: one in summer (June–September) and another in early spring (March–April), despite unfavorable dry conditions in spring.
• The spatial FDR correction effectively controls the rate of false positives when applying the test across 10,000+ grid cells.
• Fire patches were successfully constructed based on land cover homogeneity, enabling regional analysis of fire seasonality patterns.
• The results suggest that anthropogenic land management practices, such as post-harvest and pre-planting burns, are responsible for the observed multimodal seasonality in the region.